rtto Sftij,
A
«t wort®
Risk and Exposure Assessment for the Review
of the Primary National Ambient Air Quality
Standard for Sulfur Oxides
-------
-------
EPA-452/R-18-003
May 2018
Risk and Exposure Assessment for the Review of the Primary National Ambient Air Quality
Standard for Sulfur Oxides
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Health and Environmental Impacts Division
Research Triangle Park, NC
-------
DISCLAIMER
This document has been prepared by staff in the Health and Environmental Impacts
Division, Office of Air Quality Planning and Standards, U.S. Environmental Protection Agency
(EPA). Any findings and conclusions are those of the authors and do not necessarily reflect the
views of the Agency. Mention of trade names or commercial products is not intended to
constitute endorsement or recommendation for use. Questions or comments related to this
document should be addressed to Dr. Stephen Graham, U.S. Environmental Protection Agency,
Office of Air Quality Planning and Standards, C539-07, Research Triangle Park, North Carolina
27711 (email: graham.stephen@epa.gov) and Dr. Nicole Hagan, U.S. Environmental Protection
Agency, Office of Air Quality Planning and Standards, C504-06, Research Triangle Park, North
Carolina 27711 (email: hagan.nicole@epa.gov).
i
-------
TABLE OF CONTENTS
LIST OF APPENDICES iv
LIST OF FIGURES v
LIST OF TABLES vii
LIST 01 ACRONYMS AND ABBREVIATIONS xi
1 INTRODUCTION 1-1
1.1 Background 1-2
1.2 Previous Reviews and Assessments 1-4
1.3 Current Review, CAS AC Advice and Public Comment 1-7
1.3.1 REA Aspects Updated Since 2009 1-7
1.3.2 CAS AC Advice and Public Comment 1-9
REFERENCES 1-11
2 OVERVIEW OF ASSESSMENT APPROACH 2-1
2.1 Conceptual Model for SO2 Exposure and Risk 2-1
2.1.1 Sources of SO2 2-3
2.1.2 Exposure Pathways and Route 2-4
2.1.3 At-Risk Populations 2-5
2.1.4 Health Endpoints 2-6
2.1.5 Risk Metrics 2-6
2.2 Assessment Approach 2-8
REFERENCES 2-11
3 AMBIENT AIR CONCENTRATIONS 3-1
3.1 Characterization of Study Areas 3-2
3.2 Air Quality Modeling 3-4
3.2.1 General Model Inputs 3-9
3.2.2 Stationary Sources Emissions Preparation 3-13
3.2.3 Air Quality Receptor Locations 3-15
3.2.4 Concentrations Associated with Sources Not Explicitly Modeled 3-16
3.2.5 Hourly Concentrations at Air Quality Model Receptors 3-19
3.3 Selection of Air Quality Receptors for Exposure Modeling Domain 3-23
3.4 Air Quality Adjustment to Conditions Meeting the Current Standard 3-24
3.5 Five-Minute Concentrations 3-31
ii
-------
3.5.1 Preparing Monitoring Data: Assessing Completeness & Filling Missing Values
3-31
3.5.2 Estimating Continuous 5-minute Concentrations at Monitor Having Only 1-hour
Average and Hourly Maximum 5-minute Data 3-34
3.5.3 Estimating 5-minute Concentrations Across Study Areas 3-38
REFERENCES 3-50
4 POPULATION EXPOSURE AND RISK 4-1
4.1 Populations Simulated 4-2
4.1.1 Demographics 4-3
4.1.2 Asthma Prevalence 4-6
4.1.3 Personal Attributes 4-13
4.2 Meteorological Data 4-20
4.3 Construction of Human Activity Sequences 4-21
4.3.1 Consolidated Human Activity Database 4-21
4.3.2 Commuting Activity Pattern Data 4-22
4.3.3 Assigning Activity Pattern Data to Individuals 4-23
4.3.4 Method for Longitudinal Activity Sequences 4-26
4.4 Microenvironmental Concentrations 4-28
4.4.1 Air Exchange Rates for Indoor Residential Microenvironments 4-33
4.4.2 Air Conditioning Prevalence for Indoor Residential Microenvironments 4-35
4.4.3 AER Distributions for All Other Indoor Microenvironments 4-36
4.4.4 Removal Rate for Indoor Microenvironments 4-36
4.4.5 Factors for Estimating In-Vehicle/Near-Road Microenvironmental Concentrations.
4-37
4.5 Estimating Exposure 4-38
4.6 Risk Metrics 4-39
4.6.1 Comparison to Benchmark Concentrations 4-39
4.6.2 Lung Function Risk 4-42
4.7 Approach for Characterizing Uncertainty and Variability 4-48
4.7.1 Assessment of Variability and Co-variability 4-49
4.7.2 Characterization of Uncertainty 4-49
REFERENCES 4-51
5 POPULATION EXPOSURE AND RISK RESULTS 5-1
in
-------
5.1 Characteristics of the Simulated Population and Study Areas 5-2
5.2 Exposures at or above Benchmark Concentrations 5-5
5.3 Lung Function Decrements Associated with 5-minute SO2 Exposures 5-11
5.4 Study Area Differences and Population Distribution 5-14
5.4.1 Derivation of DV&POP Metric 5-14
5.4.2 Comparing the Study Areas with the DV&POP Metric 5-15
5.5 Comparison with 2009 REA Results 5-20
6 VARIABILITY ANALYSIS AND UNCERTAINTY CHARACTERIZATION 6-1
6.1 Treatment of Variability and Co-Variability 6-1
6.2 Characterization of Uncertainty 6-6
6.2.1 Characterizing Sources of Uncertainty 6-7
6.2.2 Exposure Model Sensitivity Analyses 6-23
REFERENCES 6-38
LIST OF APPENDICES
A. Surface characteristic values and meteorological data preparation for input to air
quality modeling
B. Development of hourly emissions profiles
C. Air quality modeling domains for study areas
D. Modeled air quality evaluation
E. Asthma prevalence
F. Description of the Air Pollutants Exposure Model (APEX)
G. ICF Final Memo: Joint distributions of body weight and height for use in APEX
H. ICF Final Memo: Resting metabolic rate (RMR) and ventilation rate (Ve)
algorithm refinements
I. Consolidated Human Activity Database (CHAD) data
J. Detailed exposure and risk results
K. Daytime hourly concentration estimates and measurements by season
IV
-------
LIST OF FIGURES
Figure 2-1. Conceptual model for exposure and associated health risk of SO2 in ambient air...
2-2
Figure 2-2. Overview of the assessment approach 2-10
Figure 3-1. Location of surface and upper air meteorological stations, SO2 emissions sources,
and ambient monitors used to predict ambient air quality in the Fall River study
area 3-6
Figure 3-2. Location of surface and upper air meteorological stations, SO2 emissions sources,
and ambient monitors used to predict ambient air quality in the Indianapolis study
area. Also included is source type and 2011 NEI SO2 emissions 3-7
Figure 3-3. Location of surface and upper air meteorological stations, SO2 emissions sources,
and ambient monitors used to predict ambient air quality in the Tulsa study area.
Also included is source type and 2011 NEI SO2 emissions 3-8
Figure 3-4. Comparison of AERMOD predicted SO2 concentrations (y-axis) with observed
air monitor SO2 concentrations (x-axis) during daytime of the three warmer
seasons at the highest design value monitor in each study area 3-22
Figure 3-5. Comparison of ambient air measurements from high concentration years (x-axis)
to low concentration years (y-axis) in the Fall River (top row), Indianapolis
(middle row), and Tulsa (bottom row) study areas. Left column contains the year
having the highest 99th percentile daily maximum concentration. Right column
contains the year having the 2nd highest 99th percentile daily maximum
concentration 3-26
Figure 3-6. Location of air quality receptors, emission sources, and ambient monitors in the
Fall River exposure modeling domain and receptor design values calculated from
modeled hourly SO2 concentrations adjusted to just meet the current standard
3-28
Figure 3-7. Location of air quality receptors, emission sources, and ambient monitors in the
Indianapolis exposure modeling domain and receptor design values calculated
from modeled hourly SO2 concentrations adjusted to just meet the current
standard 3-29
Figure 3-8. Location of air quality receptors, emission sources, and ambient monitors in the
Tulsa exposure modeling domain and receptor design values calculated from
modeled hourly SO2 concentrations adjusted to just meet the current standard
3-30
Figure 3-9. Comparison of estimated to measured SO2 concentrations in ambient air in Fall
River monitor 250051004: 1-hour average (top panels), maximum 5-minute
(middle panels) and continuous 5-minute (bottom panels) for 2011 (left panels)
and 2012 (right panels) 3-37
Figure 4-1. Influence of age, race, obesity, sex and family income on adult asthma prevalence
(based onNHIS 2011-2015 for four U.S. regions) 4-11
v
-------
Figure 4-2. Influence of age, race, obesity, sex and family income on child asthma prevalence
(based onNHIS 2011-2015 for four U.S. regions) 4-12
Figure 4-3. Illustration of the mass balance model used by APEX 4-31
Figure 4-4. Percent of individuals experiencing changes in sRaw > 100% (top panel) and
sRaw > 200% (bottom panel) using controlled human exposure study data (Table
4-12) fit using a probit regression (solid lines). Dashed lines indicate a 90 percent
confidence interval for the mean response 4-45
Figure 5-1. Population in the Fall River study area considering 2010 U.S. Census tracts 5-4
Figure 5-2. Population in the Indianapolis study area considering 2010 U.S. Census tracts. 5-4
Figure 5-3. Population in the Tulsa study area considering 2010 U.S. Census tracts 5-5
Figure 5-4. Percent of children's time in indoor, outdoor, and vehicle MEs while exposed to
SO2 in Fall River (top), Indianapolis (middle), and Tulsa study areas 5-10
Figure 5-5. Values of the DV&POP exposure metric in the Fall River study area 5-17
Figure 5-6. Values of the DV&POP exposure metric in the Indianapolis study area 5-18
Figure 5-7. Values of the DV&POP exposure metric in the Tulsa study area 5-19
Figure 6-1. Spatial pattern of design values using an adjustment based on the maximum
design value (left panel) and an adjustment based on the 99th percentile design
value (right panel) in the Fall River study area 6-26
Figure 6-2. Spatial pattern of design values using an adjustment based on the maximum
design value (left panel) and an adjustment based on the 99th percentile design
value (right panel) in the Indianapolis study area 6-27
Figure 6-3. Spatial pattern of design values using an adjustment based on the maximum
design value (left panel) and an adjustment based on the 99th percentile design
value (right panel) in the Tulsa study area 6-27
VI
-------
LIST OF TABLES
Table 3-1. General features of the study areas selected for the exposure and risk assessment..
3-4
Table 3-2. National Weather Service surface stations for meteorological input data in study
areas 3-10
Table 3-3. National Weather Service upper air stations for meteorological input data in study
areas 3-10
Table 3-4. Monthly seasonal assignments input to AERSURFACE 3-11
Table 3-5. Monthly surface moisture categorizations for the three study areas 3-12
Table 3-6. Facilities with point sources included in the air quality modeling domain for each
study area 3-13
Table 3-7. SO2 concentrations (ppb) used to account for source emissions not explicitly
modeled in the three study areas, stratified by season and hour of day 3-18
Table 3-8. Maximum SO2 design values modeled at air quality receptors and associated
proportional adjustment factors applied to primary source concentrations in each
study area 3-28
Table 3-9. Percent of missing values in the hourly and 5-minute ambient air monitoring data
sets for the three study areas (2011-2013) 3-34
Table 3-10. Descriptive statistics and correlations associated with measured and estimated 1-
hour average, maximum 5-minute, and continuous 5-minute SO2 concentrations,
Fall River (monitor 250051004), 2011-2012 3-38
Table 3-11. Descriptive statistics for concentrations at monitors and concentrations estimated
at air quality receptor locations, Fall River study area 2011-2013 3-43
Table 3-12. Descriptive statistics for concentrations at monitors and concentrations estimated
at model receptor locations, Indianapolis study area 2011-2013 3-44
Table 3-13. Descriptive statistics for concentrations at monitors and concentrations estimated
at model receptor locations, Tulsa study area 2011-2013 3-45
Table 3-14. Percent of air quality receptors and monitors at which 5-minute SO2
concentrations (for conditions just meeting standard) exceed concentrations of
interest on single and multiple days, Fall River study area 2011-2013 3-47
Table 3-15. Percent of air quality receptors and monitors at which 5-minute SO2
concentrations (for conditions just meeting standard) exceed concentrations of
interest on single and multiple days, Indianapolis study area 2011-2013 3-48
Table 3-16. Percent of air quality receptors and monitors at which 5-minute SO2
concentrations (for conditions just meeting standard) exceed concentrations of
interest on single and multiple days, Tulsa study area 2011-2013 3-49
Table 4-1. Distribution of the percent of total population that are children residing in the
census blocks comprising each study area 4-5
Vll
-------
Table 4-2. Estimated asthma prevalence for children and adults in census blocks of three
study areas, summary statistics 4-8
Table 4-3. Regression parameters used to estimate RMR by sex and age groups 4-16
Table 4-4. Study area meteorological stations, locations, and hours of missing data 4-21
Table 4-5. Comparison of outdoor time expenditure and exertion level by asthma status for
children and adult CHAD diaries used by APEX 4-25
Table 4-6. Microenvironments modeled and calculation method used 4-30
Table 4-7. AERs for indoor residential microenvironments (ME-1) with A/C by study area
and temperature 4-34
Table 4-8. AERs for indoor residential microenvironments (ME-1) without A/C by study
area and temperature 4-35
Table 4-9. American Housing Survey A/C prevalence from 2013 Current Housing Reports
for selected urban areas 4-35
Table 4-10. Parameter estimates of SO2 removal rate distributions in two indoor
microenvironments modeled by APEX 4-37
Table 4-11. Responses reported in controlled human exposure studies at a given benchmark
concentration 4-41
Table 4-12. Summary of controlled human exposure studies containing individual response
data: number and percent of exercising individuals with asthma who experienced
greater than or equal to a 100 or 200 percent increase in specific airway resistance
(sRaw), adjusted for effects of exercise in clean air 4-44
Table 4-13. Example of risk calculation using estimated daily maximum 5-minute exposures
of children with asthma in the Fall River study area 4-47
Table 5-1. Summary of study area features and the simulated population 5-3
Table 5-2. Percent and number of children and adults with asthma estimated to experience at
least one day per year with a SO2 exposure at or above 5-minute benchmark
concentrations while breathing at elevated rate, air quality adjusted to just meet
the existing standard 5-8
Table 5-3. Percent of children and adults with asthma estimated to experience multiple days
per year with a SO2 exposure at or above 5-minute benchmark concentrations
while breathing at elevated rate, air quality adjusted to just meet the existing
standard 5-9
Table 5-4. Percent and number of children and adults with asthma estimated to experience at
least one day per year with a S02-related increase in sRaw of 100% or more while
breathing at an elevated rate, air quality adjusted to just meet the existing
standard 5-12
Table 5-5. Percent of children and adults with asthma estimated to experience multiple days
per year with a SCh-related increase in sRaw of 100% or more while breathing at
elevated rate, air quality adjusted to just meet the existing standard 5-13
Vlll
-------
Table 5-6. Contribution of different magnitudes of 5-minute SO2 exposures to lung function
risk (sRaw increase of at least 100%) estimated for children with asthma in Fall
River 5-13
Table 6-1. Summary of how variability was incorporated into the exposure and risk
assessment 6-4
Table 6-2. Important components of co-variability in exposure modeling 6-6
Table 6-3. Characterization of Key Uncertainties in Exposure and Risk Assessments using
APEX 6-8
Table 6-4. Comparison of measured and estimated continuous 5-minute SO2 concentrations
in ambient air, Fall River monitor 250051004, 2011 6-24
Table 6-5. Comparison of simulated exposures, for children with asthma, at or above
benchmarks using measured versus estimated continuous 5-minute SO2
concentrations from monitor 250051004, Fall River, 2011 6-24
Table 6-6. Comparison of simulated lung function decrements in children with asthma using
measured versus estimated 5-minute continuous SO2 concentrations, Fall River
2011 6-25
Table 6-7. Air quality adjustment factors for main body REA and sensitivity analysis 6-26
Table 6-8. Comparison of two approaches used to adjust ambient air concentrations to just
meet the existing standard (2011-2013): Percent of children with asthma
estimated to experience at least one day per year with a SO2 exposure at or above
5-minute benchmark concentrations while at elevated exertion 6-29
Table 6-9. Comparison of two approaches used to adjust ambient air concentrations to just
meet the existing standard (2011-2013): Percent of children with asthma
estimated to experience multiple days per year with a SO2 exposure at or above 5-
minute benchmark concentrations while at elevated exertion 6-30
Table 6-10. Percent of children with asthma estimated to experience at least one day per year
with a S02-related increase in sRaw of 100% or more while breathing at elevated
rates, air quality adjusted to just meet the existing standard 6-30
Table 6-11. Percent of children with asthma estimated to experience multiple days per year
with a S02-related increase in sRaw of 100% or more while breathing at elevated
rates, air quality adjusted to just meet the existing standard 6-31
Table 6-12. Comparison of three approaches for using continuous 5-minute monitoring data to
estimate 5-minute concentrations associated with modeled 1-hour concentrations
at receptor locations: Air quality adjusted to just meet the existing standard, Fall
River study area 2011 6-33
Table 6-13. Comparison of three approaches for using continuous 5-minute ambient air
monitoring data to estimate 5-minute concentrations associated with modeled 1-
hour concentrations: Estimated exposures for air quality adjusted to just meet the
existing standard, Fall River, 2011 6-34
IX
-------
Table 6-14. Comparison of three approaches for using continuous 5-minute monitoring data to
estimate 5-minute concentrations associated with modeled 1-hour concentrations:
Estimated lung function decrements associated with exposure to air quality
adjusted to just meet the existing standard, Fall River 2011 6-35
Table 6-15. Comparison of estimated lung function risk using mean, lower bound and upper
bound of the fitted E-R function: Percent of children with asthma estimated to
experience at least one or multiple days per year with a S02-related increase in
sRaw of 100% or more while breathing at elevated rates, air quality adjusted to
just meet the existing standard, 2011-2013 6-36
x
-------
LIST OF ACRONYMS AND ABBREVIATIONS
A/C
air conditioner
ACS
American Community Survey
AER
air exchange rate
AHR
airway hyperresponsiveness
AHS
American Housing Survey
APEX
Air Pollutants Exposure model
AQS
Air Quality System
ASOS
Automated Surface Observing Stations
BASE
Building Assessment Survey and Evaluation
BSA
body surface area
CAA
Clean Air Act
CASAC
Clean Air Scientific Advisory Committee
CHAD
Consolidated Human Activity Database
DV
design value
EGU
Electricity generating unit
EPA
Environmental Protection Agency
E-R
exposure-response
EVR
equivalent ventilation rate
FEVi
forced expiratory volume in one minute
IRP
Integrated Review Plan
ISA
Integrated Science Assessment
ISH
Integrated Surface Hourly
km
kilometer
lat
latitude
Ion
longitude
m
meter
MCC
Markov-chain clustering
ME
microenvironment
MER
mixed-effects regression
MLR
multiple linear regression
MRLC
Multi-Resolution Land Characteristics
MSA
Metropolitan Statistical Area
NAAQS
National Ambient Air Quality Standard
NCEI
National Centers for Environmental Information
NED
National Elevation Data
xi
-------
NEI
National Emissions Inventory
NHIS
National Health Interview Survey
NLCD
National Land Cover Dataset
N02
nitrogen dioxide
NWS
National Weather Service
03
ozone
OAQPS
Office of Air Quality Planning and Standards
ppb
parts per billion
PA
Policy Assessment
PM
particulate matter
PMR
peak-to-mean ratio
PSD
Prevention of Significant Deterioration
REA
Risk and Exposure Assessment
RMR
resting metabolic rate
SIP
State Implementation Plan
SOx
oxides of sulfur
sRaw
specific airway resistance
Ve
activity-specific ventilation rate
WHO
World Health Organization
Xll
-------
1 INTRODUCTION
This document, Risk and Exposure Assessment for the Review of the Primary National
Ambient Air Quality Standardfor Sulfur Oxides (hereafter referred to as REA), describes the
quantitative human exposure and risk characterization conducted to inform the U.S.
Environmental Protection Agency's (EPA's) current review of the primary (health-based)1
national ambient air quality standard (NAAQS) for sulfur oxides (SOx). This document presents
the methods, key results, observations, and related uncertainties associated with the quantitative
analyses performed. The REA draws upon the Integrated Science Assessment (ISA; U.S. EPA
2017a) and reflects consideration of the Clean Air Scientific Advisory Committee's (CAS AC)
advice and public comments on the draft REA.
In this review, as in each NAAQS review, the policy implications of the REA results are
considered in the policy assessment prepared separately for the review. The policy assessment
presents analyses and staff conclusions regarding the policy implications of the key scientific and
technical information that informs the review. The policy assessment is intended to "bridge the
gap" between the relevant scientific evidence and technical information and the judgments
required of the Administrator in his consideration of the adequacy of the current standards. The
policy assessment for this review of the primary NAAQS for SOx is titled, Policy Assessment for
the Review of the Primary National Ambient Air Quality Standardfor Sulfur Oxides (PA; U.S.
EPA, 2018).
The remainder of this chapter summarizes the legislative requirements (section 1.1),
provides an overview of the history of the primary NAAQS for SOx (section 1.2), and describes
aspects of the REA that have been updated since the 2009 REA, and revisions made from the
draft to the final REA in consideration of CASAC recommendations and public comments
(section 1.3). Following Chapter 1, the REA presents an overview of the assessment approach
(Chapter 2), describes the study areas and air quality modeling (Chapter 3), describes the
exposure modeling and risk characterization (Chapter 4), presents the exposure and risk
estimates (Chapter 5), and describes the analysis of variability and characterization of
uncertainty (Chapter 6).
1 The EPA is separately reviewing the welfare effects associated with sulfur oxides and the public welfare protection
provided by the secondary SO2 standard, in conjunction with a review of the secondary standards for nitrogen
oxides and particulate matter with respect to their protection of the public welfare from adverse effects related to
ecological effects (U.S. EPA, 2017b).
1-1
-------
1.1 BACKGROUND
Sections 108 and 109 of the Clean Air Act (CAA) govern the establishment and periodic
review of the NAAQS. Section 108 [42 U.S.C. 7408] directs the Administrator to identify and
list certain air pollutants and then to issue air quality criteria for those pollutants. The
Administrator is to list those air pollutants "emissions of which, in his judgment, cause or
contribute to air pollution which may reasonably be anticipated to endanger public health or
welfare," "the presence of which in the ambient air results from numerous or diverse mobile or
stationary sources;" and "for which... [the Administrator] plans to issue air quality criteria... "
CAA section 108(a)(1). The NAAQS are established for the pollutants listed. The CAA requires
that NAAQS are to be based on air quality criteria, which are intended to "accurately reflect the
latest scientific knowledge useful in indicating the kind and extent of all identifiable effects on
public health or welfare which may be expected from the presence of [the] pollutant in the
ambient air..." CAA section 108(a)(2). Under CAA section 109 [42 U.S.C. 7409], the EPA
Administrator is to propose, promulgate, and periodically review, at five-year intervals,
"primary" (health-based) and "secondary" (welfare-based)2 NAAQS for such pollutants for
which air quality criteria are issued.3 Based on periodic reviews of the air quality criteria and
standards, the Administrator is to make revisions in the criteria and standards, and promulgate
any new standards, as may be appropriate. The CAA also requires that an independent scientific
review committee review the air quality criteria and standards and recommend to the
Administrator any new standards and revisions of existing air quality criteria and standards as
may be appropriate, a function now performed by the CASAC.
The current primary NAAQS for SOx is a 1-hour standard set at a level of 75 parts per
billion (ppb), based on the 3-year average of the annual 99th percentile of 1-hour daily maximum
SO2 concentrations. This standard was set in the last review of the primary NAAQS for SOx,
which was completed in 2010 (75 FR 35520, June 22, 2010). In comparison to the standards
existing at that time, establishment of the 1-hour standard was determined to provide increased
protection for people with asthma and other at-risk populations against an array of respiratory
2 Section 302(h) of the CAA provides that all language referring to effects on welfare includes but is not limited to,
"...effects on soils, water, crops, vegetation, man-made materials, animals, wildlife, weather, visibility, and
climate, damage to and deterioration of property, and hazards to transportation, as well as effects on economic
values and on personal comfort and well-being...."
3 Section 109(b)(1) [42 U.S.C. 7409] of the CAA defines a primary standard as one "the attainment and maintenance
of which in the judgment of the Administrator, based on such criteria and allowing an adequate margin of safety,
are requisite to protect the public health." Section 109(b)(2) of the CAA directs that a secondary standard is to
"specify a level of air quality the attainment and maintenance of which, in the judgment of the Administrator,
based on such criteria, is requisite to protect the public welfare from any known or anticipated adverse effects
associated with the presence of [the] pollutant in the ambient air."
1-2
-------
effects related to short-term exposures (as short as 5 minutes) and to maintain longer-term
concentrations below those specified by the then-existing standards (75 FR 35550, June 22,
2010).4
The EPA initiated the current review of the primary NAAQS for SOx in May 2013, with
a call for information from the public (78 FR 27387, May 10, 2013). The EPA held a workshop
on June 12-13, 2013 to discuss policy-relevant scientific and technical information to inform the
EPA's planning for the review. Following the workshop, the EPA developed the plan for the
review, which is described in the Integrated Review Plan for the Primary National Ambient Air
Quality Standardfor Sulfur Dioxide (U.S. EPA, 2014; hereafter referred to as the IRP). The IRP
includes policy-relevant questions for the review, the process and schedule for conducting the
review, and descriptions of the purpose, contents and approach for developing the key
documents for the review.
The key documents in the review include an Integrated Science Assessment (ISA), a
REA (as warranted), and a PA. In general terms, the ISA is to provide a critical assessment of the
latest available scientific information upon which the NAAQS are to be based, and the PA is to
evaluate the policy implications of the information contained in the ISA and of any policy-
relevant quantitative analyses, such as a quantitative REA performed for the current review or, as
applicable, for past reviews. Based on that evaluation, the PA presents staff conclusions
regarding policy options for the Administrator to consider in reaching decisions on the NAAQS.5
The EPA has developed this REA describing the quantitative risk and exposure
assessment being conducted by the Agency to support this review of the primary SOx standard.
This document is intended to be a concise presentation of the methods, key results, observations,
and related uncertainties associated with the analyses performed. The REA builds upon the
health effects evidence presented in the ISA, as well as CASAC advice and public comments on
the REA planning document (Review of the Primary National Ambient Air Quality Standardfor
Sulfur Oxides: Risk and Exposure Assessment Planning Document, REA Planning Document,
U.S. EPA, 2017c) following a consultation with the CASAC at a public meeting in March 2017
(82 FR 11449). In consideration of CASAC comments at that consultation and public comments,
the EPA developed the draft REA (U.S. EPA, 2017d) and the draft PA (U.S. EPA 2017e), which
4 In the 2010 decision to establish a new 1-hour standard, the EPA revoked the then-existing 24-hour and annual
primary standards.
5 The basic elements of a standard include the indicator, averaging time, form, and level. The indicator defines the
pollutant to be measured in the ambient air for the purpose of determining compliance with the standard. The
averaging time defines the time period over which air quality measurements are to be obtained and averaged or
cumulated. The form of a standard defines the air quality statistic that is to be compared to the level of the
standard in determining whether an area attains the standard. The level of a standard defines the air quality
concentration used (i.e., an ambient air concentration of the indicator pollutant).
1-3
-------
were released on August 4, 2017 (82 FR 43756, September 19, 2017). The draft REA and draft
PA were reviewed by the CAS AC on September 18-19, 2017 (82 FR 37213, August 9, 2017).
Following a CASAC teleconference on April 20, 2018 (83 FR 14638, April 5, 2018), the
CAS AC's recommendations, based on its review of the draft REA and draft PA, were provided
in a letter to the EPA Administrator (Cox and Diez Roux, 2018a,b). The EPA staff considered
these recommendations, as well as public comments provided on the draft REA and draft PA,
when developing this REA.
The ISA and REA informed the development of the PA and will inform the subsequent
rulemaking steps that will lead to final decisions on the primary NAAQS for SOx. The PA
document includes staff analysis of the scientific basis for policy options for consideration by the
Administrator prior to rulemaking. The PA integrates and interprets information from the ISA
and the REA to frame policy options for consideration by the Administrator. The PA is intended
to help "bridge the gap" between the Agency's scientific and technical assessments, presented in
the ISA and REA and the judgments required of the Administrator in determining whether it is
appropriate to retain or revise the standards. The PA is also intended to facilitate the CASAC's
advice to the Administrator on the adequacy of existing standards, and any new standards or
revisions to existing standards as may be appropriate. Concurrent with the release of this REA,
the PA (U.S. EPA, 2018) is also being released.
The schedule for completion of this review is governed by a court order, which resulted
from the entry of consent decree resolving a lawsuit that was filed in July 2016 and that
concerned, in relevant part, the timing of completion of this review. Center for Biological
Diversity etal. v. McCarthy (No. 4:16-cv-07396-VC, N.D. Cal.). The order specifies that the
Administrator shall sign a notice setting forth his proposed decision concerning the review of the
primary NAAQS for SOx no later than May 25, 2018; and sign a notice setting forth his final
decision concerning the review of the primary NAAQS for SOx no later than January 28, 2019.
1.2 PREVIOUS REVIEWS AND ASSESSMENTS
Reviews of the primary NAAQS for SOx completed in 1996 and 2010 included analyses
of potential exposure to SO2 in ambient air (61 FR 25566, May 22, 1996; 75 FR 35520, June 22,
2010). These analyses pertained to the then-existing 24-hour and annual standards, but primarily
focused on whether additional protection was necessary to protect at-risk populations (people
with asthma) against short-term (e.g., 5-minute) peak exposures while at elevated breathing rates
(e.g., while exercising). The analyses that informed the review completed in 1996 focused on
potential exposures to 5-minute concentrations at or above 600 ppb for several air quality
scenarios (61 FR 2556, May 22, 1996). The 2010 review analyses estimated the number of
individuals and percent of the modeled at-risk population that would be expected to experience
1-4
-------
5-minute exposures above several concentrations of potential concern extending down to 100
ppb ("benchmark concentrations" based on findings from controlled human exposure studies)
and also the number of individuals and percent of the population expected to experience a
doubling or greater increase in specific airway resistance (sRaw) or a reduction in forced
expiratory volume in one second (FEVi) of at least 15% (U.S. EPA, 2009 [hereafter referred to
as the 2009 REA]). As summarized in more detail in the PA, the analyses in the 2009 REA
informed the 2010 decision to establish a new 1-hour standard to protect at-risk populations from
short-term (e.g., 5-minute) peak exposures (75 FR 35520, June 22, 2010).
The multiple quantitative analyses that informed the 1996 review decision are described
in the 1986 Addendum to the 1982 OAQPS Staff Paper (U.S. EPA, 1986), the 1994 Supplement
to the 1986 OAQPS Staff Paper Addendum (U.S. EPA, 1994) and the final decision notice (61
FR 25566, May 22, 1996). A key aspect of the design for those analyses was the focus on 5-
minute concentrations at or above 600 ppb, an exposure level that the Agency judged could pose
an immediate significant health risk for a substantial portion of asthmatics at elevated breathing
rates, e.g., while exercising (61 FR 25573, May 22, 1996). The available ambient monitoring
data from 1988-1995 were analyzed to estimate the frequency of 5-minute peak concentrations
above 500, 600, and 700 ppb, the number of repeated exceedances of these concentrations, and
the sequential occurrences of peak concentrations within a given day (U.S. EPA, 1994; SAI,
1996). The analysis indicated that during that period a substantial number of 5-minute
concentrations at or above 600 ppb occurred in several locations in the vicinity of certain sources
(61 FR 25574, May 22, 1996). The probability of at-risk individuals breathing at elevated levels
with the probability of encountering such peak concentrations was assessed in several exposure
analyses (U.S. EPA, 1986, 1994; Burton et al., 1987; Rosenbaum et al., 1992; Stoeckenius et al.,
1990; Sciences International, Inc., 1995).
A series of exposure analyses informed the 1994 proposed decision. These analyses
focused on exposures of interest associated with coal-fired power utilities, all power utility
boilers, non-utility sources of SO2 emissions and such exposures associated with the projected
reduction in emissions from fossil-fueled power plants following implementation of the acid
deposition provisions (Title IV) of the 1990 Clean Air Act Amendments (U.S. EPA, 1986;
Burton et al., 1987; Stoeckenius et al., 1990; Rosenbaum et al., 1992). Subsequent to the 1994
proposal, an additional exposure analysis of non-utility sources was submitted to the rulemaking
docket (Sciences International, Inc., 1995). Together these analyses provided a range of
estimates of the number of individuals with asthma and the percent of the population with
asthma to be exposed to 5-minute concentrations of 500 and 600 ppb while at elevated exertion,
as well as estimates of such individuals to be exposed on multiple occasions in a year. These
1-5
-------
analyses generally employed the time-activity exposure modeling approaches and underlying
data that were available at that time.
Quantitative analyses performed for the review completed in 2010, and documented in
the 2009 REA, included analyses of the limited then-available ambient air monitoring data for 5-
minute concentrations in 40 U.S. counties and a population exposure assessment (75 FR 35520,
June 22, 2010; 2009 REA). The air quality analyses provided estimates of the annual number of
days that daily 5-minute maximum SO2 concentrations at a monitor exceeded 5-minute
concentrations of interest or benchmark concentrations6 (2009 REA, Chapter 7). In the exposure-
based approach, population-based estimates of human exposure were developed using an
exposure model in order to account for the time people spend in different microenvironments, as
well as for time spent at elevated breathing rates while exposed to peak 5-minute SO2
concentrations (2009 REA, Chapter 8). The analyses were performed for recent ambient air
concentrations (unadjusted, "as is" air quality), and with ambient air concentrations adjusted to
just meet the then-existing annual and daily standards and several potential alternative standards.
The 2009 REA simulated population exposure using version 4.3 of the Air Pollutant
Exposure (APEX) model, a probabilistic model that simulates the movement of individuals
through time and space and estimates their exposure to a given pollutant in indoor, outdoor, and
in-vehicle microenvironments.7 The model was used to simulate population exposures in two
study areas: Greene County, MO and a three-county portion of the St. Louis Metropolitan
Statistical Area (MSA). The simulated population included all people with asthma, with results
also presented for the subset of those who were children. Health risk was characterized by
estimating, for each air quality scenario: (1) the number and percent of people with asthma
exposed, while breathing at elevated rates, to 5-minute daily maximum SO2 concentrations that
exceeded the benchmark concentrations; and (2) the number and percent of exposed people with
asthma estimated to experience moderate or greater lung function responses (in terms of FEVi
and sRaw) at least once per year and the total number of such lung function responses estimated
to occur per year (2009 REA, Chapter 8 and 9). An extensive analysis of variability and
6 The benchmark concentrations are concentrations chosen to represent "exposures of potential concern" which were
used in the analyses to estimate exposures and risks associated with 5-minute concentrations of SO2 (75 FR
35527, June 22, 2010). Based on the evidence in the 2008 ISA and recommendations from the CASAC, staff
concluded that it was appropriate to examine 5-minute benchmark concentrations in the range of 100-400 ppb
(2009 REA, chapter 7). The comparisons of SO2 concentrations to benchmark concentrations provided
perspective on the extent to which, under various air quality scenarios, there was the potential for at-risk
populations to experience SO2 exposures that could be of concern.
7 The APEX model is designed to account for sources of variability that affect people's exposures. It stochastically
generates simulated individuals using census-derived probability distributions for demographic characteristics
based on the information from the Census at the tract, block-group, or block-level (2009 REA).
1-6
-------
characterization of uncertainty accompanied the exposure estimates (2009 REA, sections 8.11
and 9.4).
1.3 CURRENT REVIEW, CASAC ADVICE AND PUBLIC COMMENT
In preparing the planning document for this REA, we considered the scientific evidence
presented in the second draft ISA (U.S. EPA, 2016) and the key science policy issues raised in
the IRP (U.S. EPA, 2014). In February 2017, the REA Planning Document was released to the
CASAC and made available for public comment (82 FR 11356, February 22, 2017). The EPA
held a consultation with the CASAC and solicited comments on the REA Planning Document
during a March 2017 public meeting at which the CASAC also reviewed the second draft ISA
(82 FR 11356, February 22, 2017). The consultative advice from the CASAC and public
comments were considered in developing the draft REA (U.S. EPA, 2017d), which implemented
an exposure-based approach to assess population exposure and risk in three urban study areas
(Fall River, MA, Indianapolis, IN, and Tulsa, OK). The draft REA was reviewed by the CASAC,
along with the draft PA (82 FR 37213, August 9, 2017; 83 FR 14638, April 5, 2018). The EPA
also solicited comment from the public on both documents (82 FR 43756 September 19, 2017;
82 FR 48507, October 18, 2017). Comments and advice from the CASAC, and public comment
have been considered in development of this REA and the PA.
1.3.1 REA Aspects Updated Since 2009
As was also the case in the last review of the primary sulfur dioxide (SO2) standards
completed in 2010, the health effects evidence available in this review indicates that short-term
exposures to SO2 are causally linked to respiratory effects and that people with asthma are the at-
risk population. Specifically, controlled human exposure studies demonstrate an increased risk of
lung function decrements for people with asthma exposed while at increased breathing rates. The
quantitative risk and exposure assessment presented in this REA is based on these findings. The
approach to estimating health risk in this REA is similar to that in the REA conducted as part of
the last review (2009 REA), which included quantitative analyses of both exposure and risk.
Specifically, the 2009 REA included: analyses focused on short-term (5-minute) SO2
concentrations; an exposure assessment designed to estimate exposures likely to be experienced
by at-risk populations while at elevated breathing rates; and risk characterization utilizing two
types of metrics: (1) comparisons of exposures to concentrations of potential concern
(benchmark levels), and (2) lung function risk estimates.
The quantitative analyses performed for the current review and presented in this
document reflect the use of several new pieces of information that address important areas of
uncertainty identified in the last review. Perhaps most importantly, the REA uses an updated SO2
1-7
-------
ambient air monitoring dataset. Specifically, the data for 5-minute concentrations are greatly
expanded with regard to both the number of monitoring locations for which hourly maximum 5-
minute concentrations are available and the number for which all 5-minute values for each hour
are available. Limitations in the 5-minute dataset available at the time of the last review
influenced the approaches that could be used in the 2009 REA to characterize the potential for at-
risk populations to experience exposures of potential concern. The analysis approach for this
REA is based on linking the health effects information to population exposure estimates which
draws on the improved understanding of 5-minute SO2 concentrations and takes advantage of a
number of improvements and updates to the air quality, exposure, and risk models, and their
associated input data.
As in the last review, this REA uses the Air Pollutant Exposure model (APEX) to
estimate population exposures that account for the time people spend in different
microenvironments, as well as for time spent at elevated breathing rates while exposed to peak 5-
minute SO2 concentrations. The REA also reflects the new information and model improvements
that are now available including:
• A SO2 air monitoring dataset that is greatly expanded with regard to both the number
of monitoring locations for which hourly maximum 5-minute concentrations are
available and the number for which all 5-minute values for each hour are available;
• Estimated exposures associated with air quality adjusted to just meet the current
standard across a three-year averaging period;8
• Improvements in the air quality dispersion model, AERMOD, intended to reduce
uncertainties in 1-hour concentration estimates;
• Greatly expanded database of human activity patterns that provide a stronger
foundation for inhalation exposure modeling;
• Improvements to the exposure model, APEX, designed to reduce uncertainties in
personal attributes of simulated individuals (e.g., breathing rates); and,
• Use of an expanded dataset for development of a lung function exposure-response
function, intended to reduce uncertainties in the response across the range of the study
data.
Based on the new information, model improvements, exploratory data evaluations, and updated
characterization of uncertainties, the results from this REA provide an improved characterization
of exposure and risk to inform the EPA's review of the primary SO2 standard.
8 The 2009 REA estimated exposures considering air quality adjusted to just meet several alternative standards
across a single-year period.
1-8
-------
The exposure-based risk assessment for this review includes an assessment of air quality
conditions just meeting the current standard in three study areas. The study areas were selected
based on consideration of the magnitude of recent SO2 concentrations, number of monitors in the
area, including those with 5-minute monitoring data, and population size. The risk
characterization is based on comparisons of population 5-minute exposures at elevated breathing
rates to health-based benchmark levels and estimated population risk of moderate or greater SO2-
related lung function decrements. The analyses and results are documented in this REA and key
findings of the REA are considered in the broader context of the PA, which also considers the
current evidence as assessed in the ISA and characterization of SO2 concentrations in ambient air
across the U.S. based on recent monitoring data, with particular attention to peak 5-minute
concentrations.
1.3.2 CASAC Advice and Public Comment
After consultation with the CASAC on the REA Planning Document (U.S. EPA, 2017c)
in March 2017 (Diez Roux, 2017), the EPA developed the draft REA (U.S. EPA, 2017d). The
CASAC SOx Panel discussed its review of the draft REA at a public meeting on September 20-
21, 2017 and in a public teleconference on April 20, 2018. The CASAC comments and
recommendations on the draft REA are provided in a May 2018 letter to the Administrator (Cox
and Diez Roux, 2018a). A number of comments on aspects of the draft REA were also received
from the public (see the public docket for this review, EPA-HQ-OAR-2013-0566 at
www.regulations.gov).
This final REA has been produced in consideration of the comments received on the draft
REA from the CASAC and from the public. The approach used to estimate population exposure
and risk has remained largely the same as the approach used in the draft REA, with a number of
adjustments and additions to address comments. Key changes include:
• Clarification regarding key design aspects including the air quality scenario and scope of
REA (Chapter 2);
• Revised study area maps that show locations of meteorological stations, air quality
receptors, emissions sources, and ambient air monitors, that also indicate source types
and SO2 emissions (sections 3.2 and 3.4);
• Improvements in estimating ambient concentrations associated with sources not explicitly
modeled in the Indianapolis study area (section 3.2.4);
• Additional evaluations of the daytime estimated 1-hour and 5-minute ambient air
concentrations in the three study areas by season (sections 3.2.5 and section 3.5.3.3);
• Expanded discussion regarding the approach used to adjust air quality to just meet the
current standard (section 3.4);
1-9
-------
Use of newly acquired continuous 5-minute ambient air monitoring data to estimate 5-
minute concentrations at modeled air quality receptors in the Indianapolis study area
(section 3.5.1);
Analysis of asthma prevalence information regarding the influence of body mass index
and race on populations with asthma (section 4.1.2);
Reorganized, clarified and expanded discussion regarding exposure model input data (i.e.,
body weight, surface area, energy expenditure) and algorithms (i.e., resting metabolic
rate, breathing rate) (section 4.1.3);
Expanded discussion of using activity pattern data from any individual in CHAD,
regardless of whether their asthma status is known or unknown, to represent the
simulated individuals with asthma (section 4.3.3);
Additional analysis and revised study area maps to better indicate where study area
populations overlap with highest ambient SO2 concentrations (sections 5.1 and 5.4); and
Updated analysis of the microenvironments where the simulated population experiences
the highest exposures (section 5.2);
Inclusion of number (and percentage) of individuals in the estimates of population
exposure and risk of lung function decrements presented in summary tables (section 5.2
and 5.3);
Expanded discussion of previously identified uncertainties, as well as identification and
discussion of additional uncertainties (Table 6-3).
1-10
-------
REFERENCES
Burton CS, Stoeckenius TE, Stocking TS, Carr EL, Austin BS, Roberson RL (1987). Assessment
of Exposures of Exercising Asthmatics to Short-term SO2 Levels as a Result of
Emissions from U.S. Fossil-fueled Power Plants. Systems Applications Inc., San Rafael,
CA. Publication No. 87/176, September 23, 1987.
Cox, LA; Diez Roux, A. (2018a). Letter from Louis Anthony Cox, Chair, Clean Air Scientific
Advisory Committee, and Ana Diez Roux, Immediate Past Chair, Clean Air Scientific
Advisory Committee, to Administrator E. Scott Pruitt. Re: CASAC Review of the EPA's
Risk and Exposure Assessment for the Review of the Primary National Ambient Air
Quality Standard for Sulfur Oxides (External Review Draft - August 2017). April 30,
2018.
Cox, LA; Diez Roux, A. (2018b). Letter from Louis Anthony Cox, Chair, Clean Air Scientific
Advisory Committee, and Ana Diez Roux, Immediate Past Chair, Clean Air Scientific
Advisory Committee, to Administrator E. Scott Pruitt. Re: CASAC Review of the EPA's
Policy Assessment for the Review of the Primary National Ambient Air Quality Standard
for Sulfur Oxides (External Review Draft - August 2017). April 30, 2018.
Diez Roux A. (2017). Letter from Ana Diez Roux, Chair, Clean Air Scientific Advisory
Committee, to Administrator Gina McCarthy. Re: Consultation on the EPA's Review of
the Primary National Ambient Air Quality Standard for Sulfur Oxides: Risk and
Exposure Assessment Planning Document (External Review Draft - February 2017).
April 18, 2017.
Rosenbaum AS, Hudischewskyj AB, Roberson RL, Burton CS. (1992). Estimates of Future
Exposures of Exercising Asthmatics to Short-term Elevated SO2 Concentrations
Resulting from Emissions of U.S. Fossil-fueled Power Plants: Effects of the 1990
Amendments to the Clean Air Act and a 5-Minute Average Ambient S02 Standard.
Publication No. SYSAPP- 92/016. April 23, 1992. Docket No. A-84-25, IV-K-37.
SAI. (1996). Summary of 1988-1995 Ambient 5-Minute SO2 Concentration Data. Prepared for
US EPA Office of Air Quality Planning and Standards by Systems Application
International. Contract #68-D3-0101, May 1996.
Sciences International (1995). Estimate of the Nationwide Exercising Asthmatic Exposure
Frequency to Short-term Peak Sulfur Dioxide Concentrations in the Vicinity of Non-
Utility Sources. Prepared for National Mining Association by Sciences International, Inc.,
Alexandria VA. April 1995. Docket No. A-84-25, VIII-D-71.
Stoeckenius TE, Garelick B, Austin BS, O'Connor K, Pehling JR. (1990). Estimates of
Nationwide Asthmatic Exposures to Short-Term Sulfur Dioxide Concentrations in the
Vicinity of Non-Utility Sources. Systems Applications Inc., San Rafael, CA. Publication
No. SYSAPP-90/129, December 6, 1990.
1-11
-------
U.S. EPA. (1986). Review of the National Ambient Air Quality Standards for Sulfur Oxides:
Updated Assessment of Scientific and Technical Information, Addendum to the 1982
OAQPS Staff Paper. Office of Air Quality Planning and Standards, Research Triangle
Park, NC, EPA/450/5-86-13. Available from: NTIS, Springfield, VA; PB87-
200259/XAB.
U.S. EPA. (1994). Supplement to the second addendum (1986) to air quality criteria for
particulate matter and sulfur oxides (1982): Assessment of new findings on sulfur dioxide
acute exposure health effects in asthmatic individuals. Environmental Criteria and
Assessment Office, Office of Health and Environmental Assessment, Office of Research
and Development, Research Triangle Park, NC, EPA/600/FP-93/002, August 1994.
Available at: https://www3.epa. gov/ttn/naaq s/standards/so2/s so2 pr.html
U.S. EPA. (2009). Risk and Exposure Assessment to Support the Review of the SO2 Primary
National Ambient Air Quality Standard. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-09-007, July 2009. Available at:
https://www3. epa. gov/ttn/naaq s/standards/so2/data/200908 S02REAFinalReport.pdf
U.S. EPA. (2014). Integrated Review Plan for the Primary National Ambient Air Quality
Standard for Sulfur Dioxide, Final. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/P-14-007, October 2014. Available at:
https://www3. epa. gov/ttn/naaq s/standards/ so2/data/ )28so2reviewplan .pdf
U.S. EPA. (2016). Integrated Science Assessment (ISA) for Sulfur Oxides - Health Criteria
(Second External Review Draft). National Center for Environmental Assessment-RTP
Division, Office of Research and Development, Research Triangle Park, NC,
EPA/600/R-16/351, December 2016. Available at:
https://cfpub.epa.gov/ncea/isa/recordisplav.cfm?deid=326450
U.S. EPA. (2017a). Integrated Science Assessment (ISA) for Sulfur Oxides - Health Criteria
(Final). National Center for Environmental Assessment-RTP Division, Office of
Research and Development, Research Triangle Park, NC, EPA/600/R-17/451, December
2017. Available at: https://cfpub.epa.gov/ncea/isa/recordisplav.cfin?deid=338596
U.S. EPA. (2017b). Integrated Review Plan for the Secondary National Ambient Air Quality
Standard for Ecological Effects of Oxides of Nitrogen, Oxides of Sulfur and Particulate
Matter. Office of Air Quality Planning and Standards, Research Triangle Park, NC, EPA-
452/R-17-002, January 2017. Available at: https://www.epa.gov/naaqs/nitrogen-dioxide-
no2-and-sulfur-dioxide-so2-secondarvstandards-planning-documents-current
U.S. EPA. (2017c). Review of the Primary National Ambient Air Quality Standard for Sulfur
Oxides: Risk and Exposure Assessment Planning Document. EPA-452/P-17-001,
February 2017. Available at:
https://www3.epa.gov/ttn/naaqs/standards/so2/data/20170216so2rea.pdf
U.S. EPA. (2017d). Review of the Primary National Ambient Air Quality Standard for Sulfur
Oxides: Risk and Exposure Assessment Planning Document. Office of Air Quality
Planning and Standards, Research Triangle Park, NC, EPA-452/P-17-001, February
1-12
-------
2017. Available at:
https://www3.epa.gov/ttn/naaqs/standards/so2/ciata/20170216so2rea.pdf
U.S. EPA. (2017e). Policy Assessment for the Review of the Primary National Ambient Air
Quality Standard for Sulfur Oxides, External Review Draft. Office of Air Quality
Planning and Standards, Research Triangle Park, NC, EPA-452/P-17-003, August 2017.
Available at: https://www.epa.gov/naaqs/sulfur~dioxide~so2~primarv~air~qualitv~
standards
U.S. EPA. (2018). Policy Assessment for the Review of the Primary National Ambient Air
Quality Standard for Sulfur Oxides, Final. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-18-002, May 2018. Available at:
https://www.epa.gov/naaqs/siilfiir~dioxide~so2.-primarv~air~qiialitv~stan.dards
1-13
-------
2 OVERVIEW OF ASSESSMENT APPROACH
As summarized in the IRP and PA for this review of the NAAQS for SOx, the review
focuses on the presence in ambient air of sulfur oxides, a group of closely related gaseous
compounds that include sulfur dioxide and sulfur trioxide and of which sulfur dioxide (the
indicator for the current standard) is the most prevalent in the atmosphere and the one for which
there is a large body of scientific evidence on health effects. Sulfur trioxide is known to be
present in the emissions of coal-fired power plants, factories, and refineries, but it reacts with
water vapor within emission stacks or immediately after release into the atmosphere within
seconds to form H2SO4 which quickly condenses onto existing atmospheric particles or
participates in new particle formation (ISA, p. 2-18). Thus, only SO2 is present at concentrations
in the gas phase that are relevant for chemistry in the atmospheric boundary layer and
troposphere, and for human exposures (ISA, p. 2-18). The health effects of particulate
atmospheric transformation products of SOx, such as sulfates, are addressed in the review of the
NAAQS for particulate matter (U.S EPA, 2018; U.S. EPA, 2016). For these reasons, this REA is
focused on SO2.9 The conceptual model for exposure and associated health risk of SO2 in
ambient air that guides our assessment in this review is described in this section along with an
overview of the implemented approach.
2.1 CONCEPTUAL MODEL FOR SO2 EXPOSURE AND RISK
The conceptual model for our consideration of exposure and risk associated with SO2 in
ambient air is illustrated in Figure 2-1. This general model guided our assessment in the last
review and, as discussed in the REA Planning Document and draft REA, remains appropriate in
the current review. The unshaded boxes indicate components included in the assessment in this
review. Current information regarding the individual components specified in the model
(emissions sources, exposure pathways, routes of exposure, exposed populations, health
endpoints and risk metrics) is summarized in the following sections. A more detailed
characterization of this information is presented in the ISA (U.S. EPA, 2017a).
9 While there are some toxicological animal studies of SOx in mixtures, such as with co-occurring PM, that indicate
some enhanced effect on lung function parameters, there are a number of limitations with regard to appropriate
controls and relevance to ambient air exposures (ISA, pp. 5-143 to 5-144). Thus, the available information does
not support characterization in this assessment of any potential for modification SCh-related effects by
copollutants, such as PM. Uncertainties in the exposure and risk estimates generated in this REA with regard to
the potential for modification of S02-related effects by co-occurring pollutants, such as PM, are characterized in
section 6.2.1.
2-1
-------
£
Z3
m
O
CL
&
s
g
p
g
U.
U1
eg
a;
I
-------
2.1.1 Sources of SO2
Sulfur dioxide occurs in ambient air as a result of direct emissions of SO2 as well as
emissions of other compounds, such as reduced sulfur compounds or sulfides, that are converted
to SO2 through chemical reactions in the atmosphere. The largest natural sources of SO2 are
volcanoes and wildfires. Fossil fuel combustion is the main anthropogenic source of SO2 and
industrial chemical production and pulp and paper production are among the sources of reduced
sulfur compounds that are converted to SO2 in the atmosphere. Anthropogenic sources of SO2
emissions that contribute to SO2 found in ambient air are primarily large facilities and include
coal-fired electricity generating units (EGUs) as well as other industrial facilities (U.S. EPA,
2008 [hereafter referred to as the 2008 ISA], section 2.1; ISA, section 2.2.1). Because such large,
discrete sources are the primary source of SO2 (e.g., versus more prevalent, widespread sources),
ambient concentrations can vary substantially across an area and can be relatively high in areas
affected by these large sources.
Coal-fired EGUs are an important emissions source because coal contains sulfur, which
is present to some degree in all fossil fuels. The sulfur content of the most common types of coal
varies between 0.4 and 4% by mass (ISA, section 2.2). Fuel sulfur is almost entirely converted to
sulfur oxides during combustion. This makes accurate estimates of SO2 combustion emissions
possible based on fuel composition and combustion rates (ISA, section 2.2). Fuel combustion by
electric utilities as well as industrial and other sources is the largest source of anthropogenic SO2
emissions (ISA, Figure 2-1).
Although they may be fewer in number than fossil fuel-fired EGUs, other types of large
emissions facilities that may impact local air quality include copper smelters, kraft pulp mills,
Portland Cement plants, iron and steel mill plants, sulfuric acid plants, petroleum refineries, and
chemical processing plants. For example, the metal processing sector represents less than 2.3%
of total emissions from the 2014 National Emissions Inventory (NEI),10 however, monitoring
sites that have recorded some of the highest 1-hour daily maximum SO2 concentrations in the
U.S. are located near copper smelters in Arizona (ISA, sections 2.5.2 and 2.5.4, Figure 2-11).
The two smelters in this area emit appreciable quantities of SO2, estimated at 17,000 tons per
year (tpy) and 5,000 tpy (ISA, p. 2-50), but for added perspective, several EGUs in other areas
have been estimated to emit well over 50,000 tpy in the 2014 NEI.
The main indoor source of SO2 is indoor combustion of sulfur-containing fuels, such as
from space heaters that are generally used in the U.S. as emergency or supplemental sources of
heat. For example, a study in the eastern U.S. reported that kerosene heaters, but not fireplaces,
10 The National Emissions Inventory (NEI) is a comprehensive and detailed estimate of air emissions of criteria
pollutants, criteria precursors, and hazardous air pollutants from air emissions sources. For additional
information, see https://www.epa.gov/air-emissions-inventories/nationai-emissions-inventory-nei.
2-3
-------
woodstoves, or gas space heaters, resulted in increased indoor concentrations of SO2 (ISA,
section 3.4.1.1). Personal SO2 exposure measurements, however, have generally been lower than
ambient air concentrations, indicating personal exposure is generally dominated by ambient air
(outdoor) sources (ISA, section 3.4.1).
The context for the REA is exposure and associated risk of SO2 emitted into ambient air.
Accordingly, the conceptual model for the REA focuses on sources to ambient air (Figure 2-1).
2.1.2 Exposure Pathways and Route
Human exposure to SO2 involves the contact between a person and the pollutant in any of
the various locations (or microenvironments, MEs) in which people spend their time. As SO2 is a
gas, human exposure occurs through inhalation of air containing SO2. The concentrations of SO2
occurring in each ME and the associated activity performed in the ME at the time of exposure
both contribute to individual exposure events. Together, these exposure events make up an
individual's exposure (ISA, section 3.2.2).
Exposure microenvironments occur indoors (e.g., in homes, offices or stores), outdoors
(e.g., yards, parks, sidewalks) and in vehicles (e.g., automobiles, buses). All of these
microenvironments can receive ambient air that may contain SO2. Thus, the pathways by which
people are exposed to SO2 in ambient air involve inhaling air while spending time in the various
MEs.
While indoors, people can be exposed to SO2 from indoor sources as well as to SO2
associated with outdoor air that has infiltrated into indoor MEs. Studies of personal exposure
have generally found that the largest portion of a person's day is spent indoors (ISA, section
3.4.2.1). As a result of this and indoor SO2 concentrations typically being lower than SO2
concentrations measured outdoors, SO2 exposure concentrations are often much lower than SO2
concentrations in ambient air (ISA, section 3.4.1). As stated in the ISA, high correlations (>0.75)
between indoor and outdoor SO2 concentrations indicate that variations in outdoor ambient SO2
concentration11 are driving indoor SO2 concentrations, which is considered to be consistent with
the relative lack of indoor sources of SO2 (ISA, section 3.4.1.2).
Thus, personal SO2 exposure is expected to be dominated by SO2 emitted into ambient air
in outdoor microenvironments and enclosed microenvironments with high air exchange rates,
such as buildings with open windows and vehicles. This was found to be the case in exposure
"Concentrations of SO2 in ambient air are spatially highly variable compared to pollutants such as ozone (ISA,
section 3.2.3); this is due to the point source nature of SO2 emissions. Another factor in the spatial variability is
the dispersion and oxidation of SO2 in the atmosphere, processes that contribute to decreasing concentrations with
increasing distance from the source. Point source emissions of SOx create a plume of higher concentrations,
which may or may not impact large portions of surrounding populated areas depending on meteorological
conditions and terrain (ISA, section 3.2.3).
2-4
-------
modeling of recent air quality performed for the 2009 REA: more than 80% of the events by
which simulated individuals experienced elevated 5-minute exposure concentrations of interest
were in outdoor MEs (2009 REA, Figure 8-21). As was done in the 2009 REA for the last review
of the NAAQS for SOx, exposures to SO2 in ambient air outdoors, as well as to ambient air that
has infiltrated indoors, are included in the REA for the current review.
2.1.3 At-Risk Populations
As at the time of the 2009 REA, the current evidence demonstrates that the populations at
increased risk of effects from SO2 exposure continue to be people with asthma, particularly
children with asthma (ISA, section 6.3.1). Strong evidence of this comes from the controlled
human exposure studies of people with asthma exposed to SO2 when their breathing rates are
increased, such as from exercise (ISA, section 5.2.1.9). Consistent with the controlled human
exposure study findings of asthma exacerbation-related effects, some epidemiological studies in
the current evidence report associations between short-term SO2 exposure and increased risk of
asthma-related emergency department visits and hospital admissions (ISA, section 5.2.1.9).
The short-term respiratory effects that are the focus of the quantitative assessment, and
for which the evidence for respiratory effects associated with policy-relevant SO2 exposure
concentrations is strongest, are asthma exacerbation-related effects (ISA, Table 1-1). Under
resting conditions, inhaled SO2 is readily removed in the nasal passages (ISA, section 1.5.1).
However, during activities that result in elevated breathing rates, such as those associated with
exercise, and/or an increased potential for taking breaths through the mouth (versus the nose),
there is greater transport of inhaled SO2 past the nasal passages to the tracheobronchial region of
the airways where it can contribute to bronchoconstriction-related effects and asthma
exacerbation (ISA, section 1.5.1). Thus, elevated breathing rates and breathing habits that
include breathing through the mouth (oronasal), such as that occurring during exercise, play
important roles in eliciting S02-related effects in at-risk populations.
While some controlled exposure studies involving adolescents with asthma have
indicated that this age group has similar responsiveness as adults, controlled exposure study data
are not available for children younger than 12 years (ISA, section 5.2.1.2). However, some
factors indicate that children (e.g., younger than 13 years) with asthma may be at a greater risk
than adults with asthma. For example, children, particularly younger than 13 years of age, have a
greater tendency to breathe through the mouth than do adults (ISA, section 4.1.2.2). Evidence
also suggests that older adults with asthma may also be at an increased risk compared to younger
adults with asthma (ISA, section 6.5.1.2).
The evidence in controlled exposure studies documents the difference in sensitivity to
S02-related respiratory effects in individuals with and without asthma. For example, these
2-5
-------
studies document respiratory effects occurring in exercising study subjects with asthma at
exposure concentrations below 1000 ppb, while higher concentrations are needed to elicit similar
effects in healthy subjects and in some subjects with asthma (ISA, sections 5.2.1.2 and 5.2.1.7).12
The currently available information does not identify other populations at increased risk beyond
what is described here (ISA, section 6.6). As indicated in Figure 2-1, people with asthma,
including both adults and children, are specifically identified as at-risk populations in the REA
for this review.
2.1.4 Health Endpoints
The health effects causally related to SO2 exposures are effects on the respiratory system
(ISA, section 1.6). As demonstrated in long-standing evidence from controlled human exposure
studies and consistent with findings in epidemiological studies, short-term SO2 exposures (as
short as a few minutes) can result in asthma exacerbation-related effects in people with asthma.
The controlled human exposure studies have demonstrated a relationship between 5- and 10-
minute peak SO2 exposures and bronchoconstriction-related decrements in lung function in
exercising individuals with asthma; depending on the exposure level, these decrements are
accompanied by respiratory symptoms (ISA, section 5.2.1.2).
Lung function decrements were quantified in these studies by reductions in forced
expiratory volume in one second, FEVi, and increased specific airway resistance, sRaw. In
considering the magnitude of these responses, the ISA (as in the 2008 ISA) focuses on 15% or
greater reductions in FEVi and increases in sRaw of 100% or more (ISA, sections 1.6.1.1 and
5.2.1.2). Such responses have been reported in some individuals with asthma exposed to 5-
minute concentrations as low as 200 ppb while exercising. Across the range of exposure
concentrations studied, both the percentage of individuals affected to at least this degree and the
severity of the response increases with increasing SO2 concentrations. At higher concentrations
(above 400 ppb), such responses were frequently accompanied by respiratory symptoms (ISA,
section 5.2.1.2).
2.1.5 Risk Metrics
As was the case in the 2009 REA, the risk metrics included in the current REA (bottom
panels, Figure 2-1) are based on the S02-induced bronchoconstriction-related lung function
12 The evidence from controlled exposure studies has long documented the sizeable variation in sensitivity to SO2
among individuals with asthma. This was further characterized in a pooled analysis of data from five such studies
that is newly available in this review (Johns et al., 2010). This new analysis demonstrates the study population of
individuals with asthma to fall into one of two subpopulations with regard to airway responsiveness to SO2. One
subpopulation is insensitive to the bronchoconstrictive effects of SO2 even at concentrations as high as 1.0 ppm,
and it is the second subpopulation that has an increased risk for bronchoconstriction at the lower concentrations of
SO2 (ISA, section 5.2.1.2).
2-6
-------
decrements documented in the strong evidence base of controlled human exposure studies of
exercising individuals with asthma. Bronchoconstriction, an asthma-exacerbation-related effect,
is the "most sensitive indicator of SCh-induced lung function effects" and the evidence for this
effect is strong (ISA, section 5.2.1.2, p. 5-8). The first of the risk metrics included in this REA
involves characterization of the extent to which individuals with asthma were estimated to
experience 5-minute exposures at or above concentrations of potential concern while they are at
elevated breathing rates. The second metric quantifies the extent to which individuals with
asthma are estimated to experience lung function responses (in terms of a doubling, or larger
increase, in sRaw) as a result of 5-minute SO2 exposures while at elevated breathing rates.
In deriving these two risk metrics, the controlled human exposure studies are used in two
ways: (1) to identify exposure concentrations of potential concern ("benchmark concentrations")
and (2) to derive exposure-response (E-R) functions for lung function decrements. As described
in more detail in section 3.5.1, the benchmark concentrations are 5-minute exposure
concentrations chosen to represent exposures of potential concern. The first metric, the
comparison of SO2 exposures to benchmark concentrations, provides perspective on the extent to
which there is potential for sensitive individuals with asthma to experience SO2 exposures that
could be of concern at air quality just meeting the current standard.
The second metric relies on the E-R function and exposure estimates to estimate risk of
decrements in lung function based on sRaw, which is a specific measure of bronchoconstriction.
The focus on sRaw as the primary indicator of lung function response is consistent with the
emphasis on this indicator in the REA for the last review. The E-R functions for sRaw are based
on more observations from individual subjects than were E-R functions based on FEVi (2009
REA, p. 332), which provides greater confidence in the resultant quantitative relationship when
compared with that developed for the FEVi health endpoint.
Another category of metric shown in the conceptual model figure represents potential
asthma-exacerbation-related health outcomes that are reported in the epidemiological evidence.
As indicated by the shading in Figure 2-1, this category of metrics is not included in this REA as
the current evidence base does not support its inclusion. This was also the case in the 2009 REA
(REA Planning Document, section 3.2.3). As examined in detail in the ISA, the epidemiological
evidence includes studies reporting associations between short-term SO2 concentrations and
asthma-related emergency department visits or hospitalizations. The risk characterization for the
2009 REA focused on metrics for lung function decrements related to bronchoconstriction,
concluding that the epidemiological evidence did not support development of an epidemiological
study-based risk model. In considering support in the evidence available in this review, the REA
Planning Document for this REA reached the same conclusion (REA Planning Document,
2-7
-------
section 3.2.3). Thus, as shown in Figure 2-1, the incidence of respiratory health outcomes metric
is not included in this REA.
2.2 ASSESSMENT APPROACH
The approach employed for this REA generally involves estimating population exposures
to ambient air-related SO2 concentrations and associated health risk for air quality conditions
simulated to just meet the current standard (Figure 2-2). This approach, which draws on air
monitoring data, air quality modeling and exposure modeling, was applied in three study areas
(section 3.1) selected to be most informative to this review. The focus on air quality conditions
just meeting the current standard reflects the key overarching question articulated in the IRP for
this review: Does the currently available scientific evidence- and exposure/risk-based
information, as reflected in the ISA and REA, support or call into question the adequacy of the
protection afforded by the current standard (IRP, section 3; PA, section 3.2)? In considering the
final ISA and the draft REA results, the draft PA reached preliminary conclusions that the
answer to this question was no and that it is appropriate to consider retaining the current standard
without revision. The CAS AC concurred with this conclusion (Cox and Diez Roux, 2018).
Accordingly, exposure and risk analyses using alternative air quality conditions were not
warranted and have not been performed for this REA.
As indicated by the case study approach, the REA analyses are not intended to provide a
comprehensive national assessment. Rather, they are intended to provide assessments for a small
varied set of study areas, and the associated exposed at-risk populations, that will be informative
to EPA's consideration of potential exposures and risks that may be associated with the air
quality conditions occurring under the current SO2 standard. The purpose of the REA is to
assess, based on the currently available, improved and expanded tools and information, the
potential for exposures and risks beyond those indicated by the information available at the time
the current standard was established. In this way, the REA can inform the EPA's conclusions on
the public health protection afforded by the current standard.
Consistent with the health effects evidence and the health risk metrics identified in
section 2.1.5, the focus is on short-term exposures of individuals in the population with asthma
during times when they are breathing at an elevated rate. Exposure and risk is characterized for
two population groups: adults (individuals older than 18 years) with asthma and school-aged
children (aged 5 to 18 years)13 with asthma. The focus on these populations is consistent with the
ISA's identification of individuals with asthma as the population at risk of SCh-related effects,
13 As in other NAAQS reviews, this REA does not estimate exposures and risk for children younger than 5 years old
due to the more limited information contributing relatively greater uncertainty in modeling their activity patterns
and physiological processes than children between the ages of 5 to 18.
2-8
-------
and its conclusion that within this population, children with asthma may be at greater risk than
adults with asthma (ISA, section 6.6).
In order to estimate ambient air concentrations at the needed temporal scale of 5-minute
increments, the REA employs air quality modeling as informed by additional information from
5-minute ambient air monitoring data. Air quality modeling is used in order to adequately
capture the spatial variation in ambient SO2 concentrations across an urban area, which can be
relatively high in areas affected by large point sources and which the limited number of
monitoring locations in each area are unlikely to capture. Continuous 5-minute ambient air
monitoring data are used to reflect the fine-scale temporal variation in SO2 concentrations
documented by these data and for which air quality modeling is limited, e.g., by limitations in
currently available input data such as emissions estimates. Thus, 5-minute concentrations in
ambient air were estimated using a combination of 1-hour concentrations from the EPA's
preferred near-field dispersion model, the American Meteorological Society/EPA regulatory
model (AERMOD), and relationships between 1-hour and 5-minute concentrations occurring in
the local ambient air monitoring data.14
The Air Pollutants Exposure (APEX) model, a probabilistic human exposure model that
simulates the activity of individuals in the population, including their exertion levels and
movement through time and space, was then used to estimate 5-minute exposure concentrations
for individuals based on exposures in indoor, outdoor, and in-vehicle microenvironments. The
use of APEX for estimating exposures allows for consideration of factors that affect exposures
that are not addressed by consideration of ambient air concentrations alone. These factors include
1) attenuation in SO2 concentrations expected to occur in some indoor microenvironments, 2) the
influence of human activity patterns on the time series of exposure concentrations, and 3)
accounting for human physiology and the occurrence of elevated breathing rates concurrent with
SO2 exposures, all key to appropriately characterizing health risk for SO2.
The estimated exposures were then combined with findings of the controlled human
exposure studies to characterize health risk using two approaches. The first approach compares
estimated exposures to benchmark concentrations of interest and the second combines exposures
with an E-R function to estimate the expected occurrences of decrements in lung function.
14 The current information continues to support the use of an air dispersion model such as AERMOD over the use of
other models, such as photochemical models, for modeling of directly emitted SO2 concentrations for use in
assessing risk and exposure for this pollutant. Unlike dispersion models, photochemical models cannot capture
the sharp concentration gradients that can occur near SO2 sources. Also, SO2 emissions to ambient air are
dominated by point sources, such as large coal-fired utilities, and AERMOD is the EPA's preferred air quality
model for S02for State Implementation Plans (SIPs) and new source permitting purposes. For all of these
reasons, AERMOD remains the most appropriate model for predicting SO2 concentrations in ambient air.
2-9
-------
Thus, two types of risk metrics were derived from the simulated individual exposure
profiles: (1) the number and percent of the simulated subpopulation that had at least one 5-
minute exposure above the benchmark concentrations of 100, 200, 300, and 400 ppb and (2) the
number and percent per year of simulated at-risk individuals that would experience moderate or
greater lung function decrements in response to 5-minute daily maximum peak exposures while
engaged in moderate or greater exertion. Estimates were developed for three study areas. The
details and basis for each of these aspects of the assessment are described in Chapters 3 and 4.
[
4s is AQ and that
adjusted for different
AQ scenarios
Air Quality Modeling (AERMOD)
(continuous 1-hour concentrations)
I Ambient Air Monitoring Data
I (continuous 5-minute concentrations)
Air quality
model-based
approach
I
Exposure Modeling (APEX)
(5-minute exposures at elevated exertion)
Health-Based
Benchmarks
I
Controlled Human
Exposure Data
(5-10 minute exposures
at elevated exertion)
Lung Function Exposure-
Response Relationship
Exposures (at exertion) at
or above Benchmarks
Lung Function Risk
Output: Number and percent of
exposed people with asthma estimated
to experience moderate or greater lung
function responses (i e , sRaw)
Output: Number and percent of people
with asthma at moderate or greater
exertion estimated to be exposed to 5-
mmute daily maximum S02 concentrations
that exceed 5-minute benchmark values
r
Jr
Exposure and Risk-Related Considerations in Review of Standard
Figure 2-2. Overview of the assessment approach.
2-10
-------
REFERENCES
Cox, LA; Diez Roux, A. (2018). Letter from Louis Anthony Cox, Chair, Clean Air Scientific
Advisory Committee, and Ana Diez Roux, Immediate Past Chair, Clean Air Scientific
Advisory Committee, to Administrator E. Scott Pruitt. Re: CAS AC Review of the EPA's
Policy Assessment for the Review of the Primary National Ambient Air Quality Standard
for Sulfur Oxides (External Review Draft - August 2017). April 30, 2018.
Johns DO, Svendsgaard D, Linn WS. (2010). Analysis of the concentration-respiratory response
among asthmatics following controlled short-term exposures to sulfur dioxide. Inhal
Toxicol. 22:1184-1193. http://dx.doi.org/10.3109/089S8378.2( )
U.S. EPA. (2008). Integrated Science Assessment (ISA) for Sulfur Oxides - Health Criteria
(Final Report). National Center for Environmental Assessment-RTP Division, Office of
Research and Development, Research Triangle Park, NC, EPA-600/R-08/047F,
September 2008. Available at:
http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=198843
U.S. EPA. (2009). Risk and Exposure Assessment to Support the Review of the SO2 Primary
National Ambient Air Quality Standard. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-09-007, July 2009. Available at:
https://www3.epa.gOv/ttn/naaqs/standards/so2./data/2.00908S02.REAFinalReport.pdf
U.S. EPA. (2016). Integrated Review Plan for the Secondary National Ambient Air Quality
Standards for Particulate Matter. Office of Air Quality Planning and Standards, Research
Triangle Park, NC, EPA-452/R-16-005, December 2016. Available at:
https://www.epa.eov/naaqs/particulate-matter-pm-standards-plannlne-documents-curren.t-
review
U.S. EPA. (2017). Integrated Science Assessment (ISA) for Sulfur Oxides - Health Criteria
(Final). National Center for Environmental Assessment-RTP Division, Office of
Research and Development, Research Triangle Park, NC, EPA/600/R-17/451, December
2017. Available at: https://cfpub.epa.eov/ncea/isa/recordisplav.cfm?deid=338596
U.S. EPA. (2018). Policy Assessment for the Review of the Primary National Ambient Air
Quality Standard for Sulfur Oxides, Final. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-18-002, May 2018. Available at:
https://www.epa.eov/naaqs/siilfur-dioxide-so2-primary-air-qiialitv-standards
2-11
-------
3 AMBIENT AIR CONCENTRATIONS
As summarized in chapter 2, the approach for this REA is based on linking the health
effects information to estimated population-based exposures that reflect our current
understanding of 5-minute concentrations of SO2 in the ambient air. This approach is applied to
three study areas to provide a valuable perspective on exposures and risks for at-risk populations
that is informative to this review of the SO2 primary standard. This chapter describes the
methodology for developing the spatial and temporal patterns of 5-minute concentrations in
ambient air for each of the three study areas. Our overall objective for this methodology is not
necessarily to develop an air quality surface for each study area that exactly matches one that has
occurred. Rather, it is to develop a hypothetical air quality scenario that accounts for the spatial
and temporal pattern of ambient SO2 concentrations in each study area that might be expected to
occur when the current primary SO2 standard has just been met, is based on the types of SO2
sources that have existed in the area (and local or nearby sources that may also influence ambient
air concentrations) and considers the expected variability in observed meteorological conditions.
This hypothetical scenario, however, is not necessarily reflective of a specific calendar year,
even though data from specific years have been used as a basis for the development of the
hypothetical scenario. In so doing, we have implemented methods intended to capture the
appropriate spatial and temporal heterogeneity in SO2 concentrations that occur near and around
important emissions sources considering this hypothetical air quality scenario and, when
considering population demographics, to reasonably represent the population groups at risk for
S02-related health effects.
The three study areas and time periods simulated are described in section 3.1 below. Air
quality modeling is used to develop the spatially varying distributions of 1-hour concentrations,
as described in section 3.2. The definition of the extent and scale of the exposure modeling
domain and associated air quality receptor grid is described in section 3.3. The next step in the
approach is the development of an air quality scenario for each study area that reflects conditions
that just meet the current standard. This step involves adjustment of the estimates resulting from
the air quality modeling for each area. Section 3.4 summarizes the method used for the
adjustment of the air quality concentrations to a scenario that just meets the current primary SO2
standard. Development of the temporally varying 5-minute concentrations at each air quality
receptor site is described in section 3.5.
3-1
-------
3.1 CHARACTERIZATION OF STUDY AREAS
The study areas for this REA are Fall River, MA, Indianapolis, IN, and Tulsa, OK (Table
3-1). These study areas were selected to meet a number of individual and collective criteria. The
following list includes the criteria used in considering individual study areas:
• Design value1 near the current standard (75 ppb). Using recent air quality monitoring
data (2011-2015), design values ranging from 50 ppb to 100 ppb were considered
preferable in order to minimize the magnitude of the adjustment needed to generate air
quality just meeting the current standard, therefore potentially minimizing the
uncertainties in estimates of exposures associated with the adjustment approach. In
considering areas with regard to this criterion, consecutive 3-year periods as far back as
2011-2013 were considered.
• One or more air quality monitors reporting 5-minute SO2 data for the 3-year study
period. In judging whether monitors provided such a 3-year record, completeness
requirements (summarized in section 3.5) were applied for all three years to ensure the
availability of adequate data for informing the ambient air concentrations used for
exposure modeling. Study areas having continuous 5-minute data were preferable to
those with only hourly maximum 5-minute data. There are no monitoring requirements to
report continuous 5-minute data at all ambient air monitors, therefore we used this as an
additional consideration after an initial screen for the top candidate areas.
• Availability of existing air quality modeling datasets. There are many areas in the U.S.
that have chosen to model air quality for regulatory purposes, i.e., in designating areas
with regard to determining the attainment of the current standard. This criterion was
considered important for efficiency purposes and to maintain consistency between our
assessment approach and state-level modeling regarding the years selected, sources
included, emission levels and profiles, and assumptions used to predict ambient air
concentrations.
• Population size greater than 100,000. Candidate study areas having the larger
populations were given priority to provide a more robust and improved representation of
exposures and risk to key at-risk populations.
• Significant and diverse emissions sources. Preference was given to study areas with a
diverse source mix, including EGUs, petroleum refineries, and secondary lead smelting
(generally reflects battery recycling). A diverse source mix allows for capturing
1 A design value (DV) is a statistic that describes the air quality status of a given area relative to a particular
NAAQS. A design value summarizes the concentrations of a criteria pollutant in terms of the statistical form of
the standard for that pollutant, thus indicating whether the area meets or exceeds the standard. Consistent with the
form of the SO2 standard, SO2 design values are calculated as the 3-year average of the annual 99th percentile of
the daily maximum 1-hour average concentrations (see 40 CFR 50.17). By regulation, design values calculated
from monitoring data are considered to be valid if they meet specified completeness criteria, which for SO2 are
data for at least 75 percent of the sampling days in all four quarters of all three years of the period (see Appendix
T to Part 50).
3-2
-------
exposures to both large sources (e.g., emissions of 10,000-20,000 tons per year [tpy])2
and small sources (e.g., emissions of hundreds of tpy) distributed about a study area.
In consideration of the above criteria, Fall River, MA, Indianapolis, IN, and Tulsa, OK
were selected.3 In identifying this set of study areas, we also concluded it to be desirable for the
study areas, as a set, to represent different geographical regions of the U.S. The three study areas
- in Massachusetts, Indiana and Oklahoma -are in three different climate regions of the U.S.: the
Northeast, Ohio River Valley (Central), and South (Karl and Koss, 1984). These regions,
particularly the Ohio River Valley, generally have a higher concentration of EGU and non-EGU
sources of SO2 emissions than other areas of the country (ISA, Figure 2-3). Given the objective
of assessing air quality conditions that just meet the current standard, our focus, as indicated by
the first criterion above, is not on the areas in the U.S. with ambient air concentrations
substantially above the standard.4 Additionally, we minimized inclusion of study areas near the
ocean or large water bodies, such as the Great Lakes, given the potential for unusual atmospheric
chemistry and associated transformation of SO2 in those areas and our limited ability to
accurately model such events.
We considered more than one hundred areas and multiple time periods as study area
candidates. Closer examination of candidate areas and time periods led us to select the three
study areas identified above and the study period of 2011 to 2013, as they best fit the above
selection criteria.5 The study areas and time periods selected - Fall River, MA, Indianapolis, IN,
and Tulsa, OK (Table 3-1) - together represent an array of differing exposure circumstances for
5-minute peak SO2 concentrations in ambient air. This array expands on the more limited set of
study areas, focused in a single region of the U.S., that was addressed in the 2009 SO2 REA. As
described in subsequent sections, information for the 2011-2013 period in the three study areas
was used to develop the air quality scenarios that represent conditions just meeting the current
2 While there may be other sources having similar or greater SO2 emissions, design values for the ambient monitors
surrounding these other sources may not necessarily fall within that particular selection criterion. Again, having
monitor design values at or near the existing standard is considered important in limiting the magnitude of
uncertainty associated with adjusting concentrations that just meet the existing standard.
3 Further investigation of available information for potential study area locations with regard to the criteria identified
above resulted in the identification of the selected three study areas, for two of which existing air quality
modeling datasets were available. Such datasets were not available for many of the potential study areas referred
to as candidates in the REA Planning Document (e.g., Detroit and Savannah).
4 This objective of the REA and, more specifically, the design value criterion used to identify candidate study areas
for the REA differs from the criteria used in selecting the six focus areas in the ISA. The selection criteria used to
identify focus areas in the ISA did not consider ambient monitoring concentration levels, and as such, four of the
six ISA focus areas would not meet the above REA design value criterion alone (ISA, section 2.5.2.2).
5 Use of this time period (2011-2013) in these three study areas, in which concentrations were closer to the current
standard than indicated in more recent data, allowed us to apply a smaller adjustment in developing the air quality
scenarios for just meeting the current standard, thus reducing any associated uncertainty.
3-3
-------
standard for which this REA has estimated exposures and risks to at-risk populations from SO2
concentrations in ambient air.
Table 3-1. General features of the study areas selected for the exposure and risk
assessment.
Study Area
Geographic
Region
# of Monitors in
Exposure
Modeling Domaina
Reporting
5-Minute Data
(# with Continuous
Data)
2011-2013
DVb
(ppb)
Population in
Exposure
Modeling
Domainac
# of Sources
emitting >100
tons per yeard
in Exposure
Modeling
Domain
Source Typese
Fall River, MA
New England
1(1)
64
183,874
1
EGU
Indianapolis,
IN
Ohio River
Valley
3(3)
78
547,968
4
EGUs, secondary
lead smelter,
airport
Tulsa, OK
South
4(4)
55
257,423
3
EGU, petroleum
refineries
a Delineation of the exposure modeling domain is described in section 3.4; it includes the area within 10 km of the sources with
SO2 emissions above 100 tons in 2011, 2012 or 2013 and inclusive of the monitors with 5-minute data.
b Highest monitor-based design value in exposure modeling domain.
c Population sizes are drawn from 2010 U.S. Census.
d This reflects information in 2011 National Emissions Inventory. As described in section 3.2, other sources are also reflected in
the air quality modeling, either explicitly or via the addition of study-area-specific concentrations.
e This reflects sources counted in column to the left of this one. As described in section 3.2, other sources are also reflected in
the air quality modeling, either explicitly or via the addition of study-area-specific concentrations.
3.2 AIR QUALITY MODELING
The EPA's preferred model for near-field dispersion, AERMOD (U.S. EPA, 2016a, b),
was used to generate 1-hour concentrations for the 3-year period, 2011-2013, across the exposure
modeling domains for the three study areas: Fall River, MA, Indianapolis, IN, and Tulsa, OK. In
addressing the development of model inputs and specifications, as well as performing the
modeling runs themselves, the steps listed below were performed for all three study area
modeling domains.
(1) Collected and analyzed general input parameters. Meteorological data, processing
methodologies used to derive input meteorological fields (e.g., temperature, wind speed,
precipitation), and information on surface characteristics and land use were needed to
help determine pollutant dispersion characteristics, atmospheric stability and mixing
heights (section 3.2.1).
(2) Defined sources and estimated emissions. The modeled emission sources included
major stationary emission sources within the domain (section 3.2.2).
3-4
-------
(3) Defined air quality receptor locations. Receptor locations were identified for the
dispersion modeling at varying spatial scale (depending on distance from source to
receptor) from 2 km to 100 m (section 3.2.3).
(4) Calculated background concentrations. In this context the phrase "background
concentrations" refers to SO2 concentrations resulting from sources (nearby and distant)
other than those whose emissions are explicitly modeled. These concentrations were
calculated based on ambient air monitoring data that exclude hours of the day that were
most likely influenced by the modeled emission sources (section 3.2.4).
(5) Estimated concentrations at receptors. Full annual time series of hourly concentration
were estimated for 2011-2013 by summing concentration contributions from each of the
emission sources at each of the defined air quality receptors (section 3.2.5).
Details regarding both modeling approaches and input data used are provided below with
supplemental information regarding model inputs and methodology provided in Appendices A,
B, and C. To ensure use of the appropriate local data for the simulated time periods, as well as
efficiency and consistency for these areas, we drew on information for the Indianapolis and
Tulsa study areas (e.g., stack locations, building parameters, etc.) that had been developed for
regulatory purposes.6'7 Information for the Fall River study area was developed specifically for
this assessment in a manner that was technically appropriate and generally consistent with that
for the other two areas. The sections below summarize development of the information described
in the steps listed above for all study areas. Figures 3-1 to 3-3 show the locations of the upper
and surface meteorological stations, the modeled SO2 emission sources, and the ambient
monitoring sites used for predicting air quality used in this REA. Because some of the
meteorological stations and emissions sources were located outside of the general study area,8
two maps are provided for each study area: one map encompassing all of the features and the
second map focused on those features closest to or within each study area.
6 For the Indianapolis study area, we drew on the modeling performed by Indiana Department of Environmental
Management for Indiana's State Implementation Plan (SIP) for the Marion County SO2 nonattainment area. This
documentation is available at:
http://www.in.gov/idem/airqnalitv/files/affaininent so2 multl 20.1.5 demo attach k.pdf.
7 For the Tulsa study area, we drew on the modeling performed by Oklahoma Department of Environmental Quality
to address regulatory Prevention of Significant Deterioration (PSD) requirements for refineries in the Tulsa area.
This information is available for Permits 2012-1062-TVR2 M-9 and 2010-599-TVR M-7 at:
http://www.deq.state.ok.ns/aadnew/permitling/PermitsIssnedDuringPastYear.htmi.
8 For better visualization of the meteorological stations, emission sources, and the ambient monitors used to estimate
air quality for this assessment, the area highlighted is an approximation based on census tracts that encompass the
actual exposure study area (section 3.3) which is comprised of a subset of census blocks within those same census
tracts.
3-5
-------
Fall River Study Area (MA/RI)
Central HK&
Pa%vty"ck«t
Mtlrtu
CranMW]
N River Study Area (MA/RI)
Fall River Study Aiea
Nam*
¦ Uontor (t004)
a a>»^>n
-------
Indianapolis Study Area (IN) -
Indianapolis Study Area
Nam*
uo*k»
UoraUf (00TJ)
Uo««uf (Oor»t
Bet-ror*
-------
Tulsa Study Area (OK)
249C
«•* 112 ip»
ME (EGU) ".Ml Kl
>\u«t SueonffftS)
¦i AM* Sueoo (OUN>
Broken >
Miles
———
Tulsa Study Area
Name
¦ Monitor (0175)
¦ Monitor (0235)
Monitor (1127)
A East (Refinery) - 26 tpy
A West (Refinery) - 2.596 tpy
~ Sapulpa (Glass Plant) - 212 tpy
A PSO NE (EGU) -17.941 tpy
+ Surface Met Station (RVS)
+ Upper Air Met Station (OUN)
Study Area
£ Cp«rS»»:M«c «no) ce<"f.=c!cri CCfi'-St
Figure 3-3. Location of surface and upper air meteorological stations, SOi emissions sources, and ambient monitors used to
predict ambient air quality in the Tulsa study area. Also included is source type and 2011 NEI SO2 emissions.
3-8
-------
3.2.1 General Model Inputs
3.2.1.1 Meteorological Inputs
All meteorological data used for the AERMOD dispersion model simulations were
processed with the AERMET meteorological preprocessor, version 16216 (U.S. EPA, 2016c)
using regulatory options. The National Weather Service (NWS) served as the source of input
meteorological data for AERMOD. Tables 3-2 and 3-3 list the surface and upper air NWS
stations chosen for the three study areas. The NWS hourly surface data are archived in the
Integrated Surface Hourly (ISH) database for which there is a potential concern for a high
incidence of calms and variable wind conditions. This is due to how the hourly data are reported
from the Automated Surface Observing Stations (ASOS) in use at most NWS stations. Wind
speeds less than three knots are assigned a value of zero knots, and the definition used for a
variable wind observation (wind direction that varies more than 60° in a 2-minute observation)
may include wind speeds up to 6 knots, but with a wind direction that is reported as missing. The
AERMOD model currently cannot simulate dispersion under these conditions. This issue was
addressed by reducing the number of calms and missing winds in the surface data for each of the
three NWS surface stations using separately archived 1-minute averaged wind data from the
ASOS stations. Low wind speeds and wind direction are retained in the 1-minute ASOS data.
Hourly average wind speeds and directions were calculated using the 1-minute wind data to
supplement the hourly wind data in the ISH format. The 1-minute data were processed with
AERMINUTE, version 15272 (U.S. EPA, 2015a). AERMINUTE performs quality assurance
procedures on the 1-minute data files, computes the hourly averages of wind speed and direction,
and outputs the hourly averages in a data file that can be directly input into AERMET.
3-9
-------
Table 3-2. National Weather Service surface stations for meteorological input data in
study areas.
Study Area
Station
Identifier
WMO
(WBAN)
Latitude
(degrees)
Longitude
(degrees)
Elevation
(m)
GMT Offset
(hours)
Fall River, MA
Providence
PVD
725070
(14765)
41.7225
-71.4325
19
-5
Indianapolis, IN
Indianapolis
International
Airport
IND
724380
(93819)
39.725170
-86.281680
241
-5
Tulsa, OK
Tulsa R L
Jones Jr
Airport
RVS
723564
(53908)
36.042441
-95.990166
192
-6
Table 3-3. National Weather Service upper air stations for meteorological input data in
study areas.
Study Area
Station
Identifier
WMO
(WBAN)
Latitude
(degrees)
Longitude
(degrees)
Elevation
(m)
GMT
Offset
(hours)
Fall River, MA
Chatham, MA
CHH
744940
(14684)
41.67
-69.97
12
-5
Indianapolis, IN
Lincoln, IL
ILX
745600
(04833)
40.15
-89.33
178
-6
Tulsa, OK
Norman, OK
OUN
723560
(13968)
35.23
-97.47
354
-6
3.2.1.2 Surface Characteristics and Land Use Analysis
The AERSURFACE tool, version 13016 (U.S. EPA, 2013) was used to determine surface
characteristics (e.g., albedo, Bowen ratio, and surface roughness) for input to AERMET. Surface
characteristics were calculated for the location of the ASOS meteorological towers, which were
approximated by using aerial photos and the station history from the National Centers for
Environmental Information (NCEI). AERSURFACE utilizes 1992 land cover data from the
National Land Cover Dataset (NLCD). Land cover data was obtained from the Multi-Resolution
Land Characteristics (MRLC) consortium website.9 Each of the three surface meteorological
stations are located at an airport and were specified accordingly in AERSURFACE.
Though the current version of AERSURFACE is limited to processing older land cover
data for input to AERMET, a review of historical and more recent satellite imagery indicates
there have not been substantial changes in the land cover within the area immediately
9 https://www. inrlc. gov
3-10
-------
surrounding the meteorological towers for the three modeled sites between 1992 and 2011.10 For
each of the three sites that were modeled, the surface meteorological observations were collected
at NWS stations located at airports. The meteorological towers at these airports are located in
grassy areas near or between runways. Surface roughness is derived as an inverse-distance
weighted average of the land cover within a 1.0 km radius centered on the meteorological tower.
Thus, the land cover that is nearest to the tower, where there is the least amount of change over
time, has the greatest influence on the derived roughness value. Bowen ratio and albedo, on the
other hand, are derived from a 10 km x 10 km area centered on the meteorological tower. Bowen
ratio and albedo represent an average of the land cover across the 10 km x 10 km area in which
each land cover pixel is weighted equally. Isolated areas where the land cover has changed
substantially over time have little effect on the average value of Bowen ratio and albedo within
the 10 km x 10 km area.
AERSURFACE allows for the surface roughness length to be defined by up to 12 wind
sectors with a minimum arc of 30 degrees each. For each of the three ASOS stations, roughness
was estimated for each of 12 sectors, beginning at 0 degrees through 360 degrees (i.e., 0-30, 30-
60, 60-90, etc.). The wind sectors for each of the three surface stations are illustrated in
Appendix A. The AERSURFACE default month-to-season assignments were used for Tulsa, and
reassignments were performed for both Indianapolis and Fall River. The monthly seasonal
assignments input to AERSURFACE for each of the three surface stations are shown in Table 3-
4. Surface characteristics were output by month. Note that there are two winter options: 1) winter
with no snow (or without continuous snow) on the ground the entire month and 2) winter with
continuous snow on ground the entire month.11 A month was considered to have continuous
snow cover if a snow depth of one inch or more was reported for at least 75% of the days in the
month.
Table 3-4. Monthly seasonal assignments input to AERSURFACE.
Study Area
Winter
(continuous snow)
Winter
(no snow)
Spring
Summer
Fall
Fall River, MA
-
Dec, Jan, Feb, Mar
Apr, May
Jun., Jul, Aug
Sep, Oct, Nov
Indianapolis, IN
-
Dec, Jan, Feb, Mar
Apr, May
Jun., Jul, Aug
Sep, Oct, Nov
Tulsa, OK
-
Dec, Jan, Feb
Mar, Apr, May
Jun., Jul, Aug
Sep, Oct, Nov
Seasonal definitions: Winter - Late fall after frost and harvest, or winter with no snow; Spring - Transitional spring with partial
green coverage or short annuals; Summer - Midsummer with lush vegetation; Fall - Fall with unharvested cropland
10 Google Earth was used to evaluate the land use/land cover in the area immediately surrounding the meteorological
tower, out to a distance of 1 km, and similarly in the region around the airport, out to a distance of 5 km.
11 For many of the land cover categories in the 1992 NLCD classification scheme, the designation of winter with
continuous snow on the ground would tend to increase wintertime albedo (reflectivity) and decrease wintertime
Bowen ratio (sensible to latent heat flux) and surface roughness compared to the winter with no snow or without
continuous snow designation.
3-11
-------
AERSURFACE also requires information about the climate and surface moisture at the
surface station. The station has to be categorized as either arid or non-arid. Each of the three
surface stations were categorized as non-arid in AERSURFACE. Surface moisture is based on
precipitation amounts and is categorized as either wet, average, or dry. For the three surface
stations, 2010 local climatological data from the NCEI was used to look at 30 years (1981-2010)
of monthly precipitation. The 30th and 70th percentiles of precipitation amounts were calculated
separately for each of 12 months (January through December) based on the 30-year period. The
precipitation amount for each month in 2011-2013 was then compared to the 30th and 70th
percentiles for the corresponding month. Months during which precipitation was greater than the
70th percentile were considered wet, while months that were less than the 30th percentile were
considered dry. Months within the 30th and 70th percentile range were considered average.
AERSURFACE was run for each moisture condition to obtain monthly values for wet, dry, and
average conditions. Using the AERSURFACE output for each of the three moisture categories, a
separate set of monthly surface characteristics was compiled for each of the three years for input
to AERMET. The monthly categorization of the surface moisture at each of the locations is
shown in Table 3-5. Refer to Appendix A for a complete listing of the surface characteristic
values input to AERMET for each surface station and a detailed discussion of the meteorological
data preparation.
Table 3-5. Monthly surface moisture categorizations for the three study areas.
Jan Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov Dec
Fall River, MA
2011 Avg. Wet
Dry
Wet
Avg.
Wet
Wet
Wet
Wet
Wet
Wet Avg.
2012 Avg. Dry
Dry
Avg.
Wet
Wet
Avg.
Wet
Wet
Wet
Dry Wet
2013 Dry Wet
Dry
Dry
Avg.
Wet
Avg.
Wet
Wet
Dry
Wet Wet
Indianapolis, IN
2011 Wet Wet
Wet
Wet
Wet
Wet
Dry
Dry
Wet
Wet
Wet Wet
2012 Wet Avg.
Wet
Avg.
Dry
Dry
Dry
Wet
Wet
Wet
Dry Avg.
2013 Wet Wet
Wet
Wet
Wet
Wet
Dry
Dry
Wet
Wet
Wet Wet
Tulsa, OK (Moisture conditions atRVS are based on precipitation data from Tulsa International Airport, TUL)
2011 Dry Wet
Dry
Wet
Dry
Dry
Dry
Wet
Dry
Dry
Wet Avg.
2012 Dry Avg.
Wet
Avg.
Dry
Wet
Dry
Wet
Dry
Avg.
Dry Dry
2013 Wet Wet
Dry
Avg.
Avg.
Dry
Wet
Wet
Dry
Wet
Avg. Avg.
Moisture categories were defined by comparing existing year/month precipitation values with 30-year monthly
precipitation data set: Wet (>70th percentile); Dry (<30
th percentile); Avg. (within 30th and 70th percentile)
3-12
-------
3.2.2 Stationary Sources Emissions Preparation
3.2.2.1 Emitting Sources and Locations
The modeling approach in all three study areas involved modeling key sources as point
sources and accounting for other sources through the use of additional study-area-specific
concentrations (see section 3.2.4). The facilities modeled as point sources included all those
emitting more than 100 tpy of SO2 in 2011, as well as some in Indianapolis that were somewhat
smaller (Table 3-6). These facilities were selected from version 2 of the 2011 National Emissions
Inventory (NEI)12 and paired to a representative surface meteorological station. Any stacks listed
as in the same location with identical temporal profiles and identical release parameters within a
certain tolerance (typically to the nearest integer value) were aggregated into a single stack to
simplify modeling, but all emissions were retained. For facilities with an SO2 emission total
exceeding 1,000 tpy in 2011, every stack emitting more than 1 tpy was included in the modeling
inventory.
Table 3-6. Facilities with point sources included in the air quality modeling domain for
each study area.
Study Area
Facility Name
NEI ID
Fall River, MAa
Brayton Point Energy (EGU)
5058411
Belmont Advanced Wastewater Treatment Plantb
4885211
Citizens Thermal, formerly Indianapolis Power and Light
4885311
I PL - Harding Street Generating Station
7255211
Indianapolis, IN
Rolls Royce Corporation (combustion engine manufacture and testing)b
7972011
Vertellus Specialties, formerly Reilly Industries and Reilly Tar and Chemical
(chemical manufacturing)
Quemetco (lead battery recycling facility)
7972111
8235411
Public Service Co. of Oklahoma (PSO) Northeastern Power Station
8212411
Tulsa, OKc
Sapulpa Glass Plant
Tulsa Refinery West
7320611
8402711
Tulsa Refinery East
8003911
a Contributions to ambient concentrations from another facility emitting more than 100 tpy (SEMASS Partnership municipal
waste combustor [8127611]), although 30 km away, are accounted for by the additional study-area-specific concentrations for
Fall River (see section 3.2.4).
b These sources, although having 2011 NEI emissions under 100 tons, were included based on proximity to nearby
monitoring locations and previous modeling for Indianapolis and Tulsa.
c There are facilities in the region outside of the immediate study area emitting more than 100 tpy (e.g., Oklahoma Gas &
Electric Company Muskogee Generating Station [8506011], however, they are outside the nominal distance (50 km) used for
dispersion modeling. Note also, contributions to ambient concentrations from any emission sources not explicitly modeled
and potentially influencing ambient concentrations in the study area are accounted for by the additional study-area-specific
concentrations for Tulsa (see section 3.2.4).
12 See: https://www.epa.gov/air-emissions-inventories/2011-national-emissions-inventorv-nei-technical-support-
clociiment
3-13
-------
The locations of all emitting stacks that were modeled were corrected based on GIS
analysis or by using locations identified in the local information developed by the state of
Indiana for modeling for Indianapolis and the state of Oklahoma for Tulsa.13 This was necessary
because many stacks in the NEI are assigned the same location, which often corresponds to a
location in the facility rather than the actual stack locations. NEI sources were mapped to
AERMOD sources based on matching stack parameters and temporal profiles within the same
facility. The release heights and other stack parameters were taken from the values listed in the
2011 NEI. Table B-3-1 (in Appendix B) lists all stacks in all domains.
3.2.2.2 Source Terrain Characterization
With the exception of sources at Quemetco and fugitive sources at Rolls Royce in
Indianapolis, all source elevations for the three study areas were calculated in AERMAP, version
11103 (U.S. EPA, 2016d). Source elevations at Quemetco and fugitive sources at Rolls Royce
were determined by ArcGIS overlays of the sources and National Elevation Data (NED).
3.2.2.3 Emissions Data Sources
Data for the parameterization of major facility point sources in the modeling domains
comes primarily from these sources: the 2011 NEI (U.S. EPA, 2015b),14 point source
submissions to the NEI database for the years 2012 and 2013,15 the Air Markets Program data
(CAMD database) (U.S. EPA, 2017a), and temporal emission profile information from the
EPA's 201 lv6.3 Emissions Modeling Platform (U.S. EPA, 2016e). The NEI database contains
stack locations, emissions release parameters (i.e., height, diameter, exit temperature, exit
velocity), and annual SO2 emissions. The CAMD database has information on hourly SO2
emission rates for all the electric generating units (EGUs) in the U.S. where the units are boilers
or equivalent, each of which can have multiple stacks. For sources that did not have hourly data
in the CAMD database, annual total emissions data from the NEI were converted into the hourly
temporal profiles required for AERMOD according to temporal profiles that are part of the
EPA's 201 lv6.3 emissions modeling platform.
13 As noted in section 3.2 above, local information was provided by these states in documentation developed for SIP
and PSD-related purposes.
14 We consider the 2011 NEI is the most appropriate emissions data set to use for modeling the 3 -years of air quality
in this REA because the exposure period used is based on 2011 -2013 ambient monitor data (and the associated
meteorology).
15 Annual total emissions for the largest point sources are reported to the NEI each year by the State air agencies.
Every third year (e.g., 2011, 2014), emissions for all point sources are to be reported to the NEI by the State air
agencies. Submissions to the NEI may also include any needed changes to the facility information for point
sources (e.g., locations, stack parameters, control devices), as this information is stored persistently in the NEI
database between NEI submission cycles and is updated as needed.
3-14
-------
The emissions information needed for running AERMOD was drawn from this array of
information sources (detailed information is provided in Appendix B). For EGU sources, the
more detailed information (e.g., hourly emissions values) were drawn from the CAMD database
and annual estimates from the NEI. For sources other than EGUs for which hourly SO2
emissions estimates were not available in the CAMD database, temporal profiles were used to
prepare the hourly emissions factors as described in Appendix B.
The designation of sources in the three study areas as urban or rural reflected information
about the source and surrounding area. The urban/rural designation of a source is important in
determining the boundary layer characteristics that affect the model's prediction of downwind
concentrations. It is particularly important for SO2 modeling because AERMOD invokes a 4-
hour half-life for urban SO2 sources (U.S. EPA, 2016a, section 7.2.1.1) to account for SO2
removal by conversion to sulfuric acid (catalytic and photochemical) and adsorption on to
particulate matter (Turner, 1964).16 For Fall River, a rural designation was used based on land
use data, the fact that the stacks at Brayton were tall, and the AERMOD Implementation Guide
(U.S. EPA, 2016g) recommendation to use a rural designation when modeling tall stacks in
urban areas. Classifying tall stacks with buoyant releases as urban sources in urban areas may
artificially limit plume height, thus artificially increasing modeled ground level concentrations.
The use of the AERMOD urban option for these sources may not be appropriate given that the
actual plume is likely to be transported over the urban boundary layer. For Indianapolis, all
sources were classified as urban sources based on having a broadly defined urban population of
1,000,000, consistent with the classification in the SIP modeling. For Tulsa, all sources were
classified as urban based on having a broadly defined urban population of 396,466, consistent
with the classification in the PSD modeling.
Building downwash parameters for Indianapolis and Tulsa were set based on local
information available from Indiana and Tulsa state modeling work. Given the lack of building
information available in Fall River, building downwash was not used in modeling for this study
area.
3.2.3 Air Quality Receptor Locations
Among the three study areas, the sizes of the air quality modeling domain and receptor
grid varied in consideration of differences such as number, size, and distribution of the key
emissions sources. The domains and receptor grids for Indianapolis and Tulsa drew on the
approach used by Indiana and Oklahoma in modeling these areas for their SIP and for PSD
16 For urban sources, AERMOD accounts for the urban heat island effect on increasing mixing heights for hours
under atmospheric stable conditions. Details on determining the urban or rural status of sources can be found in
U.S. EPA (2016a), U.S. EPA (2016f), and U.S. EPA (2016g).
3-15
-------
purposes. Where these domains were larger than the areas of interest for the exposure
assessments, the receptor grids were subset to receptors that encompassed the census blocks of
interest for the exposure assessment as described in section 3.3 below. The full air quality
modeling domain for Indianapolis was 38 km x 32 km with receptor spacing ranging from 2 km
at the edges, to 1 km, 500 m, 250 m, and 100 m near the emission sources, with fence line
receptors included.17 The Tulsa domain was 26 km x 29 km and receptor spacing ranged from 1
km at the edges to 666.67 m, 250 m, and 100 m near the sources, with fence line receptors also
included. For Fall River, we generated a domain (20 km x 20 km receptor grid with 500 m
spacing) that specifically fit the needs of the exposure assessment. Receptor elevations and hill
heights for all three areas were obtained from AERMAP.
3.2.4 Concentrations Associated with Sources Not Explicitly Modeled
Concentrations associated with sources of SO2 that were not explicitly modeled in all
three study areas (e.g., source emissions from outside the modeling domain in addition to
emissions from sources within the domain that were not explicitly modeled with AERMOD)
were separately estimated and added to the AERMOD modeled concentrations to produce the
hourly concentrations at each receptor. For example, for Fall River these concentrations were
approximated to account for the impacts from SEMASS Partnership given its distance (-30 km)
from the Fall River emission source of interest (Brayton EGU), rather than including SEMASS
Partnership as a point source in the AERMOD modeling run.
For all three study areas, these concentrations were calculated in terms of three-year
averages of seasonal-hour-of-day concentrations.18 This approach generally relied on the use of
ambient air monitoring data from a designated monitor (i.e., one not receiving direct impact from
emission sources modeled in the domain). Measurements from this monitor were excluded, as
recommended in the EPA air quality modeling guidance (U.S. EPA, 2016a, f), during times
when the sources that were explicitly modeled were potentially impacting monitor
concentrations and were informed by monitor siting relative to the modeled sources and wind
direction.19 For Fall River, monitor 250051004 (see Figure 3-1) was used for this purpose. Hours
17 The air quality modeling receptor grids utilized varying spatial resolution within the grids, as is customary in most
regulatory modeling applications. The exact placement of receptors usually depends on individual state modeling
guidance for dispersion modeling for regulatory applications. This accounts for the varying range of receptor
grids in the assessment for Indianapolis and Tulsa. Receptors are normally placed in locations of ambient air, i.e.
where the general public has access and along fencelines of the modeled sources. Receptors are usually spaced
close together near the modeled sources to capture concentration gradients near the sources, and they are spaced
with decreasing spatial resolution farther away from the sources.
18 This approach was implemented as recommended in the EPA's modeling guidance for SO2 (U.S. EPA 2016f).
19 Wind direction data was obtained from the surface meteorological stations representing each study area.
3-16
-------
when wind direction was from the west to north (270° to 360°) were excluded from the
calculation to remove the impacts from the source that was explicitly modeled (Brayton EGU).
For Indianapolis, monitor 180970078 (northern monitor; see Figure 3-2) was used. Hours with
wind directions between 170° and 270° were excluded to eliminate impacts from the modeled
sources in that study area. For Tulsa, monitor 401431127 (located north of the refineries, see
Figure 3-3) was used. Hours when the wind direction was either 90° to 140° or 270° to 6° were
excluded to eliminate impacts from the two refineries or the PSO Northeastern power station.
Table 3-7 shows the seasonal-hour-of-day concentrations estimated to result from source
emissions not explicitly modeled in the AERMOD runs for the three study areas.20
20 Use of this approach to estimate concentrations associated with source emissions not modeled contributes to
uncertainty in the exposure and risk estimates and is summarized in Table 6-3 below.
3-17
-------
Table 3-7. SO2 concentrations (ppb) used to account for source emissions not explicitly modeled in the three study areas,
stratified by season and hour of day.
Hour
Fall
?iver
Indianapolis
Tu
sa
Winter
Spring
Summer
Fall
Winter
Spring
Summer
Fall
Winter
Spring
Summer
Fall
1
4.07
5.47
9.07
9.43
10.93
6.73
5.03
3.67
2.27
1.27
5.50
1.20
2
5.27
8.43
6.37
7.07
10.60
6.63
3.93
4.73
2.33
0.87
2.60
1.50
3
4.77
4.70
9.13
9.13
10.37
6.40
4.20
4.33
1.83
0.40
4.30
0.93
4
7.30
5.40
7.63
12.23
8.87
6.73
4.57
3.43
1.83
0.50
0.70
1.47
5
8.03
4.80
7.40
10.37
8.83
6.87
7.63
4.70
2.03
1.37
0.60
1.70
6
6.23
4.97
8.00
11.03
11.67
5.23
3.83
5.33
1.93
0.47
8.30
1.43
7
9.30
6.83
7.83
11.27
13.40
5.37
5.20
5.40
1.57
1.03
0.80
1.47
8
8.27
6.07
7.47
8.33
10.00
6.07
5.93
6.47
2.33
3.90
1.20
2.63
9
7.17
5.80
7.30
8.20
7.77
7.50
30.7
7.10
1.93
1.23
1.33
1.50
10
8.13
5.43
7.27
9.40
13.07
9.73
25.73
15.13
2.90
2.37
0.93
1.43
11
8.57
9.30
10.50
7.47
10.20
9.07
23.27
40.57
2.80
1.87
1.53
2.63
12
8.43
7.80
18.37
8.90
12.70
8.63
17.63
37.93
5.30
2.17
2.20
2.67
13
8.77
11.83
15.90
7.50
17.63
5.93
14.83
21.83
6.13
2.30
2.40
5.23
14
9.27
8.33
16.93
7.00
13.13
5.60
9.50
11.07
2.80
2.30
3.03
2.90
15
8.00
3.30
6.40
4.00
13.13
16.33
7.40
7.97
1.80
1.67
2.00
2.20
16
6.83
2.33
6.00
3.67
7.53
4.87
9.90
12.53
3.10
1.97
2.47
2.83
17
8.93
3.60
4.33
3.03
6.97
6.30
8.53
25.10
3.30
3.60
2.13
4.17
18
5.80
2.47
3.63
2.70
11.27
11.37
9.97
16.33
4.27
3.67
5.77
4.00
19
4.43
2.30
3.27
2.87
6.77
9.10
7.47
10.83
2.87
1.47
1.50
2.20
20
4.33
2.03
3.20
2.73
9.57
4.93
10.20
7.60
2.33
2.87
1.83
2.53
21
4.07
2.30
3.13
2.67
10.57
5.87
6.13
6.57
2.57
2.67
1.33
2.00
22
3.63
2.10
2.97
2.57
12.17
4.27
10.30
5.47
2.63
1.37
0.93
2.20
23
3.70
2.60
3.07
2.60
6.13
6.13
10.73
4.00
3.67
1.03
0.67
2.30
24
4.80
2.80
6.77
7.93
5.67
6.27
6.63
3.9
3.17
1.43
2.17
1.87
3-18
-------
3.2.5 Hourly Concentrations at Air Quality Model Receptors
Once all model inputs have been created, i.e. hourly meteorology, emissions, building
parameters, etc., the AERMOD dispersion model is run to estimate hourly concentrations for
each study area. AERMOD reads the hourly meteorological data files, pairs the hourly
meteorology with the appropriate emissions and building parameters for each hour and uses
Gaussian plume theory to calculate an hourly concentration at each receptor. AERMOD then
outputs the hourly concentrations to a file that can be used in the exposure assessment.21 An
initial evaluation of the modeled concentrations based on comparison to the full distribution of
monitored concentrations can be found in Appendix D.22 Briefly, modeled concentrations were
compared to ambient air measurements using two approaches: calculated design values and
simple Q-Q (quantile-quantile) plots of the 1-hour, 3-hour, and 24-hour average concentrations.
Overall, for the three modeled areas, the modeled concentrations were comparable to the ambient
air measurements, although there were instances of over- and under-prediction of concentrations
at the upper percentiles of the concentration distribution. When evaluating on an annual basis,
model-to-monitor agreement tended to be best using the 2011 concentrations.
To augment the model-monitor evaluation of hourly SO2 concentrations in ambient air
presented in Appendix D, we performed an additional evaluation focused on air quality model
estimates during time periods with relatively greater potential for population exposures.23 The
context for this air quality modeling performance evaluation24 is particular to the intended
purpose of the air quality modeling in providing estimates of 1-hour concentrations across the
exposure modeling domain that are used with spatially limited monitoring data to estimate short-
term exposure concentrations, especially those in outdoor microenvironments. The focus is on
21 For this assessment, AERMOD output the hourly concentrations resulting from emissions from each of the largest
sources in each study area separately. These concentrations were used to develop a factor for adjusting
concentrations such that total concentrations (from all sources) just meet the current standard (section 3.4). After
adjustment, modeled concentrations were combined along with the estimated concentration contribution from
source emissions not explicitly modeled and were then used in estimating population exposures.
22 In this section, "modeled concentrations" and "unadjusted model estimates" refers to the concentrations derived
by adding the concentrations estimated to result from sources not explicitly modeled (section 3.2.4) to the
AERMOD outputs.
23 While the continuous time-series of hourly concentrations estimated by the air quality modeling is not expected to
precisely reflect that of the monitor measurements, some consistency with regard to when relatively higher
concentrations occur (e.g., daytime vs nighttime) is particularly desirable for use in exposure modeling and
provides a measure of confidence with respect to the intended use of the ambient air concentration estimates and
in estimating population exposures.
24 Given the specialized use for the air quality model predictions, we recognize the importance of performance
considerations that may differ from those common in evaluations of air quality modeling for regulatory purposes.
For example, an area of interest in our evaluation described here is consideration of the occurrence of peak
concentrations during times with greater (versus lesser) population exposure potential.
3-19
-------
outdoor microenvironments (and hence, ambient air concentrations) given the overwhelming
influence of these MEs on population exposure estimates (see section 5.2). We also recognize
that participation in outdoor events is typically influenced by seasonal and diurnal variability in
activity patterns. For example, more people spend time outdoors when the weather is
comfortable (e.g., temperate spring mornings or autumn afternoons) and during daylight hours
than when conditions are opposite (e.g., cold winter nights). Accordingly, this evaluation of the
modeled and measured hourly concentrations considers important time-of-day and time-of-year
stratifications.25
While the estimated exposures in this REA do not utilize the AERMOD predictions
without adjustment, the evaluation summarized here (on the unadjusted model estimates and
monitor measurements) is considered to provide some perspective on the uncertainty that may be
contributed to the spatial and temporal pattern of hourly concentrations estimated by AERMOD
that then feeds into the air quality adjustment step (described in section 3.4) and to the
development of the 5-minute concentrations in each study area (described in section 3.5).
Because this data evaluation was performed using the unadjusted ambient air quality,
conclusions drawn from this evaluation, while informative regarding the overall model
performance, are not directly transferrable to the hypothetical air quality scenario simulated for
the REA main body results, per se. Additionally, we were not able to develop directly
comparable modeling and monitoring datasets for our hypothetical air quality scenario (i.e., air
quality adjusted to just meet the current standard) because the adjustment approach applied to the
model estimates to create this scenario uses a proportional factor to adjust the primary source
concentration contribution at each receptor, while holding all other source concentration
contributions unadjusted (section 3.4). Accordingly, this approach cannot be applied to the
ambient air monitoring concentrations. Thus, the evaluation provided here is simply intended to
be somewhat informative, particularly with regard to considering the extent to which the
relatively higher concentration events predicted by the modeling occur in the same seasons and
portion of the day (daytime vs nighttime) as the relatively higher concentration monitor events.
We focus our evaluation here on the relatively higher monitor concentrations (i.e., the
upper part of the concentration distribution) that occur during daytime hours in the spring,
summer and fall seasons, and using the monitors having the highest design value in each study
area as indicative of events with the potential for 5-minute concentrations of greatest interest in
25 Data were stratified by two times of day (daytime - 6 AM to 8PM; nighttime - all other hours) and four seasons
(winter - December, January, February; spring - March, April, May; summer - June, July, August; and fall -
September, October, November). Additionally, as the interest of this evaluation is occurrences of relatively higher
concentrations during times of day and seasons when people are most likely to encounter them and is not
regarding annual variability, the three years of data for each location are pooled before stratification.
3-20
-------
this REA.26 The model estimates are, for the most part, similar in magnitude to the monitor
measurements (Figure 3-4). Across the study areas, the closest fit of the higher-concentration
estimates to the monitor measurements for Fall River occurs in the spring, while concentrations
for the summer and fall seasons appear to be somewhat over- and under-predicted by the model,
respectively. In Indianapolis, the highest monitor concentration events in the spring are not
reflected in the model estimates, while they may be somewhat over-predicted in the summer and
fall seasons. In Tulsa, the higher concentration events observed at the monitor are not reflected
by concentrations predicted by the model for any of the three seasons.
26 The complete set of graphs for this evaluation considering all seasons and monitors are provided in Appendix K.
3-21
-------
Indianapolis 180970057 (Spring-Day)
225
—»200
-Q
CL 175
Q.
^150
O
to
"g 100
tj 75
50 75 100 125 150 175 200 225
Observed S02 (ppb)
Tulsa 401430175 (Summer-Day)
90
«. 80
bo
L
*60
I 50
| 40
i 30
i 20
• 10
0
10 20 30 40 50 60 70 80
Observed S02 (ppb)
Fall River 250051004 (Spring-Day)
Fall River 250051004 (Summer-Day)
150
50 75 100
Observed S02 (ppb)
Indianapolis 180970057 (Summer-Day)
225
—»200
-Q
a 175
Q.
^150
O
to
"g 100
tj 75
50 75 100 125 150 175 200 225
Observed S02 (ppb)
Indianapolis 180970057 (Fall-Day)
225
90
— 80
-Q
Q-70
Q.
^60
°50
"g^O
tS 30
"S 20
10
0
90
—»80
-Q
Q-70
Q.
^60
°50
"g40
u 30
~S 20
10
0
Fall River 250051004 (Fall-Day)
-D 125
Q.
Q.
moo
25 50 75 100 125 150
Observed S02 (ppb)
-D 125
Q.
Q.
^100
25 50 75 100 125 150
Observed S02 (ppb)
9-175
"O 100
50 75 100 125 150 175 200 225
Observed S02 (ppb)
Tulsa 401430175 (Spring-Day)
10 20 30 40 50 60 70 80 90
Observed S02 (ppb)
Tulsa 401430175 (Fall-Day)
10 20 30 40 50 60 70 80 90
Observed S02 (ppb)
Figure 3-4. Comparison of AERMOD predicted SOi concentrations (y-axis) with observed air monitor SO2 concentrations (x-
axis) during daytime of the three warmer seasons at the highest design value monitor in each study area.
3-22
-------
3.3 SELECTION OF AIR QUALITY RECEPTORS FOR EXPOSURE
MODELING DOMAIN
As described above, the air quality modeling was done at a fine spatial scale that in some
locations included receptor cells as small as 100 m by 100 m. Thus, the air quality modeling
domains (Appendix C) included thousands of air quality receptor points, many more than
considered practical for use by APEX in estimating exposures. APEX simulations were
performed at a census block level, which, combined with the thousands of air quality receptors
each considering the full 5-minute time-series of concentrations, presented computational
challenges. In addition, the spatial range of the modeled air quality receptors extended outwards
beyond areas expected to be influenced by the major sources present in each study area. Thus,
the number of air quality receptors included in the exposure modeling was reduced to a more
practicable number (i.e., fewer than 2,000) while still retaining the modeled receptors having the
highest design value in the particular study area.
The approach used to define the exposure model domain within the air quality modeling
domain in each study area, along with the number of air quality receptor sites included in the
exposure modeling domain, is as follows:
• Fall River: Hourly SO2 concentrations in ambient air were estimated at receptor sites
defined by a 500 m grid. For the exposure modeling, we selected receptor sites that fell
within 10 km of the Brayton EGU (latitude (lat) 41.709989, longitude (Ion) -71.192441)
and within 10 km of the continuous 5-minute monitor (lat 41.69, Ion -71.17), which
yielded 1,494 air quality receptors covering a land area of approximately 375 km2.
• Indianapolis: Hourly SO2 concentrations in ambient air were estimated at receptors
defined by a receptor grid ranging from outside to inside at 2 km, then 1 km, 500 m, 250
m, and 100 m near the two major sources. For the exposure modeling, we selected
receptor sites that fell within 10 km of the two major sources (Citizen Thermal: lat
39.762800, Ion -86.166800; IP&L Harding: lat 39.7119, Ion -86.1975) and all receptors
within 10 km of Quemetco (lat 39.755391, Ion -86.300155) and within 10 km of
Indianapolis International Airport (lat 39.716809, Ion -86.296127).27 The finest scale grid
concentrations retained were those falling within a 500 m interval, which yielded 1,917
air quality receptors covering a land area of approximately 675 km2.
• Tulsa: Hourly SO2 concentrations in ambient air were estimated at receptors defined by a
receptor grid ranging from outside to inside at 1 km, 666.67 m, 500 m, 250 m, and 100 m
near the two major sources (West Refinery: lat 36.139140 Ion -96.025440; East Refinery:
lat 36.11705271, Ion -96.00477176). For the exposure modeling, we selected receptor
27 Emissions from the Indianapolis International Airport were not explicitly modeled to remain consistent with the
modeling performed for Indiana's SIP for the Marion County SO2 nonattainment area; however, the exposure
modeling domain was expanded using this source location to make this study area more representative of a large
urban population.
3-23
-------
sites that fell within 10 km of these two sources and receptor sites within 10 km of
monitor 401431127 (lat 36.20, Ion -95.98).28 With the exception of 24 receptors modeled
at a 100 m scale (retained in order to retain locations with the highest model-estimated
DVs), the finest scale grid concentrations retained were those falling within a 500 m
interval, which yielded 1,389 total air quality receptors covering a land area of
approximately 550 km2.
These exposure modeling domains for the three study areas are shown, with adjusted air quality
per section 3.4 below, in Figures 3-6 through 3-8.
3.4 AIR QUALITY ADJUSTMENT TO CONDITIONS MEETING THE
CURRENT STANDARD
The exposure and risk analyses were conducted for air quality adjusted to just meet the
current primary SO2 standard. Use of this adjusted air quality surface is most appropriate to
quantitatively evaluate the associated exposures and health risks in this REA (section 2.2). As
described in the REA Planning Document, a proportional approach was used to adjust ambient
concentrations (not the modeled emissions) in the 2009 REA. An analysis of ambient
concentration data at that time demonstrated that the proportional adjustment of ambient
concentrations is an appropriate approach to use to generate air quality that just meets a
particular standard (Rizzo 2009). We analyzed recent air quality data in the REA Planning
Document to evaluate this assumption for several candidate areas for the purpose of justifying
the selection of this approach for use in this REA (U.S. EPA, 2017b, Figure 4-6 and Appendix
C). The results of the air quality comparisons shown in the REA Planning Document were
similar to what was observed previously (Rizzo, 2009).
We further refined these air quality analyses here to include the monitoring data from the
three REA study areas. We also extended the time period considered to encompass the most
recent year in which ambient air monitor concentrations had a 99th percentile daily maximum 1-
hour concentration at or just below the level of the current standard (i.e., 75 ppb, the air quality
scenario adjustment goal) and the past year in each study area that had the highest daily
maximum 1-hour concentrations (i.e., evaluate a maximum range in ambient air concentrations
to reasonably support the use of potentially high adjustment factors, where needed). Thus,
evaluated data included recent ambient air monitoring data from 2015 and measurements from as
far back as 1980. We also focused the analysis on the monitor having the highest recent design
28 In addition to SO2 emission from the two refineries and the glass plant, the emissions from the PSO Northeastern
Power Station were used to estimate ambient SO2 concentrations in the Tulsa exposure study area (Table 3-6).
However, the exposure study area was not expanded to include receptors near the PSO because it is located
approximately 40 km northeast of Tulsa (see Figure 3-3).
3-24
-------
values in each of the three study areas and paired two years having the two highest design values
and with two recent years having design values at or just below 75 ppb.
Figure 3-5 presents the results of this ambient air monitoring data comparison.
Concentrations are linearly related across the wide range of concentrations and, in a few
instances, exhibit proportionality across the majority of the concentration distribution (i.e., in
addition to exhibiting linearity, the regression intercept equals zero). Based on the paired 99th
percentile daily maximum 1-hour concentrations, the potential upper range of adjustment factors
supported by these comparisons would range from 2.2 to 3.1. However, there are instances of
non-proportionality as has been described previously (2009 REA; U.S. EPA 2017b), including
limited deviation from linearity, particularly at the upper percentiles, and the presence of
statistically significant linear regression intercepts. Thus, based on these analyses, we used a
largely proportional adjustment approach in this REA with a variation from the 2009 REA
approach, as described below to account for deviations from proportionality.
The process of adjusting air quality to just meet a standard of interest begins with
consideration of the design values (DVs) calculated at the various locations in the study area.
When using a proportional adjustment approach, the highest DV is used to derive a single factor
(F) to adjust the monitored concentrations across the study area. In each study area, F is then
used to adjust all SO2 concentrations in a study area by this factor to simulate just meeting the
current standard. In the case of the SO2 standard, this adjustment of air quality is based on three
years of concentrations, which is consistent with the form for the current standard.
A variation of this approach to air quality adjustment is used in this assessment. This new
approach attempts to better consider relative source contributions to the ambient air
concentrations that may or may not change given the particular air quality scenario. For instance,
in the Fall River study area, the influence of the Brayton EGU (a source having >100 tpy SO2
emissions in the area) was accounted for by air quality modeling as a point source and the
resulting surface of modeled air concentrations was combined with the set of concentrations that
account for emission sources not modeled in the study area. In considering how to derive a
concentration surface reflecting the hypothetical scenario of air quality conditions just meeting
the current standard, we concluded that adjusting just the concentrations resulting from the EGU
emissions alone (rather than the aggregate concentrations from the EGU and the mix of
concentrations from the sources not modeled) would create a scenario that better reflected how
air concentrations would change in response to actions performed to meet air quality standards.
3-25
-------
Fall River 250051004: highest vs low 99th DM1H
100
o
l/> h_
T 1,5
5
r
/
iii.
.
¦ II.
iiii
:
0 30 60 90 120 150 180 210 240 270 300
0-100 percentile DM1H S02 (ppb):
1980 - highest year
Indianapolis 180970057: highest vs low 99th DM1H
40 80 120 160 200 240
0-100 percentile DM1H S02 (ppb):
1988 - highest year
60
50
40
30
20
10
0
0 30 60 90 120 150 180 210 240 270
0-100 percentile DM1H S02 (ppb):
1981 - 2nd highest year
Indianapolis 180970057: 2nd highest vs low 99th DM1H
1/in
140
S3
CL
3 120
o
.
w— on
.Q
Q.
a. 120
o
c/> k- 100 -
_ ro
,H >¦
on
•
y = 0.4138X - 8.5418
0-100 percentile DIV
2009 - low
NJ 4^ c
o o o o c
-U- .... 1 .... 1
0-100 percentile DIV
2011 - low
N> <£> Ot C
O O O O C
•
4
z4'
' 1
¦ ill
. 1 . •
1,11
1 . 1 1
; y = 0.393x - 11.016
•
• .
4
¦ • ¦ ¦
¦ » ' i
UuLL
¦ ''¦1
80 120 160 200 240 280
0-100 percentile DM1H S02 (ppb):
1984 - 2nd highest year
Tulsa 401430175: highest vs low 99th DM1H
Tulsa 401430175: 2nd highest vs low 99th DM1H
it
Q.
O
180
160
140
120
100
80
60
40
20
0
: y = 0.422x + 1.195"
•
' 1 1 1
1 1 1 1 1 1
¦ ' ' ' ' 1
1 1 1 1 1
30 60 90 120 150 180
0-100 percentile DM1H S02 (ppb):
1988 - highest year
180
jQ
Q.
160
^Q.
M
140
o
CO
k.
X
ro
at
120
rH
>
$
100
Q
o
0)
,
80
"¦M
in
c
a>
o
o
60
<-»
IN
40
Q.
o
o
20
rH
o
0
: y = 0.5996X + 0.2189
•
, , , i
...I
i i i . i
30 60 90 120 150 180
0-100 percentile DM1H S02 (ppb):
1992 - 2nd highest year
Figure 3-5. Comparison of ambient air measurements from high concentration years (x-
axis) to low concentration years (y-axis) in the Fall River (top row), Indianapolis
(middle row), and Tulsa (bottom row) study areas. Left column contains the
year having the highest 99th percentile daily maximum concentration. Right
column contains the year having the 2nd highest 99th percentile daily maximum
concentration.
3-26
-------
Accordingly, we applied this approach to the Fall River study area air quality, with the
concentrations contributed from the EGU adjusted just enough such that the aggregate of these
modeled concentrations and the concentrations not modeled just met the current standard at the
air quality receptor having the highest design value. This concentration adjustment approach was
also applied in a similar manner to the other two study areas, with a primary source (among the
collection of sources modeled in these areas) identified for the air quality adjustment. Then,
concentrations at air quality receptors that were contributed from all other sources were left
unadjusted. In the Indianapolis study area, the air quality receptor concentrations contributed
from each of the modeled sources were evaluated; the IP&L Harding Street Facility was
identified as the primary contributor to most of the air quality receptors having the highest
concentrations, particularly those within 10 km of the facility. A similar evaluation was done for
the Tulsa study area; the West Refinery was identified as the primary contributor to the highest
concentrations at air quality receptors in the study area.
The steps involved for this adjustment approach are summarized here. First, the
maximum DV and associated air quality receptor (rmax) was identified among the DVs from the
complete collection of modeled air quality receptors in each study area that comprise the
exposure modeling domain. Then the following formula was used to calculate the single
adjustment factor to be applied to the primary source concentrations (Ci), while considering the
concentrations associated with the other sources (C0ih) as unchanged:
In order to have air quality just meet the current standard in each study area, the study
area specific adjustment factor was used to adjust all hourly concentrations at each receptor as
follows:
Table 3-8 contains the air quality receptor design values for each study area and the
proportional adjustment factor that was applied to the concentrations that reflect the primary
source emissions in each area in order to have concentrations just meet the current standard.
Figures 3-6 to 3-8 show the air quality receptors in each study area and their respective design
values following the above described approach for adjusting the hourly concentrations to just
meet the current standard.
Ql,rmax,2011 + Ql,rmax,2012 + Ql,rmax,2013
Equation 3-1
{(75 X 3) (Coth,rmax,2011 + ^oth,rmax,2012+ ^oth,rmax,2013)}
Equation 3-2
3-27
-------
Table 3-8. Maximum SO2 design values modeled at air quality receptors and associated
proportional adjustment factors applied to primary source concentrations in
each study area.
Study Area
Modeled Air Quality
Receptor Maximum
S02 DV (ppb)
Primary Source in
Study Area
Proportional
Adjustment
Factora
Fall River, MA
101.4
Brayton EGU
1.46
Indianapolis, IN
311.3
Harding EGU
4.21
Tulsa, OK
73.5
West Refinery
0.98
a The proportional adjustment factor is based on and applied only to the primary source contributing to the highest
concentrations in the study area, while other source contributions as well as background concentrations are
assumed to remain unchanged in approximating air quality conditions to just meet the current standard.
5 km
!¦¦¦¦¦¦
!¦¦¦¦¦¦!
~¦¦¦¦¦¦¦¦¦¦¦¦
¦¦¦¦¦¦¦¦¦¦¦¦¦¦
¦¦¦¦¦¦¦¦¦SBSBSSS
v^BKEVSsiisESB
¦¦¦¦¦¦¦¦¦¦¦¦¦¦
¦¦¦¦¦¦¦Ill
¦¦¦¦
¦¦¦¦¦¦
¦¦¦¦¦¦
¦¦¦I
Design Value
(ppb)
75
sssss
!¦¦¦
!¦¦¦
Figure 3-6. Location of air quality receptors, emission sources, and ambient monitors in the
Fall River exposure modeling domain and receptor design values calculated
from modeled hourly SO2 concentrations adjusted to just meet the current
standard.
3-28
-------
5 km
¦ ¦ ¦
¦ ¦ ¦
¦ ¦ ¦
¦¦¦¦¦¦ limaii M0078 •
¦¦¦¦¦¦¦¦ ¦¦¦¦¦¦¦¦¦¦¦¦¦¦
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦SSSBSBSBBSSBS
¦¦¦¦¦¦¦¦ ¦¦¦ ¦¦¦¦¦¦¦¦¦¦¦
¦¦i
SET . ¦
Quemetco*
inaaimu ¦ i ¦¦¦¦¦¦¦¦¦__
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦a
!¦
1HH
lir-
Bgsaosal
0057
0057
!¦¦¦¦¦¦¦¦¦¦¦¦¦¦
Ulllllllll
¦¦¦¦I
!¦¦¦
¦¦¦¦¦¦¦¦¦¦a
s°
¦ ¦¦¦¦!
¦ mn
Design Value ¦¦
(ppb)
¦
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦:
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦I
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦a
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦a
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦la
;¦¦¦¦¦¦¦¦¦¦ ¦¦¦¦¦¦¦¦¦¦¦¦¦ ¦¦¦¦
¦¦¦ ¦¦¦¦¦ ¦¦¦¦¦
¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦ ¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦
IN
||
HiHBqBBHBnpMBBBBr *_
¦ ¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦¦
¦
¦
¦¦
¦¦¦
¦¦
M0073 •
¦¦¦¦
¦¦¦¦
¦
¦
¦ ¦
¦ ¦
¦ ¦
75
¦ ¦ ¦ ¦ ¦
¦ ¦
Figure 3-7. Location of air quality receptors, emission sources, and ambient monitors in the
Indianapolis exposure modeling domain and receptor design values calculated
from modeled hourly SO2 concentrations adjusted to just meet the current
standard.
3-29
-------
PSO NE (EGU)
M1127*
M1127*
M0t79j.MO175
West Refine rv» «|
• M0235
East Refinery •
M01I9.
¦¦¦¦¦¦¦¦¦¦¦
West Refinery*
• M0175
¦Ugi
• M0235
East Refinery •
Sapulpa Glass Plant
Design Value
(PPb)
0 75
Figure 3-8. Location of air quality receptors, emission sources, and ambient monitors in the Tulsa exposure modeling domain
and receptor design values calculated from modeled hourly SO2 concentrations adjusted to just meet the current
standard.
3-30
-------
3.5 FIVE-MINUTE CONCENTRATIONS
In this assessment, we combined the fine-scale temporal characteristics of continuous 5-
minute monitoring data local to each study area with the fine-scale spatial characteristics of
hourly concentrations estimated by AERMOD. First, missing values within any monitoring data
set were interpolated using the measured values immediately bounding the missing values. Then,
where continuous 5-minute data were not available, an algorithm was constructed to randomly
sample 5-minute concentrations from lognormal distributions that conform to the existing 1-hour
average and maximum 5-minute measurements. Finally, the complete year pattern of 5-minute
monitored concentrations was combined with the complete year pattern of hourly concentrations
modeled at each receptor, based on matching the rank ordered 1-hour concentration distributions.
The following section details how this was done, noting specifically where the approach differs
from that described in the REA Planning Document.
3.5.1 Preparing Monitoring Data: Assessing Completeness & Filling Missing Values
Because there are years when the ambient air monitor did not report every hourly or 5-
minute concentration and because APEX needs the complete time-series of 5-minute ambient air
concentrations to estimate exposures, an approach was developed to approximate missing 5-
minute values in the ambient air monitor data sets. As described in section 3.1 above, the study
areas and years selected for this assessment corresponded to years in which the monitor datasets
met completeness requirements for calculating a design value. This completeness requirement is
typically applied to the hourly monitor concentrations and used for regulatory purposes. To best
inform our estimation of 5-minute concentrations, we did not restrict the 5-minute concentrations
using this completeness requirement for this assessment. Our intent in this REA was to utilize as
much of the 5-minute measurements as was available in each study area.29 From ambient air
monitors in the three selected study areas, the following data sets containing 5-minute
concentrations were available:
29 For the hourly measurements, the following steps were taken: (1) a 75% completeness criterion was applied to
each day monitored, with the monitored day considered valid if it contained measurements for at least 18 of the
24 hours; (2) the number of days within a quarter of the calendar year were evaluated, also using a 75%
completeness criterion such that a monitored quarter was considered valid if there were at least 68-69 valid days
and a year was considered complete if all four quarters were valid; (3) data were screened for outliers, such that
hours in which a 5-minute max hourly value (AQS parameter 42406 and duration code 1) was reported and fell
within a factor of 1 and 12 times the AQS hourly value (parameter 42401 and duration code 1) were kept. For the
continuous 5-minute measurements, the screening for outliers was as follows: only 5-minute data with a
corresponding hourly value in AQS (parameter 42401 and duration code 1) were kept and only 5-minute values
with an hourly mean value less than 120% of the hourly value in AQS (parameter 42401 and duration code 1)
were kept.
3-31
-------
• Fall River: continuous 5-minute data were available for 2011 and 2012. For 2013, the
maximum 5-minute concentrations within the hour were available.
• Indianapolis: continuous 5-minute data were available for 2011-2013.
• Tulsa: continuous 5-minute data were available for 2011-2013.
A simple approach was selected to estimate any missing 1-hour, maximum 5-minute,
and continuous 5-minute concentrations within the ambient air monitor data sets listed in Table
3-9. We used PROC EXPAND (SAS, 2017) to interpolate between missing values, using the
measured values that bound the missing data to estimate missing concentrations via the JOIN
method (SAS, 2017). This approach fits a continuous curve to the data by connecting successive
straight-line segments. While this approach does not directly calculate an average of the
concentrations surrounding data gaps and generate a single concentration to use for all hours
within a particular gap, the degree of variability assigned to concentrations within multi-hour
gaps is limited. While more complex methods exist (e.g., autoregressive models) to perhaps
increase the representation of variability that might be occurring within multi-hour data gaps, the
performance of these simple methods is similar to complex methods when filling data sets
having few (< 5-10%) missing values (Junger and de Leon, 2015).
To support the use of this method to substitute for missing values, we evaluated
monitoring data available in the three study areas. Table 3-9 provides the number of missing
values within each 1-hour, maximum 5-minute, or continuous 5-minute across the 3-year period
and the percentage that number is of the number of values in a full dataset. There were very few
instances where the gap of missing data spanned several hours to days. The percentage of the
total dataset values that were missing was at or less than 5% in nearly all instances when
considering the Fall River and Tulsa Study areas. In contrast, the percentage of missing data for
some of the ambient air monitors in the Indianapolis study area was greater. For example,
Indianapolis monitor 18090073 had 40-60% of hours missing concentrations in both the
continuous 5-minute data set and the maximum 5-minute and 1-hour data sets, and thus was not
considered useful in subsequent assessment calculations (and was not used in further analyses).
Indianapolis monitor 18090057 also had a large percentage of missing continuous 5-minute data
for two of the years (25-58% missing), although it still had robust reporting of the maximum 5-
minute and 1-hour data (1-3% missing) for each of the three monitor years. Concentration
reporting from the Indianapolis monitor 18090078 was fairly complete considering either data
set and for all three years (4-9% missing). We recognize that Indianapolis monitor 180970057
exceeds the above recommendation of having somewhere between 5 to 10% missing data when
using the simple interpolation to estimate 5-minute concentration, however, we decided that use
of continuous 5-minute concentrations from the local monitor was better than use of a surrogate
3-32
-------
monitor from Detroit to represent variability in 5-minute concentrations, as was done in the draft
REA.
For each of the three study areas, we used the PROC EXPAND interpolation approach to
fill the missing continuous 5-minute concentrations, with the exception of the Fall River monitor
250051004 in 2013 that only reported 5-minute maximum and 1-hour concentrations. Therefore,
a second approach was employed to estimate within hour 5-minute concentration variability for
this Fall River monitor (section 3.5.2). A complete set of 1-hour and continuous 5-minute data
was first needed to apply this second approach. To estimate missing 1-hour and continuous 5-
minute data for the 2013 Fall River monitor, PROC EXPAND used their respective measured
concentrations to interpolate the missing values. Because of the dependence of 1-hour
concentrations and maximum 5-minute concentrations,30 the following steps were used for
estimating missing maximum 5-minute concentrations:
• Using PROC EXPAND, estimate the missing 1-hour concentrations for each monitor and
year;
• Calculate peak-to-mean ratios (PMRs) using the measured 1-hour and maximum 5-minute
concentrations;
• Using PROC EXPAND, estimate the missing PMR values for each monitor and year;
• Calculate missing maximum 5-minute concentrations by multiplying the complete set of
PMRs by their corresponding 1-hour concentrations.
30 PROC EXPAND could have been used to estimate the missing maximum 5-minute concentrations based on using
the measured values; however, this was not done because these simulated 5-minute values would not have been
entirely consistent with the estimation of missing hourly concentrations. This lack of consistency would lead to
PMRs that fall outside of the mathematically acceptable range (i.e., 1 < PMR < 12). For this reason, measurement
related PMRs were used for the interpolation of missing PMR (with a restriction to remain between 1 and 12) to
ultimately estimate reasonable maximum 5-minute concentrations. The minimum ratio is 1 because the highest 5-
minute concentration in an hour could never be less than the hourly mean. The maximum ratio is 12 because if
the maximum 5-minute concentration (max5) was the only measured non-zero value (i.e., all other 11 5-minute
measurements are 0), the hourly mean would be (max5 + (11 x 0))/12 or simply max5/12, thus effectively
yielding a PMR = max5/(max5/12) = 12.
3-33
-------
Table 3-9. Percent of missing values in the hourly and 5-minute ambient air monitoring
data sets for the three study areas (2011-2013).
Continuous 5-minute data
1-hour and 5-minute
maximum data
Study Area
Monitor ID
Year
% Missing
Days/Year <
75% complete
% Missing
Days/Year <
75% complete
2011
3.5
4
-
-
Fall River
250051004
2012
2.9
2
-
-
2013
-
-
4.7
7
2011
58.0
33
1.2
2
18090057
2012
25.5
10
2.1
5
2013
11.6
25
2.8
9
2011
9.2
32
8.0
31
Indianapolis
18090078
2012
4.5
10
4.3
9
2013
8.2
26
7.4
22
2011
42.6
208
41.4
202
18090073
2012
52.6
281
51.4
272
2013
64.6
311
63.8
304
2011
1.2
2
-
-
401430175
2012
2013
1.1
2.6
3
9
-
-
2011
-
-
-
-
401430179
2012
-
-
-
-
Tulsa
2013
3.2
12
-
-
2011
2.7
10
-
-
401430235
2012
2013
CO CO
CO ^
12
4
-
-
2011
1.3
5
-
-
401431127
2012
7.3
31
-
-
2013
2.3
7
-
-
The symbol
indicates there were no data needed for this evaluation because there were adequate continuous
5-minute data available or there were no data available.
3.5.2 Estimating Continuous 5-minute Concentrations at Monitor Having Only 1-hour
Average and Hourly Maximum 5-minute Data
In this assessment, we are interested in estimating 5-minute exposures using the complete
time-series of 5-minute ambient air concentrations for each year. We are also interested in
utilizing, to the maximum extent possible, the local ambient air measurements to inform this
estimation. As described above, there were no 5-minute continuous measurements available for
one year (2013) for the Fall River study area. Based on the ambient air monitoring data that were
available (i.e., 1-hour average and maximum 5-minute concentrations within each hour) and
knowing that air pollutant concentrations are typically lognormally distributed (Kahn, 1973), an
approach was developed to estimate the eleven other 5-minute concentrations occurring within
each hour in the year for which continuous 5-minute measurements were not available. While
3-34
-------
early studies (e.g., Larsen, 1977) have developed models to estimate a few of the upper
percentiles of a concentration distribution using relationships between peak concentrations and
time-averaging (e.g., estimate a 2nd highest 1-hour from the 2nd highest 8-hour), they are not
considered directly applicable to estimating a complete time-series of continuous 5-minute
concentrations in a year (i.e., 105,120 values). We also note that there are maximum 5-minute
monitored concentrations associated with instances where the hourly concentrations are reported,
which already provides appropriate values for important peak 5-minute concentrations. Because
the Fall River study area had continuous 5-minute data available for two of the years of interest,
while also needing an approach to estimate continuous 5-minute concentrations for 2013, the
2011-2012 Fall River continuous 5-minute data served as a case study for developing and
evaluating this approach.
We first evaluated the 5-minute data set to confirm lognormal distributions would be
appropriate to fit the twelve measured 5-minute values in each hour and to determine the
parameters associated with that distribution. Using the set of continuous 5-minute monitor data
in Fall River (2011-2012), where all twelve31 5-minute measurements within an hour were
available, data were categorized by their 1-hour average concentrations and their peak to mean
ratios (i.e., PMRs, the maximum 5-minute concentration divided by the 1-hour average). This
categorization was done because the 2009 REA analyses indicated a relationship between the
magnitude of hourly SO2 concentrations and the magnitude of the PMRs, consistent with
conclusions made regarding this relationship (Singer, 1961). For the hourly concentrations, bins
of 10 ppb increments were used to categorize hourly concentrations upwards from 0 through 80
ppb, with a final bin containing all concentrations above 80 ppb (yielding a total of 9 hourly
concentration bins). PMR was categorized by 0.5 increments from 1 to 2, then in whole units
from 2 to 4, ending with a final PMR bin of > 4 (yielding a total of 5 PMR bins).
Then, we used PROC CAPABILITY (SAS, 2017) to evaluate the fit of eight statistical
distribution forms32 for both the varying hourly concentration and PMR binned continuous 5-
minute data. Distribution fits were evaluated using four goodness-of-fit statistics: Kolmogorov
Smirnov, Cramer von Mises, Anderson Darling, and Chi-Square (SAS, 2017). Best fit
distributions were selected based on having the lowest p-value (or highest critical value) in the
collection of fit statistics. For the low 1-hour concentration binned data (e.g., 0 to <10 ppb, 10 to
<20ppb), normal distributions were found to have the best statistical fit, while for higher 1-hour
concentration binned data, lognormal distributions had the best statistical fit (along with a few
31 One hour has 12 five-minute periods (60/5=12), thus there are a total of twelve 5-minute concentrations possible
within an hour.
32 Distributions evaluated were normal, lognormal, Weibull, gamma, Pareto, exponential, beta, and Rayleigh.
3-35
-------
having gamma and Weibull distributions as the most reasonable fit). This was not entirely
unexpected given that some of the distribution types could not be fit to the binned data set (e.g.,
the number of samples in some of the bins was too small, the prevalence of concentration values
of 0). Overall, the results indicate the within-hour 5-minute concentrations are generally
consistent with a lognormal distribution, particularly considering high concentrations of interest,
and that a lognormal distribution can be used to reasonably approximate the missing eleven
within-hour 5-minute concentrations.
To do so, the parameters of all the fitted normal distributions were transformed to
lognormal terms (geometric means and standard deviations) (Casella and Berger, 2002) and
combined with the suite of parameters estimated for all of the fitted lognormal distributions.
Series of twelve 5-minute concentrations were randomly sampled from these distributions for
thousands of iterations, creating a new data set consisting of a distribution of thousands of
datasets of twelve 5-minute concentrations, each lognormally distributed and having their own
hourly average concentration and PMR. Individual sets of twelve 5-minute concentrations were
then divided by their respective 1-hour average concentrations to create sets of normalized 5-
minute concentrations (estimated concentrations), and then categorized by their PMR in 0.1
increments. For method validation, a test data set was created from the 2011-2012 Fall River
monitor data, using only the observed 1-hour average and maximum 5-minute concentrations.
From the data set of estimated concentrations, a set of twelve mean normalized33 5-minute
concentrations were then randomly assigned to each 1-hour/maximum 5-minute concentration in
the test data set and were linked using the same categorization of PMR in 0.1 increments.
Finally, the within-hour continuous 5-minute concentrations were calculated for each hour by
multiplying the observed 1-hour average by the normalized twelve 5-minute concentrations.34
The complete set of estimated 1-hour mean, 5-minute maximum, and continuous 5-
minute concentrations were compared with the respective metric in the monitoring dataset.
Figure 3-9 illustrates the relationship, indicating excellent reproducibility of the original 1-hour
(top panels) and maximum 5-minute concentrations (middle panels) and reasonable agreement
between the estimated and measured 5-minute continuous concentrations (bottom panels). Table
3-10 provides summary statistics for comparison to further support the relationship.
33 All twelve 5-minute concentrations occurring within an hour were divided by that hourly 1-hour average
concentration.
34 Where needed, a small downward or upward adjustment was applied to the suite of 5-minute concentrations to
ensure the modeled values had a 1-hour average and maximum 5-minute concentration consistent with the
monitoring measurements. The approach was designed to precisely replicate the 1 -hour average and its associated
variability of all 12 within hour 5-minute concentrations, thus there are a few instances where the estimated and
measured 5-minute maximum deviated slightly from one another (Figure 3-9, middle panel).
3-36
-------
100
100
20 30 40 50 60 70 80 90 100
Measured Mean lhr Max S02 (ppb)
50 75 100 125 150 175 200 225
Measured 5-minute Max S02 (ppb)
50 75 100 125 150 175 200 225
Measured 5-minute S02 (ppb)
10 20 30 40 50 60 70
Measured Mean lhr S02 (ppb)
80 90 100
50 75 100 125 150 175
Measured 5-minute Max S02 (ppb)
50 75 100 125 150 175
Measured 5-minute S02 (ppb)
Figure 3-9. Comparison of estimated to measured SO2 concentrations in ambient air in Fall
River monitor 250051004: 1-hour average (top panels), maximum 5-minute
(middle panels) and continuous 5-minute (bottom panels) for 2011 (left panels)
and 2012 (right panels).
3-37
-------
Table 3-10. Descriptive statistics and correlations associated with measured and estimated
1-hour average, maximum 5-minute, and continuous 5-minute SO2
concentrations, Fall River (monitor 250051004), 2011-2012.
Variable
Year
Data Set
N
SO2 Concentrations (ppb)
Mean Std Dev Minimum
Maximum
Correlation (r)
2011
Estimated
7728
3.01
5.97
0.09
93.4
1.00000
1-hour
Measured
7728
3.01
5.97
0.09
93.4
average
2012
Estimated
8404
2.43
4.27
0.11
86.3
1.00000
Measured
8404
2.43
4.27
0.11
86.3
2011
Estimated
7728
5.62
14.11
0.2
205.7
0.99999
Maximum
Measured
7728
5.59
14.11
0.2
205.7
5- minute
2012
Estimated
8404
4.04
9.71
0.2
155.5
0.99996
Measured
8404
4.01
9.70
0.2
155.5
2011
Estimated
92736
3.01
7.03
0.02
205.7
0.97516
Continuous
Measured
92736
3.01
7.24
0
205.7
5-minute
2012
Estimated
100848
2.43
4.94
0.03
155.5
0.97922
Measured
100848
2.43
5.11
0
155.5
3.5.3 Estimating 5-minute Concentrations Across Study Areas
In the following sections we discuss the approach used to estimate 5-minute
concentrations across the exposure modeling domain encompassing each study area (section
3.5.3.1), summarize the estimated 5-minute concentrations in relation to available ambient air 5-
minute measurements (section 3.5.3.2), and include a comparison of estimated concentrations
with ambient air monitor measurements considering the occurrence of concentrations at or above
concentrations of interest during times of greater exposure potential (section 3.5.3.3).
3.5.3.1 Combining 5-minute Monitor Measurements with 1-hour AERMOD Receptor
Estimates
The complete temporal profile of each of the three years of continuous 5-minute monitor
data developed using the above approach(es) was used to approximate the within-hour variation
in 5-minute concentrations at each AERMOD air quality receptor site in each study area. The
approach used in this REA to combine the monitor data with the modeled hourly estimates is a
slight variation of that described in the REA Planning Document.35 We have adjusted the
proposed approach in the REA Planning Document to better reflect instances where the ambient
35 For the REA Planning Document, we originally proposed to match by consecutive hour, i.e., using the complete
calendar years of hourly concentrations for both the ambient monitor and each air quality receptor. Then, each
within-hour distribution of twelve 5-minute concentrations from the monitor would be adjusted using a
multiplicative factor derived from the ratio of the 1-hour average concentrations (i.e., modeled divided by
measured) (see REA Planning Document, Equation 4-4).
3-38
-------
air monitor may capture a high concentration event that may not necessarily occur at the same
clock time at a modeled air quality receptor that is located at a distance from the monitor. Events
such as these would result from varying lateral or vertical transport of pollutant plumes that may
not necessarily be captured by the air quality modeling,36 affecting both the temporal and spatial
characteristics of the air quality surface.
Considering this, the calendar-based approach (described in the REA Planning
Document) could result in a mismatching of times when peak concentration occurs across the
spatial domain and thus lead to potentially erroneous distributions of 5-minute concentrations.
For this REA, we linked the high concentration events occurring in both the monitor data set and
the modeled hourly estimates at air quality receptors by ranking their respective 1-hour
concentration distributions. Thus, all low 1-hour concentrations at each modeled air quality
receptor will be linked to the distribution of 5-minute concentrations that occur during low 1-
hour concentrations measured at the monitor, and, in a similar fashion, all high hourly
concentration events will be appropriately linked, irrespective of clock hour. A similar equation
to that provided in the REA Planning Document that replicates the pattern of the monitored 5-
minute values in an hour by scaling the 5-minute values so their hourly averages are equal to the
AERMOD predictions for that hour (Equation 3-3) is described here:
Ys,r,i = ±Ji2y Xr,i Equation 3-3
12Li=lxr,i
where,
Xr,i = the ith 5-minute value (ppb) at the monitor, having 1-hour ranked concentration r
Ys,r = the 1-hour AERMOD value (ppb) at location s, having 1-hr ranked concentration r
Ys,r,i = the ith 5-minute value (ppb) having 1-hr ranked concentration r, at location .s
s = AERMOD prediction point in space
r = rank ordered 1-hour concentration, r = 1,2, ..., 8760 (or 8784 for leap years)
i = sequence of 5-minute values within the hour, i = 1, 2,..., 12.
Thus, the complete year distribution of continuous 5-minute concentrations was applied
to the modeled receptors using the complete time-series of hourly scaling factors (unique to each
receptor) to yield the time-series of 5-minute SO2 concentrations (e.g., n= 12x24x365 = 105,120
36 There is variation in the emissions and meteorological data input to the model relative to the actual emissions and
meteorology. For example, it is possible that, given the limited number of meteorological stations and their
geographic locations relative to the hundreds of receptors modeled across a 200 km2 study area, the actual local
fine scale weather patterns will not all coincide in time and space.
3-39
-------
values) at every air quality receptor in each study area. Effectively, all spatial gradients that may
exist for each hour across the study area are maintained; the 5-minute monitoring data only add a
finer scale to the within-hour temporal variability. Because the ranked concentration distributions
for each modeled air quality receptor may have a differing order of actual clock hours, it is likely
that the within-hour 5-minute concentration variability (and hence maximum 5-minute
concentrations) differs across the air quality receptors when considering the same clock hour.
This is considered a reasonable and realistic outcome of using this approach.
For instances where a study area has more than one ambient air monitor (i.e.,
Indianapolis and Tulsa), modeled receptors were linked with 5-minute concentration data from
the nearest monitor. Again, all spatial gradients that may exist within each hour across the study
area are maintained and it is likely that there is differing within-hour 5-minute concentration
variability and occurrence of maximum 5-minute concentrations across the air quality receptors
when considering the same clock hour. The assignment of monitor to modeled air quality
receptors is as follows:
• Fall River: all air quality receptors were linked to 5-minute concentrations from the
single ambient air monitor in the study area (250051004).
• Indianapolis: monitor 180970057 is located between the two largest sources (Harding
and Citizens Thermal) and is considered to best represent local source related 5-minute
concentration variability. The 5-minute concentrations from this monitor were linked to
air quality receptors within 10 km of Harding and 5 km within Citizens Thermal, i.e.,
those receptors potentially having a strong local source influence. All other receptors
used monitor 180970078 to represent air quality receptors not having a strong local
source influence on 5-minute concentrations. Monitor 180970073 is considered outside
of the exposure modeling domain and had a large percent of missing data, thus these data
were not used at this time.
• Tulsa: monitor 401430175 is closest to the West Refinery and monitor 401430235 is
closest to the East Refinery. These monitors are considered to best represent local source
related 5-minute concentration variability. Based on the spatial pattern of DVs,
concentrations from monitor 401430175 were linked to air quality receptors within 10 km
of the West Refinery and concentrations from monitor 401430235 were linked to
receptors within 5 km of the East Refinery. All other receptors used monitor 401431127
to represent air quality receptors not having a strong local source influence on 5-minute
concentrations. Monitor 401430179 is proximal to monitor 401430175, although further
from the West Refinery. This monitor only has data for 2013 and was not used to
estimate 5-minute concentrations at this time.
3.5.3.2 Summary of Estimated 5-minute Concentrations Across Study Areas
After estimating the continuous 5-minute concentrations at each air quality receptor
location, the distributions of these 5-minute concentrations were compared to those of the 5-
minute ambient air measurements in each study area. To do so for this comparison, the ambient
3-40
-------
air monitor concentrations in each study area were first adjusted proportionally using the single
factor derived from the maximum monitor design value to reflect conditions that would just meet
the current standard. As such, the adjusted ambient air concentrations from the monitor having
the highest design value would hypothetically represent a distribution of the highest
concentrations in a study area among the monitored data set.37
We summarized the monitor continuous 5-minute concentrations by identifying the 90th
and 99th percentiles of the distribution and selecting the maximum 5-minute concentration. The
estimated continuous 5-minute concentrations at the air quality receptor sites were also
summarized by considering the upper percentiles of the distribution. The 90th and 99th percentiles
of the distribution, along with the maximum 5-minute concentration, were identified at each
modeled receptor location. Because there were over a thousand air quality receptors within each
study area, we consolidated each of these statistics to a new set of statistics. We still focused on
the 90th and 99th percentiles of the distribution and the maximum 5-minute concentration,
however now we considered the distribution of each of these upper percentile concentrations
across the entire set of air quality receptors. For example when considering the maximum 5-
minute concentrations, the maximum of all the maximum 5-minute concentrations (i.e., the single
highest air quality receptor concentration considering the entire study area), the 99th percentile of
all maximum 5-minute concentrations (i.e., 1% of the complete set of modeled receptors have a
maximum 5-minute concentration greater than this value), and the 90th percentile of all maximum
5-minute concentrations (10% of the complete set of modeled receptors have a maximum 5-
minute concentration greater than this value) would be presented. This summary sequence would
then follow for the other two statistics (the upper percentile distribution of all 90th and 99th
percentile 5-minute concentrations from the collection of receptors) generated from the
collection of air quality receptors, which are provided in Tables 3-11 through 3-13.
There is reasonable agreement at the upper percentiles between the adjusted monitored
concentrations and the estimates developed for the receptor sites, particularly considering the
99th percentile and maximum values in the Fall River and Tulsa study areas (Table 3-11 and 3-
13). For example, the range in particular percentile concentrations (e.g., the 90th, 99th, and
maximum of the estimated maximum percentile 5-minute concentrations across all receptors)
estimated for the model receptor locations bound the measured 5-minute concentrations quite
well (e.g., maximum 5-minute concentrations for 2011 and 2012 in the Fall River study area). In
some instances, the range of upper percentile concentrations for the model receptor sites extends
above the monitor upper percentile concentrations (e.g., the 99th percentile concentrations in Fall
37 Therefore, the maximum hourly design value for both the ambient monitor and modeled receptor would be 75
ppb, making the two sets of data more compatible.
3-41
-------
River for 2012 and 2013). In other cases, the range of the receptor upper percentile
concentrations is below the monitor upper percentile concentrations (e.g., the maximum 5-
minute concentrations in Fall River for 2013).
For most percentiles of the concentration distribution for the Indianapolis study area, the
receptor concentrations are greater than the monitor-based concentrations. This may be related to
the approach for estimating concentrations associated with source emissions not explicitly
modeled (Table 3-7).38 Given that the range of maximum 5-minute concentrations estimated at
the receptor locations (e.g., 452-642 ppb for the first year) extends above that of the monitor
(355 ppb), it is possible that, even when adjusted for just meeting the current standard, these
maximum 5-minute concentrations at modeled receptor sites appear somewhat high. However,
we also note that there are situations where the estimated maximum 5-minute concentrations at
receptor sites were well below that of the monitor (e.g., receptor concentrations peaked between
167-205 ppb compared to monitor concentration of 369 ppb for year two of the simulation). It
may also be that the numerous receptors situated in close proximity to the largest emissions
sources in the area are representing hourly (and hence 5-minute) variability not reflected by the
monitors. In the absence of having monitors at all the receptor sites to confirm this, the upper
range of predicted concentrations across each of the study areas remain as an important
uncertainty.
38 In this same analysis performed for the draft REA (which relied on a different approach for estimating
concentrations associated with source emissions not explicitly modeled), concentrations estimated at the lesser
primary source influenced receptors were less than that observed for monitor 180970078. While the draft REA
used a surrogate monitor to approximate 5-minute variability in Indianapolis and this REA used the continuous 5-
minute data from study area monitors, this change in approach does not appear to be a significant contributor to
the differences between the concentration distributions (data evaluation not shown).
3-42
-------
Table 3-11. Descriptive statistics for concentrations at monitors and concentrations
estimated at air quality receptor locations, Fall River study area 2011-2013.
Unadjusted or
Adjusted
Type of Statistic
2011
2012
2013a
Values
Monitor (250051004) 5-minute S02 Concentrations (ppb)
p90
4
3
4
unadjusted
p99
31
21
12
max
206
156
206
p90
5
4
4
adjusted b
p99
37
25
14
max
241
182
241
Estimated 5-minute SO2 Concentrations (ppb) at Air Quality Receptors
p90p90
11
10
11
p99p90
11
10
11
maxp90
11
10
11
p90p99
32
27
22
adjustedc
p99p99
41
31
24
maxp99
48
35
26
p90max
183
129
121
p99max
247
187
150
maxmax
268
214
180
a For 2013, only the maximum 5-minute measurement concentrations were available in Fall River, even
though this evaluation includes estimated continuous 5-minute concentrations for monitor 250051004.
b Adjusted concentrations were based on a monitor-based design value (adjustment factor =64/75 = 0.85).
c Adjusted concentrations were based on highest modeled air quality receptor and the primary source
contribution to concentrations at that receptor (see section 3.4).
Abbreviations: pN=
\lth percentile of 5-minute concentrations at monitor; pNpN = Nth percentile of the
distribution of all study area receptor Nth percentile 5-minute concentrations. For example, p90 = 90th
percentile of 5-minute concentrations at monitor and p90p99 = 90th percentile of the distribution of all study
area receptor 99th percentile 5-minute concentrations.
3-43
-------
Table 3-12. Descriptive statistics for concentrations at monitors and concentrations
estimated at model receptor locations, Indianapolis study area 2011-2013.
Unadjusted
or Adjusted
Values
Type of
Statistic
2011
2012
2013
2011
2012
2013
Monitor 5-minute SO2 Concentrations (ppb)a
Local Primary Source Influence
Less Primary Source Influence
(monitor 180970057)
(monitor 180970078)
p90
4
5
5
5
6
5
unadjusted
p99
22
38
34
29
31
37
max
370
384
255
99
108
107
p90
3
4
5
5
6
5
adjusted b
p99
21
37
33
29
30
36
max
355
369
245
99
104
103
Estimated 5-minute SO2 Concentrations (ppb) at Model Receptors
Local Primary Source Influence
Less Primary Source Influence
p90p90
21
19
20
18
18
18
p99p90
23
21
22
20
19
19
maxp90
25
23
23
20
19
20
p90p99
52
54
53
44
45
44
adjustedc
p99p99
61
62
61
45
46
45
maxp99
68
66
67
45
47
45
p90max
452
167
239
132
157
145
p99max
534
188
286
132
157
145
maxmax
642
205
343
135
157
145
a For all years monitored, continuous 5-minute measurement concentrations were available.
b Adjusted concentrations were based on a monitor-based design value (adjustment factor =78/75 = 1.04).
c Adjusted concentrations were based on highest modeled air quality receptor and the primary source contribution to
concentrations at that receptor (see section 3.4).
Abbreviations: p90 = 90th percentile of 5-minute concentrations at monitor. p90p90 = 90th percentile of the distribution of all
study area receptor 90th percentile 5-minute concentrations.
3-44
-------
Table 3-13. Descriptive statistics for concentrations at monitors and concentrations
estimated at model receptor locations, Tulsa study area 2011-2013.
Adjusted or
Unadjusted
Values
statistic
2011
2012
2013
2011
2012
2013
2011
2012
2013
Monitor 5-minute SO2 Concentrations (ppb)a
Local Primary 5
Influence (4014
Source
30175)
Local Primary Source
Influence (401430235)
Less Primary S
Influence (4014
Source
31127)
unadjusted
p90
15
11
7
2
1
1
2
2
1
p99
50
42
33
17
5
7
8
5
4
max
154
152
123
114
77
50
67
33
84
adjusted b
p90
20
15
10
3
1
1
2
2
2
p99
68
57
45
23
7
10
11
7
5
max
210
207
168
155
105
68
92
46
114
Estimated 5-minute SO2 Concentrations (ppb) at Model Receptor Locations
Local Primary Source
Influence
Local Primary Source
Influence
Local Primary Source
Influence
adjustedc
p90p90
10
10
8
7
7
6
5
5
5
p99p90
29
24
14
12
10
7
6
6
5
maxp90
41
37
17
13
11
8
6
6
5
p90p99
41
34
23
35
28
22
16
13
9
p99p99
95
84
40
48
34
24
20
16
10
maxp99
118
108
49
53
39
26
24
18
11
p90max
126
116
64
170
207
96
99
59
57
p99max
239
238
118
199
270
109
127
73
65
maxmax
297
345
157
221
311
116
163
96
75
a For all years monitored, continuous 5-minute measurement concentrations were available.
b Adjusted concentrations were based on a monitor-based design value (adjustment factor =55/75 = 0.73).
c Adjusted concentrations were based on highest modeled air quality receptor and the primary source contribution to
concentrations at that receptor (see section 3.4).
Abbreviations: p90 = 90th percentile of 5-minute concentrations at monitor. p90p90 = 90th percentile of the distribution of all
study area receptor 90th percentile 5-minute concentrations
3.5.3.3 Estimated Peak 5-minute Concentrations at Air Quality Receptor Sites During
Times of Greater Exposure Potential
Similar to the evaluation conducted on the hourly concentrations (section 3.2.5 and
Appendix K), we were interested in understanding how well the estimated 5-minute
concentrations corresponded with the available ambient measurements, while focusing on times
most likely associated with population exposure and considering all modeled receptors and the
estimated 5-minute concentrations used as input to the exposure model. Accordingly, we
3-45
-------
stratified both the estimated and measurement data sets by time-of-day and season and have
focused on the daytime hours during the three warmer seasons.39
This analysis used the set of estimated 5-minute concentrations for all air quality
receptors in each study area and compared these with the available monitor data, with both sets
based on an adjustment intended to reflect the hourly concentrations just meeting the current
standard. The 5-minute estimated concentrations for the air quality receptor sites are derived
from the hourly concentrations estimated for the current standard scenario (adjustment for hourly
concentrations described in section 3.4). As that same approach (which is based on adjusting
concentrations associated with emissions from the primary source) could not be applied for the
monitor concentrations, monitor concentrations were adjusted by a different approach.
Concentrations at the highest monitor in each study area were adjusted such that the 3-year DV
just equaled the level of the standard (75 ppb). For areas with more than one monitor, the
concentrations at all monitors were adjusted by the same factor (based on the DV monitor). This
difference in deriving somewhat conceptually comparable datasets affects our ability to precisely
judge the implications of these comparisons with regard to potential bias in the receptor
estimates and limits our conclusions accordingly.
Calculated for each data set were instances where 5-minute concentrations were at or
above 100, 200, 300, and 400 ppb, at each individual air quality receptor and for each year.
These counts developed for each air quality receptor location were then binned using the number
of days per year, i.e., a receptor had at least 1 day, 2 or more days, 5 or more days, and 10 or
more days at or above a selected level. Then the number of air quality receptor locations in each
bin was summed, indicating how many air quality receptor locations in a study area had
estimated concentrations at or above the levels of interest. Then we calculated the percentages
these numbers were of the total number of receptor sites in each study area. Similar counts and
percentages were also calculated for the monitor data. Results generated for each of the three
study areas are provided in Table 3-14 to Table 3-16.
In general, there is consistency between the estimated and measured concentrations
regarding the number of days per year that concentrations are at or above 5-minute
concentrations of interest considering the years and seasons simulated. For example, in Fall
River, 6 of the 9 season/years had at least one day with a concentration at or above 100 ppb at
the ambient air monitor, while 5 of the 9 season/years were above the same level for more than
40% of air quality receptors in the study area (Table 3-14). The occurrence of these upper
percentile concentrations seems more frequent than would be observed when considering the
39 Data were stratified by two times of day (daytime and nighttime) and four seasons (winter, spring, summer and
fall) as described in section 3.2.5.
3-46
-------
ambient air monitor alone, particularly for the Indianapolis study area (Table 3-15). Again, this
could be a function of the model representing variability in ambient concentrations not observed
at the ambient air monitors due to the siting of many modeled receptors in close proximity to the
important emission sources in each study area. Comparisons of the estimated and monitor 5-
minute concentrations in the Tulsa study area (Table 3-16) were similar to that observed for the
other two study areas, although they differed by having a much smaller percent of receptors at or
above the concentrations of interest.
Table 3-14. Percent of air quality receptors and monitors at which 5-minute SO2
concentrations (for conditions just meeting standard) exceed concentrations of
interest on single and multiple days, Fall River study area 2011-2013.
5-minute
Concentrations
of Interesta
Percent of Receptors Exceeding
Concentration of Interest on
Specified Number of Days in Yearc
Number of Days
Percent of Monitors Exceeding
Concentration of Interest on
Specified Number of Days in Yearc
Number of Days
Season b
Year
>1
>2
>5
>10
>1
>2
>5
>10
2011
41.7
7.0
0
0
100
100
0
0
Fall
2012
1.6
0.1
0
0
100
100
0
0
2013
2.3
0.1
0
0
0
0
0
0
2011
86.3
60.0
14.1
1.3
100
100
0
0
100
Spring
2012
0.9
0
0
0
100
100
0
0
2013
3.9
0.7
0
0
100
0
0
0
2011
100
99.9
71.2
12.2
100
100
100
0
Summer
2012
100
100
26.2
4.8
0
0
0
0
2013
100
100
7.0
0.1
0
0
0
0
2011
0
0
0
0
0
0
0
0
Fall
2012
0
0
0
0
0
0
0
0
2013
0
0
0
0
0
0
0
0
2011
1.3
0.3
0
0
100
0
0
0
200
Spring
2012
0
0
0
0
0
0
0
0
2013
0
0
0
0
0
0
0
0
2011
1.1
0.8
0
0
100
100
0
0
Summer
2012
0.1
0
0
0
0
0
0
0
2013
0
0
0
0
0
0
0
0
a There were no estimated or measured concentrations at or above 300 ppb.
b Daytime hours (6 AM to 8 PM) only.
c There were 1,494 receptors modeled and 1 monitor.
3-47
-------
Table 3-15. Percent of air quality receptors and monitors at which 5-minute SO2
concentrations (for conditions just meeting standard) exceed concentrations of
interest on single and multiple days, Indianapolis study area 2011-2013.
5-minute
Percent of Receptors Exceeding
Concentration of Interest on
Specified Number of Days in Yearb
Percent of Monitors Exceeding
Concentration of Interest on
Specified Number of Days in Yearb
Concentrations
of Interest
Number of Days
Number of Days
Season a
Year
>1
>2
>5
>10
>1
>2
>5
>10
2011
100
99.7
60.6
60.5
0
0
0
0
Fall
2012
100
100
63.5
60.5
100
67
33
0
2013
100
100
60.5
60.5
67
33
33
0
2011
60.4
57.3
25.4
2.0
33
33
0
0
100
Spring
2012
54.6
38.4
5.2
0.7
67
33
0
0
2013
36.9
16.0
3.0
0.5
33
33
0
0
2011
100
100
94.3
60.2
33
33
0
0
Summer
2012
100
100
62.1
60.5
67
33
0
0
2013
61.0
60.5
60.5
60.5
67
33
0
0
2011
59.5
57.1
0
0
0
0
0
0
Fall
2012
0
0
0
0
33
0
0
0
2013
60.4
58.2
0
0
0
0
0
0
2011
1.3
0.1
0
0
0
0
0
0
200
Spring
2012
0
0
0
0
33
0
0
0
2013
0.9
0
0
0
0
0
0
0
2011
5.4
1.6
0
0
0
0
0
0
Summer
2012
0.3
0
0
0
33
0
0
0
2013
25.2
0.6
0
0
33
0
0
0
2011
57.3
0.1
0
0
0
0
0
0
Fall
2012
0.1
0
0
0
0
0
0
0
2013
0.2
0
0
0
0
0
0
0
2011
0.5
0.1
0
0
0
0
0
0
300
Spring
2012
0.2
0
0
0
33
0
0
0
2013
0.1
0
0
0
0
0
0
0
2011
1.9
0.1
0
0
0
0
0
0
Summer
2012
0.1
0.1
0
0
33
0
0
0
2013
0
0
0
0
0
0
0
0
2011
2.8
0
0
0
0
0
0
0
Fall
2012
0
0
0
0
0
0
0
0
2013
0
0
0
0
0
0
0
0
2011
0.5
0
0
0
0
0
0
0
400
Spring
2012
0
0
0
0
0
0
0
0
2013
0
0
0
0
0
0
0
0
2011
0.9
0
0
0
0
0
0
0
Summer
2012
0
0
0
0
0
0
0
0
2013
0
0
0
0
0
0
0
0
a Daytime hours (6 AM to 8 PM) only.
b There were 1,917 receptors modeled and 3 monitors.
3-48
-------
Table 3-16. Percent of air quality receptors and monitors at which 5-minute SO2
concentrations (for conditions just meeting standard) exceed concentrations of
interest on single and multiple days, Tulsa study area 2011-2013.
5-minute
Concentrations
of Interesta
Season b
Year
Percent of Receptors Exceeding
Concentration of Interest on
Specified Number of Days in Yearc
Percent of Monitors Exceeding
Concentration of Interest on
Specified Number of Days in Yearc
Number of Days
Number of Days
>1
>2
>5
>10
>1
>2
>5
>10
100
Fall
2011
8.2
3.5
2.0
0.7
67
33
0
0
2012
8.0
2.8
0.9
0.6
33
0
0
0
2013
0.6
0.4
0.1
0
33
0
0
0
Spring
2011
5.3
2.1
1.2
0.8
33
33
33
0
2012
4.5
1.3
0.6
0.6
33
33
33
0
2013
0.6
0.4
0.1
0
67
33
0
0
Summer
2011
3.5
1.4
0.9
0.6
33
33
33
33
2012
2.7
1.0
0.6
0.6
67
33
33
0
2013
0.6
0.5
0.2
0.1
33
33
0
0
200
Fall
2011
0.6
0.4
0
0
33
0
0
0
2012
0.5
0.4
0.1
0
0
0
0
0
2013
0
0
0
0
0
0
0
0
Spring
2011
0.6
0.4
0.1
0
0
0
0
0
2012
0.7
0.4
0.3
0.2
33
0
0
0
2013
0
0
0
0
0
0
0
0
Summer
2011
0.6
0.4
0.1
0.1
0
0
0
0
2012
0.5
0.5
0.2
0.1
0
0
0
0
2013
0
0
0
0
0
0
0
0
a There were no estimated or measured concentrations at or above 300 ppb.
b Daytime hours (6 AM to 8PM) only.
c There were 1,389 receptors modeled and 3 monitors.
3-49
-------
REFERENCES
Casella G and Berger RL. (2002). Statistical Inference. Second Edition. Editor: Carolyn Crocket,
Duxbury/Wadsworth Group, Pacific Grove CA.
Junger WL and de Leon AP. (2015). Imputation of missing data in time series for air pollutants.
Atmospheric Environment. 102:96-104.
Kahn HD. (1973). Note on the distribution of air pollutants. Journal of the Air Pollution Control
Association. 23(11):973.
Karl TR and Koss, WJ. (1984). Regional and National Monthly, Seasonal, and Annual
Temperature Weighted by Area, 1895-1983. Historical Climatology Series 4-3, National
Climatic Data Center, Asheville, NC, 38 pp. Available at:
https://www.ncdc.noaa.gov/monitoring-references/maps/iis-climate-regions.php
Larsen R. (1977). An air quality data analysis system for interrelating effects, standards, and
needed source reductions: Part 4. a three-parameter averaging-time model. Journal of the
Air Pollution Control Association. 27(5):454-459.
Rizzo M. (2009). Memorandum to the Sulfur Dioxide Review Docket (EPA-HQ-OAR-2007-
0352). Investigation of How Distributions of Hourly Sulfur Dioxide Concentrations Have
Changed Over Time in Six Cities. Sulfur Dioxide Review Docket. Docket ID No. EPA-
HQ-OAR-2007-03 52-0017.
SAS. (2017). Base SAS® 9.4 Procedures Guide, Seventh Edition. Available at:
http://documentation.sas.eom/api/collections/pgmmvacdc/9.4/docsets/proc/content/proc.p
df?locale=en#nameddest=bookinfo. Also used were SAS/ETS 14.2 User's Guide, edited
by A. Baxter, E. Huddleston and SAS/QC 14.2 User's Guide, edited by A. Baxter, V.
Clark, E. Huddleston, S. Prabhu, R. Rodriguez, D. Sawyer, J. Simmons.
Singer I. (1961). The relationship between peak and mean concentrations. Journal of the Air
Pollution Control Association. 11(7):336-341.
Turner B. (1964). A diffusion model for an urban area. Journal of Applied Meteorology. 3(1):83-
91.
U.S. EPA. (2009). Risk and Exposure Assessment to Support the Review of the SO2 Primary
National Ambient Air Quality Standard. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-09-007, July 2009. Available at:
https://www3. epa. gov/ttn/naaq s/standards/ so2/data/200908 S02REAFinalReport.pdf
U.S. EPA. (2013). AERSURFACE User's Guide. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-454/B-08-001, Revised January 2013. Available at:
https://www3.epa.gOv/scram001/7thconf/aerm.od/aersurface userguide.pdf
3-50
-------
U.S. EPA. (2015a). AERMINUTE User's Guide. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-454/B-15-006, October 2015. Available at:
https://www3.epa.gOY/ttn/scram/7thconl7aerm.od/aerm.iniite userguide.pdf
U.S. EPA. (2015b). 2011 National Emissions Inventory (NEI) Technical Support Document,
version 2. Office of Air Quality Planning and Standards, Research Triangle Park, NC,
August 2015. Available at: https://www.epa.gov/sites/production/files/2Q15-
10/docum en.ts/nei2.011 v2 tsd 14au&
U.S. EPA. (2016a). Guideline on Air Quality Models. 40 CFR Part 51 Appendix W.
U.S. EPA. (2016b). User's Guide for the AMS/EPA Regulatory Model - AERMOD. EPA-
454/B-16-011. Office of Air Quality Planning and Standards, Research Triangle Park,
NC.
U.S. EPA. (2016c). User's Guide for the AERMOD Meteorological Processor (AERMET).
Office of Air Quality Planning and Standards, Research Triangle Park, NC, EPA-454/B-
16-010.
U.S. EPA. (2016d). User's Guide for the AERMOD Terrain Preprocessor (AERMAP). Office of
Air Quality Planning and Standards, Research Triangle Park, NC, EPA-454/B-16-012,
December 2016.
U.S. EPA. (2016e). Technical Support Document (TSD): Preparation of Emissions Inventories
for the Version 6.3, 2011 Emissions Modeling Platform. Office of Air Quality Planning
and Standards, Research Triangle Park, NC, August 2016. Available at:
https://www. epa. gov/ sites/producti on/files/2016-
Q9/documents/2*' * s 20 I ^mismod tsd aug2^' i * tuial.pdf
U.S. EPA. (2016f). SO2 NAAQS Designations Modeling Technical Assistance Document, Draft.
Office of Air Quality Planning and Standards, Research Triangle Park, NC, August 2016.
Available at: https://www.epa.gov/sites/production/files/2Q16-
06/docum ents/ so2m odelin gtad.pdf
U.S. EPA. (2016g). AERMOD Implementation Guide. Office of Air Quality Planning and
Standards, Research Triangle Park, NC, EPA-454/B-16-013, December 2016.
U.S. EPA. (2017a). Air Markets Program Data. Available at: https://ampd.epa.gov/ampd/
U.S. EPA. (2017b). Review of the Primary National Ambient Air Quality Standard for Sulfur
Oxides: Risk and Exposure Assessment Planning Document. Office of Air Quality
Planning and Standards, Research Triangle Park, NC, EPA-452/P-17-001, February
2017. Available at:
https://www3.epa.gov/ttn/naaqs/standards/so2/data/20170216so2rea.pdf
3-51
-------
4 POPULATION EXPOSURE AND RISK
This chapter describes the methods used to characterize exposure and health risk
associated with SO2 emitted into ambient air under conditions just meeting the current primary
standard. As summarized in section 2.2, the overall analysis approach is based on linking the
health effects information to estimated population-based exposures that reflect our current
understanding of 5-minute concentrations of SO2 in the ambient air.
Population exposures were estimated using the EPA's Air Pollution Exposure Model
(APEX), version 5. The APEX model is a multipollutant, population-based, stochastic,
microenvironmental model that can be used to estimate human exposure via inhalation for
criteria and toxic air pollutants. APEX is designed to estimate human exposure to these
pollutants at the local, urban, and consolidated metropolitan level. In this REA we have used
APEX to estimate exposures in three study areas, the details of which are provided in the
following subsections. Additional information not provided here regarding all of APEX modules,
algorithms, and model options can be found in the APEX User's Guide (U.S. EPA, 2017a, b).
Briefly, APEX calculates the exposure time-series for a user-specified exposure duration
and number of individuals. Collectively, these simulated individuals are intended to be a
representative random sample of the population in a given study area. To this end, demographic
data from the decennial census are used so that appropriate model sampling probabilities can be
derived, considering personal attributes such as age and sex, and used to properly weigh the
distribution of individuals in any given geographical area. For this REA, the core demographic
geographical units for estimating exposure are census blocks. Because AERMOD predicted SO2
concentrations at air quality receptor sites in a regular spaced grid (chapter 3), APEX matches
the centroid of each census block (which are irregularly spaced due to varying size) in the study
area with the closest receptor to estimate exposures for simulated individuals residing in each
census block.
For each simulated person, the following general steps are performed:
• Select attribute variables and choose values to characterize the person (e.g. age, sex, body
weight, disease status);
• Construct the activity event sequence (minute by minute time series) by selecting a
sequence of appropriate daily activity diaries for the person (using demographic and other
influential variables);
• Calculate the concentrations in the microenvironments (MEs) that simulated individuals
visit;
• Calculate the person's simultaneous breathing rate and exposure for each event and
summarize for the selected exposure metric.
4-1
-------
A simulated individual's complete time-series of exposures (i.e., exposure profile),
representing intra-individual variability in exposures, is combined with the exposure profiles for
all simulated individuals in each study area and summarized to generate the population
distribution of exposures, representing inter-individual variability in exposures. As described
above regarding air quality and in the sections that follow describing APEX model inputs and
approaches to estimating exposure, the overarching goal of the REA is to account for the most
significant factors contributing to inhalation exposure, i.e., the temporal and spatial distribution
of people and pollutant concentrations throughout the study area and among the
microenvironments. The population distributions of exposures are combined with the health
effects information to characterize associated risk via two types of metrics: comparison to
benchmark concentrations and lung function risk. The details of the methods for exposure and
risk estimation are described in the sections that follow.
4.1 POPULATIONS SIMULATED
APEX stochastically generates a user-specified number of simulated persons to represent
the population in the study area. The number of simulated individuals can vary and is dependent
on the size of the population to be represented. In these analyses, the number of simulated
individuals was set at 100,000 in each area, a more than adequate number of individuals to
represent the geographically-restricted population residing within the exposure modeling
domains (approximately 180,000 - 500,000). Each simulated person is represented by a
"personal profile." The personal profile includes characteristics such as a specific age, a specific
home sector, a specific work sector (or does not work), specific housing characteristics, specific
physiological parameters, and so on. The profile does not correspond to any particular individual
in the study area, but rather represents a simulated person. Accordingly, while a single profile
does not, in isolation, provide information about the study population, a distribution of profiles
represents a random sample drawn from the study area population. This means that the modeling
objective is for the statistical properties of the distribution of profiles to reflect statistical
properties of the population in the study area.
APEX generates population-based exposures using several population databases. Based
on the geographic boundaries defining the study areas and the study groups of interest, APEX
will simulate representative individuals using appropriate geographic, demographic, and health
status information provided by existing population-based surveys. In this REA, there is variation
in the geographic units by which some of the input data sets are organized (e.g., U.S. census
tracts or a smaller subdivision such as census blocks). For example, employment status data are
provided at the tract level while population demographics are available at the block level.
Regardless of the geographic unit of the input data, all population-based data sets were applied at
4-2
-------
the block level in the exposure simulations. Where only tract level data were available, we
assigned the tract specific information directly to the blocks that comprise a particular tract.
Several updates were made to the APEX model inputs and algorithms for use in
simulating the populations of interest in this REA and are described in the following sections: (1)
population demographic data that are based on the 2010 census (section 4.1.1), (2) asthma
prevalence rates based on the 2011-2015 National Health and Nutrition Examination Survey
(NHANES) and vary by age, sex and geographic location (section 4.1.2), and data and equations
used to approximate personal attributes such as body weight, resting metabolic rate, and
breathing rate (section 4.1.3).
4.1.1 Demographics
As described in section 3.2.3, ambient air concentrations were modeled to a fine-scale
grid (100 m - 2 km) in each study area to better capture spatial heterogeneity in ambient air SO2
concentrations. We used U.S. Census blocks, the finest geographical scale available for the
population data,1 to take full advantage of this fine-scale air quality surface and best match the
potential at-risk populations with areas having the highest SO2 concentrations. Block-level
population counts were obtained from the 2010 Census of Population and Housing Summary
File l.2 Summary Files 1 (SF1) contains what the Census program calls "the 100-percent data,"
which is the information compiled from the questions asked of all (100% of) people and housing
units in the U.S. Three standard APEX input files3 are used for the current assessment. For the
purposes of having a more tractable analysis, we restricted these population demographic files to
include the census blocks within the five states that encompass the three study areas (i.e.,
Connecticut, Indiana, Massachusetts, Oklahoma, and Rhode Island) rather than use a national-
based file that would include all 50 U.S. states.
• PopGeoLocs2010 3StudyAreas.txt: census block identifiers (ID's), latitudes and
longitudes in degrees.
1 The minimum size for census block is between 30,000 to 40,000 ft2 or approximating a grid cell of about 55 - 60
meters (see https://www.census.gov/geo/reference/earm.html). The next larger sized census geographic unit is a
census block group which is comprised of multiple blocks. When considering that, on average, there are about 30
to 85 blocks per block group in the states where the study areas are located, it is likely that census block groups
would be more amenable to a modeled air quality surface having a grid cell size of about 1.6 - 8.1 Km (see
fattps://www.eensiis.gov/geo/maps~data/data/tatties/tractbtock.fatniB.
2 Technical documentation - 2010 Census Summary File 1—Technical Documentation/prepared by the U.S. Census
Bureau, Revised 2012 - available at: http://www.census.gov/prod/cen2010/doc/sfl.pdf.
3 The names of all APEX files are provided here to link the brief description with the appropriate input file.
4-3
-------
• PopBlockFemale2010 3StudyAreas.txt: census block identifiers, block-level population
counts for females, stratified by 23 age groups.4
• PopBlockMale2010 3StudyAreas.txt: census block identifiers, block-level population
counts for males, stratified by 23 age groups.
We evaluated the spatial distribution of the population in each study area, focusing on
children, a study group identified as an important at-risk population (section 2.1.3). First, we
subset this APEX input data set to include the blocks that are a part of the study areas. Then,
because there are wide ranging numbers of people in each block, we stratified these data into
three population groups: blocks having at least 100 people, blocks having greater than 100 to at
least 500 people, and blocks having greater than 500 people. Finally, we calculated the percent of
the total population that were children (aged 0-17) in each census block of each study area and
population group, and calculated the percentiles of that distribution providing perspective on the
spatial distribution of children (and adults) in each study area (Table 4-1).
In each study area there are a number of blocks having no people residing in them (i.e.,
non-residential blocks). While these blocks are retained in the exposure simulations as they could
still serve as an area where an individual might visit and be exposed to SO2 (e.g., a workplace
location within a study area), these blocks were not used to calculate the population spatial
distribution statistics. The majority of the residential blocks (84-93%) in each study area have
fewer than 100 people, with very few blocks (<1%) having a total population greater than 500
people. From a relative perspective, the Indianapolis study area had approximately double the
percent of residential blocks having a total population greater than 500 people compared to the
other two study areas. Further, while the overall distribution of the percent of children in each
study area is general similar (comprising 15-20% on average per residential block), the
Indianapolis study area consistently has a greater percent of children across most of the
percentiles of the distribution, most notably so for the residential blocks with a population
greater than 500 people.
4 The age groups in this file are: 0-4, 5-9, 10-14, 15-17, 18-19, 20-20, 21-21, 22-24, 25-29, 30-34, 35-39, 40-44, 45-
49, 50-54, 55-59, 60-61, 62-64, 65-66, 67-69, 70-74, 75-79, 80-84, >84.
4-4
-------
Table 4-1. Distribution of the percent of total population that are children residing in the
census blocks comprising each study area.
Percent of Population that are Children Residing in Study Area Census Blocks
Census blocks with <100
Census blocks with >100
Census blocks with >500
people per block
to <500
people per block
people per block
Study areaa
FR
IN
OK
FR
IN
OK
FR
IN
OK
Mean (%)
14.7
16.9
16.7
16.4
19.7
17.7
14.6
18.4
11.9
Standard deviation
(%)
10.0
11.0
11.2
7.2
9.0
9.8
8.2
8.9
8.7
Minimum (%)
0
0
0
0
0
0
0.5
0
0.5
5th percentile (%)
0
0
0
4.3
2.6
0.9
0.5
0.2
0.65
10th percentile (%)
0
0
0
8.3
7.8
3.8
3.05
2.8
1.2
25th percentile (%)
7.75
9.1
00
CO
12.3
14.5
11.4
8.9
14
3
50th percentile (%)
14.3
17
16.7
16.1
19.7
18
14.8
19.5
11.1
75th percentile (%)
20.3
24
23.9
19.9
25.4
23.8
22
25.8
17.7
90th percentile (%)
27.3
30.4
30.9
24.5
30.5
30
23.9
28.5
24.95
95th percentile (%)
31.9
34.3
35.1
28.2
33.9
34.9
25.3
30.6
27.9
99th percentile (%)
42.3
46.2
47.5
39.4
42.3
41.7
25.3
34.5
28.5
Maximum (%)
86.2
100
83.3
51.7
61.6
61.3
25.3
34.5
28.5
Number of blocks
3883
11130
7319
471
1123
355
10
57
20
Number of blocks
with non-zero
population
2588
7650
5106
471
1123
355
10
57
20
a FR = Fall River, IN =
ndianapolis, OK = Tulsa.
The employment file for APEX contains the probability of employment separately for
males and females by age group (starting at age 16) and by Census tract (the only census unit
available for this type of data). The 2010 Census collected basic population counts and other data
using the short form, but collected more detailed socioeconomic data (including employed
persons) from a relatively small subset of people using the 5-year American Community Survey
(ACS).5 The ACS dataset provides the number of people in the labor force, which we stratified
by sex/age/tract, considering both civilian workers and workers in the Armed Forces. The data
were stratified by sex and age group, and were processed so that each sex-age group combination
is given an employment probability fraction (ranging from 0 to 1) within each census tract.
Children under 16 years of age were assumed to be unemployed. To match the population
5 2010 U.S. Census American Factfinder: htt p ://factfi nder2. ce nsus. gov/. For instance, to obtain the table ID B23001
"Sex by age by employment status for the population 16 years and over", the following steps were performed.
First, select the "guided search option", choose "information about people" and select "employment (labor force)
status", "sex" and "age". For geography type select "census tract -140" for each state. Tables containing the
employment numbers were downloaded and used to calculate the employment probabilities for each age group.
4-5
-------
demographic files, we included only the census blocks within the five states that encompass the
three study areas. To use the file at a block level, all blocks were assigned the same employment
probabilities as the parent tract. One standard APEX input file is used for the current assessment:
• EmpBlock2010 3StudyAreas.txt: census block IDs, employment probabilities (in
fractional form), stratified by 13 age groups.6
4.1.2 Asthma Prevalence
The population groups included in this exposure assessment are adults with asthma (>18
years old) and children with asthma (5 to 18 years old),7 based on their identification in this REA
as an at-risk population (section 2.1.3). To best approximate the number (and percent) of
individuals comprising each of these population groups in each study area, we considered several
influential variables that could affect asthma prevalence. It is widely recognized that there are
significant differences in asthma prevalence based on age, sex, U.S. region, and family income
level, among other factors.8 There is spatial heterogeneity in family income level across census
geographic areas (and also across age groups)9 and spatial variability in local scale ambient air
concentrations of SO2 (e.g., Figures 3-6 to 3-8). Thus, we have developed an approach to better
estimate the variability in population-based SO2 exposures by accounting for these particular
attributes of this study group and their spatial distribution across each of the study areas.
With regard to asthma prevalence, the data are used to identify if a simulated individual
residing within a modeled census geographic area has asthma - and are not used for selection of
any other personal attribute nor in the selection of activity pattern data. Thus, our primary
objective with these data was to generate census block level prevalence estimates that reflect
variability in asthma prevalence contributed by several known influential attributes (e.g., age,
sex, geographic location). Two data sets were identified and linked together to estimate asthma
prevalence used for this REA. First, asthma prevalence data were obtained from the 2011-2015
National Health Interview Survey (NHIS) and are stratified by NHIS defined regions (Midwest,
Northeast, South, and West), age, and sex.10 We explored other variables that were available in
6 The age groups in this file are: 16-19, 20-21, 22-24, 25-29, 30-34, 35-44, 45-54, 55-59, 60-61, 62-64, 65-69, 70-74,
and >75.
7 As in other NAAQS reviews, this REA does not estimate exposures and risk for children younger than 5 years old
due to the more limited information contributing relatively greater uncertainty in modeling their activity patterns
and physiological processes than children between the ages of 5 to 18.
8 For example, see the Center for Disease Control report "National Surveillance of Asthma: United States, 2001-
2010", available at: https://www.cdc.gov/nchs/data/series/sr_03/sr03_035.pdf.
9 For example, see the U.S. Census report "Income and Poverty in the United States: 2016", available at:
https ://www. census. gov/content/dam/Census/library/publications/2017/demo/P60-259.pdf.
10 Information about the NHIS is available at: http://www.cdc.gov/nchs/nhis.htm.
4-6
-------
the NHIS data set that contributed to variability in asthma prevalence and that could be used to
extrapolate the asthma prevalence to a finer geographic scale than the NHIS-provided four
regions. The linking variable had to be common with variables available in the population
demographic data. Based on this criterion, we selected family income level to poverty thresholds
(i.e., whether the family income was considered below or at/above the US Census estimate of
poverty level for the given year) and used that as an additional variable to stratify the NHIS
asthma prevalence. Then, we obtained information from the 2013 Census ACS to estimate
family income level to poverty thresholds at the census tract level and stratified by several ages
and age groups.11 By combining these two data sets, we developed census tract level asthma
prevalence estimates for children (by age in years) and adults (by age groups), also stratified by
sex (male, female) that were weighted by the individual census tract populations and family
income level. To match the population demographic and employment files, we included only the
census blocks within the same five states that encompass the three study areas. The census tract
sex- and age-specific prevalence were extrapolated to census blocks using the 11-character
identifier shared between census tracts and blocks. A detailed description of how the NHIS data
were processed to create the data set used for input to APEX is provided in Appendix E. One
standard APEX input file is used for the current SO2 assessment:
• asthma_prev 1115 block 3StudyAreas.txt: census block identifiers, block-level asthma
prevalence (in fractional form) stratified by sex, 18 single year ages (for ages <18),12 and
7 age groups (for ages > 17).
The asthma prevalence estimates vary for the different ages and sexes of children and
adults13 that reside in each census block of each study area. We evaluated the spatial distribution
of the asthma prevalence using the specific blocks that comprise the exposure modeling domain
in each study area. We first separated the estimates for children from those for adults and
calculated the distribution of asthma prevalence for the blocks, stratified by sex (Table 4-2). By
design (i.e., the use of age, sex, and family income variables), there is spatial variability in the
11 Census tract level data is the finest scale geographical unit having family income information. The family
income/poverty ratio threshold used was 1.5, that is the surveyed person's family income was considered either <
or > than a factor of 1.5 of the U. S. Census estimate of poverty level for the given year.
12 The census data set used only had children for single years up to and including age 17, after that year they are
provided in groups. The upper portion of this age range differs from those considered as children in estimating
exposures i.e., in our exposure assessment children are considered upwards to 18 years old. To simulate the
number of children with asthma age 18, estimated prevalence from the first adult group were used (i.e.,
individuals age 18-24).
13 While prevalence rates were estimated for all ages of children (in single years 5 -17), for adults they were
estimated for seven age groups: 18-24 years, 25-34 years, 35-44 years, 45-54 years, 55-64 years, 65-74 years,
and, >75 years old (see Appendix E for more information).
4-7
-------
prevalence estimates. Consistent with broadly defined national asthma prevalence (e.g., Table 3-
2 of SO2 PA), children have higher estimated rates than adults,14 male children have higher rates
than female children,15 and adult females have higher rates than adult males (e.g., compare with
mean values of Table 4-2). However, when evaluating variability contributed by study area, age,
sex, and family income level on a spatial scale, an additional degree of variability emerges across
the study areas (as presented in Tables 4-2). The Fall River study area has some of the highest
asthma prevalence for children considering most of the statistics with rates as high as 21.5% in
one or more census blocks for males of a given year of age. The Tulsa study area exhibits some
of the lowest asthma prevalence when considering adults (both sexes) with rates as low as 4.0%
in one or more blocks for males within a given age group. These summary statistics represent the
range of age- and sex-specific values for the census blocks used in each APEX simulation to
estimate the number of individuals that have asthma.
Table 4-2. Estimated asthma prevalence for children and adults in census blocks of three
study areas, summary statistics.
Study Area
(# census blocks)
and
Population group
Sex
Asthma Prevalence across a
I ages (or age groups) for all census blocksa
Mean
Standard
Deviation
Minimum
Median
Maximum
child
Fall River
female
9.3%
2.4%
5.7%
9.2%
18.6%
male
13.3%
2.3%
8.4%
13.3%
21.5%
(4'364) H If
adult
female
9.9%
1.5%
7.2%
9.8%
17.6%
male
6.3%
1.0%
5.1%
5.8%
9.0%
child
Indianapolis
female
9.1%
2.0%
5.8%
8.6%
19.4%
male
10.8%
2.3%
6.6%
10.7%
16.8%
(12,31°) . ..
adult
female
9.9%
1.8%
6.8%
10.0%
17.6%
male
6.1%
1.5%
2.5%
5.9%
10.4%
child
Tulsa
female
10.2%
1.5%
7.3%
10.2%
13.9%
male
12.0%
2.0%
7.5%
12.3%
16.1%
(7,694) adult
female
8.8%
1.7%
5.5%
8.8%
14.4%
male
5.1%
0.6%
4.0%
5.0%
6.9%
a As described in text above this table, prevalence estimates are based on age (or age group) and sex-specific prevalence
estimates for each census block derived from CDC NHIS asthma prevalence and U.S. census income/poverty ratio information.
14 Asthma prevalence, when not separated by sex, is greater for children (mean of 11.3%, 10.0%, and 11.1%) than
that of adults (mean of 8.1%, 8.0%, 7.0%) for the Fall River, Indianapolis, and Tulsa study areas, respectively.
Nationally, asthma prevalence for children is 8.4% and for adults is 7.6% (Table 3-2 of SO2 PA).
15 Asthma prevalence, when not separated by the three study areas evaluated, is greater in boys (mean of 11.6%)
than that of girls (mean of 9.4%). Nationally, asthma prevalence for boys is 9.9%, for girls is 6.9% (Table 3 -2 of
S02 PA).
4-8
-------
There are many other personal attributes that have been shown to influence asthma
prevalence, such as race, ethnicity, obesity, smoking, health insurance, and activity level (e.g.,
Zahran and Bailey, 2013). The set of variables chosen to stratify asthma prevalence for use in
this REA (i.e., age, sex, and family income level) was based on 1) maximizing the potential
range in asthma prevalence variability, 2) maximizing the number of survey respondents
comprising a representative subset study group, and 3) having the ability to link the set of
attributes to variables within the Census population demographic data sets. Many of the
additional potential influential factors identified here are not available in the census data and/or
have limited representation in the asthma prevalence data (e.g., the survey participant has health
insurance or they provide a response to a question regarding their body weight). Race is perhaps
the only attribute common to both the prevalence and population data sets that could be an
important influential factor and was not directly used in this REA to calculate asthma prevalence.
However, the use of race in calculating asthma prevalence, either alone or in combination with
family income level, would further stratify the NHIS analytical data set and appreciably reduce
the number of individuals of specific age, sex, race, and family income level, potentially
reducing the confidence in calculated asthma prevalence based on so few data. Because family
income level already strongly influences asthma prevalence across all races and stratifies the
NHIS data into only two subgroups (i.e., above or below the poverty threshold) rather than the
larger number of subgroups a race variable might yield, family income was chosen as the next
most important variable beyond age and sex to rely on for weighting the spatial distribution of
asthma prevalence.
That said, there is some uncertainty in our estimates caused by not utilizing race as a
influential variable to spatially weight asthma prevalence (e.g., in addition to family income
level). Therefore, we evaluated the influence race and obesity (as indicated by body mass index
or BMI) might have on asthma prevalence.16 While considering this, we also note that while
census demographic data are available on the spatial distribution of variables such as race, age,
sex and income level, we are unaware of useful data for spatially allocating obesity prevalence
across a geographic area.
We first evaluated the number of NHIS adult and child participants that provided race
and BMI information and compared them to the numbers of these groups that provided the
information (age, sex, family income) used in the REA approach. The adult data set had few
missing values when considering the new variables race and BMI (totaling about 160,000
16 The processing of the prevalence data is described in Appendix E and considered age, sex, and family income
level for each of four regions (i.e., Northeast- NE, Midwest-MW, South-S, and West-W). We evaluated these
same variables here, however, we also included data responses from the variable RACERPI2 (a value of "2"
represented black African Americans) and the variable BMI (or BMISC) for race and obesity, respectively.
4-9
-------
adults); however, only one-third of children had values for BMI. Thus, the analytical data for
children set was substantially smaller in survey sample size compared to the NHIS data set used
for the REA approach (i.e., 20,000 vs 60,000 participants). The smaller sample size of children
responding to this survey question indicates a potentially greater source of uncertainty with the
use of this data set as compared to the relatively more complete adult data set for these variables.
A logistic model was constructed using PROC SURVEYLOGISTIC (SAS, 2017) and
included the newly expanded list of potentially influential variables. Age was considered as a
continuous variable while four other variables were evaluated using a dichotomous form: sex
(female to not); family-income ratio (below poverty to not); race (black African American to
not); BMI (obese or BMI> 30 to not). Models were then applied to this data set using the same
four regional stratifications. We first evaluated the independent effect each variable had on
asthma prevalence (i.e., the individual responded "yes" to the question about still having
asthma). Odds ratios were calculated and plotted along with 95% confidence intervals.17 The
odds ratios can be interpreted as the percent difference between the two conditions being
compared and the variable's impact on the estimated asthma prevalence. Results of this analysis
for each of the variables is presented in Figure 4-1 (adults) and Figure 4-2 (children). We also
evaluated all possible variable interactions in the statistical model. While the results are
somewhat variable across the data sets and complicated to interpret, in general many of the
variable independent effects remained statistically significant, and there were few significant
interactions (Appendix E, Attachment 4).
Overall, the results for the adult data were more consistent across the different U.S.
regions than the child data, possibly due to having a less complete data set for the children.
Obesity, rather than race, appears to play a more important role in influencing the asthma
prevalence in adults (Figure 4-1) compared to children, while both obesity and race appear to
play an important role in explaining asthma prevalence in children (Figure 4-2). Family income
level was important alone and in interactions with other influential variables (e.g., BMI, race)
considering both children and adults and for most of the four U.S. regions. Further, family
income level consistently exhibited greater influence than race on adult asthma prevalence in
three of the four regions, while race exhibited a greater influence than family income when
considering child asthma prevalence. Sex also had a greater influence on asthma prevalence in
17 The odds ratios can be interpreted as the percent difference between the two conditions being compared and the
variable's impact to the estimated asthma prevalence. For instance, Figure 4-1 shows that in the Northeast, an
odds ratio of 1.85 was calculated for adult asthma prevalence considering the BMI variable (i.e., obese - BMI >30
vs. not obese - BMI<30). This suggests that if an individual is obese, they are 85% more likely to have asthma
than an individual that is not obese. In addition, the 95% confidence intervals that include a value of 1.00 are not
considered statistically significant, thus when considering the BMI variable, statistical significance (p<0.05) can
be assigned to the effect this variable has on the asthma prevalence.
4-10
-------
adults compared to children, though women are more likely to have asthma than men while boys
are more likely to have asthma than girls. Age was generally not a large factor as assessed in the
constructed model (i.e., the statistical test evaluates year-to-year comparative differences),
though clearly asthma varies across the full lifetime of ages (e.g., see Appendix E, Attachments 1
and 2) and is an important variable for spatially linking at risk populations and ambient air
concentrations in the exposure model.
This evaluation indicates that use of the family income variable in this exposure
assessment as an influential factor to estimate variation in asthma prevalence provides
reasonably similar estimates as race, particularly for adults. For children, while family income
was shown as an important influential variable and, in a few instances, could serve as a surrogate
variable to approximate the degree of race-related influence, family income alone may not
entirely explain spatial variability in asthma prevalence across urban areas such as those in this
REA.
Calculated Odds Ratios - Adult Asthma Prevalence
2.5
2.0
o
1.5
ro
o:
m
T3
1.0
O
Age
•—•
0.5 -
0.0
Black African
American
Obese
Sex
Family Income
t
U.S. Region
Figure 4-1. Influence of age, race, obesity, sex and family income on adult asthma
prevalence (based on NHIS 2011-2015 for four U.S. regions).
4-11
-------
Calculated Odds Ratios - Child Asthma Prevalence
3.0
2.5
2.0
ro
en
l/l
T3
T3
O
1.5
1.0 -
0.5
0.0
Age
• •
Black African
American
i
\
Obese
Sex
i
Family Income
to
{/)
{/)
{/)
-------
that would be characterized as obese. However, the resulting distribution of body mass in the
simulated population is not likely to reflect body mass variation at a local-scale in any particular
study area.
4.1.3 Personal Attributes
In addition to using the above demographic information to construct the simulated
individuals, each modeled person is assigned anthropometric and physiological attributes. All of
these variables are treated probabilistically in APEX, taking into account interdependencies
where possible, and reflecting variability in the population. It is not the intention of this
document to provide detailed description of all the model inputs in each of the files and the data
used in their derivation, however there are a few that have been recently updated for use in this
REA, namely new statistical distributions for estimating body weight, equations for estimating
resting metabolic rate, and equations for estimating activity specific ventilation rate. Brief
descriptions of the data used to develop the input files are provided in the sections below. For
additional detail, see Appendices F through H and the data within the APEX input files.
4.1.3.1 Body Weight and Surface Area
Anthropometric attributes utilized by APEX in various assessments for estimating
pollutant-specific exposures or doses include height, body weight (BW), and body surface area
(BSA). Two key personal attributes determined for each individual in this assessment are BW
and BSA, both of which are used in the calculation of a number of other variables associated
with estimating exposures (e.g., ventilation rate).
Regarding the estimation of body weight, a new APEX input file was generated using
2009-2014 NHANES data.18 Briefly, body weight and height data for surveyed individuals were
obtained and stratified by sex and single years for ages 0 - 79; all ages above 80 were combined
as a single age group. Statistical form of the age- and sex-specific body weight and height
distributions were evaluated using a log-likelihood statistic. Body weight was found to best fit a
lognormal distribution; height was found to best fit a normal distribution. Because height and
body weight are not independent, the joint distributions of height and logarithm of body weight
were fit assuming a bivariate normal distribution. Then, parameters defining the joint
distributions19 were smoothed using a natural cubic spline to have them represent continuous
functions of age rather than vary discontinuously. In addition, having the smoothed parameters
18 Original data are available in the form of the questionnaire datasets for 2009-2010, 2011-2012, 2013-2014
NHANES from pages reached from this main page: https://wwwn.cdc.gov/ncIis/nhanes/Defanit.aspx. Details
regarding the data used and the derivation of the distributions is provided in Appendix G.
19 Five parameters were used for each age and sex: mean log(BW), standard deviation of log (BW), mean (height),
standard deviation of (height), and body weight height correlation coefficients.
4-13
-------
could be used to extrapolate information obtained from the single age year distributions (ages 0 -
79) to approximate statistical distributions of body weight for ages > 80. A linear function was
fitted to ages 70 and above to extrapolate the parameter values (and hence the statistical
distributions of body weight) up to age 100. These distributions are randomly sampled to
estimate an age and sex-specific body weight for each simulated individual. Comparison of the
new distributions to the body weight distributions previously used by APEX and developed from
the 1999-2004 NHIS indicate, for both sexes and across all ages, simulated body weight is about
two percent greater using the updated distributions. This difference is expected given the
consistent trend of increasing body weight that has occurred over the past few decades.
Finally, age- and sex-specific body surface area, a variable used in conjunction with
breathing rate to approximate moderate or greater exertion (section 4.1.3.3) is estimated for each
simulated individual as shown in Equation 4-1, and is based on an analysis provided by
Burmaster (1998).
BSA = e-2-2781 x BW0,6821 Equation 4-1
One standard APEX input file is used for the SO2 assessment:
• Physiology040617 noHT Graham Glen QA.txt: Provides parameters for estimating
body weight (log BW, standard deviation of BW, lower and upper bounds of BW, by
single age years 0-100 and by two sexes) and regression coefficients used in estimating
BSA for all sexes and ages.
4.1.3.2 Energy Expenditure and Oxygen Consumption
Energy expended by different individuals engaged in different activities can have an
important role in pollutant-specific exposure and/or dose. For example, energy expenditure is
related to ventilation rate, which is an important variable in this REA given that the S02-induced
lung function response has been documented to occur under conditions of elevated ventilation
(section 2.1.4 above). In addition, because we are also interested in exposures that occur over
short durations (i.e., 5-minutes), estimating activity-specific ventilation rate (Ve) has been an
important motivation behind the development of the algorithm used by APEX. The fundamental
basis for Ve algorithm is founded in energy expenditure which, for our modeling purposes here,
can be related to an individual's resting metabolic rate (RMR) or the energy expended while an
individual is at complete rest, along with the energy expended while an individual performs
activities involving greater exertion, termed here as metabolic equivalents of work (METs). The
approaches used by APEX for estimating RMR and METs are described below, beginning first
with the update to the equations used for estimating an individual's RMR.
4-14
-------
To estimate RMR for the 2009 REA, the previous version of APEX (version 4.3) had
used an algorithm originally based on analyses by Schofield (1985). Because all of the clinical
subject data used by Schofield (1985) were from studies conducted as far back as 60 years prior
to that publication, we felt it was important to search for newly available study data to better
represent the simulated population in this REA. In addition, while using the Schofield (1985)
equations there were occasional abrupt discontinuities in the estimated RMR observed at some of
the equation boundaries (e.g., between age 59 and 60), which were largely a function of how the
data were stratified (six age groups and two sexes) and the resulting equation parameters.
Since the 2009 REA, we have reviewed recent RMR literature and other published
sources containing individual data and have compiled the associated individual RMR
measurements, along with associated influential attributes such as age, sex, and body weight,
where available. Data from these individual studies were then combined with RMR data reported
in the Oxford-Brookes database (Henry, 2005; IOM, 2005) and screened for duplicate entries. In
addition, observations missing values for RMR, BW, age, or sex were deleted, resulting in a
dataset containing 16,254 observations (9,377 males and 6,877 females).
Using this new RMR dataset, and having a goal of updating the previous RMR equations
and reducing discontinuities in RMR between age groups, new equations were developed. The
equations follow the general format of a multiple linear regression (MLR) model, using age and
body weight as independent variables to estimate each simulated individual's RMR, along with a
residual error term (f).20 It is known that RMR and BW, as well as RMR and age, are not exactly
linearly related; the algorithms developed here use BW (in kg), age (in years), and the natural
logarithms of BW and (age+1)21 as follows in Equation 4-2, with their parameter estimates
provided in Table 4-3.
RMR = /?0 + /?XBW + /?2 log(BW) + P3Age + /33\og(Age) + £t Equation 4-2
When comparing observed versus predicted values, the new RMR equations have a bias
of less than 0.5%, compared to the previously used APEX equations which had a bias of between
1-2%. Further, the discontinuities in RMR seen across particular age group boundaries using the
previous equations have been reduced when using these updated equations in APEX. Additional
20 The residual error term largely accounts for the estimation of inter-personal variability in RMR for individuals
having the same body weight and age. There are other potentially influential sources of variability that are not
explicitly accounted for by the equation (e.g., seasonal influences on RMR) and thus remain as an uncertainty.
21 The "+1" modifier allows APEX to round age upwards instead of downwards to whole years, which is necessary
to avoid undefined log(0) values.
4-15
-------
details regarding the derivation of the updated RMR equation and performance evaluation are
provided in Appendix H. One standard APEX input file is used for the SO2 assessment:
• Physiology040617 noHT Graham Glen QA.txt: Regression coefficients used to
estimate RMR (kcal day"1) for two sexes and six age groups.
Table 4-3. Regression parameters used to estimate RMR by sex and age groups.
Sex n BW log(BW) Age log(Age) Intercept Std dev
male
0-5
625
13.19
270.2
-18.34
131.3
-208.5
69.10
6-13
1355
10.21
260.2
13.04
-205.7
333.4
115.3
14-24
4123
0.207
1078.
115.1
-2794.0
3360.6
161.1
25-54
2531
2.845
729.6
3.181
-191.6
-1067
178.2
55-99
743
9.291
264.8
-5.288
181.5
-705.9
163.6
female
0-5
625
11.94
261.5
-22.31
120.9
-183.6
64.16
6-13
1618
5.296
409.1
40.37
-524.9
392.7
99.43
14-29
2657
0.968
676.9
40.89
-1002
772.7
143.1
30-53
1346
4.935
355.4
16.28
-896.0
2225
145.3
54-99
631
2.254
445.9
5.464
-489.9
944.2
124.5
Units: RMR = kilocalories/day; BW = kilograms; Age = years
Following the estimation of an age- and sex-specific RMR for simulated individuals, the
next variable used for estimating ventilation rate involved an approximation of the energy
expended for activities an individual performs throughout their day. As mentioned above,
activity-specific energy expenditure is highly variable and can be estimated using metabolic
equivalents of work (METs), or the ratios of the rate of energy consumption for non-rest
activities to the resting metabolic rate of energy consumption, as follows:
EE =MET x RMR Equation 4-3
where,
EE = Energy expenditure (kcal/minute)
MET = Metabolic equivalent of work (unitless)
RMR = Resting metabolic rate (kcal/minute)
Statistical distributions of METs were developed for simulated activities using the
physical-activity compendium (Ainsworth et al., 2011; hereafter "the compendium"). The
compendium contains a point value for the MET associated with each of several hundred
different activities. Activity-specific MET distributions were developed by cross-walking the
activities described in the compendium with the descriptions of activities in the activity pattern
data base used by APEX (US EPA, 2017c). The shape of the statistical distribution (e.g., normal,
4-16
-------
lognormal, triangular, point) for each activity was assigned based on the number of
corresponding activities in the compendium and goodness-of-fit statistics. When simulating
individuals, APEX randomly samples from the activity-specific METs distributions to obtain
values for every activity performed. Two standard APEX input files are used for the SO2
assessment:
• MET Distributions 092915.txt. MET distribution number, statistical form, distribution
parameters, lower and upper bounds, activity description
• METmapping current APEXJile.txt: activity codes, age group (where applicable),
occupation group, MET distribution number, and activity description used to link of MET
distributions to activities performed
The rate of oxygen consumption (VO2, Liters min"1) for each activity is then calculated
from the energy expended (kcal min"1) using an energy conversion factor (ECF, Liters O2 kcal"1)
as follows in Equation 4-4:
V02=EE X ECF Equation 4-4
The value of the ECF is randomly selected from a uniform distribution for each person,
U[0.20, 0.21] (Johnson et al., 2002, adapted from Esmail et al., 1995). One standard APEX input
file is used for the SO2 assessment:
• Physiology040617 noHT Graham Glen QA.txt. Parameters of the uniform distribution
representing the ECF used for all ages and both sexes.
4.1.3.3 Ventilation Rate
Human activities are variable over time, with a wide range of activities possible within
only a single hour of the day. The type of activity an individual performs, such as sleeping or
jogging (as well as individual-specific factors such as age, weight, RMR) will influence their
ventilation rate. APEX estimates minute-by-minute ventilation rates that account for the
expected variability in the activities performed by simulated individuals. Ventilation rate is
important in this assessment because the lung function responses associated with short-term peak
SO2 exposures coincide with moderate or greater exertion (ISA, section 5.2.1.2). In our exposure
modeling approach, we used APEX to generate the complete time-series of activity-specific
ventilation rates and the corresponding time-series of estimated SO2 exposures. APEX then
aggregates both the ventilation rate and exposure concentration to the averaging time of interest
(a 5-minute average). Thus, the model provides exposure estimates for the simulated individuals
that pertain to specific target levels for both ventilation rate and exposure concentration. The
approach to estimating activity-specific energy expenditure and associated ventilation rate
4-17
-------
involves several algorithms and physiological variables, with details found in the APEX User's
Guide (U.S. EPA, 2017a, b).
Using the existing measurement Ve dataset from Graham and McCurdy (2005), new Ve
algorithms were developed for predicting activity specific Ve in the individuals simulated by
APEX. The new Ve algorithms do not directly employ previously used variables to stratify the
data (age groups, sex) and explain variability (age, body weight, height) in ventilation rate,
effectively simplifying and reducing the number of equations. The new algorithms utilize a new
variable, the maximum volume of oxygen consumed (VChm) as an input.22 Body weight, height
and sex - as well as fitness level (which is often represented by VChm) - influence oxygen
consumption for a particular activity. However, variability for each of these influential variables
are already captured in the algorithm used to estimate each simulated individual's RMR, and
subsequently, the estimation of their activity specific VO2.23 Thus, the only input variables
needed for the new Ve algorithm are VO2 and VChm,24 both of which are estimated by APEX.
Details for the derivation of and performance evaluation of the new equation that APEX
uses to estimate ventilation rate are provided in Appendix H. Briefly, the Ve dataset contains
6,636 observations, with 4,565 males and 2,071 females. Similar to the earlier ventilation
equation by Graham and McCurdy (2005), a mixed-effects regression (MER) model was fit
because the MER separates residuals into within-person (ew) and between-person (eb) effects,
known as intrapersonal and interpersonal effects, respectively.25 It was found that the actual
values of VO2 and VChm are less relevant than the fraction of maximum capacity, represented by
fi = VO2/ VCtem. The variable fi may operate non-linearly (for example, fi = 0.9 is likely more
than twice as encumbering as fi = 0.45). A transformation regression approach (PROC
TRANSREG - SAS, 2017) was used to determine the most appropriate variable transformation,
indicating a power of 4 to 5 be used when only the log transformed VO2 was used as the
independent variable and described in Equation 4-5.
22 Use of VO2111 as an explanatory variable in separate related research on metabolic equivalents of task (MET)
values for persons with unusual maximum capacity for work suggests that their MET distributions are modified in
a predictable way by their maximum MET (or, equivalently, by VChm), thus providing support for use of this
variable in the new VE algorithms.
23 Oxygen consumption associated with activities performed is based on the activity specific metabolic equivalents
for work (METs), an individual's estimated RMR, and an energy to oxygen conversion factor (U.S. EPA, 2017b).
24 Distributions of VC^m used by APEX were derived from 20 published studies reporting individual data and
grouped mean (and standard deviation) data obtained from 136 published studies. Details are provided in Isaacs
and Smith (2005).
25 N(0, eb) is a normal distribution with mean zero and standard deviation C/,=0.09866 meant to capture /'wterpersonal
variability, which is sampled once per person. N(0, ew) is an /'n/rapersonal residual with standard deviation of
c»=0.07852. which is resampled daily due to natural /n/rapcrsonal fluctuations in VE that occur daily.
4-18
-------
y _ e (3.300 + 0.8128xln(7Oz)+ 0.5126 x (702^I/02m)4+N(0,efc)+N(0,ew))
Equation 4-5
In comparing the statistical fit of the new equation with the equations used by APEX
previously to estimate ventilation rate, the resulting coefficient of determination (r2 values) for
the new equation (r2 = 0.94) indicates an improved fit compared to that of the previous equations
(r2 = 0.89-0.92). Further, because the data were not stratified by age groups (or any other
groupings), there are no discontinuities in predictions made across age boundaries as was
observed when employing the previous equation. Information used in estimating ventilation rate
is found in the following APEX two input files:
• Physiology040617 noHT Graham Glen QA.txt: parameters describing statistical
distributions of normalized maximum oxygen consumption rate (NVChm) for two sexes
by single age years (0-100) (see, Isaacs et al., 2005).
• Ventilation VEMethod2 102816 new. txt: minimum and maximum age ranges, regression
coefficients, between and within error terms used to estimate individual activity-specific
ventilation.
To use this information to estimate health risks for children, the ventilation rates observed
for the adult study subjects need to be converted into rates that best reflect the different
physiology of children. Consistent with prior REAs (U.S. EPA, 2009, 2014b; Whitfield et al.,
1996) we used an equivalent ventilation rate (EVR), which is essentially an allometrically
normalized ventilation rate, to estimate instances when a simulated individual reaches a
ventilation rate relatively as high as that of the study subjects (i.e., termed here as moderate or
greater exertion).
EVR = Equation 4-6
BSA M
In the controlled human exposure studies evaluating the health effects of SO2, the
ventilation rates for study subjects (i.e., male and female adults) experiencing effects from 5- to
10-minute SO2 exposures are generally within 40-50 L/min, with most set at or around 40 L/min
(ISA, Table 5-2 and Table 4-12 below).26 However, body surface area was not measured in the
controlled human exposure studies and the relevant ventilation data were not separated by sex.
We approximated BSA of the study subjects as 1.82 m2 based on data for adult males and
26 In these studies, subjects were breathing freely during exercise; thus, it is expected that there was a mixture of
nasal, oral, and oro-nasal breathing that occurred across the study subjects. Without information regarding the
precise breathing method used by any subject corresponding with their health response, we assumed that the
mixture in breathing method used by study subjects is representative for the simulated population.
4-19
-------
females from U.S. EPA (1989).27 Based on these data, we estimate EVR for the study subjects to
be 40/1.82 ~ 22 L/min-m2. Accordingly, we have used this EVR as the target EVR in this
assessment and simulated individuals at or above an EVR of 22 L/min-m2 (children or adult)
during a 5-minute exposure event were characterized as performing activities at or above
moderate exertion. This is essentially the same target EVR value as that used in the 2009 REA
(i.e., > 22 L/min-m2), approximated at that time based on data from U.S. EPA (1997). This value
used for EVR would represent a mean value based on the data used in its estimation and is
considered reasonable to apply and approximate when, on average, individuals in a population
might experience periods of moderate or greater exertion. There is uncertainty in this mean based
on the broad scale of data used in it derivation, as well as by not having any information on how
to characterize intra- or inter-personal variability, where existing, and in its direct extrapolation
from adults to children.
4.2 METEOROLOGICAL DATA
Temperature data are used by APEX in selecting human activity data and in estimating
AERs for indoor residential MEs. Hourly surface temperature measurements were obtained from
the National Weather Service Integrated Surface Hourly (ISH) data files (described in section
3.2.1.1). The weather stations used for each study area are given in Table 4-4. Given the limited
geographic area of each study area, data from a single station was used to represent the ambient
air temperature in each study area. The occurrence of missing temperature data was limited to a
few instances (Table 4-4). Temperature values for the hours missing data were estimated using
SAS PROC EXPAND, a simple linear interpolation technique. Because of the small number of
missing values, the impact of the filled values to estimated exposures is assumed negligible.
Multiple unique APEX input files are used, one for each year and study area, and generally in
two formats:
• METdata[studyarea]Y[year].txt: MET station IDs, hour of day, hourly temperature (°F)
for each MET station, by study area and year
• METlocs[studyarea]Y[year], txt: MET station ID's, latitudes and longitudes, start and stop
dates of temperature data
27 Most of the controlled human exposure studies were conducted in the 1980s, thus use of the 1989 EPA Exposure
Factors Handbook is considered the most representative source to use in estimating BSA for the study subjects
compared with the 1997 and 2011 versions of that document given that body weight distributions (and hence
BSA) have changed over time.
4-20
-------
Table 4-4. Study area meteorological stations, locations, and hours of missing data.
Study Area
Station Name
Station
Number
Latitude
Longitude
Number of hours with
missing temperature
2011 2012 2013
Fall River, MA
PROVIDENCE TF
GREEN ARPT
14765
41.7225
-71.4325
0
0 5
Indianapolis, IN
INDIANAPOLIS
INTERNATIONAL APT
93819
39.72517
-86.28168
0
0 0
Tulsa, OK
RICHARD LLOYD
JONES JR APT
53908
36.0396
-95.9846
10
0 0
4.3 CONSTRUCTION OF HUMAN ACTIVITY SEQUENCES
Exposure models use human activity pattern data to predict and estimate exposure to
pollutants. Different human activities, such as outdoor exercise, indoor reading, or driving a
motor vehicle can lead to different pollutant exposures. This may result from differences in the
amount of the pollutant in the different locations where the activities are performed as well as
from differences in the energy expended in performing the different activities (because energy
expenditure influences inhalation and ingestion and thus may influence pollutant intake). To
accurately model exposures to ambient air pollutants, it is critical to have a firm understanding of
the locations where people spend time and the activities performed in such locations. The
following subsections describe the activity pattern data, population commuting data, and the
approaches used to simulate where individuals might be and what they might be doing.
After the basic demographic variables are identified by APEX for a simulated individual
in the study area, values for the other variables are selected as well as the development of the
activity patterns that account for the places the simulated individual visits and the activities they
perform. The following subsections describe the population data we used in the assessment to
assign key features of the simulated individuals, and approaches used to simulate the basic
physiological functions important to the exposure estimates for this REA.
4.3.1 Consolidated Human Activity Database
The Consolidated Human Activity Database (CHAD) provides time series data on human
activities through a database system of collected human diaries, or daily time location activity
logs (U.S. EPA, 2017c). The purpose of CHAD is to provide a basis for conducting multi-route,
multi-media exposure assessments (McCurdy et al., 2000). The data contained within CHAD
come from multiple surveys with variable study-specific structure (e.g., real time minute-by-
minute recording of diary events versus a recall method using time-block-averaging). Common
4-21
-------
to all studies, individuals provided information on their locations visited and activities performed
for each survey day. Personal attribute data for these surveyed individuals, such as age and
gender, are included in CHAD as well. The latest version of CHAD contains data for nearly
180,000 person-days, however for this assessment, APEX uses about 55,000 of these.28 Most of
the CHAD data are from studies conducted since 2000, several of which are newly included
since the 2009 REA. See Appendix I for a list of the studies available, study dates, and number
of diaries included from each. Three standard APEX input files are used for the current
assessment to create the activity pattern profiles for all simulated individuals.
• Activity diaries events no ATUS BLS.txt: CHAD ID, clock hour (hhmm), duration of
event (minutes), CHAD activity code, and CHAD location code, serving as a daily
sequence of locations visited, activities performed, and their duration for individuals in
CHAD
• Activity diaries questionnaire no ATUS BLS.txt: CHAD ID, day-of-week, sex, race,
employment status, age, maximum daily temperature, average temperature, occupation,
missing time (minutes), record count, commute time (see also section 4.3.2)
• Activity diaries statistics no ATUS BLS.txt. CHAD ID, total daily time spent outdoors
(minutes) (see also section 4.3.4)
4.3.2 Commuting Activity Pattern Data
Exposures can vary across a study area based on spatial heterogeneity in ambient air
concentrations and how that corresponds with a simulated individual's activity pattern and
geographic location. APEX approximates home-to-work commuting flows between census
designated areas for each employed individual, and thus accounts for differing ambient air
concentrations that may occur in these geographic locations. APEX has a national commuting
database originally derived from 2010 Census tract level data collected as part of the U.S. DOT
Census Transportation Planning Package. The data used to generate the APEX commuting file
are from the "Part 3-The Journey to Work" files.29 The Census files contain counts of individuals
commuting from home to work locations at a number of geographic scales. These data have been
28 Data from the U.S. Bureau of Labor Statistics American Time Use Survey (ATUS) are in CHAD master 071113,
but they are not used by APEX in our simulations because of an important survey coding issue. Time spent at
home for ATUS participants was not distinguished as indoors or outdoors, an important distinction for accurately
estimating SO2 exposures. It could be possible to approximate the time expenditure of the ATUS diaries using an
independent source of information, such as using the other CHAD diaries that recorded indoor and outdoor time
(e.g., the 55,000 CHAD diaries used for estimating exposure would be the best source of information). However,
it is unlikely that the representation of time expenditure would change/improve nor would the estimated
exposures differ when including modified ATUS diaries that would reflect the same pattern in indoor and outdoor
time as the 55,000 CHAD diaries already used in our exposure simulations.
29 These data are available from the U.S. DOT Bureau of Transportation Statistics http://transtats.bts.gov/) at the web
site: https://www.transtats.bts.gov/Fields.asp.
4-22
-------
processed to calculate fractions (and hence commute probabilities) for each tract-to-tract flow to
create the national commuting data distributed with APEX. This database contains commuting
data for each of the 50 states and Washington, D.C. This data set does not differentiate people
that work at home from those that commute within their home tract. A companion file to the
commuting flow file is the commuting times file, i.e., an estimate of the usual amount of time in
minutes it takes for commuters to get from home to work each day and tract-to-tract commuting
distances.30 The commuting times file information is used to select CHAD activity pattern data
from individuals having time spent inside vehicles similar to the census commute times and
associated distances travelled. To use these tract level files at the block level for this REA, all
blocks were assumed by APEX to have the same commuting probabilities as the parent tract for
commuting to the blocks within other tracts by using the 11-character identifier common to both
IDs. Intra-block (within a tract) commuting is unknown and thus not simulated. Two standard
APEX input files are used for the current assessment, as listed here.
• CommutingTimesBlock2010 3StudyAreas.txt: census block ID's, count of all employed
individuals, count of employed individuals that do not work at home, 7 groups of block-
level one-way commuting times (in minutes)31
• CommutingJlow US 2010 tracts.txt. census tract IDs, tract-to-tract commute cumulative
probabilities (in fractional form), commute distance (km)
4.3.3 Assigning Activity Pattern Data to Individuals
Once APEX identifies the basic personal attributes of a simulated individual (section 4.1)
and daily temperatures (section 4.2), activity pattern data from CHAD are selected based on
age,32 sex, temperature category, and day of the week. These attributes are considered first-order
attributes in selecting CHAD diaries when modeling human exposures (Graham and McCurdy,
2004). The maximum daily temperature range is used to select activity pattern data that best
match the study area meteorological data for the simulated individual. This information is found
in the following APEX input file, varying by study area and simulation year:
• Functions [studyarea]Y[year].txt: probabilities and interval definitions associated with a
few input variables. For activity diary selection - day of week intervals (weekend or
weekday) by three temperature ranges (<55, 55-83, or >83 °F).
30 These data are from the U.S. Census data portal (http://dafaferrett.censns.gov/') and are found in Table P31,
variables P031001-P031015.
31 The nine commuting time groups in this file are: 0-4, 5-14, 15-19, 20-29, 30-44, 45-59, and >60 minutes.
32 Rather than select using exact ages, APEX allows the user to expand the pool of available diaries using the
variable ' AgeCutpct' which allows for diaries to meet the simulated individuals required age within a certain
percent of that age. A value of 15% was selected (with a default minimum of 1 year). For instance, CHAD diaries
from people ages 51 to 69 would be available to simulate a person aged 60.
4-23
-------
While there may be other important attributes that may influence activity patterns (e.g.,
obesity, disease status), there are limits to our ability to link to all the possible personal attributes
that may be of interest in modeling an individual's activities to the CHAD data. This is largely
because CHAD is a compilation of data collected from numerous individual activity pattern
studies conducted over several decades, many of which had a unique survey design. As a result,
there is a varying amount of missing personal attribute data for the CHAD diaries. For instance,
there are only a limited number of CHAD diaries with survey-requested health information (e.g.,
the health status of respondents). Specifically regarding whether or not a survey participant had
asthma, about 70% of the available diaries used by APEX in this REA had either a 'yes' or 'no'
response to this health condition, of which there were 5,107 diary days representing individuals
having asthma (of which 3,734 were children). This may appear to be a large number of diaries,
however, following a grouping of the diaries by their first-order attributes (i.e., stratifying and
reducing the available data for these diary groups of 'like individuals' by about a factor of 20 or
so), there would be fewer than 200 diaries available for simulating a single day for that particular
individual. Accordingly, the selection of diaries to use for APEX-simulated individuals does not
consider health status (e.g., whether they were for people specifying they did or did not have
asthma, or whether such information was indicated by the survey participant).
This restriction in the number of diaries from individuals having asthma is not considered
to be a significant limitation for estimating exposures for simulated individuals with asthma in
this REA. In general, modeling people with asthma similarly to healthy individuals (i.e., using
the same time-location-activity profiles) is supported by the activity analyses reported by van
Gent et al. (2007) and Santuz et al. (1997). Other researchers, for example, Ford et al. (2003),
have shown significantly lower leisure time activity levels in asthmatics when compared with
individuals who have never had asthma. Based on these conflicting conclusion, we evaluated this
issue in the 2014 O3 REA33 and, using the available activity pattern data in the CHAD database,
we compared participation in afternoon outdoor activities at elevated exertion levels among
people having asthma, people not having asthma, and unknown health status. The 2014 O3 REA
analysis and associated conclusions are described below.
As is of interest in this current SO2 REA, we wanted to focus on instances when
individuals would experience their highest O3 exposures. As has been shown in SO2 and O3
exposure assessments (U.S. EPA, 2009; U.S. EPA, 2014), the highest exposures occur when
33 See 2014 O3 REA sections 5.4.1.5 and 5G-1.4 for details (U.S. EPA, 2014). While there are about 8,300 more
diaries in the CHAD used for this SO2 REA, about 5,800 of the additional diaries added since the 2014 O3 REA
have an unknown health status. Note, the percent of diaries from people with asthma is nearly identical in both
data sets: children with asthma - 20.6% in 2014 O3 CHAD vs 20.3% in 2018 SO2 CHAD; adults with asthma -
7.5% in 2014 O3 CHAD vs 7.5% in 2018 SO2 CHAD. Therefore, rather than generate a new evaluation in this
REA, conclusions drawn from the prior analysis are considered reasonable for this REA.
4-24
-------
individuals spend time outdoors, particularly during the afternoon hours. To prepare the data set
for analysis, afternoon hours were characterized as the time between 12 PM and 8 PM and only
those persons that spent some time outdoors were retained. As is done by APEX in simulating
individuals, level of exertion was estimated by sampling from the specific METS distributions
assigned for each person's activity performed. Then, we identified activities having a METS
value of greater than 3 as times where a person was at moderate or greater exertion levels (US
DHHS, 1999). Afternoon outdoor time was then stratified by exertion level, summed for two
study groups of interest (children and adults), and presented in percent form within Table 4-5.
Of the CHAD diaries for children, about 18% are from an individual with asthma and
69% are from those who do not have asthma. About 5% of CHAD diaries for adults are from
individuals with asthma, and about 65% are from those who do not have asthma. Far fewer
children's diaries are from persons whose asthma status is unknown (12%) compared to adults
(30%>), and the proportions are smaller still in terms of the total available person-days. On
average, about 43% of all children of known asthma status spent some afternoon time outdoors,
and the percent is actually higher for children with asthma (48.5%) than for children not having
asthma (41.2%). About half of the adults whose asthma status was known spent afternoon time
outdoors with a participation rate generally similar for adults having asthma and adults not
having asthma. Participation in outdoor events for persons having unknown asthma status varied
from that of persons with known asthma status; about 60% of the children's diaries with
unknown asthma status and 31% of the adult diaries indicate some afternoon time was spent
outdoors.
Table 4-5. Comparison of outdoor time expenditure and exertion level by asthma status for
children and adult CHAD diaries used by APEX.
CHAD: Children (4to18)a
CHAD: Adults (19 to 95)b
Has Asthma?
Yes
No
Unknown
Yes
No
Unknown
Total Person Days (n)
3,206
12,346
2,128
1,254
15,465
7,075
Number of Person Days with Time
Spent Outdoors (% participation)
1,564
(48.8%)
5,092
(41.2%)
1,267
(59.5%)
602
(48.0%)
7,949
(51.4%)
2,176
(30.8%)
Percent of Afternoon Hours Spent
Outdoors (%)
28.5%
27.5%
28.9%
26.2%
27.2%
22.2%
Percent of Afternoon Time Outdoors at
Moderate or Greater Exertion (%)
80.3%
78.2%
79.2%
62.7%
63.8%
60.3%
From Table 5G-2 of 2014 03 REA (U.S. EPA, 2014)
a CHAD studies for where a survey questionnaire response of whether or not child was asthmatic include CIN, ISR, NHA,
NHW, OAB, and SEA (see Appendix I for study names).
b CHAD studies for where survey a questionnaire response of whether or not adult was asthmatic include CIN, EPA, ISR,
NHA, NHW, NSA, and SEA.
4-25
-------
The amount of time spent outdoors by the persons that did so varied little across the two
study groups and three asthma classifications. On average, diaries from children indicate
approximately 7}A hours of afternoon time is spent outdoors, 80% of which is at a moderate or
greater exertion level, regardless of their asthma status. For individuals whose asthma status is
known, slightly less afternoon time is spent outdoors by adults (about 125-130 minutes) than
children and the percent of afternoon time adults perform moderate or greater exertion level
activities is also lower (about 63%). As noted above regarding the reduced participation in
outdoor events for adults whose asthma status is unknown, diaries for these adults also have
about 20 fewer minutes of afternoon time spent outdoors compared with those persons whose
asthma status is known. Based on this analysis and additional comparisons of CHAD diary days
with literature reported values of outdoor time participation at varying activity levels (see U.S.
EPA, 2014), the 2014 O3 REA evaluation of the CHAD data indicates there are strong
similarities in outdoor time, outdoor event participation, and activity levels among the three
study groups and with those reported in independent studies of people with asthma. Thus, we
conclude the use of any CHAD diary, regardless of asthma status, is reasonable for purposes of
simulating people with asthma in this exposure assessment.
4.3.4 Method for Longitudinal Activity Sequences
In order to estimate population exposure over a full year, a year-long activity sequence
needed to be created for each simulated individual based on CHAD, which is largely a cross-
sectional activity database of 24-hour records. The typical surveyed subject in the time location
activity studies in CHAD provided about two days of diary data. For this reason, the construction
of a season-long activity sequence for each individual requires some combination of repeating
the same data from one subject and using data from multiple subjects. The best approach would
reasonably account for the day-to-day and week-to-week repetition of activities common to
individuals, and recognizing even these diary sequences are not entirely correlated, while
maintaining realistic variability among individuals comprising each study group.
APEX provides three methods of assembling composite diaries. We have selected the
method for this assessment based on our consideration of the assessment objectives,
consideration of an evaluation of differences in results produced by the three methods and
consideration of flexibility provided by each approach with regard to specifying key variable
values. Based on all of these considerations, we have selected the D&A method.
The D&A method is a complex algorithm for assembling longitudinal diaries that
attempts to realistically simulate day-to-day (within-person correlations) and between-person
variation in activity patterns (and thus exposures). This method was designed to capture the
tendency of individuals to repeat activities, based on reproducing realistic variation in a key
4-26
-------
diary variable, which is a user selected function of diary variables. The method targets two
statistics: a population diversity statistic (D) and a within-person autocorrelation statistic (A).
The D statistic reflects the relative importance of within and between-person variance in the key
variable. The A statistic quantifies the lag-one (day-to-day) key variable autocorrelation. Values
of D and A for the key variable are selected by the model user and set in the APEX parameters
file, and the method algorithm constructs longitudinal diaries that preserve these parameters.
Further details regarding this methodology can be found in Glen et al. (2008).
Besides the D&A method, there are two additional methods of compiling diaries
provided by APEX: a more basic method and a similarly complex method. The more basic
method involves randomly selecting an appropriate activity diary for the simulated individual
from the available diary pool. While this more basic method is adequate for providing a mean
short-term exposure estimate, it is less useful for this assessment for which the objective is to
estimate how often individuals may experience particular peak SO2 exposures over a year. The
more complex method uses a Markov-chain clustering (MCC) approach in attempting to recreate
realistic patterns of day-to-day variability. First, cluster analysis is employed to divide the daily
activity pattern records into three groups based on time spent in five microenvironments: indoor-
residence, other indoors, outdoor-near roads, other outdoors, and inside vehicles. For each
simulated individual, a single time-activity record is randomly selected from each cluster. Then
the Markov process determines the probability of a given time-activity pattern occurring on a
given day based on the time-activity pattern of the previous day and cluster-to-cluster transition
probabilities (and are estimated from the available multi-day time-activity records), thus
constructing a long-term sequence for a simulated individual. Details regarding the MCC method
and supporting evaluations are provided in the 2009 REA Appendix B, Attachments 4 and 5.
Che et al. (2014) performed an evaluation of the impact of the three APEX methods on
PM2.5 exposure estimates. As expected, little difference was observed across the methods with
regard to estimates of the mean exposures of simulated individuals. Differences were observed,
however, in the number of multiday exposures exceeding a selected benchmark concentration.
With regard to the number of simulated individuals experiencing 3 or more days above
benchmark concentrations, the MCC method estimates were approximately 12-14% greater than
either the random or D&A methods. For the number of persons experiencing at least one
exposure of concern, however, the MCC method estimates were approximately 4% lower than
those of the other two methods. For additional context, we note that, using all methods, there is
an order of magnitude difference in the number of persons exposed at least once versus three or
more times, indicating that, overall, the occurrence of simulated multiday exposures are rare
events regardless of method selection.
4-27
-------
Che et al. (2014) concludes that while the MCC method produces a higher number of
multiday exposures, there remains a question whether the MCC method has greater accuracy
relative to the other two methods. We note this conclusion applies to both the estimations of
single day and multiday exposures, as there is an inverse relationship between the two when
simulating exposures using APEX and a finite set of activity pattern data. Thus, the MCC
method produces a smaller number of single day exposures above benchmarks relative to the
other two methods, estimations also subject to a degree of uncertainty.
In the absence of having a robust data set (e.g., multiday/week personal exposure
information from a random population) to better evaluate the accuracy of any of the methods, we
considered selection of the longitudinal approach for this assessment from a practical
perspective, guided by a balancing of the single day and multiday exposures that can be
estimated by each method. In so doing, we selected the D&A approach, recognizing that the
D&A method allows for flexibility in the selection of the key influential variable and its setting
values, and also the ability to directly observe the impact of changes to these values on model
outputs.
The key variable selected for this REA is the amount of time an individual spends
outdoors each day, as that is one of the most important determinants of exposure to high levels of
SO2 (see section 2.1.2 above). In their evaluation, Che et al. (2014) varied the values of D and A
for this variable to determine the impact to estimated exposures. Compared to the base level
simulation (i.e., D=0.19 and A=0.22),34 increasing both D and A by 100% increased the number
of persons having at least three exposures above the selected benchmark by about 4%, while also
reducing the percent of persons experiencing at least one day above benchmarks by less than 1%
(Che et al., 2014). In recognizing uncertainty in the parameterization of D and A (i.e., based on a
limited field study of a small subset of the population, children 7-12) and that the base level
simulation D&A values produced a lower estimate of repeated exposures compared with the
MCC method, we have used values of 0.38 for D and 0.44 for A for all ages to potentially
increase representation of multiday exposures without significant reducing the percent of the
population experiencing at least one day at or above benchmark concentrations.
4.4 MICROENVIRONMENTAL CONCENTRATIONS
In APEX, exposure of simulated individuals occurs in microenvironments. To best
estimate personal exposures, it is important to maintain the spatial and temporal sequence of
microenvironments people inhabit and appropriately represent the time series of concentrations
34 Longitudinal diary data from a limited field study of children ages 7-12 (Geyh et al. 2000; Xue et al. 2004)
provide support for estimates of approximately 0.19 for D and 0.22 for A for the amount of time spent outdoors.
4-28
-------
that occur within them. Two methods available in APEX for calculating pollutant concentrations
within microenvironments are a mass balance model and a transfer factor approach. In both
approaches, ME concentrations depend on the ambient (outdoor) air SO2 concentrations and
temperatures, as well as distributions of the key parameters for each approach. Further, the
distributions of some of the key parameters depend on values of other variables in the model. For
example, the distribution of air exchange rates inside an individual's residence depends on the
type of heating and air conditioning present, which are also stochastic inputs to the model. The
value of a stochastic parameter can be set as a constant for the entire simulation (e.g., house
volume would remain identical throughout the exposure period), or APEX can be directed to
sample a new value hourly, daily, or seasonally from specified distributions. APEX also allows
the user to specify diurnal, weekly, or seasonal patterns for certain ME parameters.
Based on findings from the 2009 REA, we have specified five MEs for use in this
assessment, largely based on two factors: the expectation of an ME having exposure
concentrations of interest and the availability of factors to reasonably model the ME. The 2009
REA results indicated that the majority (70-90%) of 5-minute daily maximum SO2 exposures
between 100 and 800 ppb35 occurred while individuals were within outdoor microenvironments
(2009 REA, Figure 8-21). Given that finding and the objective for this assessment (i.e.,
understanding how often and where short-term peak SO2 exposures occur), we recognized the
added efficiency of minimizing the number of MEs, particularly lower-exposure indoor MEs,
that were parameterized and included in the modeling.
Accordingly, we aggregated the number of MEs to address exposures of ambient air
origin that occur within a core group of indoor, outdoor, and vehicle MEs. It was expected that
the exposures occurring near roads would also be associated with high exposures as they would
be modeled identically to all of the other outdoor MEs, only that these outdoor events occur near
a road. Thus, the near road ME was modeled separately in case time spent in that ME and its
associated exposures was of specific interest. An inside-vehicle ME was also modeled based on
the expectation that it would lead to some instances of high exposures, particularly considering
the high air exchange rate that occurs inside vehicles while moving and having a limited SO2
decay rate, effectively reflecting similar concentration levels as in outdoor MEs. Two indoor
MEs (indoor-residence and indoor-other) were modeled individually based on having specific air
exchange rate data available for each (4.4.1 and 4.4.3, respectively). The indoor-other ME is
comprised of all non-residential MEs, thus could include workplaces or office buildings, stores
35 Although these results were associated with a different air quality scenario than is evaluated in this REA, the
similarity in the scenario leads us to conclude the results are relevant for judgments made here. Air quality in the
2009 REA results referenced here was adjusted to just meet a 99th percentile 1-hour daily maximum single year
standard level of 150 ppb (U.S. EPA, 2009).
4-29
-------
for shopping, medical offices, and so on. Table 4-6 lists the five microenvironments selected for
this analysis and the exposure calculation method for each. The variables used and their
associated parameters to calculate ME concentrations are summarized in subsequent subsections
below. Details on the calculation of ME concentrations in APEX are presented in Appendix F,
section F.7.
Table 4-6. Microenvironments modeled and calculation method used.
Microenvironment (ME)
APEX ME
Number
Calculation
Method
Variablesa
Indoor - Residence
1
Mass balance
AER & RM
Indoor-Other
2
Mass balance
AER & RM
Outdoor
3
Factors
None
Near-road
4
Factors
None
Vehicle
5
Factors
PE
a AER = air exchange rate, RM = removal rate, PE = fraction of ambient
pollutant entering microenvironment,
None = ME concentration is equal to ambient concentration
The mass balance method, used for the indoor MEs, assumes that an enclosed
microenvironment (e.g., a room within a home) is a single well-mixed volume in which the air
concentration is approximately spatially uniform (Figure 4-3). The concentration of an air
pollutant in such a microenvironment is estimated using (1) inflow of air into the
microenvironment, (2) outflow of air from the microenvironment, (3) removal of a pollutant
from the microenvironment due to deposition, filtration, and chemical degradation, and (4)
emissions from sources of a pollutant inside the microenvironment (not used for this REA).
4-30
-------
Microenvironment
Air
outflow
Air
inflow
Indoor sources
Removal due to:
•Chemical reactions
•Deposition
•Filtration
Figure 4-3. Illustration of the mass balance model used by APEX.
Considering the microenvironment as a well-mixed fixed volume of air, the mass balance
equation for a pollutant in the microenvironment can be written in terms of concentration:
dC(t)
dt
= c — c — c
in out removal
Equation 4-7
where,
C(t) = Concentration in the microenvironment at time t
C m = Rate of change in C(t) due to air entering the microenvironment
C out = Rate of change in C(t) due to air leaving the microenvironment
C removal = Rate of change in C(t) due to all internal removal processes
The method used for the outdoor MEs uses a factors model and is simpler than the mass
balance model. In this method, the value of the concentration in a microenvironment is not
dependent on the concentration during the previous time step. Rather, this model uses the
Equation 4-8 to calculate the concentration in a microenvironment from the user-provided hourly
air quality data:
4-31
-------
Cmean ~ Cambient ^ fproximity ^ fpollutant Equation 4-8
where,
Cmean = Mean concentration over the time step in a microenvironment (ppb)
Cambient= The concentration in the ambient (outdoor) environment (ppb)
fproximity = Proximity factor (unitless)
fpollutant= fraction of ambient pollutant entering microenvironment (unitless)
The five microenvironments were mapped to the 115 CHAD location codes,36 many of
which go beyond the scale of the microenvironmental modeling. The ambient air concentration
used in calculating ME concentration for each event varies temporally and spatially. For
example, commuters (i.e., employed individuals who do not work at home) are assigned to either
their home grid or work grid concentrations, depending on whether the population probabilities
and commuting data base produce either a home or work event. Additionally, depending on the
particular microenvironment (i.e., other than home or work), the mapping of CHAD locations to
the five microenvironments also uses an identifier that designates the relative location in the air
quality surface from which the ambient air concentration (used to calculate the ME
concentration) is selected. For this assessment, such locations would include the blocks from a
simulated individual's home (H), work (W), near work (NW), near home (NH), last (L, either
NH or NW), other (O, average of all), or unknown (U, last ME determined) census block
location. Specific designations are provided in the ME mapping file, with selection based on
known factors and professional judgement. For example, when an individual is in their home, the
ambient concentration in the home block is used to calculate their ME concentration. When the
individual is at work, the block the individual commuted to is used to calculate their ME
concentration. Travel inside vehicles used the ambient concentration data from the block used to
calculate the prior ME concentration. Most other MEs (both indoor and outdoor) used ambient
concentration data selected from near home blocks.
Status attribute variables are also important in estimating ME concentrations, and can
include, but are not limited to, housing type, whether the house has air conditioning, and whether
the car has air conditioning. Because outdoor MEs are expected to contribute the most to an
individuals' highest SO2 exposure (and potential health risk) and the status attribute variables
pertain to indoor MEs, the setting of these particular variables will have limited impact to this
REA's exposure results. In this assessment, a number of temperature ranges are used in selecting
36 The location codes indicate specific MEs that extend beyond simple aggregations of indoor, in-vehicle, and
outdoor locations where people spend time. For example, CHAD has a location code for when individuals spent
time inside their residence while in the kitchen.
4-32
-------
the particular distribution for estimating air exchange rates (AERs). Maximum daily temperature
is also used in diary selection to best match the study area meteorological data for the simulated
individual (Graham and McCurdy, 2004), and air conditioning use prevalence data.
Multiple APEX ME input files (the first and third in the list below), of the same general
format, are used for each study area in the REA. A single ME mapping file is used for all study
areas. These files contain the specific setting of all variables described in this section.
• ME descriptions [studyarea]5MEs.txt: defines ME calculation method, conditional
variables used (e.g., temperature categories - see functions file), distribution type,
distribution parameters (mean, standard deviation, minimum, maximum) used for
estimating AER, decay rates, proximity factors, and PE fraction to estimate
microenvironmental concentrations.
• MicroEnvJMapping CHAD to APEX 5MEs.txt: maps 115 CHAD locations to 5 APEX
simulated microenvironments and assigns block-level ambient concentrations to use for
each location. Contains CHAD location code, CHAD description, APEX ME number,
and ambient concentration location identifier
• Functions [studyarea]Y[year].txt: variables used for selecting air exchange rates (AER)
- air conditioning (A/C) prevalence (home has A/C, does not have A/C) by five
temperature ranges for air exchange rate (<50, 50-67, 68-76, 77-85, or >85 °F). (see
section 4.4.1 and 4.4.2)
4.4.1 Air Exchange Rates for Indoor Residential Microenvironments
Distributions of AERs for the indoor residential MEs were developed previously using
data from several studies. The analysis of these data and the development of most of the
distributions used in the modeling, originally described in detail in U.S. EPA (2007) Appendix A
and recently updated by Cohen et al. (2012), are provided in U.S. EPA (2014) Appendix 5E.
Briefly, these prior analyses indicated that the AER distributions for the residential MEs
depend on the presence or absence of mechanical air conditioning (A/C) and the outdoor
temperature (and a few other variables37 for which sufficient data are not available). Further, the
AER distributions vary across the U.S. study locations,38 such that the selected AER
distributions for the modeled study areas should also depend on these influential factors. For
each combination of air conditioner (A/C) prevalence, U.S. geographic region (and hence
climate zone), and temperature (where data were available), lognormal distributions were fit.
There were a number of limitations in generating study-area-specific AERs, stratified by
temperature range and A/C type. For example, the AER data collected and the distributions
37 For example, there were insufficient data available across the studies to indicate the specific A/C unit type
(central, window, or both), whether windows were closed or open, or whether a mechanical fan was in operation.
38 The studies were conducted in several U.S. cities (e.g., Detroit, Houston, Los Angeles, New York), likely
accounting for AER differences due to local climate and variability in overall housing stock (e.g., types of
residences, year built).
4-33
-------
subsequently derived from them were available only for selected cities that had limited numbers
of samples collected at varying ambient air temperatures, and yet the summary statistics and
comparisons demonstrate that the AER distributions depend upon the city as well as the
temperature range and A/C type. Because specific AER data are not available for the study areas
in this assessment, we used AER data from Cohen et al. (2012) for a city within the same
geographic region as the particular study area, and considering the same temperature ranges on
which the AER distributions were originally based. The AER distributions used for the exposure
modeling are given in Table 4-7 (for residences with A/C) and Table 4-8 (for residences without
AJC). Upper and lower bounds were selected to guard against the generation of extreme AER
values. In general, the AER distributions are highest for the Fall River study area, while the AER
distributions used for the Tulsa study area are lowest. This implies indoor residential exposures
would tend to be greatest for the simulated individuals in the Fall River study area when
compared with the other two study areas, though again, the expectation is that outdoor exposures
would contribute to the highest exposures in all three study areas and limit the importance of
these observed differences in AER.
Table 4-7. AERs for indoor residential microenvironments (ME-1) with A/C by study area
and temperature.
Study Area
Daily Mean
Temperature (°C)
Lognormal Distribution
{GM, GSD, min, max}
Original AER Study Data Used
< 10
{0.711,2.108, 0.1, 10}
Fall River, MA
10-25
>25
{1.139, 2.677, 0.1, 10}
{1.244, 2.177, 0.1, 10}
New York, NY
< 10
{0.744, 1.982, 0.1, 10}
Indianapolis, IN
10-20
20-25
>25
{0.811,2.653, 0.1, 10}
{0.785, 2.817, 0.1, 10}
{0.916, 2.671,0.1, 10}
Detroit, Ml and New York, NY
<20
{0.407, 2.113, 0.1, 10}
Tulsa, OK
20-25
25-30
>30
{0.467, 1.938, 0.1, 10}
{0.422, 2.258, 0.1, 10}
{0.499, 1.717, 0.1, 10}
Houston, TX
4-34
-------
Table 4-8. AERs for indoor residential microenvironments (ME-1) without A/C by study
area and temperature.
. . Daily Mean
Temperature (°C)
Lognormal Distribution
{GM, GSD, min, max}
Original AER Study Data Used
< 10
{1.016,
2.138, 0.1,
10}
Fall River, MA
10-20
{0.791,
2.042, 0.1,
10}
New York,
NY
>20
{1.606,
2.119, 0.1,
10}
<0
{1.074,
1.772, 0.1,
10}
0-10
{0.760,
1.747, 0.1,
10}
Indianapolis, IN
10-20
{1.447,
2.950, 0.1,
10}
Detroit, Ml
and New York, NY
20-25
{1.531,
2.472, 0.1,
10}
>25
{1.901,
2.524, 0.1,
10}
Tulsa, OK
< 10
10-20
>20
{0.656,
{0.625,
{0.916,
1.679, 0.1,
2.916, 0.1,
2.451,0.1,
10}
10}
10}
Houston, TX
4.4.2 Air Conditioning Prevalence for Indoor Residential Microenvironments
The selection of an AER distribution is dependent on the presence or absence of A/C. We
assigned this housing attribute to indoor residential microenvironments using A/C prevalence
data from the 2013 American Housing Survey (AHS).39 A/C prevalence (specified in terms of
does or does not have mechanical air conditioning) is distinct from usage (i.e., mechanical air
conditioning is on or off), the latter ultimately represented by values selected from the AER
distribution (relatively lower values would be associated with greater recirculation of indoor air)
and dependent on the daily temperature. The A/C prevalence data were assigned to our study
areas where the AHS data best matched our exposure simulation years (Table 4-9). In all three
study areas, the sum of room unit and central A/C prevalence was used.
Table 4-9. American Housing Survey A/C prevalence from 2013 Current Housing Reports
for selected urban areas.
Study Areaa
Total Occupied
Housing Units
(x1000)
Number of Occupied Housing Units
(x1000)
% of Occupied Housing Units
Central
A/C
>1 Central
A/C
1 Room
Unit
2 Room
Units
3+ Room
Units
Central
A/C
Window
Units
Central &
Window A/C
Fall River, MA
780.3
296.6
20.1
129.6
131.0
146.0
38
52
90
Indianapolis, IN
359.7
319.3
21.5
11.9
14.7
8.4
89
10
99
Tulsa, OK
262.0
233.3
7.1
12.1
6.9
61.2
89
10
99
a Data used were from the 2013 Metropolitan Area using a geography filter of 'not in central cities'. Because there were no data for the
study areas data reported for nearby cites was used as follows: Fall River, MA - Boston, MA; Indianapolis - Louisville, KY; Tulsa, OK -
Oklahoma City OK.
39 Available at https://www.censns.gov/programs-snrveYs/ahs/data/interactive/ahstablecreator.htniL
4-35
-------
4.4.3 AER Distributions for All Other Indoor Microenvironments
To estimate AER distributions for all non-residential, indoor environments (e.g., offices,
libraries, schools, etc.), we relied on data generated as part of the U.S. EPA Building Assessment
Survey and Evaluation (BASE) study (Persily and Gorfain, 2004; Persily et al., 2005), as was
also done for the 2009 REA and REAs for other recent NAAQS reviews (e.g., U.S. EPA, 2014).
In the BASE study, a total of 390 AER measurements were collected from 96 randomly selected
office buildings throughout the U.S. using two methods, a volumetric and a carbon dioxide ratio
method. In the vast majority of cases, the reported best estimate was generated using the
volumetric method. The AER values for each office space were averaged, rather than using the
individual measurements, because of the limited degree of variability in AER measurements for
the same office space over a relatively short sampling period. We fitted exponential, lognormal,
normal, and Weibull distributions to the 96 office space average AER values, and the best fitting
of these was the lognormal. The fitted parameters for this distribution are a geometric mean of
1.109, geometric standard deviation of 3.015, and bounded by the lower and upper values of the
sample data set {0.07, 13.8}.
4.4.4 Removal Rate for Indoor Microenvironments
To estimate pollutant removal rates from air within indoor microenvironments, we first
evaluated the removal rates that had been estimated in the 2009 REA using data collected by
Grontoft and Raychaudhuri (2004) on SO2 deposition to a variety of building material surfaces
under differing conditions of relative humidity. In the 2009 REA, this information was used to
derive estimates for five indoor microenvironments: residences, office buildings, schools,
restaurants, and other buildings (see 2009 REA Appendix B section 4.1). For the current REA,
we simulated only two indoor microenvironments: residences and an aggregate ME representing
all other indoor microenvironments. Therefore, we used the same removal rates that were
derived for the 2009 REA for the residential ME and aggregated the estimated removal rates
from the other four indoor MEs as follows. One thousand values were randomly sampled from
the geometric means and standard deviations representing the removal rates for each of the four
indoor MEs. Parameters describing a lognormal distribution for the new aggregate ME (for other
indoor locations) were calculated using the 4,000 sampled values and are provided in Table 4-10.
4-36
-------
Table 4-10. Parameter estimates of SO2 removal rate distributions in two indoor
microenvironments modeled by APEX.
Microenvironment
Removal (hr1) when Heating or Air
Conditioning in Use
Removal (hr1) when Heating or Air
Conditioning Not in Use
Geometric
Mean
Standard
Deviation
Lower
Limit3
Upper
Limit3
Geometric
Mean
Standard
Deviation
Lower
Limit
Upper
Limit
Indoor Residence
3.14
1.11
2.20
5.34
13.4
1.11
10.3
26.0
Indoor Other
3.32
1.37
1.53
5.07
N/A
N/A
N/A
N/A
a Lower and Upper Limits were approximated by the 10th and 90th percentile values.
b N/A not applicable, assumed to always have mechanical building ventilation in operation.
c From Table B.4-6 of 2009 REA.
d Derived from 4,000 values sampled from removal distributions representing four indoor microenvironments (Table B.4-6 of
2009 REA).
4.4.5 Factor for Estimating In-Vehicle/Near-Road Microenvironmental Concentrations
As was the case for the 2009 REA, there are no SO2 measurement data available to
develop a factor for estimating SO2 concentrations inside vehicles resulting from the ambient air
pollutant entering the microenvironment (and termed PE factor). The ratio of inside-vehicle ME
concentrations to outdoor concentrations is commonly used to develop this PE factor. Thus,
based on the outdoor concentration, one can estimate the inside-vehicle concentrations.
Therefore, as was done for the 2009 REA, the PE factors used were developed from NO2 data
provided in Chan and Chung (2003) and used in the 2008 NO2 REA (U.S. EPA, 2008a). As both
SO2 and NO2 are gaseous, and data for PE factors are not broadly available for other gases, this
was concluded to be a reasonable approach.
We note that pollutant removal rates inside vehicles might be different because SO2 is
more water soluble than NO2, although we could not find removal rate data specific to motor
vehicles. A comparison of indoor residential removal rates used for NO2 (2008 NO2 REA) to that
of the 2009 SO2 REA suggests that there might be greater removal of SO2 within indoor
microenvironments, indicating that use of the same PE factor for SO2 as was used for NO2 could
lead to overestimation of inside-vehicle SO2 concentrations.40 Further, however, the in-vehicle
NO2 measurements on which the in-vehicle-to-outdoor-ratios were based might have included a
small amount of in-vehicle emissions, potentially yielding a discrepancy between effective PE
factors for NO2 and SO2. The additional uncertainty from this influential factor is expected to be
40 NO2 removal rates for the 2010 REA were assumed to range from 1.02 to 1.45 h1, based on six measurements
obtained from a single house provided by Spicer et al. (1993). SO2 removal rates for the 2009 REA were
approximated by using SO2 deposition collected data Grontoft and Raychaudhuri (2004) for a variety of building
material surfaces under differing conditions of relative humidity and configured to five indoor
microenvironments. The lower and upper limits of the removal rates ranged from 1.64 to 5.34 h1.
4-37
-------
small compared to the overall uncertainty implied by using a uniform distribution that assumes
all factors that influence variability and that are not directly accounted for have the same impact.
Chan and Chung (2003) measured inside-vehicle and outdoor NO2 concentrations for
three ventilation conditions: air-recirculation, fresh air intake, and with windows open. Mean in-
vehicle-to-outdoor ratio values ranged from about 0.6 to just over 1.0, with higher values
associated with increased ventilation (i.e., window open). A uniform distribution U{0.6, 1.0}
was selected for the PE factor due to the simplified manner it is applied in this REA. For
example, we could not consider influential characteristics such as use of vehicle ventilation
systems due to the lack of data available to reasonably assign values for each study area.
4.5 ESTIMATING EXPOSURE
Based on the event-specific exposures estimated for each individual as described in the
preceding sections, APEX identifies the occurrence of daily maximum 5-minute SO2 exposures
at or above specific levels, while at or above the target ventilation rate (i.e., an EVR > 22 L/min-
m2). More specifically, this is the count of individuals (with asthma) experiencing a specific
number of days per year (e.g., one or more, two or more, etc.) with exposures at or above
specified 5-minute SO2 concentrations (i.e., falling within bins representing different magnitudes
of exposure) while at elevated ventilation.
The daily maximum 5-minute exposure concentrations (of people with asthma at elevated
ventilation) are binned considering the overall features expected for the distribution of ambient
SO2 concentrations and population-based SO2 exposures. Observed ambient concentrations are
generally lognormally distributed - on average, 1-hour daily maximum concentrations are about
5 ppb, 90th percentile 1-hour daily maximum concentrations are typically below 20 ppb, while
99th and maximum 1-hour daily maximum concentrations can be a factor of 10 to 20 times
higher than the mean (ISA, Table 2-13). It follows that because of this distribution of ambient air
concentrations, it is likely that most simulated individuals will experience low daily maximum
exposures (between 5 and 20 ppb), some will experience a daily maximum exposure between 20
and 100 ppb, while few will experience exposures above 100 ppb. Considering this and the
relationship documented in the controlled human exposure studies between exposure
concentration and percent of individuals estimated to experience a lung function decrement
(section 4.6.2), exposure bins were as follows.
For exposure concentrations below 150 ppb, the exposure bins are set at 10 ppb
increments (e.g., 10-20 ppb, 20-30 ppb, etc.); exposure concentrations at or above 150 and
below 250 ppb are at 20 ppb increments though also including a bin for 200 ppb; and exposure
concentrations at or above 250 are at 50 ppb increments, totaling 29 exposure bins. The smaller
bin increments are used for lower exposure concentrations given the relatively greater number of
4-38
-------
exposure events expected to occur in that range and the desire to reduce potential for
overestimation through use of larger size bins (see REA Planning Document, section 4.2.4.2).
From this we summarize the number of days with maximum exposures within each exposure bin,
such that the exposure model outputs are summarized as (1) counts of people exposed at least
one day per year to a range of short-term peak SO2 concentrations while at or above the target
exertion level, and (2) counts of people experiencing multiple days per year with the maximum
5-minute exposure at or above a particular level while at or above the target exertion level.
4.6 RISK METRICS
Using the population exposure estimates, we derived two types of metrics to characterize
potential population health risk: (1) comparison to benchmark concentrations; and, (2) lung
function risk. As in the last review, these approaches are based on the body of evidence from the
controlled human exposure studies reporting lung function decrements (as measured by changes
in sRaw), as well as changes in other measures of lung function, respiratory symptoms, and
various markers of inflammation, in adult study subjects having asthma. For both approaches,
estimates are developed for two groups of individuals with asthma living in the study areas:
adults with asthma (individuals older than 18 years), and children with asthma (individuals aged
5 to 18 years).
4.6.1 Comparison to Benchmark Concentrations
One of the two types of risk metrics in this assessment is based on comparison of
estimated 5-minute exposures experienced while at an elevated ventilation rate to benchmark
concentrations based on the controlled human exposure studies. In addition to its use in the 2009
SO2 REA, the benchmark approach was used in past NO2 and O3 REAs (e.g., U.S. EPA, 2014),
although ventilation rate does not play a role in the approach for the NO2 REA. For this metric,
the time-series of exposures for each APEX-simulated individual is used to identify the daily
maximum 5-minute SO2 concentrations that occur while at moderate or greater exertion. Based
on all of the instances a daily maximum 5-minute exposure (while at or above the target EVR) is
at or above a benchmark concentration, summaries of the individual-level exposures are
produced and combined to generate a statistic for the simulated at-risk population in each study
area. This statistic indicates the number (and percent) of simulated persons experiencing
exposures at or above the benchmark concentrations, while at moderate or greater exertion.41
41 A 'person-day' metric can be generated, indicating the total number of exceedances across the study area as a
whole, but this metric is less informative for this review. The metric conflates variability in individual exposures,
which can vary widely depending on the occurrence of peak concentrations and the distribution of time spent
outdoors, and from a physiological perspective, creates an uninterpretable aggregate population exposure metric.
4-39
-------
As in the 2009 REA, we have identified a set of benchmark concentrations to represent
"exposures of potential concern" (75 FR 35527, June 22, 2010), 5-minute exposure
concentrations for which there is potential for a respiratory response indicative of some level of
bronchoconstriction to occur in an exposed individual, with the potential and the severity varying
with the magnitude of the benchmark concentration. These levels are derived solely from the
controlled human exposure studies, which can examine the health effects of SO2 in the absence
of copollutants that typically can confound results in epidemiologic analyses; thus, health effects
observed in such controlled studies can confidently be attributed to a defined SO2 exposure level.
Considering this information on variation in SO2 exposures and severity of respiratory
response, as described in the ISA and summarized in section 2.2.3 of the REA Planning
Document, we concluded that it is appropriate, as in the last review, to use four benchmark
concentrations: 100, 200, 300 and 400 ppb. As recognized in the last review, we consider
exposures with respect to the 200 and 400 ppb 5-minute benchmark concentrations to be of
particular interest because: (1) 400 ppb represents the lowest exposure concentration in
controlled human exposure studies where moderate or greater lung function decrements occurred
that were often statistically significant at the group mean level and frequently accompanied by
respiratory symptoms; and (2) 200 ppb is the lowest exposure concentration in controlled human
exposure studies at which moderate or greater lung function decrements were found in some
individuals, although these lung function changes were not statistically significant when
evaluated at the group mean level (75 FR 35527, June 22, 2010) (Table 4-11). Additionally,
analyses of pooled datasets for study subjects with asthma that are responsive to SO2 at
concentrations below 1000 ppb found statistically significant increases in lung function
decrements at 300 ppb (ISA, p. 5-19 to 5-20, 5-153; Johns et al., 2010). The lowest benchmark
concentration (100 ppb) is one half the lowest exposure concentration tested by studies in which
the exposure conditions allowed the study subjects to breathe freely.42 We have included this
benchmark concentration in consideration of the nonzero, albeit low (fewer than 10%),
percentage of subjects with asthma experiencing moderate decrements in lung function at the
200 ppb exposure concentration and the lack of specific study data for some groups of
individuals with asthma, such as primary-school-age children (ages 5 to 11) and those with
severe asthma.43
42 Studies of free-breathing subjects generally make use of small rooms in which the atmosphere is experimentally
controlled such that study subjects are exposed by freely breathing the surrounding air (e.g., Linn et al., 1987).
43 We have considered the evidence with regard to the response of individuals with severe asthma that are not
generally represented in the full set of controlled human exposure studies. There is no evidence to indicate such
individuals would experience moderate or greater lung function decrements at lower SO2 exposure concentrations
than individuals with moderate asthma. With regard to the severity of the response, the limited data that are
4-40
-------
Table 4-11. Responses reported in controlled human exposure studies at a given
benchmark concentration.
Benchmark
Concentration
(ppb)
Responses Reported in Controlled Human Exposure Studiesa
Decrements in Lung Function
Respiratory Symptoms,
Supporting Studies
400
Across studies of exposures at/above this concentration (400-
500 ppb), 13-60% of exposed exercising study subjects with
asthma experienced moderate decrements in lung function,
and 4-40% experienced more severe responsesa bc
"Stronger evidence, with
some statistically significant
increases in respiratory
symptoms" (ISA, Table 5-2)d
300
Across studies of exposure at this concentration, 10-33% of
exposed exercising study subjects with asthma experienced
moderate decrements in lung function, and 0-40%
experienced more severe responsesa ef
"Limited evidence of SO2-
induced increases in
respiratory symptoms in
some people with asthma"
(ISA, Table 5-2)
200
Across studies of exposures at this concentration, 7-9% of
exposed exercising study subjects with asthma experienced
moderate decrements in lung function, and up to 3%
experienced more severe responsesa a
100
This is one half the lowest concentration tested in free-breathing exposure conditionsh
a Drawn from Table 5-2 of the ISA.
b Bronchoconstriction in individuals with asthma is the most sensitive indicator of S02-induced lung function effects and is
characteristic of an asthma attack, and airway hyperresponsiveness (AHR) is a characteristic feature of individuals with
asthma (ISA, section 5.2.1.2). As in the last review, the ISA describes as moderate decrements in lung function that involve at
least a doubling in sRaw or at least a 15% reduction in FEV1; increases in sRaw of 200% or more and FEV1 reductions of
20% or more are indicated as more severe (ISA, section 1.6.1.1 and Table 5-2).
c Linn etal., 1983,1987; Bethel et al., 1983; Roger etal., 1985; Magnussen etal., 1990; Horstman etal., 1986; ISA, Table 5-2.
d Lowest exposure finding both statistically significant lung decrements and respiratory symptoms (2008 ISA, section 3.1.3.1).
e Linn etal., 1988,1990; ISA, Table 5-2.
f Statistically significant increases in lung function decrements in study subjects with asthma that are responsive to SO2 at
concentrations below 1000 ppb (ISA, pp. 5-19 to 5-20; Johns et al 2010).
9 Linn etal., 1983,1987; ISA, Table 5-2.
h Very limited data are available for this exposure concentration from five studies utilizing a mouthpiece to deliver pollutant
concentrations (PA, section 3.2.1.3). In these studies, nasal absorption of SO2 is bypassed during oral breathing, thus
allowing a greater fraction of inhaled SO2 to reach the tracheobronchial airways. As a result, individuals exposed to SO2
through a mouthpiece are likely to experience greater respiratory effects from a comparable SO2 exposure using a free
breathing protocol (ISA, p. 5-23).AIthough few of these studies included an exposure to clean air while exercising that would
have allowed for determining the effect of SO2 versus that of exercise in causing bronchoconstriction, in those cases, the
magnitudes of change in affected subjects appeared to be smaller than responses reported from studies at 200 ppb or more,
with none indicating as much as a doubling in sRaw (PA, section 3.2.1.3).
available indicate a similar magnitude SCh-specific response (in sRaw) as that for individuals with less severe
asthma, although the individuals with more severe asthma are indicated to have a greater response to exercise
prior to SO2 exposure, indicating that those individuals "may have more limited reserve to deal with an insult
compared with individuals with mild asthma" (ISA, p. 5-22).
4-41
-------
4.6.2 Lung Function Risk
For lung function risk, we have focused on estimating the risk of experiencing SO2-
related increases in sRaw44 that correspond to moderate decrements in lung function as described
in the ISA.45 The assessment estimates the number of people (and percent of the population)
expected to experience such a decrement and the total number of occurrences of these effects per
individual across the simulation period. Results include the number of people (and percent of
population) estimated to experience at least one such decrement in a year and the number
estimated to experience multiple decrements. Estimates are generated for each of two lung
function response definitions: an increase in sRaw by at least 100% (A sRaw > 100%), and an
sRaw increase of at least 200% (A sRaw > 200%). These measures of lung function risk are
derived from the E-R function (discussed below) and the number of exposures (concomitant with
moderate or greater exertion) among the population that are at or above each of a set of exposure
concentrations estimated from the exposure modeling.
The E-R function is based on the controlled human exposure studies of decrements in
lung function experienced by exercising individuals exposed to a range of 5-minute SO2
concentrations. Table 4-12 presents all study summary data for changes in sRaw from all
references from which individual study data are available (ISA, Table 5-2). Because the health
response variable is binary, a generalized linear model (GLiM) was used to construct the E-R
function (SAS, 2017), represented by the following
g([f) = /?0 + (3-lX Equation 4-9
Briefly explained, one important feature of GLiM is the function (g) used to link the
structural component (i.e., the standard portion of a linear model, fio + fhX) to the mean of a
conditional response distribution (ju). There are several types of link functions to use in fitting
these regression models, the selection of which is generally guided by the empirical fit of the
data, practical considerations, and knowledge of the form of the response distribution.
Two link functions (i.e., probit and logistic) were evaluated for developing the E-R
function used in the 2009 REA (2009 REA, section 9.2 and Appendix C). Both functions are
symmetrical, yielded similar model fits and had nearly similar functional shape, indicating either
44 Although risk of lung function decrements in terms of both FEVi and sRaw were estimated in the REA for the last
review, the risk related to increases in sRaw, a direct indicator of bronchoconstriction for which data are available
across a more extensive set of exposure concentrations than FEVi, was given greater emphasis and is the focus
here.
45 The ISA describes a doubling in sRaw (or a 15% reduction in FEVi) to be a moderate lung function decrement
(ISA, p. 1-17).
4-42
-------
link function could be used to approximate risk. However, as is commonly observed for logistic
functions, the lower and upper tails tend to be flatter when compared to a probit function, that is,
the approach of the curve towards the horizontal axis is more gradual (see Figures 9-2 to 9-5 of
2009 REA). It followed that, when using the logistic function, the estimated health risk differed
by a factor of 2 to 3 from that estimated using the probit function (see Tables 9-5 and 9-8 of 2009
REA). These differences in estimated risk were largely the result of combining the slightly
higher probability of risk at low exposure concentrations when using the logistic function
combined with the large number of simulated individuals having low 5-minute exposures; this is
particularly the case for exposures less than 100 ppb (2009 REA, Figures 9-7 and 9-8). Thus, the
probit model reduces the contribution of these exposures to risk estimates, for which E-R
information is lacking, relative to that provide by the logistic function, thus better addressing the
uncertainty in the E-R extrapolation to such low concentrations that appears magnified when
using a logistic function. Further, as noted by the CASAC comments on a draft of the 2009 REA,
assumptions regarding the distribution of individual thresholds for response support use of a
probit function, which is based on the inverse of the cumulative standard normal distribution
function, rather than a logistic function which assumes a logistic distribution, for estimating risk
associated with population-based SO2 exposures (Samet, 2009, pp. 14 and 60-63).
Based on the above factors, we used a probit model for this risk analysis as in the 2009
REA.46 We used all of the data available47 to fit the two separate E-R functions (for AsRaw >
100% and AsRaw > 200%), generating both the best fit regression as well as using variability
associated with the predicted regression coefficients to provide lower and upper bounds of the
risk estimation. To illustrate the E-R relationship indicated by these data, the percent of the study
populations experiencing increases in sRaw is plotted in Figure 4-4. Further details regarding the
E-R function, its application, and the interpretation of the estimated risk is provided below.
46 The SAS procedure, PROC LOGISTIC, is used to fit the discrete response data by the method of maximum
likelihood and using link=probit model option (SAS, 2017).
47 As mentioned in the REA Planning Document, the concentration levels included in the regression can influence
the model fit, in particular the area of particular interest in this REA (low concentration related predicted
responses).
4-43
-------
Table 4-12. Summary of controlled human exposure studies containing individual response
data: number and percent of exercising individuals with asthma who
experienced greater than or equal to a 100 or 200 percent increase in specific
airway resistance (sRaw), adjusted for effects of exercise in clean air.
so2
(ppb)
Exposure
Ventil-
sRaw
sRaw
sRaw
sRaw
duration
N
ation
>100
>200
>100
>200
Reference
(minutes)
(l/min)
(N)
(N)
(%)
(%)
200
5
23
OO
l
2
0
8.7%
0.0%
Linn et al. (1983)a
200
10
40
O
l
3
1
7.5%
2.5%
Linnet al. (1987)b
250
5
19
-50-60
6
3
31.6%
15.8%
Bethel etal. (1985)
250
5
9
-80-90
2
0
22.2%
0.0%
Bethel etal. (1985)
250
10
27
-42
0
0
0.0%
0.0%
Horstman et al. (1986)a
250
10
28
O
l
1
0
3.6%
0.0%
Roger et al. (1985)
300
10
20
-50
2
1
10.0%
5.0%
Linn et al. (1988)
300
10
21
-50
7
2
33.3%
9.5%
Linn et al. (1990)
400
5
23
OO
l
3
1
13.0%
4.3%
Linnet al. (1983)a
400
10
40
0
l
9.5
3.5
23.8%
8.8%
Linnet al. (1987)b
500
5
10
-50-60
6
4
60.0%
40.0%
Bethel etal. (1983)
500
10
27
-42
6
1
22.2%
3.7%
Horstman et al. (1986)a
500
10
28
O
l
5
1
17.9%
3.6%
Roger et al. (1985)
600
5
23
CO
l
9
6
39.1%
26.1%
Linnet al. (1983)a
600
10
40
0
l
13.5
9.5
33.8%
23.8%
Linnet al. (1987)b
600
10
20
-50
12
7
60.0%
35.0%
Linn et al. (1988)
600
10
21
-50
13
6
61.9%
28.6%
Linn et al. (1990)
1000
10
10
O
l
6
2
60.0%
20.0%
Kehrl et al. (1987)
1000
10
28
O
l
14
7
50.0%
25.0%
Roger et al. (1985)
1000
10
27
-42
15
7
55.6%
25.9%
Horstman et al. (1986)a
Data presented are from all studies from which individual data were available (ISA Table 5-2 and Figure 5-1) on percentage of
individuals who experienced greater than or equal to a 100 or 200% increase in specific airway resistance (sRaw). Lung
function decrements are adjusted for the effects of exercise in clean air (calculated as the difference between the percent
change relative to baseline with exercise|S02 and the percent change relative to baseline with exercise|clean air).
a Data were not available for use in developing the E-R function for the 2009 SO2 REA.
b Responses of mild and moderate asthmatics reported in Linn et al. (1987) are the average of the first and second round
exposure responses following the first 10 min period of exercise.
4-44
-------
100%
90%
0s
O
o
80%
70%
60%
s
c
o
50%
40%
Z3
Q_
O
CL
30%
<4—
o
4—'
c
100% (top panel) and
s Raw > 200% (bottom panel) using controlled human exposure study data
(Table 4-12) fit using a probit regression (solid lines). Dashed lines indicate a 90
percent confidence interval for the mean response.
4-45
-------
The intent of the REA approach described in this section is to calculate population risk,
not individual risk. Thus, it is considered appropriate to focus on the mean response calculated
from the limited number of subjects in the collection of independently performed controlled
human exposure studies. Using the study subject data, we approximated a best fit function that
represents a mean response for any daily maximum 5-minute exposure concentration and derived
a confidence interval for it in order to present a range of estimated population response. We
applied the best fit function to the exposures estimated for the entire simulated population,
assuming the simulated at-risk population (people with asthma in the three study areas) is
comprised of individuals that have a similar response frequency as the controlled human
exposure study test subjects. The 90% confidence interval for the mean response was used to
approximate lower and upper bounds of the E-R function and used estimate lower and upper
bounds of the population risk as part of the uncertainty characterization (section 6.2.2.4). Given
the objective of estimating risk associated with population exposures, this confidence interval is
considered a reasonable approach for estimating the range of the estimated population risk.
An alternative approach for developing a range for the estimated risk might be a
prediction interval that incorporates the spread of the individual study responses at each exposure
concentration.48 We concluded that such an approach - that can be biased by one particular study
having responses outside of the curves representing the 90% confidence interval (e.g., see Figure
4-4) - would not provide an appropriate representation of population risk for sRaw responses.
Given the wide-ranging responses in the individual studies, the very small number of subjects
tested in each study leads us to conclude that such an approach that emphasized one or a few
individual studies would be less likely to represent the response frequency for the entire
population of people with asthma in each study area.
Using the exposure model counts of individuals with daily maximum 5-minute
concentrations falling into the different bins (as described in section 4.5 above), the number of
occurrences of lung function response is calculated by multiplying the number of exposures in an
exposure bin by the response probability (given by the probit E-R function for the specified
definition of lung function response) associated with the midpoint of that bin. Provided in Table
4-13 are single-year exposure estimates for children with asthma in the Fall River study area to
demonstrate this calculation.
48 In general, a prediction interval for a regression is useful in estimating a random future value of the dependent
variable (y), while a confidence interval is useful in estimating the average (or expected) value of y variable given
the same value of the independent variable (x).
4-46
-------
Table 4-13. Example of risk calculation using estimated daily maximum 5-minute
exposures of children with asthma in the Fall River study area.
Exposure Results
ER Function
Estimated Risk
5-minute SO2
Exposure Bina
(ppb)
Number of
Children b
(n)
Bin
Midpoint
(ppb)
Calculated Response
(fraction of population)
Number of Individuals
Responding c
(n)
0
44
5
2.49E-07
0
10
103
15
4.02E-05
0
20
149
25
2.92E-04
0
30
143
35
9.45E-04
0
40
190
45
2.12E-03
0
50
249
55
3.90E-03
0
60
345
65
6.28E-03
2
70
346
75
9.26E-03
3
80
477
85
1.28E-02
6
90
396
95
1.69E-02
6
100
379
105
2.15E-02
8
110
271
115
2.66E-02
7
120
196
125
3.21E-02
6
130
149
135
3.80E-02
5
140
70
145
4.41E-02
3
150
75
160
5.40E-02
4
170
36
180
6.80E-02
2
190
8
195
7.90E-02
0
200
5
205
8.65E-02
0
210
3
220
9.80E-02
0
230
0
240
1.14E-01
0
250
0
275
1.42E-01
0
300
0
325
1.82E-01
0
350
0
375
2.22E-01
0
400
0
425
2.60E-01
0
Total
3633
52
a The exposure bin includes daily maximum 5-minute exposures of at least that value, but less than that of the
next exposure bin.
b This is the number of children with asthma experiencing the exposure while at moderate or greater exertion. In
the Fall River study area, the total population of children with asthma is 3,641.
c Multiplying number of children by the calculated response, then rounded down to the nearest integer, gives the
number of individuals responding.
For example, the midpoint of the 10-20 ppb bin is 15 ppb (Table 4-13). The 15 ppb
exposure bin contains a total of 103 individuals who experienced a daily maximum 5-minute
concentration in the simulated year of at least 10 ppb, but less than 20 ppb. The
frequency/probability obtained from the probit function at 15 ppb (i.e., 4.02 E-05) is then used to
4-47
-------
estimate the number of the 103 persons that respond. To avoid accounting for and summing
(numerically calculated) fractions of people, all risk estimates obtained by combining the number
of individuals with the percent responding within each bin (i.e., the count of individuals
responding) are truncated at the integer level. Therefore, in the Table 4-13 example for the 10-20
ppb bin, the number of individuals estimated to experience a response (i.e., 0.004 persons) is
zero. After calculating the number of whole individuals estimated to respond in each bin, these
are summed to generate the total estimated population risk (i.e., 52 children with asthma). Thus,
1.4% of children with asthma (52 divided by 3641) are estimated to experience at least one day
in the simulated year with an sRaw increase of 100% or more) as a result of their daily maximum
5-minute SO2 exposure.
Additionally, the contribution to risk estimates from each exposure bin is developed
based on the apportionment of the risk estimates to the exposure bins. In this example, nearly
90% of the estimated risk is attributed to 5-minute concentrations at or above 50 ppb and less
than 150 ppb. No children were estimated to experience a response at a 5-minute concentration
below 50 ppb.49
4.7 APPROACH FOR CHARACTERIZING UNCERTAINTY AND
VARIABILITY
An important issue associated with any population exposure and risk assessment is the
assessment of variability and characterization of uncertainty. Variability refers to the inherent
heterogeneity in a population or variable of interest (e.g., residential air exchange rates). The
degree of variability cannot be reduced through further research, only better characterized with
additional measurement. Uncertainty refers to the lack of knowledge regarding the values of
model input variables (i.e., parameter uncertainty), the physical systems or relationships used
(i.e., use of input variables to estimate exposure or risk or model uncertainty), and in specifying
the scenario that is consistent with purpose of the assessment (i.e., scenario uncertainty).
Uncertainty is, ideally, reduced to the maximum extent possible through improved measurement
of key parameters and iterative model refinement. The following two sections describe the
approaches we have used to assess variability (section 4.7.1) and to characterize uncertainty
(section 4.7.2) in this REA. The primary outcome is a summary of variability and uncertainty
evaluations conducted to date of our SO2 exposure assessments and APEX exposure modeling,
and the identification of the elements or areas of the assessment with which is associated the
greatest uncertainty.
49 Thus, it can be observed that in this example for this study area and air quality scenario, 50 ppb represents a
limiting value for the response function among the lowest exposure level bins. Such values would be expected to
differ with population and air quality scenario characteristics.
4-48
-------
4.7.1 Assessment of Variability and Co-variability
The goal in addressing variability in the REA is to ensure that the estimates of exposure
and risk reflect the variability of SO2 concentrations in ambient air, population characteristics,
associated SO2 exposures, physiological characteristics of simulated individuals, and potential
health risk across the study areas and for the simulated at-risk populations. In the REA, there are
several algorithms that are used to account for variability when generating the two risk metrics.
For example, variability may arise from differences in the population residing within census
tracts (e.g., age distribution) and the activities that may affect population exposure to SO2 (e.g.,
time spent outdoors, performing moderate or greater exertion level activities outdoors). The
range of exposure and associated risk estimates are intended to reflect such sources of variability,
although we note that the range of values obtained reflects the input parameters, algorithms, and
modeling system used, and may not necessarily reflect the complete range of the true exposure or
risk values.
We note also that correlations and non-linear relationships between variables input to the
model can result in the model producing inaccurate results if the inherent relationships between
these variables are not preserved. APEX is designed to account for co-variability, or linear and
nonlinear correlation among the model inputs, provided that enough is known about these
relationships to specify them. This is accomplished by providing inputs that enable the
correlation to be modeled explicitly within APEX. For example, there is a non-linear relationship
between the outdoor temperature and air exchange rate in homes. One factor that contributes to
this non-linear relationship is that windows tend to be closed more often when temperatures are
at either low or high extremes than when temperatures are moderate. This relationship is
explicitly modeled in APEX by specifying different probability distributions of air exchange
rates for different ambient air temperatures.
Important sources of the variability and co-variability accounted for by APEX and used
for this SO2 exposure analysis have been identified and summarized in section 6.1. Where
possible, we identified and incorporated the observed variability in input data sets rather than
employing standard default assumptions and/or using point estimates to describe model inputs.
4.7.2 Characterization of Uncertainty
While it may be possible to capture a range of exposure or risk values by accounting for
variability inherent to influential factors, the true exposure or risk for any given individual within
a study area may be unknown, although it can be estimated. To characterize health risks,
exposure and risk assessors commonly use an iterative process of gathering data, developing
models, and estimating exposures and risks, given the goals of the assessment, scale of the
assessment performed, and limitations of the input data available. However, significant
4-49
-------
uncertainty often remains and emphasis is then placed on characterizing the nature of that
uncertainty and its impact on exposure and risk estimates.
In section 6.2, we have summarized the most important uncertainties potentially affecting
the exposure estimates derived for this assessment. In so doing, we recognize that uncertainties
associated with APEX exposure modeling are also characterized in the REAs conducted for
recent reviews of the primary NAAQS for NO2, carbon monoxide, and O3, along with other
pollutant-specific issues (U.S. EPA, 2008a, 2010, 2014). Conclusions drawn from each of these
characterizations are considered in light of new information and of the approaches used in this
REA. Additionally, the new evaluations performed in the current REA have been synthesized
following the approach outlined by WHO (2008) and used to identify, evaluate, and prioritize the
most important uncertainties relevant to the estimated exposure and risk outcomes. The
characterization presented in section 6.2 uses a predominantly qualitative approach
supplemented by various model sensitivity analyses and input data evaluations, all
complementary to quantitative uncertainty characterizations conducted for the 2007 O3 REA by
Langstaff (2007).
The approach used for this REA varies from that described by WHO (2008) in that a
greater focus has been placed on evaluating the direction and the magnitude50 of the uncertainty.
This refers to qualitatively rating how the source of uncertainty, in the presence of alternative
information, may affect the estimated exposures and health risk results. Following the
identification of key uncertainties, we have subjectively scaled the overall impact of the
uncertainty by considering the relationship between the source of uncertainty and the exposure
concentrations (e.g., low, moderate, or high potential impact). Also to the extent possible, we
have included an assessment of the direction of influence, indicating how the source of
uncertainty may be affecting exposure or risk estimates (e.g., the uncertainty could lead to over-
or under-estimates). Further, and consistent with the WHO (2008) guidance, section 6.2
discusses the uncertainty in the knowledge base (e.g., the accuracy of the data used,
acknowledgement of data gaps) and, where possible, particular assessment design decisions (e.g.,
selection of particular model forms). The output of the uncertainty characterization is the
summary in section 6.2 that describes, for each identified source of uncertainty, the magnitude of
the impact and the direction of influence the uncertainty may have on the exposure and risk
characterization results.
50 This is synonymous with the "level of uncertainty" discussed in WHO (2008), section 5.1.2.2.
4-50
-------
REFERENCES
Ainsworth BE, Haskell WL, Herrmann SD, Meckes N, Bassett DR Jr, Tudor-Locke C, Greer JL,
Vezina J, Whitt-Glover MC, Leon AS. (2011). 2011 Compendium of Physical Activities:
a second update of codes and MET values. Med Sci Sports Exerc. 43(8): 1575-81. doi:
10.1249/MSS.0b013e31821ecel2. See also:
https://sites.eoogle.com/site/com.pendiiimofphvsicalactivities/references
Burmaster DE. (1998). LogNormal distributions for skin area as a function of body weight. Risk
Analysis. 18(l):27-32.
Chan AT and Chung MW. (2003). Indoor-outdoor air quality relationships in vehicle: effect of
driving environment and ventilation modes. Atmos Environ. 37:3795-3808.
Che WW, Frey HC, Lau AK. (2014). Assessment of the effect of population and diary sampling
methods on estimation of school-age children exposure to fine particles. Risk Analysis.
34(12):2066-2079.
Cohen J, Mallya H, Rosenbaum A. (2012). Updated Analysis of Air Exchange Rate, ICF
International Memo to John Langstaff Available in Appendix 5E of O3 HREA.
Esmail S, Bhambhani Y, and Brintnell S. (1995). Gender differences in work performance on the
Baltimore therapeutic equipment work simulator. Amer J Occup Therapy. 49:405-411.
Geyh AS, Xue J, Ozkaynak H, Spengler JD. (2000). The Harvard Southern California chronic
ozone exposure study: assessing ozone exposure of grade-school-age children in two
Southern California communities. Environmental Health Perspectives. 108:265-270.
Glen G, Smith L, Isaacs K, McCurdy T, Langstaff J. (2008). A new method of longitudinal diary
assembly for human exposure modeling. J Expo Sci Environ Epidemiol. 18:299-311.
Ford ES, Heath GW, Mannino DM, Redd SC. (2003). Leisure-time physical activity patterns
among U.S. adults with asthma. Chest. 124:432-437.
Graham SE and McCurdy T. (2004). Developing meaningful cohorts for human exposure
models. J Expos Anal Environ Epidemiol. 14(l):23-43.
Graham SE and McCurdy T. (2005). Revised Ventilation Rate (VE) Equations for Use in
Inhalation-Oriented Exposure Models. Appendix A and D of EPA/600/R-06/129F.
Available at: http://cfpub.epa. gov/ncea/cfm/recordisplav.cfm?deid=202543
Grontoft T and Raychaudhuri MR. (2004). Compilation of tables of surface deposition velocities
for O3, NO2 and SO2 to a range of indoor surfaces. Atmos Environ. 38:533-544.
Henry CJK. (2005). Basal metabolic rate studies in humans: measurement and development of
new equations. Public Health Nutrition. 8(7A): 1133-1152.
4-51
-------
IOM. (2005). Dietary Reference Intakes for Energy, Carbohydrate, Fiber, Fat, Fatty Acids,
Cholesterol, Protein and Amino Acids. Food and Nutrition Board. U.S. Institute of
Medicine. National Academies Press. Available at:
http://www.nationalacademies.org/hmd/Activities/Nutrition/SummarvDRIs/DRI-
Tables.aspx. Table: Doubly Labeled Water Data Set: 10_DLW_Database.xls
(downloaded 8/24/16)
Johnson T. (2002). A Guide to Selected Algorithms, Distributions, and Databases Used in
Exposure Models Developed by the Office of Air Quality Planning and Standards.
Revised Draft. Prepared for U.S. Environmental Protection Agency. Available at:
https://www.epa.eov/sites/production/files/2013-08/dociiments/report052202.pdf
Isaacs K and Smith L. (2005). New Values for Physiological Parameters for the Exposure Model
Input File Physiology.txt. Memorandum submitted to the U.S. Environmental Protection
Agency under EPA Contract EP-D-05-065. NERL WA 10. Alion Science and
Technology. Available in the 2009 SO2 REA, Appendix B at:
https://www3. epa. gov/ttn/naaq s/standards/ so2/s so2 cr rea.html
Langstaff JE. (2007). Memorandum to Ozone NAAQS Review Docket (EPA-HQ-OAR-2005-
0172). Analysis of Uncertainty in Ozone Population Exposure Modeling. Docket ID No.
EPA-HQ-OAR-2005-0172-0174.
Linn W, Avol E, Peng R-C, Shamoo D, Hackney J. (1987). Replicated dose-response study of
sulfur dioxide effects in normal, atopic, and asthmatic volunteers. American Review of
Respiratory Disease. 136:1127-1134.
McCurdy T, Glen G, Smith L, Lakkadi Y. (2000). The National Exposure Research Laboratory's
Consolidated Human Activity Database. J Expo Anal Environ Epidemiol. 10:566-578.
Persily A and Gorfain J. (2004). Analysis of Ventilation Data from the U.S. Environmental
Protection Agency Building Assessment Survey and Evaluation (BASE) Study. National
Institute of Standards and Technology, NISTIR 7145, December 2004.
Persily A, Gorfain J, Brunner G. (2005). Ventilation Design and Performance in U.S. Office
Buildings. ASHRAE Journal. April 2005, 30-35.
Samet J. (2009). Letter from Johnathan M. Samet, Chair, Clean Air Scientific Advisory
Committee, to Administrator Lisa P. Jackson. Re: Clean Air Scientific Advisory
Committee's (CASAC) Review of EPA's Risk and Exposure Assessment (REA) to
Support the Review of the SO2 Primary National Ambient Air Quality Standards: Second
Draft. May 18, 2009.
Santuz P, Baraldi E, Filippone M, Zacchello F. (1997). Exercise performance in children with
asthma: is it different from that of healthy controls? Eur Respir J. 10:1254-1260.
SAS. (2017). Base SAS® 9.4 Procedures Guide, Seventh Edition. Available at:
http://dociimentation.sas.eom/api/collections/pgmmvacdc/9.4/docsets/proc/content/proc.p
df?locale=en#nameddest=bookinfo. Also used were SAS/ETS 14.2 User's Guide, edited
4-52
-------
by A. Baxter, E. Huddleston and SAS/QC 14.2 User's Guide, edited by A. Baxter, V.
Clark, E. Huddleston, S. Prabhu, R. Rodriguez, D. Sawyer, J. Simmons.
Schofield WN. (1985). Predicting basal metabolic rate, new standards, and review of previous
work. Hum Nutr Clin Nutr. 39C(S1):5-41.
Spicer CW, Kenny DV, Ward GF, Billick IH. (1993). Transformations, lifetimes, and sources of
NO2, HONO, and HN03 in indoor environments. JAWMA. 43(11): 1479-1485.
U.S. DHHS. (1999). Promoting physical activity: a guide for community action. Department of
Health and Human Services, Centers for Disease Control and Prevention (see Table 2-1).
Available at:
http://www.cdc.gov/nccdphp/dnpa/phvsical/pdf/ tensity table 2 l.pdf
U.S. EPA. (1989). Exposure Factors handbook. Office of Health and Environmental Assessment,
Washington, DC, EPA/600/8-89/043.
U.S. EPA. (1997). Exposure Factors Handbook (Final Report, 1997). Office of Research and
Development, National Center for Environmental Assessment, Washington, DC,
EPA/600/P-95/002F a-c. Available at:
https://cfpub.epa. eov/ncea/risk/recordisplav.cfm?deid=12464
U.S. EPA. (2007). Ozone Population Exposure Analysis for Selected Urban Areas. Office of Air
Quality Planning and Standards, Research Triangle Park, NC, EPA-452-R-07-010, July
2007. Available at: faftp://www.epa.gov/ttn/naaq s/standards/ozone/s o3 cr td.html
U.S. EPA. (2008a). Risk and Exposure Assessment to Support the Review of the NO2 Primary
National Ambient Air Quality Standard. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-08-008a, November 2008. Available at:
https://www3. epa. gov/ttn/naaq s/standards/nox/s nox cr rea.html
U.S. EPA. (2008b). Integrated Science Assessment (ISA) for Sulfur Oxides - Health Criteria
(Final Report). National Center for Environmental Assessment-RTP Division, Office of
Research and Development, Research Triangle Park, NC, EPA-600/R-08/047F,
September 2008. Available at:
http: //cfpub .epa. gov/n cea/cfm/recordi splay, cfm ? dei 143
U.S. EPA. (2009). Risk and Exposure Assessment to Support the Review of the SO2 Primary
National Ambient Air Quality Standard. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-09-007, July 2009. Available at:
https://www3. epa. gov/ttn/naaq s/standards/ so2/data/200908 SQ2REAFinalReport.pdf
U.S. EPA. (2010). Quantitative Risk and Exposure Assessment for Carbon Monoxide -
Amended. Office of Air Quality Planning and Standards, Research Triangle Park, NC,
EPA-452/R-10-006, July 2010. Available at: https://www.epa.gov/naaqs/carbon-
monoxide-co-standards-risk-and-exposure-assessments-current-review
4-53
-------
U.S. EPA. (2014). Health Risk and Exposure Assessment for Ozone. Office of Air Quality
Planning and Standards, Research Triangle Park, NC, EPA-452/R-14-004a, August 2014.
Available at: https://www.epa.gov/naaqs/ozone-o3-standards-risk-and-exposure-
assessments-current-review
U.S. EPA. (2017a). Air Pollutants Exposure Model Documentation (APEX, Version 5)
Volume I: User's Guide. Office of Air Quality Planning and Standards, Research
Triangle Park, NC, January 2017. EPA-452/R-17-001a. Available at:
https://www.epa.gov/fera/apex-user-giiides
U.S. EPA. (2017b). Air Pollutants Exposure Model Documentation (APEX, Version 5)
Volume II: Technical Support Document. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, January 2017. EPA-452/R-17-001b. Available at:
https://www.epa.gov/fera/apex-user-guides
U.S. EPA. (2017c). The Consolidated Human Activity Database - Master Version (CHAD-
Master). Technical Memorandum. Office of Research and Development, National
Exposure Research Laboratory, Research Triangle Park, NC, 27711. In preparation.
Previous version (09/15/2014). Available at:
https://www.epa.gov/healthresearch/consolidated-human-activitv-database-chad-use-
human-exposure-and-health-studies-and
U.S. EPA. (2017d). Integrated Science Assessment (ISA) for Sulfur Oxides - Health Criteria
(Final). National Center for Environmental Assessment-RTP Division, Office of
Research and Development, Research Triangle Park, NC, EPA/600/R-17/451, December
2017. Available at: https://cfpub.epa.gov/n.cea/isa/recordisplav.cfm.?deid=338596
van Gent R, van der Ent K, van Essen-Zandvliet LEM, Rovers MM, Kimpen JLL; de Meer G,
Klijn PHC. (2007). No difference in physical activity in (Un)diagnosed asthma and
healthy controls. Pediatric Pulmonology. 42:1018-1023.
Whitfield R, Biller W, Jusko M, Keisler J. (1996). A Probabilistic Assessment of Health Risks
Associated with Short- and Long-Term Exposure to Tropospheric Ozone. Argonne, IL:
Argonne National Laboratory.
WHO. (2008). WHO/IPCS Harmonization Project Document No. 6. Part 1: Guidance Document
on Characterizing and Communicating Uncertainty in Exposure Assessment.
International Programme on Chemical Safety, World Health Organization, Geneva,
Switzerland. Available at:
http://www.who.int/ipcs/methods/harmonization/areas/exposure/en/
Xue J, McCurdy T, Spengler J, and Ozkaynak H. (2004). Understanding variability in time spent
in selected locations for 7-12 year old children. J Expo Anal Environ Epidemiol. 14:222-
233.
Zahran HS and Bailey C. (2013). Factors associated with asthma prevalence among racial and
ethnic groups - United States, 2009-2010 Behavioral Risk Factor Surveillance System.
Journal of Asthma. 50(6):583-589.
4-54
-------
5 POPULATION EXPOSURE AND RISK RESULTS
Exposure and risk results are presented here for simulated populations residing in the
three study areas - Fall River, MA, Indianapolis, IN, and Tulsa OK - for a three-year air quality
scenario in which air quality conditions just meet the current primary SO2 standard. The
approaches used to link air quality modeling, ambient concentration measurements, exposure
modeling, and controlled human exposure study data in this assessment are summarized in
Figure 2-2. Briefly, and as described in more detail in chapter 3, first AERMOD predicts hourly
SO2 concentrations at air quality receptors within a spatial grid for each study area. Then, the
complete annual temporal pattern of 5-minute continuous ambient monitor concentrations local
to each study area was combined with the AERMOD-predicted 1-hour concentrations to generate
5-minute concentrations at every air quality receptor. As described in Chapter 4, APEX used the
5-minute air quality surface in each study area along with U.S. census block population
demographics to estimate the number of days per year each simulated individual in a particular
study area experiences a daily maximum 5-minute SO2 exposure at or above 5-minute
benchmark levels of 100, 200, 300, and 400 ppb. These short-term exposures were evaluated for
children (5-18 years old) and adults (>18 years old) with asthma when the exposure
corresponded with moderate or greater exertion (i.e., the individual's EVR was >22 L/minute-
m2). And finally, simulated individuals expected to experience a lung function decrement (i.e.,
doubling or larger increase in sRaw) were estimated by linking the population-based daily
maximum 5-minute exposures with an exposure-response function derived from controlled
human exposure study data (section 4.6.2)
Study area characteristics and the composition of the simulated population are provided
in section 5.1. Exposure results are presented in a series of tables that allow for simultaneous
comparison of the exposure and risk metrics across the three study areas and three simulation
years. Two types of results are provided for each modeling domain: (1) the percent of the
simulated subpopulation exposed at or above selected benchmarks, stratified by the number of
occurrences (i.e., days) in a year (section 5.2) and (2) the percent of the simulated subpopulation
experiencing a doubling or larger increase in sRaw, also stratified by the number of days in a
year (section 5.3). Tables summarizing all of the exposure and risk results for each study area,
exposure and response level,1 and simulated at-risk population are provided in Appendix J.
1 As described in section 4.5, exposure model output includes the number and percent of individuals at or above the
benchmark levels and several other exposure levels used for estimating lung function risk.
5-1
-------
5.1 CHARACTERISTICS OF THE SIMULATED POPULATION AND
STUDY AREAS
The three study areas differ in population, geographic size, and demographic features (as
summarized in Table 5-1 and Figures 5-1 through 5-3).2 In each study area, APEX simulated
SO2 exposures for thousands of individuals,3 the demographic features of which were based on
the information associated with the thousands of census blocks within each area (as described in
section 4.1 above).
Asthma prevalence in each modeling domain was estimated based on the NHIS asthma
prevalence data and the demographic characteristics for each study area (e.g., age, sex and family
income) using the methodology summarized in section 4.1.2. Accordingly, the percent of the
simulated populations with asthma within the exposure modeling domain varied by study area
(Table 5-1). The exposure modeling domain for Tulsa had the lowest percent of adults with
asthma (7.2%), while Indianapolis had the lowest percent of children with asthma (9.7%). Fall
River had the highest percent of children with asthma (11.2%), while Indianapolis had the
highest percent of adults with asthma (8.3%). The statistics presented here are the aggregate of
the study area as a whole, within which asthma prevalence varied widely as the modeling
approach fully accounted for the variation in asthma prevalence across census blocks with
demographic factors such as poverty, age, and sex (described in section 4.1.2).4 Nationally,
asthma prevalence is 7.8%; for children it is 8.4% and for adults it is 7.6% (PA, Table 3-2). The
asthma prevalence for children, adults, and the total population estimated for each of the three
study areas are all greater than that of the National asthma prevalence, except for adults in Tulsa
which has a slightly lower asthma prevalence. This suggest that overall, the at-risk population
simulated in the three study areas could represent at-risk populations in other U.S. area that have
a similarly above average asthma prevalence.
2 Specific census block (or tract) identifiers used for the simulations are documented in the APEX 'sites' files for
these simulations.
3 While precisely 30,000 children and 70,000 adults were simulated as part of each APEX model run, the number of
individuals estimated to be exposed are appropriately weighted to reflect the actual population residing within the
census blocks that comprise each respective study area.
4 Representing the variation in asthma prevalence that occurs at the census block level provides a level of resolution
for identification of at-risk individuals that is generally comparable with the resolution of the spatially variable
ambient air concentrations at air quality receptors. In this way, the population in census blocks with higher-
concentration air quality receptors is represented appropriately with regard to asthma prevalence and exposures of
the at-risk individuals with asthma are not under-represented.
5-2
-------
Table 5-1. Summary of study area features and the simulated population.
Study Area
(# census tracts | #
census blocks)
Population
Group
(age range)
Total
Population
Population with
Asthma
% of Population
with Asthma
Fall River
(56 | 4,364)
Children (5-18)
32,424
3,641
11.2%
Adults (19-95)
151,450
12,304
8.1 %
All (5-95)
183,874
15,945
8.7 %
Indianapolis
(172 112,310)
Children (5-18)
112,366
10,851
9.7 %
Adults (19-95)
435,602
36,217
8.3 %
All (5-95)
547,968
47,068
8.6 %
Tulsa
(114 | 7,694)
Children (5-18)
49,482
5,484
11.1 %
Adults (19-95)
207,941
15,049
7.2 %
All (5-95)
257,423
20,533
8.0 %
There are also differences among the study areas with regard to the spatial distribution of
the population (Figures 5-1 to 5-3).5 In the Fall River study area (Figure 5-1), the most highly
populated census tracts (6,000 to 9,000 people per tract) were generally toward the outer edges
of the study area, with the exception of one highly populated tract encompassing the primary
source. Most census tracts in the Fall River study area (86%) had a population of fewer than
6,000 people per tract, with a few tracts (23%) having fewer than 3,000 people per tract. In the
Indianapolis study area (Figure 5-2), most tracts also had fewer than 6,000 people per tract
(84%>), though several tracts had greater than 9,000 people, one of which is located just south of
the collection of modeled emission sources. The census tracts in the Tulsa study area were the
least populated when compared to tract populations in the other two study areas, with all but one
tract having fewer than 6,000 people and nearly 60%> of tracts having fewer than 3,000 people
per tract (Figure 5-3).
5 Data used for these figures were obtained from https://www.censns.gOv/geo/maps-dafa/ciata/gazetteer20.l.0.html.
An identical scale was used for the three figures, progressing by increments of 3,000 people to allow for
appropriate comparisons. For illustrative purposes, census tract population data were used for these maps to better
view the overall population distribution across the study area rather than using census block level data because
many of the smallest sized blocks were not viewable.
5-3
-------
Fall River Study Area (MA/RI)
0 2 4
12
Fall River Study Area
Site
~ Monitor (1004)
A Bray ton (EGU)
2010 Population
0 - 3000
~ 3001 -6000
I I 6001 -9000
| 9001 -12000
| 12001 -15000
¦ 15001 -18000
Figure 5-1. Population in the Fall River study area considering 2010 U.S. Census tracts.
Indianapolis Study Area (IN)
Indianapolis Study Area
Name
H Monitor (0057)
~ Monitor (0073)
~ Monitor (0078)
~ Belmont (VWVTP)
A Citizen (EGU)
I PL-Harding (EGU)
A Quemetco (Lead)
A Rolls Royce (Eng)
A Vert ell us (Chem)
2010 Population
| 0 - 3000
~ 3001 -6000
II || 6001 -9000
| 9001 - 12000
12001 -15000
¦ 15001 - 18000
Figure 5-2. Population in the Indianapolis study area considering 2010 U.S. Census tracts.
5-4
-------
Tulsa Study Area (OK)
Tulsa Study Area
Name
~ Monitor (0175)
¦ Monitor (0235)
PI Monitor (1127)
A East (Refinery)
A West (Refinery)
2010 Population
I I 0 - 3000
3001 -6000
6001 -9000
9001 - 12000
12001 - 15000
¦ 15001 - 18000
Figure 5-3. Population in the Tulsa study area considering 2010 U.S. Census tracts.
5.2 EXPOSURES AT OR ABOVE BENCHMARK CONCENTRATIONS
There were few simulated individuals estimated to experience 5-minute exposures at or
above the three highest benchmark levels (200, 300, and 400 ppb), in any of the study areas
(Tables 5-2 and 5-3). Regarding the two highest benchmarks of 300 ppb and 400 ppb, neither
children nor adults with asthma had any 5-minute exposures at or above these levels in the Fall
River and Tulsa study areas. In the Indi anapolis study area, a small fraction (<1%) of the
simulated population of children with asthma was estimated to experience exposures at or above
300 ppb and 400 ppb in the first year of the 3-year simulation. The relatively few exposures at or
above 300 ppb is consistent with the limited number of occurrences of these high 5-minute
concentrations in the air quality data set (Tables 3-14 to 3-16) for the air quality scenario
modeled.6 The next highest benchmark, 200 ppb, was also rarely exceeded, and when it was, a
daily maximum 5-minute exposure only occurred at or above this level on no more than one day
in the year and for a small fraction of the simulated at-risk population (<0.1% to 1.0%).
6 Air quality was adjusted to just meet the existing standard of 75 ppb, as a 3 -year average of 99th percentile annual
daily maximum 1 -hour concentrations.
5-5
-------
Sensitivity analyses described in section 6.2.2 illustrate some variation from the estimates
presented in this section. The use of alternative exposure model inputs, such as an alternative
approach to adjust ambient concentrations to meet the existing standard and an alternative
method to combine patterns of monitored 5-minute concentrations with modeled receptors can,
in some instances, contribute to somewhat higher estimates. Overall, such differences based on
the use of alternative approaches evaluated are not large and mainly affected estimated exposures
at or above the 100 ppb benchmark (i.e., ranging from no difference to a few percentage points).
Given the findings noted above for the higher benchmark levels, discussion here of
differences across air quality years and simulated populations focuses on the lowest benchmark
level. Regarding this benchmark level (100 ppb), the Tulsa study area did not have more than
0.2% of either simulated at-risk population estimated to experience one or more days with a 5-
minute exposure at or above 100 ppb (Tables 5-2 and Table 5-3). Thus, the discussion here
focuses primarily on the Fall River and Indianapolis study area results.
Across the three years modeled, the highest population exposures were estimated in the
first year. This is seen with the yearly estimates of the percent of the simulated populations
expected to experience one, two or more days with exposures above benchmark levels (Tables 5-
2 and Table 5-3). For example, in considering exposure results for children with asthma having
at least one daily maximum 5-minute exposure at or above 100 ppb in Fall River during the first
year, the percent was 32.7%, while air quality for the subsequent years yielded a lower percent
(13.2%) and 12.3%>, respectively). Such year-to-year variability in the estimated exposures can be
expected given variability in ambient concentrations across sequential years (e.g., Table 3-11 to
Table 3-13), largely resulting from actual variability in emissions and meteorology in the air
quality modeling.7 Year-to-year variability was also observed for the Indianapolis results for 100
ppb, although the range (nine percentage points) was smaller.
Across the three areas, a greater proportion of simulated children with asthma were
estimated to experience exposures at or above benchmark levels compared to adults with asthma.
For example, for the three years in the Fall River study area, as many as 12.3 to 22.1% of
children with asthma were estimated to experience at least one daily maximum 5-minute
exposure at or above 100 ppb, while the range in the percent of adults with asthma exposed was
from 1.3 to 5.1%> (Table 5-2). The number of days per year with exposure above benchmarks
was also greater for children with asthma compared to adults with asthma. For example, no
7 The emissions for the main source in Fall River declined appreciably over the 3-year simulation period, also likely
contributing to the variation observed in the annual exposure estimates. Note, the air quality adjustment used to
create the hypothetical air quality scenario of conditions just meeting the existing standard (with its three-year
form) maintains the year-to-year variability in emissions and meteorology, yielding high and low ambient
concentration years within the 3-year period.
5-6
-------
simulated adults with asthma in the Fall River study area were estimated to have more than three
days in a year with a daily maximum 5-minute exposure at or above 100 ppb, while on average
across the 3-year period, 0.9% of children with asthma were estimated to have four or more days
at or above that benchmark (Table 5-3). Such differences between these two populations are
expected given that higher exposures are more frequent outdoors (see sections 2.1.2 and below)
and that children spend more time outdoors and at a greater frequency compared to adults.
5-7
-------
Table 5-2. Percent and number of children and adults with asthma estimated to
experience at least one day per year with a SO2 exposure at or above 5-minute
benchmark concentrations while breathing at elevated rate, air quality
adjusted to just meet the existing standard.
Study area
Population
group
Benchmark
concentration
Percent (and number) of population with asthma
having at least one day per year when 5-minute
SO2 exposure_>_benchmark
(ppb)
Year 1
Year 2
Year 3
Average
100
32.7
13.2
12.3
19.4
children
(1,192)
(480)
(447)
(706)
200
0.2
0a
0
<0.1 a
Fall River
(8)
(3)
300
No exposures at or above this benchmark
100
5.1
1.9
1.3
2.8
(625)
(229)
(162)
(339)
adults
200
<0.1
(2)
0
0
<0.1
(1)
300
No exposures at or above this benchmark
100
27.0
22.3
18.0
22.4
(2,932)
(2,419)
(1,947)
(2,433)
200
1.0
0
0.9
0.7
children
(112)
(101)
(71)
0.8
(89)
0.3
(30)
300
0
0
400
0.3
0
0
0.1
Indianapolis
(33)
(11)
100
4.3
3.8
2.9
3.7
(1,549)
1,369)
(1,051)
(1,323)
200
0.1
0
0.2
0.1
adults
(43)
(62)
(35)
300
<0.1
(31)
0
0
<0.1
(10)
400
<0.1
(24)
0
0
<0.1
(8)
100
0.2
0.2
<0.1
0.1
children
(13)
(8)
(1)
(7)
Tulsa
200
No exposures at or above this benchmark
100
0.1
<0.1
0
<0.1
adults
(14)
(8)
(7)
200
No exposures at or above this benchmark
a < 0.1 represents nonzero estimates below 0.1%. A zero (0) indicates there were no individuals having the
specified exposure.
5-8
-------
Table 5-3. Percent of children and adults with asthma estimated to experience multiple
days per year with a SO2 exposure at or above 5-minute benchmark
concentrations while breathing at elevated rate, air quality adjusted to just
meet the existing standard.
Benchmark
concentration
(ppb)
Percent of population with asthma having multiple days per year when 5-minute SO2
exposure > benchmarka
Fall River
Indianapolis
Tulsa
>2 days
>4 days
>6 days
>2 days
>4 days
>6 days
>2 days
>4 days
>6 days
Children, aged 5 to 18 years
100
5.5
(1.6-12.2)
0.9
(<0.1b - 2.6)
0.2
(0 - 0.6)
6.8
(4.7 - 8.0)
0.8
(0.3-1.0)
0.1
(<0.1-0.2)
0"
0
0
200
no study area results included multiple days per year at or above this benchmark level
Adults, aged 19 to 95 years
100
0.2
(<0.1-0.4)
0
0
0.5
(0.4-0.6)
<0.1
(0 - <0.1)
<0.1
(0 - <0.1)
0
0
0
200
no study area results included multiple days per year at or above this benchmark level
a These estimates are summarized from the single year data provided in Appendix J. The first value in each cell is the average
across the three years; the range is provided in parentheses.
3 < 0.1 represents nonzero estimates below 0.1 %. Zero (0) indicates there were no individuals having the specified exposure.
We also evaluated the microenvironments where the highest exposures occurred in the
three study areas, as was done for the 2009 REA. With the summary information APEX provides
for the simulated population is the total time spent in each microenvironment and for every
exposure level across the entire simulation period.8 9 We summed the total time that simulated
individuals in the population spent at or above each of the exposure levels and calculated the
percent of this time occurring in each of the microenvironments (Figure 5-4). Consistent with
findings from the 2009 REA, the majority (about 90% or more) of the time that the population is
exposed at or above 100 ppb occurs in outdoor MEs for all three study areas.
8 For all of these APEX simulations, children are the simulated population group, of which a subset are children with
asthma. The APEX ME summary output is directly for that base population (i.e., all children) and cannot be
edited to reflect a subset of that population (e.g., specific age groups of children). Because there are no
modifications made to simulate children with asthma (i.e., all children use the same physiological and activity
pattern data), inferences made regarding exposures for the total population of children in this analysis are
applicable to the subset of simulated children with asthma.
9 This default ME summary output summarizes all exposure time for the entire simulation, thus it reflects instances
where individuals are at any exertion level (e.g., resting, vigorous, etc.). Nevertheless, the presentation here
remains informative in this assessment, particularly considering that it is likely the vigorous exertion level
activities are also linked to particular MEs such as those outdoors.
5-9
-------
Percent of Time Children are Exposed to S02 by
Microenvironment: Fall River
100
90
80
70
(U
| 60
° 50
c
0)
£ 40
v
a.
30
20
10
0
—
ndoor
Outdoor
venicie
^>1 ¦ 1
>¦• « ¦ m
1 1 1 1
150 200 250 300 350
S02 Exposure (ppb)
100
90
80 --
70
i
| 60
! 50
! 40
! 30
20
10
0
Percent of Time Children are Exposed to S02 by
Microenvironment: Indianapolis
-•-Indoor
-¦-Outdoor
-•-Vehicle
¦H
100
150
200 250 300 350
S02 Exposure (ppb)
100
90
60
50
40
30
20
10
0
Percent of Time Children are Exposed to S02 by
Microenvironment: Tulsa
ndoor
Outdoor
venicie
t. . J
i 2d
UJs* 1 •
LA't 1 1
1 1 1 1
150 200 250 300 350
S02 Exposure (ppb)
Figure 5-4. Percent of children's time in indoor, outdoor, and vehicle MEs while exposed to
SO2 in Fall River (top), Indianapolis (middle), and Tulsa study areas.
5-10
-------
5.3 LUNG FUNCTION DECREMENTS ASSOCIATED WITH 5-MINUTE
SO2 EXPOSURES
There were few simulated individuals estimated to experience S02-related increases in
sRaw of at least 100% in any of the three study areas under air quality conditions just meeting
the existing standard (Tables 5-4 and 5-5). We additionally note that as mentioned above for the
benchmark comparisons, sensitivity analyses using alternative exposure model inputs described
in section 6.2.2 indicate that the percent of individuals estimated to experience lung function
decrements of interest can vary from these estimates, although such differences are not large.
Additionally, as discussed in section 5.4 below, differences among the three study areas with
regard to the extent of areas within each study area with higher DVs and greater population
appears to contribute to the relatively higher estimates for the Fall River and Indianapolis study
areas. As recognized in the PA, such exposure circumstances are particularly informative to
consideration of public health protection provided by the current SO2 standard.
In the Fall River and Indianapolis study areas, on average across the three-year period, as
many as 1.3% of children with asthma were estimated to experience at least one day per year
with an SCh-related increase in sRaw of 100% or more; in a single year, the percent is as high as
1.5% (Table 5-4). The percent of children with asthma estimated to experience two or more such
days with an SCh-related increase in sRaw of 100% or more ranged as high as 0.8% in a single
year, while on average across the three years it was as high as 0.7% of children with asthma
(Table 5-5). When considering SCh-related increases in sRaw of 200% or more, on average as
many as 0.3% of children with asthma were estimated to experience this lung function
decrement. The percent of adults estimated to experience lung function decrements was lower
than that of children, due to adults having a lesser amount of time spent outdoors and lower
frequency of outdoor events, leading to lower exposures relative to those estimated for children.
Based on the design of the exposure assessment and how estimated exposures are
summarized for the risk calculation (i.e., use of exposure concentration bins), the number of
individuals falling within each exposure concentration bin is used to derive the number of
individuals estimated to experience the lung function decrement from their daily maximum 5-
minute exposure estimates based on the E-R function (see section 4.6.2). The extent to which
differing magnitudes of exposure concentrations contribute to the total risk estimates in each
year is shown in Table 5-6 for the children with asthma in Fall River and Indianapolis study
areas and days with a S02-related increase in sRaw of 100% or more. The majority (83-100%) of
the simulated individuals estimated to experience at least one day with such a lung function
decrement had their 5-minute daily maximum exposure between 50 and 150 ppb. In all three
study areas, there were no simulated individuals with a S02-related increase in sRaw of 100% or
5-11
-------
more when 5-minute daily maximum exposures were less than 40 ppb, effectively serving as a
threshold for the ER function at this exposure level.
Table 5-4. Percent and number of children and adults with asthma estimated to
experience at least one day per year with a SCh-related increase in sRaw of
100% or more while breathing at an elevated rate, air quality adjusted to just
meet the existing standard.
0. . Population
Study area r
J group
Increase
in sRaw
(0/\
Percent (and number) of population with asthma
having at least one day per year with specified
increase in sRaw
[/OJ
Year 1
Year 2
Year 3
Average
1.4
0.8
0.5
0.9
children
100
(52)
(28)
(20)
(33)
0.2
0.1
<0.1 a
0.1
200
(9)
(5)
(2)
(5)
i an i \iv6i
0.3
0.2
<0.1
0.2
adults
100
(42)
(21)
(9)
(24)
<0.1
<0.1
0a
<0.1
200
(6)
(1)
(2)
100
1.5
1.3
1.1
1.3
(161)
(140)
(121)
(141)
children
200
0.4
0.3
0.3
0.3
(39)
(35)
(29)
(34)
100
0.4
0.4
0.4
0.4
(158)
(147)
(128)
(144)
Indianapolis adults
200
0.1
(36)
<0.1
(32)
<0.1
(26)
<0.1
(31)
100
0
<0.1
<0.1
<0.1
children
(1)
(1)
(1)
200
0
0
0
0
Tulsa
100
<0.1
<0.1
<0.1
<0.1
adults
(1)
(2)
(1)
(1)
200
0
0
0
0
a < 0.1 represents nonzero estimates below 0.1 %. A zero (0) indicates there were no individuals having the
increase in sRaw.
5-12
-------
Table 5-5. Percent of children and adults with asthma estimated to experience multiple
days per year with a SCh-related increase in sRaw of 100% or more while
breathing at elevated rate, air quality adjusted to just meet the existing
standard.
Lung function
decrement
(increase in
sRaw)
Percent (and number) of population with asthma having multiple days per year
with specified increase in_sRawa
Average per year (minimum/year- maximum/year)
Fall River, MA
Indianapolis, IN
Tulsa, OK
# Days
# Days
# Days
>2
>4
>6
>2
>4
>6
IV
ro
IV
IV
0>
Children, aged 5 to 18 years
>100%
0.4
(<0.1b - 0.7)
0.2
(<0.1-0.4)
0.1
(0 - 0.2)
0.7
(0.6 - 0.8)
O O
CO CO
0 0^
no individuals
experiencing multiple
days with this size
increase in sRaw
>200%
<0.1
(0-0.1)
0b
0
0.2
(0.1-0.2)
<0.1
(<0.1)
<0.1
(<0.1)
Adults, aged 19 to 95 years
>100%
<0.1
(0 - <0.1)
0
0
0.2
(0.1-0.2)
<0.1
(<0.1)
<0.1
(<0.1)
no individuals
experiencing multiple
days with this size
increase in sRaw
>200%
no individuals experiencing
multiple days with this size
increase in sRaw
<0.1
(<0.1)
<0.1
(<0.1)
0
a These estimates are summarized from the single year data provided in Appendix J.
b < 0.1 represents nonzero estimates below 0.1%. Zero (0) indicates there were no individuals having the specified
increase in sRaw.
Table 5-6. Contribution of different magnitudes of 5-minute SO2 exposures to lung
function risk (sRaw increase of at least 100%) estimated for children with
asthma in Fall River.
5-minute SO2 exposure
concentration bins
Percent contribution of exposure c
Fall River
Dncentration to total risk estimatea
Indianapolis
Year 1
Year 2
Year 3
Year 1
Year 2
Year 3
0 to <50 ppb
0.0%
0.0%
5.0%
0.6%
0.7%
0.8%
50 to <100 ppb
32.7%
67.9%
45.0%
37.9%
47.1%
51.2%
100 to <150 ppb
55.8%
32.1%
50.0%
45.3%
39.3%
38.8%
150 to <200 ppb
11.5%
0.0%
0.0%
2.5%
12.9%
1.7%
>=200 ppb
0.0%
0.0%
0.0%
13.7%
0.0%
7.4%
a These results are generated from the same data used to estimate the percent of children experiencing at least one day with
an increase in sRaw > 100% provided in Table 5-4.
5-13
-------
5.4 STUDY AREA DIFFERENCES AND POPULATION DISTRIBUTION
To gain a better understanding of the role of two study area characteristics in differences
of exposure estimates among the three study areas, we derived a metric that combines the census
tract population counts (Figures 5-1 to 5-3) with the spatial distribution of the modeled air
quality receptor design values for the air quality scenario assessed (Figures 3-6 to 3-8).10 By
merging the two variables that most influence population-based exposures - ambient
concentrations and number of people - this metric can be used to indicate spatial variability in
exposures and be useful in broadly comparing relative differences across the three study areas.
The first section below summarizes how the exposure metric (labeled DV&POP) was derived
and the subsequent section describes use of the metric in comparing the study areas.
5.4.1 Derivation of DV&POP Metric
First, the air quality receptors used for estimating exposures were identified using the
APEX sites file, a file that contains the IDs for both the air quality receptors and the population
census blocks used in each exposure simulation. This set of IDs was then used to link the air
quality receptor design values with the census tract population data (i.e., the first 11 characters of
the block IDs are the tract IDs). Second, all design values were normalized to the maximum
design value in each study area, creating a set of data ranging in value of 0 to 1, with a value of 1
given to the receptor that had an original design value of 75 ppb. Because there can be multiple
census blocks (and air quality receptors) within each census tract, two new ambient
concentration variables were calculated using this normalized data set - the arithmetic average of
all normalized receptor design values falling within each tract (nDVavg) and the maximum
normalized design value within each tract (nDVmax). This was done to represent the overall
relative concentration in each tract while also recognizing the importance of the upper percentile
concentrations. Third, two new population variables were created. In each study area, the tract
populations were normalized by its own maximum tract population to generate values for the
first population variable (and thus having a value ranging from 0 to 1 for tracts in each study
area). The purpose of this population variable (nPOPmtm) was to discern intra-study area spatial
differences in population. The second population variable was created similarly, though the tract
populations in each study area were normalized using the maximum population in any of the
three study areas (nPOPmter). The Indianapolis study area had the tract with the greatest
population, thus for this study area, the values for this second population variable were identical
to values for the first population variable. However, in the other two study areas, the second
10 Note that the degree of spatial heterogeneity of SO2 concentrations across each study area is a function of the
emission source(s) characteristics (e.g. emission rate, stack height), meteorological conditions (e.g., wind speed),
among other factors.
5-14
-------
population variable had its values relative to that of Indianapolis, thus accounting for inter-study
area differences in population.
These four variables were then combined to represent the final metric (DV&POP) in each
tract and weighted such that its range of values for this new metric extends from 0 to 1, as
follows in equation 5-1:
DV&POP = 0.333 x nDVavg + 0.333 x nDVmax + 0.222 x nPOPintra + 0.112 x nPOPinter
Equation 5-1
We note that the weighting scheme gives more weight to the design value information
(2/3) than the population data (1/3) given the significance of high concentrations for this
exposure assessment. More weight was given to the intra-population variable than to the inter-
population variable to allow for between study area comparisons, but focus more on within study
area population variability.
5.4.2 Comparing the Study Areas with the DV&POP Metric
Census tract values of the DV&POP metric are presented in Figures 5-5 to 5-7 for the
three study areas. A scale of 0 to 1 is used, varying by equivalent increments of 0.2. The highest
values indicate census tracts where the confluence of population and ambient concentrations is
greatest, the lowest values indicate tracts having little influence from either the population and
ambient concentrations. Therefore, of greatest interest are DV&POP values at the upper end of
the scale.
As a general observation, it can be seen that DV&POP values are similar in the Fall River
and Indianapolis study areas and the values for those two areas differ from Tulsa (Figures 5-5
through 5-7). For example, all of the census tracts in the Fall River and Indianapolis study areas
(Figures 5-5 and 5-6) have a DV&POP value greater than 0.4, while 89% of the tracts in Tulsa
have DV&POP values less than 0.4, indicating that most tracts in the Fall River and Indianapolis
study areas have relatively higher design values and/or populations than Tulsa census tracts. This
overall observation using the metric reflects the study area-specific design value and population
information for the three areas. For example, in Fall River, over 70% of receptors had hourly
design values between 31 to 45 ppb (Figure 3-6) and 77% of tracts have a population greater
than 3,000 people (Figure 5-1).11 In Tulsa, by comparison, 83% of receptor sites have hourly
design values less than 30 ppb (Figure 3-8) and 59% of tracts have fewer than 3,000 people
(Figure 5-3).
11 Similarly, in the Indianapolis study area, over 63% of receptors had moderately high hourly design values
between 31 to 45 ppb (Figure 3-7) and 58% of tracts have a population greater than 3,000 people (Figure 5-2).
5-15
-------
Most importantly to the exposure/risk results are tracts having the higher DV&POP
values (e.g., greater than 0.6), as these are most likely locations within the study area where the
highest exposures occur for the greatest number of simulated people. In Fall River, a total of 13
tracts (comprising about 73,000 people) have DV&POP values greater than 0.6, with one of
these encompassing the primary source and having the highest DV&POP value of 0.82 (Figure
5-5). Similarly, in the Indianapolis study area (Figure 5-6), there are a total of 10 tracts
(comprising about 86,000 people) with DV&POP values above 0.6, with the one that
encompasses one of the largest sources (IPL-Harding) having the highest DV&POP value of
0.88. In contrast, the Tulsa study area (Figure 5-7) has only one tract with a DV&POP value
above 0.6, and its value is 0.61.
These broad spatial differences in population size and where that might overlap with
higher DVs likely contribute to the greater number of exposures at or above the benchmark
levels in the Fall River and Indianapolis study areas compared with the Tulsa study area.
Additionally, both Fall River and Indianapolis had a greater spatial extent of air quality receptor
sites with 5-minute ambient air concentrations at or above 100 ppb than the Tulsa study area
(Tables 3-14 to 3-16).
5-16
-------
1
2
Fall River Study Area (MA/RI)
12
16
I Miles
Fall River Study Area
Site
~ Monitor (1004)
~ Bray ton (EGU)
DV& POP
[ I 0 - 0.2
| 0.2-0.4
|04-0 6
0.6 - 0 8
¦ 0.8-1.0
Figure 5-5. Values of the DV&POP exposure metric in the Fall River study area.
5-17
-------
Indianapolis Study Area (IN)
12
16
I M iles
1
2
Indianapolis Study Area
Name
~
Monitor (0057)
~
Monitor (0073)
~
Monitor (0078)
~
Belmont (WWTP)
A
Citizen (EGU)
1 PL-Harding (EGU)
~
Quemetco (Lead)
Rolls Royce (Eng)
~
Vertellus (Chem)
DV&
POP
0-0.2
0.2-0.4
II
0.4-0.6
0.6-0.8
0.8 - 1.0
IH
Figure 5-6. Values of the DV&POP exposure metric in the Indianapolis study area.
5-18
-------
1
2
Tulsa Study Area (OK)
i,
12
16
I Miles
Tulsa Study Area
Name
IH Monitor (0175)
| Monitor (0235)
~ Monitor (1127)
~ East (Refinery)
West (Refinery)
DV & POP
I | 0 - 0.2
0.2 - 0 4
0.4 - 0.6
0.6-0.8
0 8 -1.0
Figure 5-7. Values of the DV&POP exposure metric in the Tulsa study area.
5-19
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
5.5 COMPARISON WITH 2009 REA RESULTS
The results presented in this chapter and discussed above provide estimates for air quality
conditions associated with just meeting the now-current 1-hour standard of 75 ppb (evaluated as
3-year average of annual 99th percentiles), an air quality scenario that was not included in the
2009 REA. As summarized in section 1.2 above, the 2009 REA included single-year air quality
scenarios for 99th percentile levels of 50 ppb and 100 ppb in two study areas (St. Louis and
Greene County, Missouri). For each air quality scenario, the exposure estimates for these two
areas differed, and it is plausible that population and spatial heterogeneity explain those observed
differences, although the type of analysis of these factors discussed in section 5.4 above was not
done in the 2009 REA. Further, while the range of the exposures at or above benchmark levels
estimated here is roughly consistent with the range of estimates in the 2009 REA study areas for
the air quality scenarios bracketing the current standard, there are complications associated with
a direct comparison of these results given the many ways in which these analyses differ from
those available in the last review. In addition to the expansion in the number, type, and
geographic regions of study areas assessed, there have been many improvements to input data
and modeling approaches used in this assessment compared to the prior assessment, including
the availability of continuous 5-minute air monitoring data at monitors within each of the three
study areas. The air quality scenario in the current REA extends the time period of exposure
simulations by covering a 3-year period, consistent with the statistical form established for the
now-current standard. The current air quality scenario additionally focuses on the existing
standard level of 75 ppb. Further, there are also differences between the current REA and the
2009 REA with regard to the air quality adjustment approach, and the methods for estimating 5-
minute concentrations. Also, the years simulated in this assessment reflect more recent emissions
and circumstances subsequent to adoption of the standard in 2010.
As described in section 2.2, these REA analyses are intended to be informative to EPA's
consideration of potential exposures and risks that may be associated with the air quality
conditions occurring under the current SO2 standard. This is reflected in the attributes of the
study areas, including the criteria used in their selection (section 3.1), the identification of
specific source emissions and characteristics, local meteorological conditions, and distribution of
at-risk populations. The presence in the U.S. of these areas and others having similar attributes
make the findings reported here important in considering the protection provided by the SO2
standard, as discussed in the PA.
5-20
-------
6 VARIABILITY ANALYSIS AND UNCERTAINTY
CHARACTERIZATION
An important issue associated with any population exposure or risk assessment is the
characterization of variability and uncertainty. Variability refers to the inherent heterogeneity in
a population or variable of interest (e.g., residential air exchange rates). The degree of variability
cannot be reduced through further research, only better characterized with additional
measurement. Uncertainty refers to the lack of knowledge regarding the values of model input
variables (i.e., parameter uncertainty), the physical systems or relationships used (i.e., use of
input variables to estimate exposure or risk or model uncertainty), and in specifying the scenario
that is consistent with purpose of the assessment (i.e., scenario uncertainty). Uncertainty is,
ideally, reduced to the maximum extent possible through improved measurement of key
parameters and iterative model refinement.
This chapter focuses on the general characteristics of the assessment performed,
including the data and approaches used to evaluate exposures and risk associated with air quality
conditions that just meet the existing standard in the three study areas. The approaches used to
assess variability and to characterize uncertainty in this REA are discussed in the following two
sections. The primary purpose of this characterization is to provide a summary of variability and
uncertainty evaluations conducted to date regarding our SO2 exposure assessments and APEX
exposure modeling and to identify the most important elements of uncertainty in need of further
characterization. Each section contains a concise tabular summary of the identified components
and how, for elements of uncertainty, each source may affect the estimated exposures.
6.1 TREATMENT OF VARIABILITY AND CO-VARIABILITY
The purpose for addressing variability in this REA is to ensure that the estimates of
exposure and risk reflect the variability of ambient SO2 concentrations, population
characteristics, associated SO2 exposure, and potential health risk across the study area and for
the simulated at-risk populations. In this REA, there are numerous algorithms that account for
variability of input data when generating the exposures or risk estimates of interest. For example,
variability may arise from differences in the population residing within census blocks (e.g., age
distribution) and the activities that may influence population exposure to SO2 (e.g., time spent
outdoors, performing moderate exertion-level activities outdoors). A complete range of potential
exposure levels and associated risk estimates can be generated when appropriately addressing
variability in exposure and risk assessments; note however that the range of values obtained
6-1
-------
would be within the constraints of the input parameters, algorithms, or modeling system used,
not necessarily the complete range of the true exposure or risk values.
Where possible, we identified and incorporated the observed variability in input data sets
rather than employing standard default assumptions and/or using point estimates to describe
model inputs. The details regarding many of the variability distributions used in data inputs are
described in Chapter 4, while details regarding the variability addressed within its algorithms and
processes are found in the APEX User Guides (U.S. EPA, 2017a, b).
Briefly, APEX has been designed to account for variability in most of the input data,
including the physiological variables that are important inputs to determining exertion levels and
associated ventilation rates. APEX simulates individuals and then calculates SO2 exposures for
each of these simulated individuals. The simulated individuals are selected to represent a random
sample from a defined population. The collection of individuals represents the variability of the
target population, and accounts for several types of variability, including demographic,
physiological, and human behavior. In this assessment, APEX simulated 100,000 individuals
(70,000 adults and 30,000 children) to reasonably capture the variability expected in the
population exposure distribution for each study area. APEX incorporates stochastic processes
representing the natural variability of personal profile characteristics, activity patterns, and
microenvironment parameters. In this way, APEX is able to represent much of the variability in
the exposure estimates resulting from the variability of the factors effecting human exposure.
We note also that correlations and non-linear relationships between variables input to the
model can result in the model producing incorrect results if the inherent relationships between
these variables are not preserved. That is why APEX is also designed to account for co-
variability, or linear and nonlinear correlation among several of the model inputs, provided that
enough is known about these relationships to specify them. This is accomplished by providing
inputs that enable the correlation to be modeled explicitly within APEX. For example, there is a
non-linear relationship between the outdoor temperature and air exchange rate in homes. One
factor that contributes to this non-linear relationship is that windows tend to be closed more often
when temperatures are at either low or high extremes than when temperatures are moderate. This
relationship is explicitly modeled in APEX by specifying different probability distributions of air
exchange rates for different ambient temperatures. In any event, APEX models variability and
co-variability in two ways:
• Stochastically. The user provides APEX with probability distributions characterizing the
variability of many input parameters. These are treated stochastically in the model and
the estimated exposure distributions reflect this variability. For example, the rate of SO2
removal in houses can depend on a number of factors which we are not able to explicitly
model at this time, due to a lack of data. However, we can specify a distribution of
removal rates that reflects observed variations in SO2 decay. APEX randomly samples
6-2
-------
from this distribution to obtain values that are used in the mass balance model. Further,
co-variability can be modeled stochastically through the use of conditional distributions.
If two or more parameters are related, conditional distributions that depend on the values
of the related parameters are input to APEX. For example, the distribution of air
exchange rates (AERs) in a house depends on the outdoor temperature and whether or not
air conditioning (A/C) is in use. In this case, a set of AER distributions is provided to
APEX for different ranges of temperatures and A/C use, and the selection of the
distribution in APEX is driven by the temperature and A/C status at that time.
• Explicitly. For some variables used in modeling exposure, APEX models variability and
co-vari ability explicitly and not stochastically. For example, the complete series of 5-
minute ambient air SO2 concentrations for each hour and hourly temperatures are used in
model calculations. These are input to the model continuously in the time period modeled
at different spatial locations, and in this way the variability and co-variability of 5-minute
concentrations and hourly temperatures are modeled explicitly.
Important sources of the variability and co-variability accounted for by APEX and used
for this exposure analysis are summarized in Table 6-1 and Table 6-2 below, respectively.
6-3
-------
Table 6-1. Summary of how variability was incorporated into the exposure and risk
assessment.
Component
Variability Source
Summary
Meteorological data
Spatial: local surface and upper air NWS stations used.
Temporal: 1-hour NWS wind data for 2011-2013, supplemented
with 1-minute ASOS wind data (Appendix A).
Emission source types
and profiles
Important SO2 emission sources include EGUs and petroleum
refineries. Hourly emission profiles derived from CEMS data,
where available or using EPA's 2011v6.3 emissions modeling
platform combined with the SMOKE modeling system (Appendix
B).
Ambient Input
AERMOD modeled 1-hour
ambient SO.2.
concentrations
Spatial: ambient SO.2. predicted to 1,400 - 1,900 air quality
receptors in three geographically representative study areas
Temporal: hourly SO2 for each of three years (2011-2013).
Ambient air monitor 5-
minute concentrations
Spatial: local ambient air monitors used. Where multiple monitors
available, receptors used 5-minute patterns from the closest
monitor.
Temporal: patterns of 5-minute continuous SO2 concentrations
within each hour used to estimate 5-minute continuous SO2
concentrations at modeled air quality receptors.
Population data
Individuals are randomly sampled from U.S. census blocks used in
each model study area, stratified by age (single years) and sex
probability distributions (U.S. Census Bureau, 2012).
Employment
Work status is randomly generated from U.S. census data at the
tract level by age and sex (U.S. Census Bureau, 2012).
Simulated Individuals
Activity pattern data
Data diaries used to represent locations visited and activities
performed by simulated individuals are randomly selected from
CHAD master (>55,000 diaries) using six diary pools stratified by
two day-types (weekday, weekend) and three temperature ranges
(< 55.0 °F, between 55.0 and 83.9 °F, and >84.0 °F). CHAD
diaries capture real locations that people visit and the activities
they perform, ranging from 1 minute to 1 hour in duration (U.S.
EPA, 2017c).
Commuting data
Employed individuals are probabilistically assigned ambient air
concentrations originating from either their home or work block
based on U.S. Census derived tract-level commuter data (U.S.
DOT, 2012; U.S. Census Bureau, 2012).
Longitudinal profiles
A sequence of diaries is linked together for each individual that
preserves both the inter- and intra-personal variability in human
activities (Glen et al., 2008).
Asthma prevalence
Asthma prevalence is stratified by sex, single age years for
children (5-17), seven adult age groups, (18-24, 25-34, 35-44, 45-
54, 55-64, 65-74, and, >75), three regions (Midwest, Northeast,
and South), and U.S. Census tract level poverty ratios (Appendix
E).
Physiological Factors
Relevant to Ventilation
Rate
Resting metabolic rate
Five age-group and two sex-specific regression equations, use
body mass and age as independent variables (Appendix H).
Metabolic equivalents by
activity (METS)
Randomly sampled from distributions developed for specific
activities (some age-specific) (U.S. EPA, 2017c).
Oxygen uptake per unit of
energy expended
Randomly sampled from a uniform distribution to convert energy
expenditure to oxygen consumption (U.S. EPA, 2017a, b).
6-4
-------
Component
Variability Source
Summary
Body mass
Randomly selected from population-weighted lognormal
distributions with age- and sex-specific geometric mean (GM) and
geometric standard deviation (GSD) derived from the National
Health and Nutrition Examination Survey (NHANES) for the years
2009-2014 (Appendix G).
Body surface area
Sex-specific exponential equations using body mass as an
independent variable (Burmaster, 1998).
Height
Randomly sampled from population-weighted normal distributions
stratified by single age years and two sexes developed from 2009-
2014 NHANES data (Appendix G).
Ventilation rate
Event-level activity-specific regression equation using oxygen
consumption rate (VO.2) and maximum VO2 as independent
variables, and accounting for intra and interpersonal variability
(Appendix H).
Fatigue and EPOC
APEX approximates the onset of fatigue, controlling for unrealistic
or excessive exercise events in an individual's activity time-series
while also estimating excess post-exercise oxygen consumption
(EPOC) that may occur following vigorous exertion activities using
several equations and input variable distributions (Isaacs et al.,
2007; U.S. EPA, 2017a, b).
Microenvironmental
Approach
Microenvironments:
General
Five total microenvironments are represented, including those
expected to be associated with high exposure concentrations (i.e.,
outdoors and outdoor near-road). Where this type of variability is
incorporated within particular microenvironmental algorithm inputs,
this results in differential exposure estimates for each individual
(and event) as persons spend varying time frequency within each
microenvironment and ambient air concentrations vary spatially
within and between study areas.
Microenvironments:
Spatial Variability
Ambient air concentrations used in microenvironmental algorithms
vary spatially within and among study areas.
Microenvironments:
Temporal Variability
All exposure calculations are performed at the event-level when
using either factors or mass balance approach (durations can be
as short as one minute). For the indoor microenvironments, using
a mass balance model accounts for SO2 concentrations occurring
during a previous hour (and of ambient origin) to calculate a
current event's indoor SO2 concentrations.
Air exchange rates
Several lognormal distributions are sampled based on five daily
mean temperature ranges, study area region (Chapter 4) and
study-area specific A/C prevalence rates from AHS survey data
(U.S. Census Bureau, 2013).
Removal rates
Values randomly selected for microenvironment-specific
distributions, stratified by air conditioning usage (Chapter 4).
Penetration factors
Indoor/outdoor ratios randomly sampled from a uniform distribution
(Chapter 4).
Exposure Response
Function
Regression estimates
A central tendency, along with upper and lower confidence
intervals were derived using a probit function to generate a range
of risk estimates.
Exposure bins
Fine-scale bins (10-50 ppb) stratifying the population exposures
were linked to the continuous E-R function.
6-5
-------
Table 6-2. Important components of co-variability in exposure modeling.
Type of Co-variability
Modeled
by APEX?
Treatment in APEX / Comments
Within-person correlationsa
Yes
Sequence of activities performed, microenvironments
visited, and general physiological parameters (body
mass, height, ventilation rates).
Between-person correlations
No
Perhaps not important, assuming the same likelihood of
the population of individuals either avoiding or
experiencing an exposure event based on a social
(group) activity.
Correlations between profile variables and
microenvironment parameters
Yes
Profiles are assigned microenvironment parameters.
Correlations between demographic
variables and activities
Yes
Census block demographic variables, appropriately
weighted and stratified by age and sex, are used in
activity diary selection.
Correlations between activities and
microenvironment parameters
No
Perhaps important, but do not have data. For example,
frequency of opening windows when cooking or smoking
tobacco products.
Correlations among microenvironment
parameters in the same microenvironment
Yes
Modeled with joint conditional variables.
Correlations between demographic
variables and air quality
Yes
Modeled with the spatially varying census block
demographic variables (age and sex) and fine-scale (100
m to 2 km) air quality input to APEX.
Correlations between meteorological
variables and activities
Yes
Temperature is used in activity diary selection.
Correlations between meteorological
variables and microenvironment parameters
Yes
The distributions of microenvironment parameters can be
functions of temperature.
Correlations between drive times in CHAD
and commute distances traveled
Yes
CHAD diary selection is weighted by commute times for
employed persons during weekdays.
Consistency of occupation/school
microenvironmental time and time spent
commuting/busing for individuals from one
working/school day to the next.
No
Simulated individuals are assigned activity diaries
longitudinally without regard to occupation or school
schedule (note though, longitudinal variable used to
develop annual profile is time spent outdoors).
a The term correlation is used to represent linear and nonlinear relationships.
6.2 CHARACTERIZATION OF UNCERTAINTY
While it may be possible to capture a range of exposure or risk estimates by accounting
for variability inherent to influential factors, the true exposure or risk for any given individual
within a study area is unknown. To characterize health risks, exposure and risk assessors
commonly use an iterative process of gathering data, developing models, and estimating
exposures and risks, given the goals of the assessment, scale of the assessment performed, and
limitations of the input data available. However, uncertainty remains and emphasis is then placed
6-6
-------
on characterizing the nature and potential magnitude of that uncertainty and its impact on
exposure and risk estimates. A summary of the overall characterization is provided in section
6.2.1. The summary is followed by exposure model sensitivity analyses in section 6.2.2 that
provide additional support to the characterization of four elements of uncertainty: (1) the
proportional approach applied to the primary emission source to adjust ambient air
concentrations to just meet the current standard, (2) estimating continuous 5-minute
concentrations at ambient air monitors, (3) estimating 5-minute concentrations at modeled air
quality receptors, and (4) estimating exposure and risk estimated using upper and lower bounds
of the E-R function.
6.2.1 Characterizing Sources of Uncertainty
The REAs for the previous O3, NO2, SO2, and CO NAAQS reviews each presented a
characterization of uncertainty of exposure modeling (Langstaff, 2007; U.S. EPA, 2008, 2009a,
2010, 2014). The qualitative approach used in this and other REAs, also informed by quantitative
sensitivity analyses, is described by WHO (2008). Briefly, we identified the key aspects of the
assessment approach that may contribute to uncertainty in the exposure and risk estimates and
provided the rationale for their inclusion. Then, we characterized the magnitude and direction of
the influence on the assessment results for each of these identified sources of uncertainty.
Consistent with the WHO (2008) guidance, we scaled the overall impact of the
uncertainty by considering the degree of uncertainty as implied by the relationship between the
source of uncertainty and the exposure concentrations. A qualitative characterization of low,
moderate, and high was assigned to the magnitude of influence and knowledge base uncertainty
descriptors, using quantitative observations relating to understanding the uncertainty, where
possible. Where the magnitude of uncertainty was rated low, it was judged that large changes
within the source of uncertainty would have only a small effect on the assessment results. A
designation of moderate implies that a change within the source of uncertainty would likely have
a moderate (or proportional) effect on the results. A characterization of high implies that a small
change in the source would have a large effect on results. We also included the direction of
influence, indicating how the source of uncertainty was judged to potentially affect the
exposure/risk estimates; this included whether the estimates were likely over-estimated ("over")
or under-estimated ("under") or the direction was unknown. A summary of the key findings of
those prior uncertainty characterizations that are most relevant to the current SO2 exposure
assessment are also provided in Table 6-3 (i.e., Langstaff, 2007; U.S. EPA, 2008, 2009a, 2010,
2014).
6-7
-------
Table 6-3. Characterization of Key Uncertainties in Exposure and Risk Assessments using APEX.
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Aspects of
Assessment Design
Representation of
S0.2 emission
source types
having substantial
emissions
Unknown
Low-
Moderate
Moderate
The three study areas include the most prevalent source type (i.e., EGUs)
emitting at least 15,000 tons of SO2 per year (95% of all U.S. facilities emitting
SO2 in 2011; U.S. EPA, 2015). There are only three other source types having
emissions at least as large: copper/lead smelters (2 facilities), pulp and paper
mills (2 facilities), and chemical plants (1 facility) (U.S. EPA, 2015). The limited
occurrence of these large non-EGU facilities and their occurrence in locations
with small populations and/or for which ambient air monitoring data for SO2 are
not available hampered their selection as study areas evaluated in this REA. To
the extent that the temporal patterns of emissions and/or emissions
characteristics for these source types differ in a way that would lead to greater
variability in ambient SO.2 concentrations than that associated with EGU
emissions, it is possible the risk/exposure estimates associated with these
particular sources (if having substantial emissions) could vary from those
estimated in this REA. However, risk and exposure estimates for areas with
such sources would likely have limited applicability nationally due to limited
prevalence of such areas across the U.S.
No
Representation of
population
subgroups with
asthma
Unknown
Low-
Moderate
Moderate
Consistent with the ISA identification of people with asthma (and children with
asthma in particular) as an important at-risk population for SO2 in ambient air,
risk estimates are developed for people with asthma and are reported
separately for children and adults. Exposure and risk were not estimated for
more targeted population groups with asthma based on additional personal
attributes associated with increased asthma prevalence (e.g., obesity or African
American or Hispanic ethnicity) generally due to limitations in the data needed
to simulate such subgroups. Such data limitations affect our ability to
characterize SO2 exposure and associated health risks for different population
subgroups of children and adults with asthma, some of which may have higher
exposure/risk and others lower.
No
6-8
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Algorithms
(section 3.2)
Unknown
Low
Low
Multiple historical model evaluations consistently demonstrate unbiased
ambient air concentrations under variety of conditions. Some potential
dispersion scenarios may not be adequately represented and in some
instances, concentration variability could be under-represented by model
algorithms, however, it is largely unknown as to how this uncertainty might
apply in this application.
No
AERMOD
Inputs and
Algorithms
Meteorological
Data
(section 3.2.1.1
and Appendix A)
Unknown
Low-
Moderate
Low
A limited number of missing hours of wind data remain in dataset, potentially
leading to under-estimation. Model predictions have low to medium sensitivity
to surface roughness characteristics, as long as they are appropriate for the site
of the meteorological data inputs. Data are from a well-known and quality-
assured source. One minute ASOS wind data used to supplement 1-hour data
for improved completeness, reducing the number of calms and missing data.
Two meteorological stations (one upper and one surface) are used to represent
meteorological conditions in each study area, some of which are located a few
to several km from ambient air monitor sites and the modeled air quality
receptors. There is uncertainty in the extent to which conditions measured at
these stations represent study area meteorological conditions, particularly wind
speed and direction, and how this could affect the estimation of hourly and 5-
minute concentration variability.
No
Stationary Source
Emissions and
Profiles (section
3.2.2 and
Appendix B)
Both
Low
Low
Temporal emission characteristics are well represented for most modeled point
sources. Most temporal data are from a well-known quality-assured source of
direct measurements.
No
Ambient Air Monitor
Concentrations
Database Quality
Both
Low
Low
All ambient pollutant measurements available from AQS are comprehensive
and subject to quality control. Completeness criteria applied to hourly
concentrations ensure air quality representativeness.
No
6-9
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Missing Data
Substitution
(section 3.5.1)
Under
Low
Low
Missing ambient air concentration values (hourly, 5-minute maximum, 5-minute
continuous) were interpolated using a statistical technique. Use of this type of
approach is appropriate for data sets having a limited missing number of total
values (<5-10%), though will constrain substituted values within the bounds of
the measured concentrations. In addition, there are a few monitors missing
concentrations for several hours/minutes per day and for several days in the
year, most notably in the Indianapolis study area (Table 3-9), potentially
missing a few high concentration events (if actually occurred) that would not be
estimated using the interpolation technique. However, completeness of the
maximum 5-minute concentrations was reasonable (<10%) for all monitors
used in estimating 5-minute concentrations, thus the missing within hour 5-
minute concentrations are of lesser importance and likely contribute less
uncertainty.
No
Estimation of
Continuous 5-
minute
Concentrations
(section 3.5.2)
Under
Low
Low
For one year in Fall River (2013), only the 5-minute maximum measurements
within each hour were reported. A series of lognormal distributions were used to
estimate the 5-minute continuous patterns occurring with each hour for these
monitors (Section 3.5.2). Excellent agreement was observed comparing the
estimated versus the measured values for each of the hourly and 5-minute
maximum concentrations. Agreement between the estimated and measured 5-
minute continuous concentrations was also excellent, though exhibiting some
deviations (Figure 3-6). In addition, the estimated 5-minute continuous
concentrations had less overall variability compared to the measurement data
(Table 3-10). However, there was negligible difference in exposures when
comparing an APEX simulation that used measured continuous 5-minute
concentrations versus one that used estimated values.
Yes, section
6.2.2.1
Temporal
Representation
(section 3.5.2 and
3.5.3)
Both
Low
Low
Temporal scale (5-minutes) is appropriate for analysis performed. Monitored
hourly and 5-minute maximum data are screened for temporal completeness
and considered appropriate. While 5-minute continuous data were not screened
for completeness, the number of missing values were limited in most study
areas and for most years (Table 3-9).
No
6-10
-------
Sources of Uncertainty
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Uncertainty
Comments
Category
Element
Direction
Magnitude
Spatial
Representation
(section 3.5.3.1)
Both
Moderate
Moderate
There were few ambient air monitors available to approximate 5-minute
patterns across study area: Fall River, one monitor available and used to
estimate 5-minute concentrations; Indianapolis, three monitors available (two
were used); Tulsa, four monitors available (three were used). Where more than
one monitor was available, the air quality receptors used 5-minute
concentration patterns from closest monitor.
No
Air Quality
Receptor
Concentrations
Concentration
Used to Represent
Sources Not
Modeled
(section 3.2.4)
Both
Moderate
Low-
Moderate
There is uncertainty in the estimates of hourly concentration associated with
SO.2 emission sources not explicitly modeled in the three study areas. While
temporal variability in these estimates is accounted for by calculating diurnal
and seasonal values, year to year variability is not considered, thus not
accurately accounting for instances where the contribution may vary by year.
The value used for each hour/season is the 3-year average of the 99th
percentile concentration (section 3.2.4), an approach that at most times would
generally tend to overestimate these concentrations. Further, monitor hours that
may have concentrations influenced by modeled sources were identified for
exclusion using wind direction data from nearby airports. This provided a
consistent approach across study areas as local wind direction data were not
reported at all monitors. However, uncertainty is contributed in circumstances
where the airport wind direction does not reflect conditions occurring at a
monitor. The magnitude of such uncertainty may be sizeable at some monitors.
No
6-11
-------
Sources of Uncertainty
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Uncertainty
Comments
Category
Element
Direction
Magnitude
AERMOD
Predicted Hourly
Concentrations
Both
Low-
Moderate
Moderate
Overall, comparisons of model predicted hourly values and ambient air
measurements in each study area indicate good agreement when considering
concentration magnitude alone. The first of the three years tended to exhibit the
highest concentrations in addition to having the best agreement across all three
study areas. In the Fall River study area, AERMOD over-predicted low to mid
percentile concentrations and under-predicted upper percentile concentrations.
In the Indianapolis study area, AERMOD over-predicted all percentiles of the
concentration distribution at the monitor closest to the primary emissions
source, while under-predicting mid to high percentile concentrations at the
monitors more distant from the primary emissions source. In the Tulsa study
area, AERMOD under-predicted mid to high percentile concentrations at the
monitor closest to the primary emissions source, while over-predicting most
percentile concentrations at monitors more distant from the primary emissions
source (Appendix D). Such differences are of lesser importance to the
assessment estimates given the focus on air quality after adjustment to just
meet the existing standard.
No
Hourly Ambient Air
Concentration
Estimates during
Times of
Relatively Greater
Exposure Potential
Both
Low-
Moderate
Low-
Moderate
As separately concluded for the generalized performance evaluation
summarized in Appendix D, these comparisons that consider both spatial and
temporal variability (i.e., where and when peak concentrations occur) also
indicate reasonable agreement between the model estimates and
measurements at the nearby monitor site(s). Similarity in the paired
concentrations across much of the respective distributions for times when
exposure potential may be greatest provide additional positive support for
concluding the modeled air quality surfaces are likely useful for estimating
exposures in this REA (section 3.2.5 and Appendix K). However, having limited
monitoring data available in each study area and the inability to directly
evaluate the concentrations for the air quality scenario that is the focus of this
REA limits the extent by which conclusions can be made regarding model
performance in estimating spatial variability in hourly concentrations for that
scenario.
No
6-12
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Adjustment of
Hourly
Concentrations to
Just Meet the
Existing Standard:
Proportional
Approach for
Primary Source
(section 3.4)
Under
Low
Moderate
Performance of this approach in this REA depends in part on the degree of
proportionality in the air quality distribution and the magnitude of the ambient air
concentration adjustment. A proportional approach was judged adequate for
such a use (section 3.4 above; REA Planning Document; Rizzo, 2008). The
approach used in this REA is a modification of 2009 REA adjustment approach
in that the proportional adjustment was applied only to the concentration
contribution from the primary emission source in each study area, holding
concentrations contributed from all other sources as is. The sharpness of the
concentration gradient from the primary emission source relative to the other
emission sources could be an important factor in determining the impact to the
adjusted air quality surface. However, sensitivity analyses that modified the air
quality receptor having the maximum design value (section 6.2.2.2) indicate
there was negligible (in two areas) or somewhat limited (in the third area)
impact to the estimated exposures by varying the magnitude of the adjustment
and the number of receptors to which the adjustment was applied.
Yes, section
6.2.2.2
Approach Used to
Estimate 5-minute
Concentrations:
Linking 5-minute
Monitor to Hourly
Receptor
Concentrations
(section 3.5.3)
Both
Low-
Moderate
Moderate
Hourly concentrations modeled at the air quality receptors were linked to the 5-
minute monitor concentrations using the rank order of the hourly
concentrations. Two alternative approaches were developed and evaluated.
The first, a calendar based approach, linked the modeled receptor
concentrations to the monitor by date and hour of day. The second used hourly
concentration bins (i.e., 5 ppb increments). There were differences when
comparing the upper percentiles of the 5-minute concentration distributions,
particularly when comparing the calendar based approach to the rank order and
binning approaches. There were also notable differences to the percent of the
at-risk population exposed at or above benchmarks when comparing results
from the three adjustment approaches. However, little difference was observed
when comparing risk of lung function decrements estimated using each of these
three approaches.
Yes, section
6.2.2.3
6-13
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Estimated Peak 5-
minute
Concentrations
during Times of
Relatively Greater
Exposure Potential
Both
Low-Moderate
Moderate
As concluded in section 3.5.3.3, there is reasonable agreement between the 5-
minute concentrations estimated at air quality receptors and measurement
concentrations at the local ambient air monitor site(s), considering
concentrations of interest and the number of days at or above these
concentrations. Having limited monitoring data available in each study area
however limits our ability to assess the reasonableness of the degree to which
these concentrations may be present spatially. It appears that the spatial extent
of receptors having the highest 5-minute concentrations could be over-
estimated in the Indianapolis study area and less so (to possibly not at all) in
the Fall River and Tulsa study areas.
No
Population
Demographics and
Commuting
(sections 4.1.1
and 4.3.2)
Both
Low
Low
Comprehensive and subject to quality control. Differences in 2010 population
data versus modeled years (2011-2013) are likely small when estimating
percent of population exposed.
No
APEX: General
Input Databases
Activity Patterns
(CHAD)
(sections 4.3.1
and 4.3.3)
Both
Low-
Moderate
Low-
Moderate
Comprehensive and subject to quality control. Increased number of diaries
used to estimate exposure from 2009 SO2 REA. Thoroughly evaluated trends
and patterns in historical activity pattern data - no major issues noted with use
of historical data to represent current patterns (Figures 5G-1 and 5G-2 of U.S.
EPA, 2014). Compared outdoor event participation and outdoor time of CHAD
diary data with larger American Time Use Survey (ATUS) data - CHAD
participation is higher than ATUS, likely due to ATUS survey methods.
Comparison of activity data (outdoor events and exertion level) for people with
asthma generally similar to individuals without asthma (Table 4-5) (see also
Tables 5G2-to 5G-5 of U.S. EPA, 2014). There is little indication of differences
in time spent outdoors comparing activity patterns across U.S. regions, though
sample size may be a limiting factor in drawing significant conclusions (U.S.
EPA, 2014). Remaining uncertainty exists for other influential factors that
cannot be accounted for (e.g., SES, region/local participation in outdoor events
and associated amount of time).
No
6-14
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Meteorological
(NWS)
(section 4.2)
Both
Low
Low
Comprehensive and subject to quality control, having very few missing values.
Limited use in selecting CHAD diaries for simulated individuals and AERs that
may vary with temperature. However, while using three years of varying
meteorological conditions, the 2011-2013 MET data set may not reflect the full
suite of conditions that could exist in future hypothetical air quality scenarios or
across periods greater than 3-years.
No
Asthma
Prevalence
Weighted by
Poverty Status
(section 4.1.2 and
Appendix E)
Both
Low
Low-
Moderate
Data used are from peer-reviewed quality controlled sources. Use of these data
accounts for variability in important influential variables (poverty status, as well
as age, sex, and region). It is possible that variability in microscale prevalence
is not entirely represented when considering other potentially influential
variables such as race and obesity, two attributes that can influence asthma
prevalence and can vary spatially (section 4.1.2). Family income level was used
in this REA to represent spatial variability in asthma prevalence and may, in
some instances, capture spatial variability in race and obesity (Ogden et al.,
2010), and thus to some extent, reasonably represent the potential influence
race and obesity have on asthma prevalence. However, instances where these
influential variables are not fully represented in simulating the at-risk population,
and where populations identified by such variables are associated with
increased asthma prevalence that may spatially intersect with the highest
ambient concentrations, could lead to uncertainty in estimated exposures and
health risk. Further characterization could be appropriate by comparing with
local prevalence rates stratified by a similar collection of influential variables,
where such data exist.
No
APEX:
Microenvironmental
Concentrations
Vehicle PE
Factors
(Section 4.4.5)
Both
Low
Moderate
Input distribution is from an older measurement study and for a different
pollutant (section 4.4.5). Considering that the exposures of interest need to be
concomitant with elevated exertion, the accurate estimation of 5-minute
exposures occurring inside vehicles is considered relatively unimportant.
No
6-15
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Indoor: Air
Exchange Rates
(sections 4.4.1
and 4.4.3)
Both
Low
Moderate
Uncertainty due to random sampling variation via bootstrap distribution analysis
indicated the AER geometric mean (GM) and standard deviation (GSD)
uncertainty for a given study area tends to range from ±1.0 GM and ± 0.5 GSD
hr1 (Langstaff, 2007). Each of the three study areas (Fall River, Indianapolis,
and Tulsa) used AER from a geographically similar city (New York, Detroit/New
York, and Houston, respectively). Non-representativeness remains an important
issue as city-to-city variability can be wide ranging (GM/GSD pairs can vary by
factors of 2-3) and data available for city-specific evaluation are limited
(Langstaff, 2007). There is uncertainty associated with the use of an AER
derived from a different city than the REA study areas. That said, indoor
microenvironments are considered less likely to contribute to an individual's
daily maximum 5-minute SO2 exposure while at elevated exertion levels and
likely does not contribute substantially to uncertainty in the exposure and risk
estimates.
No
Indoor: A/C
Prevalence
(section 4.4.2)
Both
Low
Low
Data were obtained from a reliable source, are comprehensive, and subject to
quality control (US Census Bureau, 2013). For two of the three study areas
(Fall River and Indianapolis), data from a geographically related city were used
(Boston and Louisville, respectively). There is uncertainty associated with the
use of an AC prevalence derived from a different city than the REA study areas.
That said, indoor microenvironments are considered less likely to contribute to
an individual's daily maximum 5-minute SO2 exposure while at elevated
exertion levels and likely does not contribute substantially to uncertainty in the
exposure and risk estimates.
No
Indoor: Removal
Rate
(section 4.4)
Unknown
Low
Moderate
In the 2009 REA it was found that indoor exposures may be underestimated
when not using all 5-minute concentrations within the hour, an issue resolved in
this current REA by using estimates of all 5-minute values. Data used to
develop removal rates were obtained from a comprehensive review, though
many assumptions were needed in developing the distributions. However, most
peak exposures concomitant with elevated exertion are expected to occur
outdoors, thus accurate estimation of indoor concentrations is of reduced
importance.
No
6-16
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
APEX: Simulated
Longitudinal
Profiles
(section 4.3.4)
Under
Low-
Moderate
Moderate
The magnitude of potential influence for this uncertainty would be mostly
directed toward estimates of multiday exposures. Simulations indicate the
number of single day and multiday exposures of interest can vary based on the
longitudinal approach selected (Che et al., 2014). As discussed in chapter 4,
the D&A method provides a reasonable balance of this exposure feature. Note
however, long-term diary profiles (i.e., monthly, annual) do not exist for a
population, thus limiting the evaluation. Further, the general population-based
modeling approach used for main body REA results does not assign rigid
schedules, for example explicitly representing a 5-day work week for employed
people.
No
Activity Profiles
Commuting
(section 4.3.2)
Both
Low
Moderate
Method used in this assessment is designed to link Census commute distances
with CHAD vehicle drive times. Considered an improvement over the prior
approach that did not match commute distance and activity time. While vehicle
time is accounted for through diary selection, it is not rigidly scheduled.
However, accurate estimation of exposures occurring while inside vehicles is
considered unimportant because it is unlikely to occur at elevated exertion.
No
Activity Patterns
for At-Risk
Population
(section 4.3.3)
Both
Low
Low-
Moderate
Analyses of activity patterns of people with asthma are similar to that of
individuals not having asthma (section 4.3.3; see also Tables 5G-2 to 5G-5 of
U.S. EPA, 2014).
No
APEX:
Physiological
Processes
Body Weight
(NHANES)
(section 4.1.3.1
and Appendix G)
Unknown
Low
Low
Comprehensive and subject to quality control, appropriate years (2009-2014)
selected for simulated population, though possible small regional variation is
not represented by national data.
No
6-17
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
RMR
(section 4.1.3.2,
Appendix H)
Unknown
Low
Low
New, improved algorithm used for this assessment. Comprehensive literature
review resulted in construction of large data base used to derive algorithm.
Algorithm considers variables most influential to RMR (i.e., age, body weight,
and sex). There are other factors that could affect intra-personal variability in
RMR such as time-of-day (Haugen et al., 2003) or seasonal/temperature
influences (van Ooijen et al., 2004; Leonard et al., 2014). Variability from these
and other potentially influential factors may be indirectly accounted for by the
residual error term used in the RMR Equation 4-2 depending on the extent to
which these influential factors varied across the clinical study data that were
used to create the RMR analytical data set. However, because there is
inadequate information regarding the presence of multiple RMR measurements
for individual study subjects, we could not estimate intra-personal variability nor
could we use these influential factors, other than age and sex, as explanatory
variables in the RMR equation. Therefore, any influences on spatial variability in
RMR, both within and among the three study areas, would largely be driven by
the spatial distribution of age and sex.
No
METS
Distributions
(section 4.1.3.2)
Over
Low-
Moderate
Moderate
APEX estimated daily mean METs range from about 0.1 to 0.2 units (between
about 5-10%) higher than independent literature reported values (Table 15 of
Langstaff, 2007). However, shorter-term values are of greater importance in this
assessment, thus METs could be better characterized where short-term METS
data are available.
No
Ventilation Rates
(section 4.1.3.3
and Appendix H)
Unknown
Low
Low-
Moderate
Predictions made using the prior algorithm showed excellent agreement with
independent measurement data, particularly when considering simulated study
group (Graham and McCurdy, 2005; Figure 5-23 and Figure 5-24 of U.S. EPA,
2014). New algorithm derived using the same data observed to have improved
predictability (Appendix H). However, a shorter-term comparison (5-minutes or
a single hour rather than daily) of predicted versus measured ventilation rates,
while more informative, cannot be performed due to lack of ventilation rate data
at this duration and considering influential factors (e.g., age, particular activity
performed).
No
6-18
-------
Sources of Uncertainty
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Uncertainty
Comments
Category
Element
Direction
Magnitude
EVR
Characterization of
Moderate or
Greater Exertion
(section 4.1.3.3)
Both
Moderate
Moderate
Given that the EVR serves as a cut point for selecting individuals performing moderate
or greater exertion activities and is an approximated mean value applied to the
population as a whole, the simulated number of people achieving this level of exercise
(along with having a response) could be either under or overestimated. This is because
EVR is calculated as a function of body surface area as a means to extrapolate the
ventilation rates achieved by adults in the controlled human exposure studies to that of
our simulated children. Fundamentally, the EVR assumes that differences between
adults and children with regard to the target ventilation rate denoting ventilation for
moderate or greater exertion can be described by differences in body surface area
(and hence body weight). On average, body surface area for adults is approximately
1.95 m2 and for children is approximately 1.33 m2 (U.S. EPA, 2011), indicating that on
average, the ventilation rate required to meet an EVR of 22 (i.e., moderate or greater
exertion) is about 43 L/min and 29 L/min for adults and children, respectively.
Recommended ventilation rates representing moderate intensity exercise (330) would need to have higher VE than non-obese people performing the
same activity to achieve an EVR of 22 given their relatively greater body surface area
(on average, approximately 30-60% greater). APEX estimates ventilation rates for
obese individuals that are approximately 10-30% greater than non-obese individuals
largely the result of obese individuals having, on average, a 10-40% greater resting
metabolic rate than non-obese individuals. Accordingly, S0.2-related health risk
estimates derived for simulated obese people with asthma could be underestimated,
holding all other potential influential factors constant.
No
6-19
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Risk Estimation for
Exposures Below
100-200 ppb
Over
Low-
Moderate
Low-
Moderate
While there is very strong support for SO2 being causally linked to lung function
responses within the range of tested exposure levels (i.e., > 200 ppb), data are limited
or lacking for lower concentrations. Data available at 100 ppb are limited to studies in
which SO2 was administered by mouthpiece, some of which also do not include a
control exposure to clean air while exercising (Sheppard et al. 1981; Sheppard et al.,
1984; Koenig et al., 1989; Koenig et al., 1990; Trenga et al., 2001). These studies
indicate smaller responses (in adults and adolescents) than is observed in the 200 ppb
chamber exposures. No data are available at lower exposure levels below 100 ppb.
Since this assessment assumes there is a causal relationship at levels below 100 ppb,
the influence of this source of uncertainty would be to over-estimate risk.
No
Lung Function Risk
Estimation
(section 4.6.2)
Probit Model Used
to Estimate E-R
Function
Unknown
Low
Low
It was necessary to estimate responses at SO2 levels both within the range of
exposure levels tested (i.e., 200 to 1,000 ppb) as well as below the lowest exposure
levels used in free-breathing controlled human exposure studies (i.e., below 200 ppb).
We have developed probabilistic exposure-response relationships using a probit form,
considered appropriate for this assessment. However, the regression model assumes
a positive response occurring at any exposure concentration, of particular relevance to
the lowest exposures.
No
Use of E-R data
from Studies of
Individuals having
Mild/Moderate
Asthma to
Represent Any
Asthma Severity
Unknown
Unknown
Moderate
The data set that was used to estimate exposure-response relationships included
people with mild and/or moderate asthma. There is uncertainty with regard to how well
the population of people with mild and moderate asthma included in the series of SO2
controlled human exposure studies represent responses that might be expected across
the entire distribution of people with asthma in the U.S. population. As indicated in the
ISA (section 5.2.1.2), the subjects studied do not generally include people with asthma
that would be classified as severe by today's classification standards. The available
studies "suggest that adults with moderate/severe asthma may have more limited
reserve to deal with an insult compared with individuals with mild asthma" (ISA, p. 5-
22).
No
6-20
-------
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Sources of Uncertainty
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Comments
Category
Element
Direction
Magnitude
Uncertainty
Reproducibility of
S0.2-induced Lung
Function
Response
Unknown
Unknown
Low
The risk assessment assumes that the SOz-induced responses for individuals are
reproducible. We note that this assumption has some support in that one study (Linn et
al., 1987) exposed the same subjects on two occasions to 0.6 ppm and the authors
reported a high degree of correlation (r > 0.7 for people with mild asthma and r > 0.8
for people with moderate asthma, p < 0.001), while observing much lower and
nonsignificant correlations (r = 0.0 - 0.4) for the lung function response observed in the
clean air with exercise exposures.
No
Use of E-R
Derived from
Adults for Children
Unknown
Unknown
Low-
Moderate
Because the vast majority of controlled human exposure studies investigating lung
function responses were conducted with adult subjects, the risk assessment relies on
data from adult subjects with asthma to estimate exposure-response relationships that
have been applied to all individuals with asthma, including children aged 5-18. The
available evidence includes some studies of adolescents (aged 12-18) with asthma
that indicate generally similar effects as observed for adults, although precise
comparisons are not feasible with the available data (ISA, pp. 5-22 to 5-23). The
studies involving adolescents administered SO2 via inhalation through a mouthpiece
rather than an exposure chamber. This technique bypasses nasal absorption of SO.2
and can result in an increase in lung SO.2 uptake. Given this is a limited dataset and
the lack of any such studies for children younger than 12,, the uncertainty in the risk
estimates for children with asthma is greater than those for adults.
No
SO.2 Exposure
History
Both
Low
Moderate
The risk assessment assumes that the SCh-induced response on any given day is
independent of previous SO2 exposures and only the highest daily 5-minute exposure
(under moderate or greater exertion) is assessed. The limited evidence related to this
source indicates effects from a subsequent-day exposure to not be statistically
significantly different from the first day. Further, responses to repeated exposures
within an hour have been found to be diminished responses from initial ones, although
data are limited or lacking regarding exposures repeated after multiple hours but within
the same 24-hour period (ISA, section 5.2.1.2).
No
6-21
-------
Sources of Uncertainty
Uncertainty Characterization
Sensitivity
Analysis
Performed?
Influence of Uncertainty on
Exposure | Risk Estimates
Knowledge-
base
Uncertainty
Comments
Category
Element
Direction
Magnitude
Assumed No
Interaction of other
Co-pollutants on
SOz-related Lung
Function
Responses
Under
Low
Moderate
There are a few studies regarding the potential for an increased response to SO2 when
exposure is in the presence of other common pollutants such as PM (potentially
including particulate sulfur compounds), nitrogen dioxide and ozone, although the
studies are limited (e.g., with regard to relevance to ambient exposure concentrations)
and/or provide inconsistent results (ISA, p. 5-25; 2008 ISA, section 3.1.4.7; ISA, pp. 5-
143 to 5-144). For example, "studies of mixtures of particles and sulfur oxides indicate
some enhanced effects on lung function parameters, airway responsiveness, and host
defense," however, "some of these studies lack appropriate controls and others involve
[sulfur-containing species] that may not be representative of ambient exposures" (ISA,
p.5-144).
No
6-22
-------
6.2.2 Exposure Model Sensitivity Analyses
6.2.2.1 Continuous 5-minute Concentrations - Estimated versus Measured
Analyses evaluating the approach used to estimate the twelve 5-minute concentrations for
each hourly concentration in the assessment is summarized in section 3.5.2. These analyses
utilized datasets at monitors for which continuous 5-minute data are available; the analyses
indicate reasonable agreement between the estimated and measured concentrations. By design,
the estimated hourly and within-hour 5-minute maximum concentrations were identical to the
measured hourly and 5-minute maximum concentrations, though sampling from lognormal
distributions led to instances where the within-hour pattern of the eleven other estimated within-
hour 5-minute concentrations varied from that measured (Figure 3-6).
We evaluated the impact this difference may have on exposures in the Fall River study
area, the only study area that used this method to estimate continuous 5-minute concentrations
for the single year that continuous 5-minute measurements were not available (2013). Two
identical APEX simulations were performed in the Fall River study area that differed only by the
ambient air concentrations used for input to the model. Both simulations used a single air quality
district, the center of which was the location of monitor 250051004, and employed a 10 km
radius of influence to select the census blocks comprising the exposure modeling domain. One
simulation used the continuous 5-minute concentrations measured in 2011 at the ambient air
monitor and the other using the pattern of 5-minute continuous concentrations estimated for that
same year and location (and initiated by the monitor's measured hourly and daily maximum 5-
minute concentrations). All other model settings were the same as that used for the APEX
simulations performed for the main REA, though only children with asthma were simulated.
We first evaluated statistics of interest beyond those presented in Table 3-10. Of interest
were the upper percentile concentrations and number of times the 5-minute ambient air
concentrations were at or above the benchmark concentrations. Table 6-4 provides the results of
this analysis. Consistent with results provided in chapter 3, there are differences between
estimated and measured values at the upper percentile concentrations shown here (i.e., 99th
percentile of the distribution and the number of values at or above 100 ppb), with the estimated
percentile concentrations slightly lower than the percentile concentrations for the measured
values. However, in the APEX simulation results there is little to no difference in either the
estimated exposures at or above the benchmarks (Table 6-5) or in the percent of the children
expected to experience a lung function decrement (Table 6-6) when considering the varying
concentration input.
6-23
-------
Table 6-4. Comparison of measured and estimated continuous 5-minute SO2
concentrations in ambient air, Fall River monitor 250051004, 2011.
Monitor ID 250051004
Continuous 5-minute SO2 concentrations (ppb)
Percentile of
distribution
Estimated
Measured
pO
0.0
0.0
P1
0.0
0.0
p5
0.1
0.1
p10
0.4
0.5
p25
1.0
1.1
P50
1.8
1.9
p75
2.8
2.7
p90
5.5
5.2
p95
9.4
9.0
p99
34.1
36.6
p100
241.1
241.1
Number of times per year 5-minute concentration at or above benchmark
Benchmark
Concentration (ppb)
Estimated
Measured
100
144
147
200
5
5
300
0
0
400
0
0
Table 6-5. Comparison of simulated exposures, for children with asthma, at or above
benchmarks using measured versus estimated continuous 5-minute SO2
concentrations from monitor 250051004, Fall River, 2011.
benchmark
(ppb)
5-minute
ambient air
concentrations
Percent of children with asthma having exposures at or above 5-
minute benchmark concentration
number of days per year at or a
sove benchmark concentration
>1
>2
>3
>4
>5
>6
100
Measured
43.9
20.0
9.0
3.9
1.6
0.7
Estimated
43.2
19.2
8.4
3.7
1.5
0.7
200
Measured
8.3
0.6
<0.1a
0a
0
0
Estimated
8.3
0.6
<0.1
0
0
0
300
Measured
no individuals estimated to experience any days at or above 300 ppb
Estimated
a < 0.1 represents nonzero estimates below 0.1%. A zero (0) indicates there were no individuals having the specified
exposure.
6-24
-------
Table 6-6. Comparison of simulated lung function decrements in children with asthma
using measured versus estimated 5-minute continuous SO2 concentrations,
Fall River 2011.
sRaw
5-minute
ambient air
concentration
input
Percent of children with asthma estimated to experience one or
more days with an increase of sRaw of specified amount
number of days per year
>1
>2
>3
>4
>5
>6
100%
Measured
2.4
0.8
0.4
0.3
0.2
0.1
Estimated
2.3
0.9
0.4
0.3
0.2
<0.1a
200%
Measured
0.6
0.1
0a
0
0
0
Estimated
0.6
0.1
0
0
0
0
a < 0.1 represents nonzero estimates below 0.1%. A zero (0) indicates there were no individuals having the specified
exposure.
6.2.2.2 Adjustment of Hourly Concentrations to Just Meet the Existing Standard
In this assessment, a proportional approach was used to adjust air quality to just meet the
current standard. For the exposure and risk results presented in Chapter 5, as described in section
3.4, we adjusted concentrations for the source contributing the most to the air quality receptor
concentrations, and that single receptor having the maximum design value in each study area.
Thus, all other design values calculated for the modeled receptors in the study area following the
air quality adjustment were less than 75 ppb, with one receptor having a design value of 75 ppb.
In light of the variation in adjustment factors (Table 3-8), the fact that the factor is
derived from the highest design value, and the finding that, while the model predicted hourly
concentrations were found generally comparable with monitor measurements, there were a few
instances where the highest upper percentile concentrations could be overestimated (see
Appendix D, Table D-3), we have evaluated the impact on the estimated population exposures of
an alternative adjustment approach. The alternative approach is intended to address the potential
for overestimation at the few highest-concentration receptors that could result in the application
of an overly large adjustment factor for a number of the receptors in the modeling domain. This
alternative adjustment procedure modifies the selection of the receptor that is used to calculate
the adjustment factor. Rather than select the single maximum design value to determine the
adjustment factor for all receptor concentrations within a study area, we chose the 99th percentile
design value to determine the adjustment factor for the receptor with that design value, and for
receptors with lower values. Thus, all receptors having design values less than the 99th percentile
following the air quality adjustment would have a design value less than 75 ppb. All study area
receptors having design values above the 99th percentile design value were adjusted using their
own individual adjustment factors that resulted in each of them having adjusted concentrations
that also yielded a design value of 75 ppb.
6-25
-------
Table 6-7 summarizes the adjustment factors used in this alternative approach. The air
quality scenario created by this alternative approach, just like the base approach used for the
exposure and risk results in Chapter 5, reflects air quality conditions that just meet the existing
standard. However, this alternative adjustment procedure using the 99th percentile design value
results in a greater spatial distribution of relatively higher concentrations across the study area
compared with the scenario created using the maximum design value, which leads to higher
percentages of children with asthma having exposures above benchmark concentrations and lung
function decrements. Figures 6-1 to 6-3 illustrate this in each of the study areas, showing the
overlay of the population distribution and the design values resulting from the two different
adjustment approaches.
Table 6-7. Air quality adjustment factors for main body REA and sensitivity analysis.
Study area
Approach for
Main body REA
Alternative Approach for Sensitivity Analysis
Maximum
Design value
(ppb)
Factor
applied to all
receptors
99th
percentile
design
value (ppb)
Factor applied to
Receptors < 99th
percentile design value
Factor applied to
Receptors > 99th
percentile design
Fall River
101.4
1.46
83,2
1.12
1,14-1.46
Indianapolis
311.3
4.21
205.2
2.77
2.85-4.21
Tulsa
73.5
0.98
63.1
0.82
0.81-0.98
¦*#= ar
i"Ris^WB::is::s£ss»is:s::»:ss"
"5IBWBBKK3"i5"ifa«s5«
'KBSH55mw,:q,i— /
1 ¦¦!¦¦¦¦¦¦¦¦ ¦ \
Design Value
(ppb)
Design Value
(PPb)
Figure 6-1. Spatial pattern of design values using an adjustment based on the
maximum design value (left panel) and an adjustment based on the 99th
percentile design value (right panel) in the Fall River study area.
6-26
-------
Figure 6-2. Spatial pattern of design values using an adjustment based on the
maximum design value (left panel) and an adjustment based on the 99th
percentile design value (right panel) in the Indianapolis study area.
¦ ¦ ¦ ¦
V ¦ ¦ ¦ ¦
I ¦ ¦ ¦ ¦
M1127*
¦*¦¦¦¦¦
M0179i *M0175
West Refinery*
East Refinery •
Design Value
(PPb)
75
M1127*
¦¦¦¦¦¦¦¦¦I"5 5;
is:::::::::;:
iSESUMU .MQ175.
West Refinery* Nnm
Isssssssiiiiu
¦¦¦¦
¦¦¦¦¦
East Refinery •
¦¦ ¦!¦¦¦ ¦¦ WM1IIS*!
¦¦¦¦
¦¦¦¦<—
¦¦¦¦!
BSSBB
Design Value
(PPb)
¦¦¦¦¦¦¦¦¦¦¦¦¦¦
¦¦¦¦¦¦¦¦¦¦¦¦¦¦
75
Figure 6-3. Spatial pattern of design values using an adjustment based on the
maximum design value (left panel) and an adjustment based on the 99th
percentile design value (right panel) in the Tulsa study area.
6-27
-------
We performed APEX simulations using these air quality data sets derived with the
alternative adjustment approach, and holding all model settings identical to those used to
generate the exposures presented in Chapter 5. Exposures and risk of lung function decrements
were estimated for children with asthma in the three study areas for all three years. Tables 6-8
through 6-11 present the results of these alternative simulations, including a comparative
summary of the results provided in Chapter 5. In general, there is a greater percent of children
expected to experience at least one daily maximum exposure at or above the benchmark
concentrations when using the alternative adjustment based on the 99th percentile design value
compared to that estimated using the adjustment based on the maximum design value (Table 6-
8). The difference was most noticeable for the Fall River study area, particularly considering the
100 ppb benchmark (i.e., 7 to 14 percentage points at the mean and maximum, respectively). The
difference was smaller when considering the 200 ppb benchmark in the Fall River study area and
both benchmarks in the two other study areas (i.e., mainly fractions of a percentage point
difference for any simulation). Further, there was also a greater percent of multiple exposures at
or above the 100 ppb benchmark in the Fall River study area using the alternative adjustment
approach, although the difference was limited to a few percentage points (Table 6-9). Only a
fractional difference in the percent of children experiencing multiple exposures at or above the
100 ppb benchmark was observed for the Indianapolis study area, and there was little to no
difference observed in any study area or when considering multiple exposures at or above the
200 ppb benchmark.
When considering lung function risk estimated using the two different adjusted air quality
surfaces, results using the 99th percentile design value for the adjustment are similar to those
estimated using the adjustment approach employing the maximum design value, although
differing slightly for the Fall River study area (Table 6-10 and 6-11). On average, about 1% of
children are estimated to experience at least one or multiple days with an increase in sRaw at or
above 100% in the Fall River and Indianapolis study areas, regardless of the adjustment
approach. Results for the Tulsa study area indicate few (<0.1%) to no children estimated to
experience any increase in sRaw of interest, neither single nor multiple days. Little to no
difference was observed for increases in sRaw at or above 200% in any study area when
considering the alternative adjustment approach, neither single nor multiple days.
6-28
-------
Table 6-8. Comparison of two approaches used to adjust ambient air concentrations to
just meet the existing standard (2011-2013): Percent of children with asthma
estimated to experience at least one day per year with a SO2 exposure at or
above 5-minute benchmark concentrations while at elevated exertion.
Benchmark
Concentration
(ppb)»
Percent of children with asthma having at least one day per
Study area
year > benchmark concentration: mean (min - max)
Max DV used to adjust air
qualityb
Max 99th DV used to adjust air
quality
100
19.4
26.7
Fall River
(12.3-32.7)
(13.8-46.8)
200
A
O
0
0.7
O
0
1
O
k>
CM
csj
I
O
100
22.4
23.0
(18.0-27.0)
(18.8-26.1)
200
0.7
0.6
Indianapolis
O
I
O,
O
I
O,
300
0.3
0.2
(0-0.8)
(0-0.7)
400
0.1
<0.1
O
I
O
io
O
I
O
100
0.1
0.4
Tulsa
A
O
I
O
So
(0-0.8)
200
0
0
a There were no daily maximum 5-minute exposures at or above 300 ppb benchmark in any study area.
b Data from Table 5-2.
c < 0.1 represents nonzero estimates below 0.1%. A value of zero (0) indicates there were no individuals having the
selected exposure in any year.
6-29
-------
Table 6-9. Comparison of two approaches used to adjust ambient air concentrations to
just meet the existing standard (2011-2013): Percent of children with asthma
estimated to experience multiple days per year with a SO2 exposure at or
above 5-minute benchmark concentrations while at elevated exertion.
Study area
Benchmark
Concentration
(ppb)
Percent of children with asthma having multiple days per year > benchmark
concentration: mean (min - max)
Max DV used to adjust air qualitya
Max 99th DV used to adjust air quality
>2 days
>4 days
>6 days
>2 days
>4 days
>6 days
Fall River
100
5.5
(1.6-12.2)
0.9
(<0.1 b-2.6)
0.2
(0b-0.6)
10.5
(2.0 - 24.0)
2.8
(0.1 -7.7)
1.0
(0 - 2.8)
200
no results included multiple days per year
at or above this benchmark concentration
<0.1
(0 - <0.1)
0
0
Indianapolis
100
6.8
(4.7-8.0)
0.8
(0.3-1.0)
0.1
(<0.1 -0.2)
6.9
(5.3-7.9)
1.0
(0.6-1.3)
0.2
(0.1-0.3)
200
no results included multiple days per year at or above this benchmark concentration
Tulsa
100
no results included multiple days per year
at or above this benchmark concentration
<0.1
(0 - <0.1)
0
0
200
no results included multiple days per year at or above this benchmark concentration
a Data from Table 5-3.
3 < 0.1 represents nonzero estimates below 0.1%. A value of zero (0) indicates there were no individuals having the selected
exposure in any year.
Table 6-10. Percent of children with asthma estimated to experience at least one day per
year with a SOi-related increase in sRaw of 100% or more while breathing at
elevated rates, air quality adjusted to just meet the existing standard.
Study area
sRaw
(%)
Percent of children with asthma having at least
one day per year > sRaw level: mean (min - max)
Max DV used to adjust air
quality3
Max 99th DV used to
adjust air quality
Fall River
100
0.9
(0.5-1.4)
1.1
(0.6-1.9)
200
0.1
(<0.1"-0.2)
0.2
(<0.1-0.4)
Indianapolis
100
1.3
(1.1-1.5)
1.3
(1.1-1.5)
200
0.3
(0.3-0.4)
0.3
(0.3-0.4)
Tulsa
100
<0.1
(0 b- <0.1)
<0.1
(<0.1 -<0.1)
200
There were no individuals that experienced a day with
this size increase in sRaw
a Data from Table 5-4.
b < 0.1 represents nonzero estimates below 0.1 %. A value of zero (0) indicates there were
no individuals having the selected exposure in any year.
6-30
-------
Table 6-11. Percent of children with asthma estimated to experience multiple days per
year with a SOi-related increase in sRaw of 100% or more while breathing at
elevated rates, air quality adjusted to just meet the existing standard.
Study area
Lung function
decrement
(increase in
sRaw)
Percent of children with asthma having multiple days per year >_sRaw
level: mean (min-max)
Max DV used to adjust air
qualitya
Max 99th DV used to adjust air
quality
>2
>4
>6
>2
>4
>6
Fall River
>100%
A
O
o
I
o
0.2
(<0.1 -0.4)
0.1
(0b-0.2)
0.6
(0.2-1.0)
0.2
(<0.1-0.4)
0.1
(<0.1-0.3)
>200%
<0.1
(0-0.1)
0
0
<0.1
(0 - 0.2)
0
0
Indianapolis
>100%
0.7
(0.6 - 0.8)
0.4
(0.4)
0.3
(0.3)
0.7
(0.7-0.8)
0.4
(0.4-0.5)
0.3
(0.3)
>200%
0.2
(0.1-0.2)
<0.1
(<0.1)
<0.1
(<0.1)
0.2
(0.1 -0.2)
<0.1
(<0.1)
<0.1
(<0.1)
Tulsa
There were no individuals experiencing an sRaw at or above any level of interest for multiple
days
a Data from Table 5-5.
b < 0.1 represents nonzero estimates below 0.1%. A value of zero (0) indicates there were no individuals having the
selected exposure in any year.
6.2.2.3 Estimating 5-minute Concentrations at Air Quality Receptors
The approach that has been used to relate the continuous 5-minute concentrations based
on ambient air measurement data to the 1-hour modeled air quality receptor concentrations is the
rank order of the hourly concentration distributions (rank-order distribution approach), as
summarized in section 3.5.2 (and also evaluated in section 6.2.2.1). To inform our consideration
of uncertainty associated with this approach, we have also evaluated two alternative approaches:
a calendar-based and concentration bin-based approach. Sensitivity analyses comparing this
approach to the two alternatives that were considered are described here.
The calendar-based approach uses the actual date and time of each of the two
concentration datasets (monitor and modeled) as the linking variable. Thus, the temporal patterns
in hourly (and hence 5-minute patterns) would be the same at all the modeled air quality
receptors, though normalized by their respective hourly concentrations that occur during that
same hour (effectively employing equation 3-3, though instead of the rank order to match hourly
concentrations, the consecutive calendar date and hour-of-day are used). We did not use the
calendar-based approach to develop the air quality surfaces used in generating the main body
exposure and risk estimates because we felt it would not appropriately represent the patterns in
5-minute concentrations, given the relationship between the within-hour 5-minute concentration
variability and the magnitude of the hourly concentrations. Often times, there is greater
6-31
-------
variability in the 5-minute concentrations occurring at low hourly concentrations (particularly
hourly values less than 1 ppb) than at higher hourly concentrations (U.S EPA, 2009, section
7.2.3.2). Further, we also expected that the monitor(s) would not necessarily reflect the exact
temporal pattern that could occur at all receptors simultaneously, given the generally sporadic
nature of peak concentrations driven by temporal and spatial variability in meteorology. That
said, this mismatching of the temporal patterns observed at the monitor with the air quality
receptors using the calendar-based approach would likely lead to instances where the 5-minute
concentrations at the upper percentiles of the distribution are overestimated (i.e., assigning
greater variability in 5-minute concentrations obtained from low hourly ambient air monitor
concentrations to the highest hourly modeled concentrations). Alternatively, 5-minute
concentrations at the lower percentiles would tend to be underestimated in certain instances. As
the estimated risk is largely a function of the highest exposures, avoiding the potential
assignment of increased variability to higher modeled concentrations was a factor in not
selecting this approach for the REA. Nevertheless, how this selection affected the exposure and
risk results warranted additional evaluation as given here.
The second alternative approach to assigning 5-minute variability to the hourly
concentrations, the concentration bin-based approach, used the actual concentration levels in
each of the two concentration datasets. Both monitor and modeled hourly concentrations were
binned by 5 ppb increments, except for the lowest hourly concentrations (i.e., three bins were
used - a 0 concentration bin, the second for hourly concentrations between 0 and 1 ppb, then a
third for hourly concentrations between 1 and 5). This concentration bin-based approach is
similar to that using the rank order distribution approach, though likely improves the matching of
the hourly concentrations between the two concentration data sets, where different (i.e.,
structurally the monitor hourly concentration distribution becomes more like the receptor hourly
concentration distribution). One limitation to the concentration bin-based approach is that there
could be limits to the monitor data set in providing measurement data to all of the bins,
particularly the highest hourly concentrations in the air quality scenario of interest for this REA
(conditions just meeting the existing standard). This was the case in the monitoring data for the
Fall River and Tulsa study areas, where the 2011-2013 monitor design values were 64 and 55
ppb, respectively. Hence, nearly all the hourly concentrations were also below the existing
standard level of 75 ppb in these two study areas. Therefore, the pattern of the 5-minute
concentrations associated with the highest hourly concentrations in those areas would all rely on
very few measurements, leading to uncertainty in their estimation.
Table 6-12 provides the statistics calculated for the upper percentiles of the 5-minute
concentrations, for air quality adjusted to just meet the existing standard, derived using each of
the three methods: (1) the rank order distribution approach (used in the assessment); (2) the
6-32
-------
calendar-based approach; and (3) the concentration bin-based approach. The table presents the 5-
minute concentrations estimated at all air quality receptor locations, along with statistics
calculated using the ambient air monitor measurement data (also adjusted to meet the standard).
Consistent with what was described above, the calendar-based approach results in unusually high
5-minute concentrations, with several receptors exhibiting concentrations at or above 300 ppb.
Neither the monitor nor the receptors using the rank order distribution-based approach had
concentrations at or above 300 ppb, while the concentration bin-based approach yielded a few
receptors (i.e., about 15 or more) with estimated 5-minute concentrations at or above that level.
Table 6-12. Comparison of three approaches for using continuous 5-minute monitoring
data to estimate 5-minute concentrations associated with modeled 1-hour
concentrations at receptor locations: Air quality adjusted to just meet the
existing standard, Fall River study area 2011.
Statistic
5-minute SO2 Concentrations in Ambient Air (ppb)
Estimation Approach
Adjusted
monitoring data
at monitor
location
Calendar
Rank Order
Distribution
Binned
p90p90
12
11
11
5
p99p90
12
11
11
maxp90
12
11
11
p90p99
29
32
31
37
p99p99
38
41
40
maxp99
45
48
48
p90max
303
183
236
241
p99max
459
247
338
maxmax
662
268
386
Abbreviations: p90 = 90th percentile of 5-minute concentrations at monitor.
p90p90 = 90th percentile of the distribution of all study area receptor 90th percentile
5-minute concentrations. Etc.
For this sensitivity analysis, all three of these approaches were used to generate an air
quality surface of 5-minute concentrations in the Fall River study area and used to simulate
exposures of children with asthma for 2011. All other model settings and input data were held
the same as in the main analysis in Chapter 5; the only difference among these three simulations
was the 5-minute concentration input described above. Table 6-13 shows the resulting estimated
exposures at or above the selected benchmarks. The largest differences among the three
approaches are estimates for the 100 ppb benchmark. There are greater percentages of children
with asthma estimated to experience at least one day with an exposure at or above 100 ppb using
the calendar-based and concentration bin-based approaches than using the rank order distribution
6-33
-------
approach. There is less variability across the three approaches when considering three or more
days with exposures at or above this benchmark. Consistent with the greater number of estimated
5-minute ambient concentrations at or above the higher benchmarks (200 through 400 ppb), the
calendar-based approach is the only approach estimating any days with exposures above these
benchmarks. Given the discussion provided above regarding this particular approach, these
results using the calendar-based approach are likely overestimates of exposure.
Table 6-13. Comparison of three approaches for using continuous 5-minute ambient air
monitoring data to estimate 5-minute concentrations associated with modeled
1-hour concentrations: Estimated exposures for air quality adjusted to just
meet the existing standard, Fall River, 2011.
benchmark
concentration
(ppb)
5-minute concentration
approach
Percent of children with asthma estimated to experience one or
more days with exposures at or above 5-minute benchmark
concentration, while breathing at elevated rates
N
umber of c
ays per year
>1
>2
>3
>4
>5
>6
100
Calendar-based
37.2
15.4
6.9
3.6
1.8
1.0
Rank order distributiona
32.7
12.2
5.5
2.6
1.3
0.6
Concentration bin-based
36.9
14.7
6.6
2.9
1.4
0.8
200
Calendar-based
4.7
0.5
0.1
0b
0
0
Rank order distribution
0.2
0
0
0
0
0
Concentration bin-base
1.2
_Q
o
V
0
0
0
0
300
Calendar
1.4
<0.1
0
0
0
0
Rank order distribution
0
0
0
0
0
0
Concentration bin-based
0
0
0
0
0
0
400
Calendar-based
0.3
0
0
0
0
0
Rank order distribution
0
0
0
0
0
0
Concentration bin-based
0
0
0
0
0
0
a Data from Table 5-2, Table 5-3, and Appendix J.
b < 0.1 represents nonzero estimates below 0.1 %. A value of zero (0) indicates there were no individuals having the selected
exposure in any year.
Table 6-14 shows the percent of children with asthma estimated to experience at least one
or more days per year with a SCb-related increase in sRaw of 100% or more while breathing at
elevated rates, using the three different approaches. The general pattern of results is similar as for
the benchmark comparison, and indicates low frequency of occurrence of lung function
decrements on at least one day or multiple days (all < 2%), at both levels of interest.
6-34
-------
Table 6-14. Comparison of three approaches for using continuous 5-minute monitoring
data to estimate 5-minute concentrations associated with modeled 1-hour
concentrations: Estimated lung function decrements associated with exposure
to air quality adjusted to just meet the existing standard, Fall River 2011.
Lung function
decrement
(increase in sRaw)
5-minute concentration
approach
Percent of children with asthma estimated to experience
one or more days with specified responsea
number of days per year
>1
>2
>3
>4
>5
>6
100%
Calendar-based
2.0
0.7
0.4
0.2
0.2
0.1
Rank order distribution
1.4
0.7
0.5
0.4
0.3
0.2
Concentration bin-based
1.6
0.7
0.4
0.3
0.2
0.2
200%
Calendar-based
0.4
<0.1b
0b
0
0
0
Rank order distribution
0.2
0.1
<0.1
0
0
0
Concentration bin-based
0.3
0.1
<0.1
0
0
0
a Data from Table 5-4, Table 5-5, and Appendix J.
b < 0.1 represents nonzero estimates below 0.1%. A value of zero (0) indicates there were no individuals having the selected
exposure in any year.
6.2.2.4 E-R Function for Lung Function Risk Estimates
The E-R functions for lung function risk were generated from the controlled human study
data provided in Table 4-12 using a probit regression (as described in section 4.6.2 above). In
addition to estimates for the risks of increases in sRaw of at least 100% and 200% based on the
best fit (mean) probit model regression coefficients, we also generated lower and upper bounds
for estimated risks using the 5th and 95th percentile predictions of the regression coefficients (see
section 4.6.2 and Appendix J, Table J-28).
For the presentation here, the lower and upper bound E-R functions were combined with
the distribution of exposures estimated in each study area, as was done using the mean regression
estimates to generate the risk estimates presented in section 5.3. As for many of the sensitivity
analyses in this chapter, the focus of this presentation is on risks for children with asthma
experiencing 5-minute exposures while breathing at elevated rates. The estimated risks using
each of the three E-R functions (for each of the two severities of response) averaged across the 3-
year study period are provided in Table 6-15.
The risks estimated for the three functions vary as expected. The highest risks (both for
single occurrences as well as multiple occurrences) are derived using the upper bound function
and the lowest with the lower bound function. With regard to the Fall River and Indianapolis risk
estimates, the differences in risk estimated using the upper bound function versus that estimated
using the mean E-R function, in terms of percent of the population, are about 2 to 3 percentage
points when considering the estimate of children experiencing at least one day per year with an
increase in sRaw of at least 100%. The differences are smaller for multiple such occurrences
6-35
-------
(e.g., a 1.4 to 1.8 percentage point difference at most considering two or more days in a year),
and also for occurrences of a 200% increase in sRaw (at most a 1.2 percentage point difference
considering at least one day per year).
Table 6-15. Comparison of estimated lung function risk using mean, lower bound and
upper bound of the fitted E-R function: Percent of children with asthma
estimated to experience at least one or multiple days per year with a SO2-
related increase in sRaw of 100% or more while breathing at elevated rates,
air quality adjusted to just meet the existing standard, 2011-2013.
Study Area
Lung function
decrement
(increase in
sRaw)
E-R Functiona
Percent of children with asthma estimated to experience
one or more days with an increase of sRaw of specified
amount (average across 3-year period)
Number of c
ays per year
>1
>2
>3
>4
>5
>6
Fall River
100%
LB
0.2
_Q
O
V
<0.1
0b
0
0
Meanc
0.9
0.4
0.3
0.2
0.1
0.1
UB
2.7
1.8
1.3
1.1
0.9
0.8
200%
LB
There were no c
increase in sRaw
lildren tha
of at least
experienc
00% usinc
ed a day with an
this E-R function
Mean
0.1
<0.1
<0.1
0
0
0
UB
1.1
0.7
0.5
0.4
0.4
0.3
Indianapolis
100%
LB
0.5
0.2
0.1
<0.1
<0.1
<0.1
Mean
1.3
0.7
0.5
0.4
0.4
0.3
UB
3.5
2.5
2.0
1.8
1.6
1.4
200%
LB
<0.1
0
0
0
0
0
Mean
0.3
0.2
0.1
<0.1
<0.1
<0.1
UB
1.5
1.1
0.9
0.8
0.7
0.7
Tulsa
100%
LB
There were no c
increase in sRaw 0
lildren tha
f at least 1
experienced a day w
D0% using either E-R
th an
linction
Mean
<0.1
0
0
0
0
0
UB
0.5
0.3
0.2
0.2
0.2
0.1
200%
LB
There were no children that experienced a day with an
increase in sRaw of at least 200% using either E-R function
Mean
UB
0.2
0.1
0.1
<0.1
<0.1
<0.1
a LB is a lower bound E-R function derived using the 5th percentile for the mean regression coefficient, Mean is the
E-R function representing the best fit (mean) regression estimate, UB is an upper bound E-R function derived using
the 95th percentile for the mean regression coefficient, each derived using the controlled human exposure-response
study data in Table 4-12. See also section 4.6.2 and Figure 4-4.
b < 0.1 represents nonzero estimates below 0.1%. A value of zero (0) indicates there were no individuals having the selected
exposure in any year.
c From main body REA results Tables 5-4 and 5-5 and Appendix J.
6-36
-------
Risk estimated using the lower bound E-R function yields a smaller percent of children
compared to that using the mean E-R function, with at most a 0.8 percentage point difference for
children experiencing at least one day per year with an increase in sRaw of at least 100% in the
Fall River and Indianapolis study areas.
Regarding the Tulsa study area, there were no children estimated to experience a 100%
increase in lung function decrement when using the lower bound function; with the mean E-R
functions, fewer than 0.1% were estimated to experience at least one day with an SCh-related
increase in sRaw of 100%. When using the upper bound function to estimate risk, a fraction of a
percent (all <0.5%) of children with asthma were estimated to experience at least one or multiple
days per year with a SCb-related increase in sRaw of 100%.
6-37
-------
REFERENCES
Burmaster DE. (1998). Lognormal distributions for skin area as a function of body weight. Risk
Analysis. 18(l):27-32.
Glen G, Smith L, Isaacs K, McCurdy T, Langstaff J. (2008). A new method of longitudinal diary
assembly for human exposure modeling. J Expos Sci Environ Epidem. 18:299-311.
Haugen HA, Melanson EL, Tran ZV, Kearney JT, Hill LO. (2003). Am J Clin Nutr. 78:1141-
1144.
Isaacs K, Glen G, McCurdy T., and Smith L. (2007). Modeling energy expenditure and oxygen
consumption in human exposure models: Accounting for fatigue and EPOC. J Expos Sci
Environ Epidemiol. 18(3):289-98.
Langstaff JE. (2007). Memorandum to Ozone NAAQS Review Docket (EPA-HQ-OAR-2005-
0172). Analysis of Uncertainty in Ozone Population Exposure Modeling. Docket ID No.
EPA-HQ-OAR-2005-0172-0174.
Leonard WR, Levy SB, Tarskaia LA, Klimova TM, Fedorova VI, Baltakhinova ME,
Krivoshapkin VG, Snodgrass JJ. (2014). Seasonal Variation in Basal Metabolic Rates
Among the Yakut (Sakha) of Northeastern Siberia. American Journal of Human Biology.
26(4):437-445.
Ogden CL, Lamb MM, Carroll MD, Flegal KM. (2010). Obesity and Socioeconomic Status in
Children and Adolescents: United States, 2005-2008. NCHS Data Brief. Number 51.
Available at: https://files.eric.ed.eov/fulltext/ED530165.pdf
Rizzo M. (2008). Memorandum to Sulfur Dioxide Review Docket (EPA-HQ-OAR-2007-0352).
Investigation of How Distributions of Hourly Sulfur Dioxide Concentrations Have
Changed Over Time in Six Cities. Document ID No. EPA-HQ-OAR-2007-03 52-0017
U.S. Census Bureau. (2012). Technical documentation - 2010 Census Summary File 1—
Technical Documentation/prepared by the U.S. Census Bureau, Revised 2012. Available
at: http://www.census.gov/prod/cen2010/doc/sfl.pdf. Employment Status from the 5-year
American Community Survey (ACS) data, 2010 U.S. Census American FactFinder.
Available at: http://factfinder2.censiis.eov/. Commuting times file from U.S. Census data
portal (http://dataferrettcensus.gov/). Table P31, variables P031001-P031015.
U.S. Census Bureau. (2013). 2013 American Housing Survey (AHS). Available at:
https://www.census.gov/programs-survevs/ahs/data/interactive/ahstablecreator.html
U.S. DOT (2012). Bureau of Transportation Statistics, Census Transportation Planning Package,
Part 3-The Journey to Work. Available at: http://tran.stats.bts.gov/
U.S. EPA. (2007). Ozone Population Exposure Analysis for Selected Urban Areas. Office of Air
Quality Planning and Standards, Research Triangle Park, NC, EPA-452-R-07-010, July
2007. Available at: http://www.epa.eov/ttn/naaqs/standards/ozone/s o3 cr td.html
6-38
-------
U.S. EPA. (2008). Risk and Exposure Assessment to Support the Review of the NO2 Primary
National Ambient Air Quality Standard. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-08-008a, November 2008. Available at:
https://www3.epa.gOv/ttn/naaas/standards/nox/s nox cr rea.html
U.S. EPA. (2009). Risk and Exposure Assessment to Support the Review of the SO2 Primary
National Ambient Air Quality Standard. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, EPA-452/R-09-007, July 2009. Available at:
https://www3. epa. gov/ttn/naaq s/standards/ so2/data/200908 SQ2REAFinalReport.pdf
U.S. EPA. (2010). Quantitative Risk and Exposure Assessment for Carbon Monoxide -
Amended. Office of Air Quality Planning and Standards, Research Triangle Park, NC,
EPA-452/R-10-006, July 2010. Available at: https://www.epa.gov/naaqs/carbon-
monoxide-co-standards-risk-and-exposure-assessments-current-review
U.S. EPA. (2011). Highlights of the Exposure Factors Handbook (Final Report). Office of
Research and Development, National Center for Environmental Assessment,
Washington, DC, EPA/600/R-10/030, 2011. Available at:
http s: //cfpub .epa. gov/n cea/ri sk/recordi splay, cfm ? del d=221023
U.S. EPA. (2014). Health Risk and Exposure Assessment for Ozone. Office of Air Quality
Planning and Standards, Research Triangle Park, NC, EPA-452/R-14-004a, August 2014.
Available at: https://www.epa.gov/naaqs/ozone-o3-standards-risk-and-exposure-
assessments-current-review
U.S. EPA. (2015). 2011 National Emissions Inventory, version 2 Technical Support Document
Documentation, available at: https://www.epa.gov/sites/production/files/2Q15-
10/documents/nei2011 v2 tsd 14aug2015 pif. Data available at:
https://www.epa.gQv/air-eroissiQins-4 inventories/^ iti on al-em is si ons-inventory-nei-
data
U.S. EPA. (2017a). Air Pollutants Exposure Model Documentation (APEX, Version 5)
Volume I: User's Guide. Office of Air Quality Planning and Standards, Research
Triangle Park, NC, January 2017. EPA-452/R-17-001a. Available at:
https://www.epa.gQv/fera/apex-user-giiides
U.S. EPA. (2017b). Air Pollutants Exposure Model Documentation (APEX, Version 5)
Volume II: Technical Support Document. Office of Air Quality Planning and Standards,
Research Triangle Park, NC, January 2017. EPA-452/R-17-001b. Available at:
https://www.epa.gQv/fera/apex-user-giiides
U.S. EPA. (2017c). The Consolidated Human Activity Database - Master Version (CHAD-
Master). Technical Memorandum. Office of Research and Development, National
Exposure Research Laboratory, Research Triangle Park, NC, 27711. In preparation.
Previous version (09/15/2014). Available at:
https://www.epa.gov/health.research/consolidated-huro.an-activitv-database-chad-iise-
huroan-expQsure-and-health-studies-and
6-39
-------
van Ooijen AMJ, van Marken Lichtenbelt WD, van Steenhovenb AA, Westerterp KR. (2004).
Seasonal changes in metabolic and temperature responses to cold air in humans.
Physiology & Behavior. 82:545-553.
WHO. (2008). WHO/IPCS Harmonization Project Document No. 6. Part 1: Guidance Document
on Characterizing and Communicating Uncertainty in Exposure Assessment.
International Programme on Chemical Safety, World Health Organization, Geneva,
Switzerland. Available at:
http://www.who.int/ipcs/methods/harmonization/areas/exposure/en/
6-40
-------
APPENDIX A
SURFACE CHARACTERISTIC VALUES AND METEOROLOGICAL DATA
PREPARATION FOR INPUT TO AIR QUALITY MODELING
A.l Introduction
Air quality dispersion modeling was performed for three study areas to support the SO2
Risk and Exposure Assessment, including: Fall River, MA; Indianapolis, IN; and Tulsa, OK.
Each of the three study areas was modeled for the same three-year period, 2011-2013. National
Weather Service (NWS) meteorological data were used as meteorological input to AERMOD
(U.S. EPA, 2016a), preprocessed with AERMET (v. 16216) (U.S. EPA, 2016b), the
meteorological preprocessor for AERMOD.
AERMET requires continuous hourly surface meteorological observations and concurrent
twice daily upper air sounding data. The surface and upper air data should be representative of
the modeling domain. The NWS and the Federal Aviation Administration (FAA) jointly operate
and maintain a network of Automated Surface Observing Systems (ASOS) at airports throughout
the U.S. Upper air data are collected by the NWS at 69 stations across the conterminous U.S.
Table A-l and Table A-2 lists the NWS surface and upper air stations selected for each of the
study areas. Figure A-l through Figure A-5 show the locations of the ASOS and upper air
stations selected for each study area, relative to emission sources that were modeled.
Table A-l. National Weather Service surface stations.
Study Area
Station
Identifier
WMO
(WBAN)
Latitude
(degrees)
Longitude
(degrees)
Elevation
(m)
GMT Offset
(hours)
Fall River
Providence
PVD
725070
(14765)
41.7225
-71.4325
19
-5
Indianapolis
Indianapolis
International
Airport
IND
724380
(93819)
39.725170
-86.281680
241
-5
Tulsa
Tulsa R. L.
Jones Jr. Airport
RVS
723564
(53908)
36.042441
-95.990166
192
-6
A-l
-------
Table A-2. National Weather Service upper air stations.
Study Area
Station
Identifier
WMO
(WBAN)
Latitude
_Jdegrees)
Longitude
(degrees)
Elevation
(m)
GMT Offset
(hours)
Fall River
Chatham, MA
CHH
744940
(14684)
41.67
-69,97
12
-5
Indianapolis
Lincoln, IL
ILX
745600
(04833)
40.15
-89.33
178
-6
Tulsa
Norman, OK
OUN
723560
(13968)
35.23
-97,47
354
-6
Fall River, MA: Surface and Upper Air Stations and Emission Sources
0 10 20 30 40 km
~ ASOS Surface Station ^ ~\
f Upper Air Station
~ Emission Source
Prniary Roads
County Boundaries
I I State Boundaries
Figure A-l. Location of surface and upper air meteorological stations and emission sources
for Fall River, MA.
A-2
-------
Marion County, IN: Surface and Upper Air Stations
~ ASOS Surface Station
f Upper Air Station
~ Emission Source
Primary Roads
County Boundaries
I I State Boundaries
Figure A-2. Location of surface and upper air meteorological stations selected for
Indianapolis, IN.
~ ASOS Surface Station
| Upper Air Station
~ Emission Source
Primary Roads
County Boundaries
I I State Boundaries
Marion County, IN: Emission Sources
Figure A-3. Location of emission sources for Indianapolis, IN.
A-3
-------
~ ASOS Surface Station
f Upper Air Station
~ Emission Source
Primary Roads
County Boundaries
I I State Boundaries
Figure A-4. Location of surface and upper air meteorological stations selected for Tulsa,
OK.
Tulsa, OK: Emission Sources
~ ASOS Surface Station
f Upper Air Station
~ Emission Source
Primary Roads
County Boundaries
I I State Boundaries
Figure A-5. Location of emission sources for Tulsa, OK.
A-4
-------
In addition to surface and upper air meteorological data, AERMET also requires the user
to input values of surface albedo, Bowen ratio, and roughness length that are representative of
the location where the surface observations are taken. Surface characteristic values were
estimated using the AERSURFACE (v. 13016) (U.S. EPA, 2013).
The remainder of this document describes the preparation of the meteorological data files
input to AERMOD for each of the three study areas. Section A.2 describes the preparation of the
surface and upper air data for input to AERMET. Section A.3 describes the estimation of surface
characteristic values using AERSURFACE, and Section A.4 describes the AERMET processing
with a brief analysis of the AERMET output for each of the study areas.
A.2 Preparation of the Surface and Upper Air Meteorological Data
A.2.1 Surface Data
Three years of surface data for 2011-2013 were downloaded from the Integrated Surface
Hourly (ISH) archive maintained by the National Oceanic and Atmospheric Administration
(NOAA) National Centers for Environmental Information (NCEI), formerly the National
Climatic Data Center (NCDC). The data are accessible for download via File Transfer Protocol
(FTP) at ftp: //ftp. n cdc. n oaa. gov.
A potential concern related to the use of NWS meteorological data for dispersion
modeling is the often high incidence of calms and variable wind conditions in the Integrated
Surface Hourly (ISH) data. This is due to the implementation of the ASOS program to replace
observer-based data beginning in the mid-1990's, and the adoption of the METAR standard for
reporting NWS observations in July 1996. Currently, the wind speed and direction used to
represent the hour in AERMOD is based on a single two-minute average, usually reported about
10 minutes before the hour. The METAR system reports winds of less than three knots as calm
(coded as 0 knots), and winds up to six knots will be reported as variable when the variation in
the 2-minute wind direction is more than 60 degrees. This variable wind is reported as a non-zero
wind speed with a missing wind direction. The number of calms and variable winds can
influence concentration calculations in AERMOD because concentrations are not calculated for
calms or variable wind hours. Significant numbers of calm and variable hours may compromise
the representativeness of NWS surface data for AERMOD applications. This is especially of
concern for applications involving low-level releases since the worst-case dispersion conditions
for such sources are associated with low wind speeds, and the hours being discarded as calm or
variable are biased toward this condition.
The NCEI maintains a separate archive of 1-minute wind data for each of the ASOS
surface stations. These wind data represent 2-minute average wind speeds calculated for each
minute of the hour. To reduce the number of calms and missing winds, these wind data were
A-5
-------
used to calculate hourly average wind speed and direction to replace the standard archive of
winds in the ISH dataset. The 1-minute data were processed with AERMINUTE (v. 15272) (U.S.
EPA, 2015), which calculates the hourly wind speed and wind direction and generates a file
formatted for input directly to AERMET, where the ISH wind data are replaced during
processing. The NCEI archives the 1-minute ASOS wind data as monthly files. Monthly 1-minute
data files were downloaded for the 2011-2013 period for each ASOS surface stations listed in
Table A-1.
A.2.2 Upper Air Data
Three years (2011-2013) of upper air sounding data were downloaded for each of the
upper air stations listed in Table A-2 from the NOAA/ Earth System Research Laboratory
(ESRL) Radiosonde Database (https://mc.noaa. gov/raobs/). The upper air data are archived in
the Forecast System Laboratory (FSL) format and maintained by the Global Systems Division,
formerly the FSL. Data for each station was downloaded as a separate file as required by
AERMET.
A.3 Estimation of Surface Characteristics Using AERSURFACE
As previously stated, surface values for albedo, Bowen ratio, and roughness length were
estimated using the AERSURFACE tool. As noted in the AERSURFACE User's Guide (U.S.
EPA, 2013), surface characteristics that are input to AERMET should be representative of the
location of the meteorological tower. AERSURFACE was run for the location of each of the
three ASOS stations using the geographic coordinates of the meteorological towers in Table A-l.
The current version of AERSURFACE utilizes 1992 land cover data from the National
Land Cover Database (NLCD) in GeoTIFF format. NLCD data files for the three ASOS stations
were downloaded from the Multi-Resolution Land Characteristics consortium website
(https://www.mrlc.gov).
AERSURFACE can generate annual, seasonal, or monthly surface characteristic values
in a format for input directly into AERMET. Monthly values were generated for each of the
locations. To properly interpret some of the land cover categories in the 1992 NLCD data,
AERSURFACE requires the user to specify whether or not the location of the weather station is
at an airport. All three ASOS stations were specified as airport locations. AERSURFACE also
allows for the surface roughness length to be defined by up to 12 wind sectors with a minimum
arc of 30 degrees each. For each of the three locations, roughness was estimated for each of 12
sectors, beginning at 0 degrees through 360 degrees {i.e., 0-30, 30-60, 60-90, etc.). The
roughness length sectors at each of the three ASOS stations are illustrated in Figure A-6 through
A-6
-------
Figure A-8. The sectors extend from the location of the meteorological tower out to 1 km, the
distance over which the roughness length is estimated.
Er.
Figure A-6. Surface roughness sectors for Providence Airport (PVD).
A-7
-------
Figure A-7. Surface roughness sectors for Indianapolis International Airport (1ND).
Figure A-8. Surface roughness sectors for Tulsa R. L. Jones Airport (RVS).
Values for the three surface characteristics are defined within AERSURFACE by season
but are computed monthly based on the assignment of months to seasons. Monthly values are
A-8
-------
then rolled up to seasonal or annual values based on the option specified by the user. The user
has the option to use default month-to-season assignments or input user-defined assignments.
Seasonal surface characteristic values are defined based on five season definitions: spring,
summer, autumn, winter with no snow, and winter with continuous snow cover. Note, there are
two winter options: 1) winter with no snow (or without continuous snow) on the ground the
entire month and 2) winter with continuous snow on ground the entire month.1 AERSURFACE
was run for Tulsa using the default month-to-season assignments, while months were reassigned
for both Indianapolis and Fall River. The month-to-season assignments used for each of the three
surface stations are shown in Table A-3, along with the seasonal definitions. A month was
considered to have continuous snow cover if a snow depth of one inch or more was reported for
at least 75% of the days in the month.
Table A-3. AERSURFACE month-to-season assignments.
Winter
Winter
Station (continuous snow)
(no snow)
Spring
Summer Autumn
PVD Feb (2015 only)
Dec, Jan, Feb, Mar
Apr, May
Jun., Jul, Aug Sep, Oct, Nov
IND
Dec, Jan, Feb, Mar
Apr, May
Jun., Jul, Aug Sep, Oct, Nov
RVS
Dec, Jan, Feb
Mar, Apr, May
Jun., Jul, Aug Sep, Oct, Nov
Seasonal definitions: Winter: Late autumn after frost and harvest, or winter with no snow; Spring: Transitional spring with
partial green coverage or short annuals; Summer: Midsummer with lush vegetation; Autumn: Autumn with unharvested
cropland
AERSURFACE also requires information about the climate and surface moisture at the
surface station. The climate at the station location is categorized as either arid or non-arid. Each
of the three surface station locations was categorized as non-arid in AERSURFACE. Surface
moisture is based on precipitation amounts and is categorized as either wet, average, or dry. For
the three surface stations, 2010 local climatological data from the NCEI was used to look at 30
years (1981-2010) of monthly precipitation. The 30th and 70th percentiles of precipitation
amounts were calculated for each of 12 months (Jan. - Dec.) based on the 30-year period. The
precipitation amount for each month in 2011-2013 was then compared to the 30th and 70th
percentiles for the corresponding month. Months during which precipitation was greater than the
70th percentile were considered wet while months that were less than the 30th percentile were
considered dry. Months within the 30th and 70th percentile range were considered average.
AERSURFACE was run for each moisture condition to obtain monthly values for wet, dry, and
average conditions. Using the AERSURFACE output for each of the three moisture categories, a
1 For many of the land cover categories in the 1992 NLCD classification scheme, the designation of winter with
continuous snow on the ground would tend to increase wintertime albedo (reflectivity) and decrease wintertime
Bowen ratio (sensible to latent heat flux) and surface roughness compared to the winter with no snow or without
continuous snow designation.
A-9
-------
separate set of monthly surface characteristics was compiled for each of the three years for input
to AERMET. The monthly categorization of the surface moisture at each of the locations is
shown in Table A-4. The resulting surface characteristic values input to AERMET, by sector,
month, and year, are listed in Table A-6 through Table A-8 at the end of this document.
Table A-4. Monthly surface moisture categorizations.
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
PVD
2011
Avg
Wet
Dry
Wet
Avg
Wet
Wet
Wet
Wet
Wet
Wet
Avg
2012
Avg
Dry
Dry
Avg
Wet
Wet
Avg
Wet
Wet
Wet
Dry
Wet
2013
Dry
Wet
Dry
Dry
Avg
Wet
Avg
Wet
Wet
Dry
Wet
Wet
IND
2011
Wet
Wet
Wet
Wet
Wet
Wet
Dry
Dry
Wet
Wet
Wet
Wet
2012
Wet
Avg
Wet
Avg
Dry
Dry
Dry
Wet
Wet
Wet
Dry
Avg
2013
Wet
Wet
Wet
Wet
Wet
Wet
Dry
Dry
Wet
Wet
Wet
Wet
RVS (Moisture conditions at RVS are based on precipitation data from Tulsa International Airport, TUL)
2011
Dry
Wet
Dry
Wet
Dry
Dry
Dry
Wet
Dry
Dry
Wet
Avg
2012
Dry
Avg
Wet
Avg
Dry
Wet
Dry
Wet
Dry
Avg
Dry
Dry
2013
Wet
Wet
Dry
Avg
Avg
Dry
Wet
Wet
Dry
Wet
Avg
Avg
A.4 AERMET Processing
The meteorological data files (upper air, ISH data, and 1-minute hourly averaged wind
data) for each station were processed in AERMET. Each year was processed separately using the
monthly surface characteristics specific to each year. AERMET processes the meteorological
data in three "Stages." Stage 1 reads in the upper air and ISH data files and performs an initial
QA on the values. Stage 2 reads the 1-minute averaged wind data and merges the three data sets
into a single file. Stage 3 performs data replacements and substitutions as specified by the user,
computes the boundary layer parameters, and generates data files formatted for input to
AERMOD. Surface characteristics were input during Stage 3. When 1-minute hourly averaged
winds were available, those winds were used for the hour, while all other surface data are from
the ISH data (temperature, cloud cover, precipitation, etc.).
Table A-5 shows the percentage of calm and missing winds in the AERMET output for
the combined three years (2011-2013) for each of the surface stations. These values take into
account the replacement of the ISH wind data with the 1-minute hourly averaged wind data
during the AERMET Stage 3 processing. Figure A-9 through Figure A-l 1 are wind roses
generated from the 2011-2013 surface data files output by AERMET for three surface stations.
A-10
-------
Table A-5. Percent calm and missing winds in AERMET surface file.
Station
% Calm
% Missing
PVD
0.49
0.06
IND
0.37
0.10
RVS
3.90
0.22
NORTH"
SOUTH
WIND SPEED
(m/s)
~ -HI
I I a.8-11.1
LB 5.7- B.8
[J 3.6- 5.7
~1 2.1 - 36
I I 0.5-2.1
Calms: 0.49%
Figure A-9. Wind rose for Providence Airport (PVD), 2011-2013 (direction blowing from).
A-ll
-------
NORTH'
WEST
SOUTH
WIND SPEED
(m/s)
~ >-11,1
[ 1 8.8-11.1
[] 5.7- 6.3
| 3.6- 5.7
rj 2.1 - 3.6
~ Q.S-2.1
Calms: 0.36%
Figure A-10. Wind rose for Indianapolis International Airport (IND), 2011-2013 (direction
blowing from).
VJ-TH
.. E':
r . --PE-J
5.7- E.5
sou -
3.6 - 5.7
~ Q.5- 2
3]"" s: o.iu'i
Figure A-11. Wind rose for Tulsa R. L. Jones Jr. Airport (RVS), 2011-2013 (direction
blowing from).
A-12
-------
Table A-6. Surface characteristics for Providence Airport (PVD) by month and year.
Station = PVD
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Jan
0-30
0.16
0.64
0.023
0.16
0.64
0.023
0.16
1.24
0.023
Jan
30-60
0.16
0.64
0.022
0.16
0.64
0.022
0.16
1.24
0.022
Jan
60-90
0.16
0.64
0.026
0.16
0.64
0.026
0.16
1.24
0.026
Jan
90-120
0.16
0.64
0.036
0.16
0.64
0.036
0.16
1.24
0.036
Jan
120-150
0.16
0.64
0.041
0.16
0.64
0.041
0.16
1.24
0.041
Jan
150-180
0.16
0.64
0.027
0.16
0.64
0.027
0.16
1.24
0.027
Jan
180-210
0.16
0.64
0.018
0.16
0.64
0.018
0.16
1.24
0.018
Jan
210-240
0.16
0.64
0.038
0.16
0.64
0.038
0.16
1.24
0.038
Jan
240-270
0.16
0.64
0.038
0.16
0.64
0.038
0.16
1.24
0.038
Jan
270-300
0.16
0.64
0.053
0.16
0.64
0.053
0.16
1.24
0.053
Jan
300-330
0.16
0.64
0.081
0.16
0.64
0.081
0.16
1.24
0.081
Jan
330-360
0.16
0.64
0.030
0.16
0.64
0.030
0.16
1.24
0.030
Feb
0-30
0.16
0.40
0.023
0.16
1.24
0.023
0.16
0.40
0.023
Feb
30-60
0.16
0.40
0.022
0.16
1.24
0.022
0.16
0.40
0.022
Feb
60-90
0.16
0.40
0.026
0.16
1.24
0.026
0.16
0.40
0.026
Feb
90-120
0.16
0.40
0.036
0.16
1.24
0.036
0.16
0.40
0.036
Feb
120-150
0.16
0.40
0.041
0.16
1.24
0.041
0.16
0.40
0.041
Feb
150-180
0.16
0.40
0.027
0.16
1.24
0.027
0.16
0.40
0.027
Feb
180-210
0.16
0.40
0.018
0.16
1.24
0.018
0.16
0.40
0.018
Feb
210-240
0.16
0.40
0.038
0.16
1.24
0.038
0.16
0.40
0.038
Feb
240-270
0.16
0.40
0.038
0.16
1.24
0.038
0.16
0.40
0.038
Feb
270-300
0.16
0.40
0.053
0.16
1.24
0.053
0.16
0.40
0.053
Feb
300-330
0.16
0.40
0.081
0.16
1.24
0.081
0.16
0.40
0.081
Feb
330-360
0.16
0.40
0.030
0.16
1.24
0.030
0.16
0.40
0.030
Mar
0-30
0.16
1.24
0.023
0.16
1.24
0.023
0.16
1.24
0.023
Mar
30-60
0.16
1.24
0.022
0.16
1.24
0.022
0.16
1.24
0.022
Mar
60-90
0.16
1.24
0.026
0.16
1.24
0.026
0.16
1.24
0.026
Mar
90-120
0.16
1.24
0.036
0.16
1.24
0.036
0.16
1.24
0.036
Mar
120-150
0.16
1.24
0.041
0.16
1.24
0.041
0.16
1.24
0.041
Mar
150-180
0.16
1.24
0.027
0.16
1.24
0.027
0.16
1.24
0.027
Mar
180-210
0.16
1.24
0.018
0.16
1.24
0.018
0.16
1.24
0.018
Mar
210-240
0.16
1.24
0.038
0.16
1.24
0.038
0.16
1.24
0.038
Mar
240-270
0.16
1.24
0.038
0.16
1.24
0.038
0.16
1.24
0.038
Mar
270-300
0.16
1.24
0.053
0.16
1.24
0.053
0.16
1.24
0.053
Mar
300-330
0.16
1.24
0.081
0.16
1.24
0.081
0.16
1.24
0.081
Mar
330-360
0.16
1.24
0.030
0.16
1.24
0.030
0.16
1.24
0.030
Apr
0-30
0.15
0.37
0.029
0.15
0.53
0.029
0.15
1.05
0.029
Apr
30-60
0.15
0.37
0.029
0.15
0.53
0.029
0.15
1.05
0.029
A-13
-------
Station = PVD
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Apr
60-90
0.15
0.37
0.034
0.15
0.53
0.034
0.15
1.05
0.034
Apr
90-120
0.15
0.37
0.047
0.15
0.53
0.047
0.15
1.05
0.047
Apr
120-150
0.15
0.37
0.052
0.15
0.53
0.052
0.15
1.05
0.052
Apr
150-180
0.15
0.37
0.036
0.15
0.53
0.036
0.15
1.05
0.036
Apr
180-210
0.15
0.37
0.025
0.15
0.53
0.025
0.15
1.05
0.025
Apr
210-240
0.15
0.37
0.051
0.15
0.53
0.051
0.15
1.05
0.051
Apr
240-270
0.15
0.37
0.045
0.15
0.53
0.045
0.15
1.05
0.045
Apr
270-300
0.15
0.37
0.062
0.15
0.53
0.062
0.15
1.05
0.062
Apr
300-330
0.15
0.37
0.088
0.15
0.53
0.088
0.15
1.05
0.088
Apr
330-360
0.15
0.37
0.037
0.15
0.53
0.037
0.15
1.05
0.037
May
0-30
0.15
0.53
0.029
0.15
0.37
0.029
0.15
0.53
0.029
May
30-60
0.15
0.53
0.029
0.15
0.37
0.029
0.15
0.53
0.029
May
60-90
0.15
0.53
0.034
0.15
0.37
0.034
0.15
0.53
0.034
May
90-120
0.15
0.53
0.047
0.15
0.37
0.047
0.15
0.53
0.047
May
120-150
0.15
0.53
0.052
0.15
0.37
0.052
0.15
0.53
0.052
May
150-180
0.15
0.53
0.036
0.15
0.37
0.036
0.15
0.53
0.036
May
180-210
0.15
0.53
0.025
0.15
0.37
0.025
0.15
0.53
0.025
May
210-240
0.15
0.53
0.051
0.15
0.37
0.051
0.15
0.53
0.051
May
240-270
0.15
0.53
0.045
0.15
0.37
0.045
0.15
0.53
0.045
May
270-300
0.15
0.53
0.062
0.15
0.37
0.062
0.15
0.53
0.062
May
300-330
0.15
0.53
0.088
0.15
0.37
0.088
0.15
0.53
0.088
May
330-360
0.15
0.53
0.037
0.15
0.37
0.037
0.15
0.53
0.037
Jun
0-30
0.15
0.36
0.035
0.15
0.36
0.035
0.15
0.36
0.035
Jun
30-60
0.15
0.36
0.035
0.15
0.36
0.035
0.15
0.36
0.035
Jun
60-90
0.15
0.36
0.040
0.15
0.36
0.040
0.15
0.36
0.040
Jun
90-120
0.15
0.36
0.056
0.15
0.36
0.056
0.15
0.36
0.056
Jun
120-150
0.15
0.36
0.061
0.15
0.36
0.061
0.15
0.36
0.061
Jun
150-180
0.15
0.36
0.043
0.15
0.36
0.043
0.15
0.36
0.043
Jun
180-210
0.15
0.36
0.031
0.15
0.36
0.031
0.15
0.36
0.031
Jun
210-240
0.15
0.36
0.059
0.15
0.36
0.059
0.15
0.36
0.059
Jun
240-270
0.15
0.36
0.050
0.15
0.36
0.050
0.15
0.36
0.050
Jun
270-300
0.15
0.36
0.068
0.15
0.36
0.068
0.15
0.36
0.068
Jun
300-330
0.15
0.36
0.094
0.15
0.36
0.094
0.15
0.36
0.094
Jun
330-360
0.15
0.36
0.042
0.15
0.36
0.042
0.15
0.36
0.042
Jul
0-30
0.15
0.36
0.035
0.15
0.49
0.035
0.15
0.49
0.035
Jul
30-60
0.15
0.36
0.035
0.15
0.49
0.035
0.15
0.49
0.035
Jul
60-90
0.15
0.36
0.040
0.15
0.49
0.040
0.15
0.49
0.040
Jul
90-120
0.15
0.36
0.056
0.15
0.49
0.056
0.15
0.49
0.056
Jul
120-150
0.15
0.36
0.061
0.15
0.49
0.061
0.15
0.49
0.061
A-14
-------
Station = PVD
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Jul
150-180
0.15
0.36
0.043
0.15
0.49
0.043
0.15
0.49
0.043
Jul
180-210
0.15
0.36
0.031
0.15
0.49
0.031
0.15
0.49
0.031
Jul
210-240
0.15
0.36
0.059
0.15
0.49
0.059
0.15
0.49
0.059
Jul
240-270
0.15
0.36
0.050
0.15
0.49
0.050
0.15
0.49
0.050
Jul
270-300
0.15
0.36
0.068
0.15
0.49
0.068
0.15
0.49
0.068
Jul
300-330
0.15
0.36
0.094
0.15
0.49
0.094
0.15
0.49
0.094
Jul
330-360
0.15
0.36
0.042
0.15
0.49
0.042
0.15
0.49
0.042
Aug
0-30
0.15
0.36
0.035
0.15
0.36
0.035
0.15
0.36
0.035
Aug
30-60
0.15
0.36
0.035
0.15
0.36
0.035
0.15
0.36
0.035
Aug
60-90
0.15
0.36
0.040
0.15
0.36
0.040
0.15
0.36
0.040
Aug
90-120
0.15
0.36
0.056
0.15
0.36
0.056
0.15
0.36
0.056
Aug
120-150
0.15
0.36
0.061
0.15
0.36
0.061
0.15
0.36
0.061
Aug
150-180
0.15
0.36
0.043
0.15
0.36
0.043
0.15
0.36
0.043
Aug
180-210
0.15
0.36
0.031
0.15
0.36
0.031
0.15
0.36
0.031
Aug
210-240
0.15
0.36
0.059
0.15
0.36
0.059
0.15
0.36
0.059
Aug
240-270
0.15
0.36
0.050
0.15
0.36
0.050
0.15
0.36
0.050
Aug
270-300
0.15
0.36
0.068
0.15
0.36
0.068
0.15
0.36
0.068
Aug
300-330
0.15
0.36
0.094
0.15
0.36
0.094
0.15
0.36
0.094
Aug
330-360
0.15
0.36
0.042
0.15
0.36
0.042
0.15
0.36
0.042
Sep
0-30
0.15
0.40
0.029
0.15
0.40
0.029
0.15
0.40
0.029
Sep
30-60
0.15
0.40
0.029
0.15
0.40
0.029
0.15
0.40
0.029
Sep
60-90
0.15
0.40
0.034
0.15
0.40
0.034
0.15
0.40
0.034
Sep
90-120
0.15
0.40
0.048
0.15
0.40
0.048
0.15
0.40
0.048
Sep
120-150
0.15
0.40
0.053
0.15
0.40
0.053
0.15
0.40
0.053
Sep
150-180
0.15
0.40
0.036
0.15
0.40
0.036
0.15
0.40
0.036
Sep
180-210
0.15
0.40
0.025
0.15
0.40
0.025
0.15
0.40
0.025
Sep
210-240
0.15
0.40
0.051
0.15
0.40
0.051
0.15
0.40
0.051
Sep
240-270
0.15
0.40
0.045
0.15
0.40
0.045
0.15
0.40
0.045
Sep
270-300
0.15
0.40
0.062
0.15
0.40
0.062
0.15
0.40
0.062
Sep
300-330
0.15
0.40
0.088
0.15
0.40
0.088
0.15
0.40
0.088
Sep
330-360
0.15
0.40
0.037
0.15
0.40
0.037
0.15
0.40
0.037
Oct
0-30
0.15
0.40
0.029
0.15
0.40
0.029
0.15
1.24
0.029
Oct
30-60
0.15
0.40
0.029
0.15
0.40
0.029
0.15
1.24
0.029
Oct
60-90
0.15
0.40
0.034
0.15
0.40
0.034
0.15
1.24
0.034
Oct
90-120
0.15
0.40
0.048
0.15
0.40
0.048
0.15
1.24
0.048
Oct
120-150
0.15
0.40
0.053
0.15
0.40
0.053
0.15
1.24
0.053
Oct
150-180
0.15
0.40
0.036
0.15
0.40
0.036
0.15
1.24
0.036
Oct
180-210
0.15
0.40
0.025
0.15
0.40
0.025
0.15
1.24
0.025
Oct
210-240
0.15
0.40
0.051
0.15
0.40
0.051
0.15
1.24
0.051
A-15
-------
Station = PVD
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Oct
240-270
0.15
0.40
0.045
0.15
0.40
0.045
0.15
1.24
0.045
Oct
270-300
0.15
0.40
0.062
0.15
0.40
0.062
0.15
1.24
0.062
Oct
300-330
0.15
0.40
0.088
0.15
0.40
0.088
0.15
1.24
0.088
Oct
330-360
0.15
0.40
0.037
0.15
0.40
0.037
0.15
1.24
0.037
Nov
0-30
0.15
0.40
0.029
0.15
1.24
0.029
0.15
0.40
0.029
Nov
30-60
0.15
0.40
0.029
0.15
1.24
0.029
0.15
0.40
0.029
Nov
60-90
0.15
0.40
0.034
0.15
1.24
0.034
0.15
0.40
0.034
Nov
90-120
0.15
0.40
0.048
0.15
1.24
0.048
0.15
0.40
0.048
Nov
120-150
0.15
0.40
0.053
0.15
1.24
0.053
0.15
0.40
0.053
Nov
150-180
0.15
0.40
0.036
0.15
1.24
0.036
0.15
0.40
0.036
Nov
180-210
0.15
0.40
0.025
0.15
1.24
0.025
0.15
0.40
0.025
Nov
210-240
0.15
0.40
0.051
0.15
1.24
0.051
0.15
0.40
0.051
Nov
240-270
0.15
0.40
0.045
0.15
1.24
0.045
0.15
0.40
0.045
Nov
270-300
0.15
0.40
0.062
0.15
1.24
0.062
0.15
0.40
0.062
Nov
300-330
0.15
0.40
0.088
0.15
1.24
0.088
0.15
0.40
0.088
Nov
330-360
0.15
0.40
0.037
0.15
1.24
0.037
0.15
0.40
0.037
Dec
0-30
0.16
0.64
0.023
0.16
0.40
0.023
0.16
0.40
0.023
Dec
30-60
0.16
0.64
0.022
0.16
0.40
0.022
0.16
0.40
0.022
Dec
60-90
0.16
0.64
0.026
0.16
0.40
0.026
0.16
0.40
0.026
Dec
90-120
0.16
0.64
0.036
0.16
0.40
0.036
0.16
0.40
0.036
Dec
120-150
0.16
0.64
0.041
0.16
0.40
0.041
0.16
0.40
0.041
Dec
150-180
0.16
0.64
0.027
0.16
0.40
0.027
0.16
0.40
0.027
Dec
180-210
0.16
0.64
0.018
0.16
0.40
0.018
0.16
0.40
0.018
Dec
210-240
0.16
0.64
0.038
0.16
0.40
0.038
0.16
0.40
0.038
Dec
240-270
0.16
0.64
0.038
0.16
0.40
0.038
0.16
0.40
0.038
Dec
270-300
0.16
0.64
0.053
0.16
0.40
0.053
0.16
0.40
0.053
Dec
300-330
0.16
0.64
0.081
0.16
0.40
0.081
0.16
0.40
0.081
Dec
330-360
0.16
0.64
0.030
0.16
0.40
0.030
0.16
0.40
0.030
A-16
-------
Table A-7. Surface characteristics for Indianapolis Int'l (IND) by month and year.
Station = IND
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Jan
0-30
0.18
0.52
0.032
0.18
0.52
0.032
0.18
0.52
0.032
Jan
30-60
0.18
0.52
0.033
0.18
0.52
0.033
0.18
0.52
0.033
Jan
60-90
0.18
0.52
0.046
0.18
0.52
0.046
0.18
0.52
0.046
Jan
90-120
0.18
0.52
0.030
0.18
0.52
0.030
0.18
0.52
0.030
Jan
120-150
0.18
0.52
0.031
0.18
0.52
0.031
0.18
0.52
0.031
Jan
150-180
0.18
0.52
0.040
0.18
0.52
0.040
0.18
0.52
0.040
Jan
180-210
0.18
0.52
0.027
0.18
0.52
0.027
0.18
0.52
0.027
Jan
210-240
0.18
0.52
0.016
0.18
0.52
0.016
0.18
0.52
0.016
Jan
240-270
0.18
0.52
0.022
0.18
0.52
0.022
0.18
0.52
0.022
Jan
270-300
0.18
0.52
0.022
0.18
0.52
0.022
0.18
0.52
0.022
Jan
300-330
0.18
0.52
0.019
0.18
0.52
0.019
0.18
0.52
0.019
Jan
330-360
0.18
0.52
0.041
0.18
0.52
0.041
0.18
0.52
0.041
Feb
0-30
0.18
0.52
0.032
0.18
0.89
0.032
0.18
0.52
0.032
Feb
30-60
0.18
0.52
0.033
0.18
0.89
0.033
0.18
0.52
0.033
Feb
60-90
0.18
0.52
0.046
0.18
0.89
0.046
0.18
0.52
0.046
Feb
90-120
0.18
0.52
0.030
0.18
0.89
0.030
0.18
0.52
0.030
Feb
120-150
0.18
0.52
0.031
0.18
0.89
0.031
0.18
0.52
0.031
Feb
150-180
0.18
0.52
0.040
0.18
0.89
0.040
0.18
0.52
0.040
Feb
180-210
0.18
0.52
0.027
0.18
0.89
0.027
0.18
0.52
0.027
Feb
210-240
0.18
0.52
0.016
0.18
0.89
0.016
0.18
0.52
0.016
Feb
240-270
0.18
0.52
0.022
0.18
0.89
0.022
0.18
0.52
0.022
Feb
270-300
0.18
0.52
0.022
0.18
0.89
0.022
0.18
0.52
0.022
Feb
300-330
0.18
0.52
0.019
0.18
0.89
0.019
0.18
0.52
0.019
Feb
330-360
0.18
0.52
0.041
0.18
0.89
0.041
0.18
0.52
0.041
Mar
0-30
0.15
0.36
0.038
0.15
0.36
0.038
0.15
0.36
0.038
Mar
30-60
0.15
0.36
0.039
0.15
0.36
0.039
0.15
0.36
0.039
Mar
60-90
0.15
0.36
0.051
0.15
0.36
0.051
0.15
0.36
0.051
Mar
90-120
0.15
0.36
0.036
0.15
0.36
0.036
0.15
0.36
0.036
Mar
120-150
0.15
0.36
0.038
0.15
0.36
0.038
0.15
0.36
0.038
Mar
150-180
0.15
0.36
0.046
0.15
0.36
0.046
0.15
0.36
0.046
Mar
180-210
0.15
0.36
0.034
0.15
0.36
0.034
0.15
0.36
0.034
Mar
210-240
0.15
0.36
0.022
0.15
0.36
0.022
0.15
0.36
0.022
Mar
240-270
0.15
0.36
0.029
0.15
0.36
0.029
0.15
0.36
0.029
Mar
270-300
0.15
0.36
0.028
0.15
0.36
0.028
0.15
0.36
0.028
Mar
300-330
0.15
0.36
0.025
0.15
0.36
0.025
0.15
0.36
0.025
Mar
330-360
0.15
0.36
0.046
0.15
0.36
0.046
0.15
0.36
0.046
Apr
0-30
0.15
0.36
0.038
0.15
0.53
0.038
0.15
0.36
0.038
Apr
30-60
0.15
0.36
0.039
0.15
0.53
0.039
0.15
0.36
0.039
A-17
-------
Station = IND
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Apr
60-90
0.15
0.36
0.051
0.15
0.53
0.051
0.15
0.36
0.051
Apr
90-120
0.15
0.36
0.036
0.15
0.53
0.036
0.15
0.36
0.036
Apr
120-150
0.15
0.36
0.038
0.15
0.53
0.038
0.15
0.36
0.038
Apr
150-180
0.15
0.36
0.046
0.15
0.53
0.046
0.15
0.36
0.046
Apr
180-210
0.15
0.36
0.034
0.15
0.53
0.034
0.15
0.36
0.034
Apr
210-240
0.15
0.36
0.022
0.15
0.53
0.022
0.15
0.36
0.022
Apr
240-270
0.15
0.36
0.029
0.15
0.53
0.029
0.15
0.36
0.029
Apr
270-300
0.15
0.36
0.028
0.15
0.53
0.028
0.15
0.36
0.028
Apr
300-330
0.15
0.36
0.025
0.15
0.53
0.025
0.15
0.36
0.025
Apr
330-360
0.15
0.36
0.046
0.15
0.53
0.046
0.15
0.36
0.046
May
0-30
0.15
0.36
0.038
0.15
1.47
0.038
0.15
0.36
0.038
May
30-60
0.15
0.36
0.039
0.15
1.47
0.039
0.15
0.36
0.039
May
60-90
0.15
0.36
0.051
0.15
1.47
0.051
0.15
0.36
0.051
May
90-120
0.15
0.36
0.036
0.15
1.47
0.036
0.15
0.36
0.036
May
120-150
0.15
0.36
0.038
0.15
1.47
0.038
0.15
0.36
0.038
May
150-180
0.15
0.36
0.046
0.15
1.47
0.046
0.15
0.36
0.046
May
180-210
0.15
0.36
0.034
0.15
1.47
0.034
0.15
0.36
0.034
May
210-240
0.15
0.36
0.022
0.15
1.47
0.022
0.15
0.36
0.022
May
240-270
0.15
0.36
0.029
0.15
1.47
0.029
0.15
0.36
0.029
May
270-300
0.15
0.36
0.028
0.15
1.47
0.028
0.15
0.36
0.028
May
300-330
0.15
0.36
0.025
0.15
1.47
0.025
0.15
0.36
0.025
May
330-360
0.15
0.36
0.046
0.15
1.47
0.046
0.15
0.36
0.046
Jun
0-30
0.18
0.44
0.045
0.18
1.76
0.045
0.18
0.44
0.045
Jun
30-60
0.18
0.44
0.045
0.18
1.76
0.045
0.18
0.44
0.045
Jun
60-90
0.18
0.44
0.056
0.18
1.76
0.056
0.18
0.44
0.056
Jun
90-120
0.18
0.44
0.046
0.18
1.76
0.046
0.18
0.44
0.046
Jun
120-150
0.18
0.44
0.048
0.18
1.76
0.048
0.18
0.44
0.048
Jun
150-180
0.18
0.44
0.063
0.18
1.76
0.063
0.18
0.44
0.063
Jun
180-210
0.18
0.44
0.051
0.18
1.76
0.051
0.18
0.44
0.051
Jun
210-240
0.18
0.44
0.040
0.18
1.76
0.040
0.18
0.44
0.040
Jun
240-270
0.18
0.44
0.043
0.18
1.76
0.043
0.18
0.44
0.043
Jun
270-300
0.18
0.44
0.042
0.18
1.76
0.042
0.18
0.44
0.042
Jun
300-330
0.18
0.44
0.032
0.18
1.76
0.032
0.18
0.44
0.032
Jun
330-360
0.18
0.44
0.055
0.18
1.76
0.055
0.18
0.44
0.055
Jul
0-30
0.18
1.76
0.045
0.18
1.76
0.045
0.18
1.76
0.045
Jul
30-60
0.18
1.76
0.045
0.18
1.76
0.045
0.18
1.76
0.045
Jul
60-90
0.18
1.76
0.056
0.18
1.76
0.056
0.18
1.76
0.056
Jul
90-120
0.18
1.76
0.046
0.18
1.76
0.046
0.18
1.76
0.046
Jul
120-150
0.18
1.76
0.048
0.18
1.76
0.048
0.18
1.76
0.048
A-18
-------
Station = IND
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Jul
150-180
0.18
1.76
0.063
0.18
1.76
0.063
0.18
1.76
0.063
Jul
180-210
0.18
1.76
0.051
0.18
1.76
0.051
0.18
1.76
0.051
Jul
210-240
0.18
1.76
0.040
0.18
1.76
0.040
0.18
1.76
0.040
Jul
240-270
0.18
1.76
0.043
0.18
1.76
0.043
0.18
1.76
0.043
Jul
270-300
0.18
1.76
0.042
0.18
1.76
0.042
0.18
1.76
0.042
Jul
300-330
0.18
1.76
0.032
0.18
1.76
0.032
0.18
1.76
0.032
Jul
330-360
0.18
1.76
0.055
0.18
1.76
0.055
0.18
1.76
0.055
Aug
0-30
0.18
1.76
0.045
0.18
0.44
0.045
0.18
1.76
0.045
Aug
30-60
0.18
1.76
0.045
0.18
0.44
0.045
0.18
1.76
0.045
Aug
60-90
0.18
1.76
0.056
0.18
0.44
0.056
0.18
1.76
0.056
Aug
90-120
0.18
1.76
0.046
0.18
0.44
0.046
0.18
1.76
0.046
Aug
120-150
0.18
1.76
0.048
0.18
0.44
0.048
0.18
1.76
0.048
Aug
150-180
0.18
1.76
0.063
0.18
0.44
0.063
0.18
1.76
0.063
Aug
180-210
0.18
1.76
0.051
0.18
0.44
0.051
0.18
1.76
0.051
Aug
210-240
0.18
1.76
0.040
0.18
0.44
0.040
0.18
1.76
0.040
Aug
240-270
0.18
1.76
0.043
0.18
0.44
0.043
0.18
1.76
0.043
Aug
270-300
0.18
1.76
0.042
0.18
0.44
0.042
0.18
1.76
0.042
Aug
300-330
0.18
1.76
0.032
0.18
0.44
0.032
0.18
1.76
0.032
Aug
330-360
0.18
1.76
0.055
0.18
0.44
0.055
0.18
1.76
0.055
Sep
0-30
0.18
0.52
0.040
0.18
0.52
0.040
0.18
0.52
0.040
Sep
30-60
0.18
0.52
0.040
0.18
0.52
0.040
0.18
0.52
0.040
Sep
60-90
0.18
0.52
0.053
0.18
0.52
0.053
0.18
0.52
0.053
Sep
90-120
0.18
0.52
0.041
0.18
0.52
0.041
0.18
0.52
0.041
Sep
120-150
0.18
0.52
0.043
0.18
0.52
0.043
0.18
0.52
0.043
Sep
150-180
0.18
0.52
0.059
0.18
0.52
0.059
0.18
0.52
0.059
Sep
180-210
0.18
0.52
0.046
0.18
0.52
0.046
0.18
0.52
0.046
Sep
210-240
0.18
0.52
0.033
0.18
0.52
0.033
0.18
0.52
0.033
Sep
240-270
0.18
0.52
0.037
0.18
0.52
0.037
0.18
0.52
0.037
Sep
270-300
0.18
0.52
0.036
0.18
0.52
0.036
0.18
0.52
0.036
Sep
300-330
0.18
0.52
0.027
0.18
0.52
0.027
0.18
0.52
0.027
Sep
330-360
0.18
0.52
0.051
0.18
0.52
0.051
0.18
0.52
0.051
Oct
0-30
0.18
0.52
0.040
0.18
0.52
0.040
0.18
0.52
0.040
Oct
30-60
0.18
0.52
0.040
0.18
0.52
0.040
0.18
0.52
0.040
Oct
60-90
0.18
0.52
0.053
0.18
0.52
0.053
0.18
0.52
0.053
Oct
90-120
0.18
0.52
0.041
0.18
0.52
0.041
0.18
0.52
0.041
Oct
120-150
0.18
0.52
0.043
0.18
0.52
0.043
0.18
0.52
0.043
Oct
150-180
0.18
0.52
0.059
0.18
0.52
0.059
0.18
0.52
0.059
Oct
180-210
0.18
0.52
0.046
0.18
0.52
0.046
0.18
0.52
0.046
Oct
210-240
0.18
0.52
0.033
0.18
0.52
0.033
0.18
0.52
0.033
A-19
-------
Station = IND
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Oct
240-270
0.18
0.52
0.037
0.18
0.52
0.037
0.18
0.52
0.037
Oct
270-300
0.18
0.52
0.036
0.18
0.52
0.036
0.18
0.52
0.036
Oct
300-330
0.18
0.52
0.027
0.18
0.52
0.027
0.18
0.52
0.027
Oct
330-360
0.18
0.52
0.051
0.18
0.52
0.051
0.18
0.52
0.051
Nov
0-30
0.18
0.52
0.040
0.18
2.26
0.040
0.18
0.52
0.040
Nov
30-60
0.18
0.52
0.040
0.18
2.26
0.040
0.18
0.52
0.040
Nov
60-90
0.18
0.52
0.053
0.18
2.26
0.053
0.18
0.52
0.053
Nov
90-120
0.18
0.52
0.041
0.18
2.26
0.041
0.18
0.52
0.041
Nov
120-150
0.18
0.52
0.043
0.18
2.26
0.043
0.18
0.52
0.043
Nov
150-180
0.18
0.52
0.059
0.18
2.26
0.059
0.18
0.52
0.059
Nov
180-210
0.18
0.52
0.046
0.18
2.26
0.046
0.18
0.52
0.046
Nov
210-240
0.18
0.52
0.033
0.18
2.26
0.033
0.18
0.52
0.033
Nov
240-270
0.18
0.52
0.037
0.18
2.26
0.037
0.18
0.52
0.037
Nov
270-300
0.18
0.52
0.036
0.18
2.26
0.036
0.18
0.52
0.036
Nov
300-330
0.18
0.52
0.027
0.18
2.26
0.027
0.18
0.52
0.027
Nov
330-360
0.18
0.52
0.051
0.18
2.26
0.051
0.18
0.52
0.051
Dec
0-30
0.18
0.52
0.032
0.18
0.89
0.032
0.18
0.52
0.032
Dec
30-60
0.18
0.52
0.033
0.18
0.89
0.033
0.18
0.52
0.033
Dec
60-90
0.18
0.52
0.046
0.18
0.89
0.046
0.18
0.52
0.046
Dec
90-120
0.18
0.52
0.030
0.18
0.89
0.030
0.18
0.52
0.030
Dec
120-150
0.18
0.52
0.031
0.18
0.89
0.031
0.18
0.52
0.031
Dec
150-180
0.18
0.52
0.040
0.18
0.89
0.040
0.18
0.52
0.040
Dec
180-210
0.18
0.52
0.027
0.18
0.89
0.027
0.18
0.52
0.027
Dec
210-240
0.18
0.52
0.016
0.18
0.89
0.016
0.18
0.52
0.016
Dec
240-270
0.18
0.52
0.022
0.18
0.89
0.022
0.18
0.52
0.022
Dec
270-300
0.18
0.52
0.022
0.18
0.89
0.022
0.18
0.52
0.022
Dec
300-330
0.18
0.52
0.019
0.18
0.89
0.019
0.18
0.52
0.019
Dec
330-360
0.18
0.52
0.041
0.18
0.89
0.041
0.18
0.52
0.041
A-20
-------
Table A-8. Surface characteristics for Tulsa R. L. Jones Jr. (RVS) by month and year.
Station = RVS
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Jan
0-30
0.18
0.48
0.055
0.18
0.87
0.055
0.18
1.96
0.055
Jan
30-60
0.18
0.48
0.031
0.18
0.87
0.031
0.18
1.96
0.031
Jan
60-90
0.18
0.48
0.043
0.18
0.87
0.043
0.18
1.96
0.043
Jan
90-120
0.18
0.48
0.039
0.18
0.87
0.039
0.18
1.96
0.039
Jan
120-150
0.18
0.48
0.030
0.18
0.87
0.030
0.18
1.96
0.030
Jan
150-180
0.18
0.48
0.059
0.18
0.87
0.059
0.18
1.96
0.059
Jan
180-210
0.18
0.48
0.048
0.18
0.87
0.048
0.18
1.96
0.048
Jan
210-240
0.18
0.48
0.110
0.18
0.87
0.110
0.18
1.96
0.110
Jan
240-270
0.18
0.48
0.083
0.18
0.87
0.083
0.18
1.96
0.083
Jan
270-300
0.18
0.48
0.057
0.18
0.87
0.057
0.18
1.96
0.057
Jan
300-330
0.18
0.48
0.083
0.18
0.87
0.083
0.18
1.96
0.083
Jan
330-360
0.18
0.48
0.044
0.18
0.87
0.044
0.18
1.96
0.044
Feb
0-30
0.18
0.48
0.055
0.18
0.87
0.055
0.18
1.96
0.055
Feb
30-60
0.18
0.48
0.031
0.18
0.87
0.031
0.18
1.96
0.031
Feb
60-90
0.18
0.48
0.043
0.18
0.87
0.043
0.18
1.96
0.043
Feb
90-120
0.18
0.48
0.039
0.18
0.87
0.039
0.18
1.96
0.039
Feb
120-150
0.18
0.48
0.030
0.18
0.87
0.030
0.18
1.96
0.030
Feb
150-180
0.18
0.48
0.059
0.18
0.87
0.059
0.18
1.96
0.059
Feb
180-210
0.18
0.48
0.048
0.18
0.87
0.048
0.18
1.96
0.048
Feb
210-240
0.18
0.48
0.110
0.18
0.87
0.110
0.18
1.96
0.110
Feb
240-270
0.18
0.48
0.083
0.18
0.87
0.083
0.18
1.96
0.083
Feb
270-300
0.18
0.48
0.057
0.18
0.87
0.057
0.18
1.96
0.057
Feb
300-330
0.18
0.48
0.083
0.18
0.87
0.083
0.18
1.96
0.083
Feb
330-360
0.18
0.48
0.044
0.18
0.87
0.044
0.18
1.96
0.044
Mar
0-30
0.16
0.36
0.088
0.16
0.56
0.088
0.16
1.37
0.088
Mar
30-60
0.16
0.36
0.056
0.16
0.56
0.056
0.16
1.37
0.056
Mar
60-90
0.16
0.36
0.070
0.16
0.56
0.070
0.16
1.37
0.070
Mar
90-120
0.16
0.36
0.072
0.16
0.56
0.072
0.16
1.37
0.072
Mar
120-150
0.16
0.36
0.052
0.16
0.56
0.052
0.16
1.37
0.052
Mar
150-180
0.16
0.36
0.068
0.16
0.56
0.068
0.16
1.37
0.068
Mar
180-210
0.16
0.36
0.063
0.16
0.56
0.063
0.16
1.37
0.063
Mar
210-240
0.16
0.36
0.219
0.16
0.56
0.219
0.16
1.37
0.219
Mar
240-270
0.16
0.36
0.133
0.16
0.56
0.133
0.16
1.37
0.133
Mar
270-300
0.16
0.36
0.095
0.16
0.56
0.095
0.16
1.37
0.095
Mar
300-330
0.16
0.36
0.133
0.16
0.56
0.133
0.16
1.37
0.133
Mar
330-360
0.16
0.36
0.083
0.16
0.56
0.083
0.16
1.37
0.083
Apr
0-30
0.16
0.36
0.088
0.16
0.56
0.088
0.16
1.37
0.088
Apr
30-60
0.16
0.36
0.056
0.16
0.56
0.056
0.16
1.37
0.056
A-21
-------
Station = RVS
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Apr
60-90
0.16
0.36
0.070
0.16
0.56
0.070
0.16
1.37
0.070
Apr
90-120
0.16
0.36
0.072
0.16
0.56
0.072
0.16
1.37
0.072
Apr
120-150
0.16
0.36
0.052
0.16
0.56
0.052
0.16
1.37
0.052
Apr
150-180
0.16
0.36
0.068
0.16
0.56
0.068
0.16
1.37
0.068
Apr
180-210
0.16
0.36
0.063
0.16
0.56
0.063
0.16
1.37
0.063
Apr
210-240
0.16
0.36
0.219
0.16
0.56
0.219
0.16
1.37
0.219
Apr
240-270
0.16
0.36
0.133
0.16
0.56
0.133
0.16
1.37
0.133
Apr
270-300
0.16
0.36
0.095
0.16
0.56
0.095
0.16
1.37
0.095
Apr
300-330
0.16
0.36
0.133
0.16
0.56
0.133
0.16
1.37
0.133
Apr
330-360
0.16
0.36
0.083
0.16
0.56
0.083
0.16
1.37
0.083
May
0-30
0.16
0.36
0.088
0.16
0.56
0.088
0.16
1.37
0.088
May
30-60
0.16
0.36
0.056
0.16
0.56
0.056
0.16
1.37
0.056
May
60-90
0.16
0.36
0.070
0.16
0.56
0.070
0.16
1.37
0.070
May
90-120
0.16
0.36
0.072
0.16
0.56
0.072
0.16
1.37
0.072
May
120-150
0.16
0.36
0.052
0.16
0.56
0.052
0.16
1.37
0.052
May
150-180
0.16
0.36
0.068
0.16
0.56
0.068
0.16
1.37
0.068
May
180-210
0.16
0.36
0.063
0.16
0.56
0.063
0.16
1.37
0.063
May
210-240
0.16
0.36
0.219
0.16
0.56
0.219
0.16
1.37
0.219
May
240-270
0.16
0.36
0.133
0.16
0.56
0.133
0.16
1.37
0.133
May
270-300
0.16
0.36
0.095
0.16
0.56
0.095
0.16
1.37
0.095
May
300-330
0.16
0.36
0.133
0.16
0.56
0.133
0.16
1.37
0.133
May
330-360
0.16
0.36
0.083
0.16
0.56
0.083
0.16
1.37
0.083
Jun
0-30
0.17
0.38
0.250
0.17
0.57
0.250
0.17
1.33
0.250
Jun
30-60
0.17
0.38
0.114
0.17
0.57
0.114
0.17
1.33
0.114
Jun
60-90
0.17
0.38
0.131
0.17
0.57
0.131
0.17
1.33
0.131
Jun
90-120
0.17
0.38
0.138
0.17
0.57
0.138
0.17
1.33
0.138
Jun
120-150
0.17
0.38
0.098
0.17
0.57
0.098
0.17
1.33
0.098
Jun
150-180
0.17
0.38
0.075
0.17
0.57
0.075
0.17
1.33
0.075
Jun
180-210
0.17
0.38
0.107
0.17
0.57
0.107
0.17
1.33
0.107
Jun
210-240
0.17
0.38
0.389
0.17
0.57
0.389
0.17
1.33
0.389
Jun
240-270
0.17
0.38
0.318
0.17
0.57
0.318
0.17
1.33
0.318
Jun
270-300
0.17
0.38
0.265
0.17
0.57
0.265
0.17
1.33
0.265
Jun
300-330
0.17
0.38
0.325
0.17
0.57
0.325
0.17
1.33
0.325
Jun
330-360
0.17
0.38
0.244
0.17
0.57
0.244
0.17
1.33
0.244
Jul
0-30
0.17
0.38
0.250
0.17
0.57
0.250
0.17
1.33
0.250
Jul
30-60
0.17
0.38
0.114
0.17
0.57
0.114
0.17
1.33
0.114
Jul
60-90
0.17
0.38
0.131
0.17
0.57
0.131
0.17
1.33
0.131
Jul
90-120
0.17
0.38
0.138
0.17
0.57
0.138
0.17
1.33
0.138
Jul
120-150
0.17
0.38
0.098
0.17
0.57
0.098
0.17
1.33
0.098
A-22
-------
Station = RVS
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Jul
150-180
0.17
0.38
0.075
0.17
0.57
0.075
0.17
1.33
0.075
Jul
180-210
0.17
0.38
0.107
0.17
0.57
0.107
0.17
1.33
0.107
Jul
210-240
0.17
0.38
0.389
0.17
0.57
0.389
0.17
1.33
0.389
Jul
240-270
0.17
0.38
0.318
0.17
0.57
0.318
0.17
1.33
0.318
Jul
270-300
0.17
0.38
0.265
0.17
0.57
0.265
0.17
1.33
0.265
Jul
300-330
0.17
0.38
0.325
0.17
0.57
0.325
0.17
1.33
0.325
Jul
330-360
0.17
0.38
0.244
0.17
0.57
0.244
0.17
1.33
0.244
Aug
0-30
0.17
0.38
0.250
0.17
0.57
0.250
0.17
1.33
0.250
Aug
30-60
0.17
0.38
0.114
0.17
0.57
0.114
0.17
1.33
0.114
Aug
60-90
0.17
0.38
0.131
0.17
0.57
0.131
0.17
1.33
0.131
Aug
90-120
0.17
0.38
0.138
0.17
0.57
0.138
0.17
1.33
0.138
Aug
120-150
0.17
0.38
0.098
0.17
0.57
0.098
0.17
1.33
0.098
Aug
150-180
0.17
0.38
0.075
0.17
0.57
0.075
0.17
1.33
0.075
Aug
180-210
0.17
0.38
0.107
0.17
0.57
0.107
0.17
1.33
0.107
Aug
210-240
0.17
0.38
0.389
0.17
0.57
0.389
0.17
1.33
0.389
Aug
240-270
0.17
0.38
0.318
0.17
0.57
0.318
0.17
1.33
0.318
Aug
270-300
0.17
0.38
0.265
0.17
0.57
0.265
0.17
1.33
0.265
Aug
300-330
0.17
0.38
0.325
0.17
0.57
0.325
0.17
1.33
0.325
Aug
330-360
0.17
0.38
0.244
0.17
0.57
0.244
0.17
1.33
0.244
Sep
0-30
0.17
0.48
0.250
0.17
0.87
0.250
0.17
1.96
0.250
Sep
30-60
0.17
0.48
0.114
0.17
0.87
0.114
0.17
1.96
0.114
Sep
60-90
0.17
0.48
0.131
0.17
0.87
0.131
0.17
1.96
0.131
Sep
90-120
0.17
0.48
0.138
0.17
0.87
0.138
0.17
1.96
0.138
Sep
120-150
0.17
0.48
0.098
0.17
0.87
0.098
0.17
1.96
0.098
Sep
150-180
0.17
0.48
0.075
0.17
0.87
0.075
0.17
1.96
0.075
Sep
180-210
0.17
0.48
0.107
0.17
0.87
0.107
0.17
1.96
0.107
Sep
210-240
0.17
0.48
0.389
0.17
0.87
0.389
0.17
1.96
0.389
Sep
240-270
0.17
0.48
0.318
0.17
0.87
0.318
0.17
1.96
0.318
Sep
270-300
0.17
0.48
0.265
0.17
0.87
0.265
0.17
1.96
0.265
Sep
300-330
0.17
0.48
0.325
0.17
0.87
0.325
0.17
1.96
0.325
Sep
330-360
0.17
0.48
0.244
0.17
0.87
0.244
0.17
1.96
0.244
Oct
0-30
0.17
0.48
0.250
0.17
0.87
0.250
0.17
1.96
0.250
Oct
30-60
0.17
0.48
0.114
0.17
0.87
0.114
0.17
1.96
0.114
Oct
60-90
0.17
0.48
0.131
0.17
0.87
0.131
0.17
1.96
0.131
Oct
90-120
0.17
0.48
0.138
0.17
0.87
0.138
0.17
1.96
0.138
Oct
120-150
0.17
0.48
0.098
0.17
0.87
0.098
0.17
1.96
0.098
Oct
150-180
0.17
0.48
0.075
0.17
0.87
0.075
0.17
1.96
0.075
Oct
180-210
0.17
0.48
0.107
0.17
0.87
0.107
0.17
1.96
0.107
Oct
210-240
0.17
0.48
0.389
0.17
0.87
0.389
0.17
1.96
0.389
A-23
-------
Station = RVS
2011
2012
2013
Month
Sector
(degrees)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Albedo
Bowen
Ratio
Roughness
(m)
Oct
240-270
0.17
0.48
0.318
0.17
0.87
0.318
0.17
1.96
0.318
Oct
270-300
0.17
0.48
0.265
0.17
0.87
0.265
0.17
1.96
0.265
Oct
300-330
0.17
0.48
0.325
0.17
0.87
0.325
0.17
1.96
0.325
Oct
330-360
0.17
0.48
0.244
0.17
0.87
0.244
0.17
1.96
0.244
Nov
0-30
0.17
0.48
0.250
0.17
0.87
0.250
0.17
1.96
0.250
Nov
30-60
0.17
0.48
0.114
0.17
0.87
0.114
0.17
1.96
0.114
Nov
60-90
0.17
0.48
0.131
0.17
0.87
0.131
0.17
1.96
0.131
Nov
90-120
0.17
0.48
0.138
0.17
0.87
0.138
0.17
1.96
0.138
Nov
120-150
0.17
0.48
0.098
0.17
0.87
0.098
0.17
1.96
0.098
Nov
150-180
0.17
0.48
0.075
0.17
0.87
0.075
0.17
1.96
0.075
Nov
180-210
0.17
0.48
0.107
0.17
0.87
0.107
0.17
1.96
0.107
Nov
210-240
0.17
0.48
0.389
0.17
0.87
0.389
0.17
1.96
0.389
Nov
240-270
0.17
0.48
0.318
0.17
0.87
0.318
0.17
1.96
0.318
Nov
270-300
0.17
0.48
0.265
0.17
0.87
0.265
0.17
1.96
0.265
Nov
300-330
0.17
0.48
0.325
0.17
0.87
0.325
0.17
1.96
0.325
Nov
330-360
0.17
0.48
0.244
0.17
0.87
0.244
0.17
1.96
0.244
Dec
0-30
0.18
0.48
0.055
0.18
0.87
0.055
0.18
1.96
0.055
Dec
30-60
0.18
0.48
0.031
0.18
0.87
0.031
0.18
1.96
0.031
Dec
60-90
0.18
0.48
0.043
0.18
0.87
0.043
0.18
1.96
0.043
Dec
90-120
0.18
0.48
0.039
0.18
0.87
0.039
0.18
1.96
0.039
Dec
120-150
0.18
0.48
0.030
0.18
0.87
0.030
0.18
1.96
0.030
Dec
150-180
0.18
0.48
0.059
0.18
0.87
0.059
0.18
1.96
0.059
Dec
180-210
0.18
0.48
0.048
0.18
0.87
0.048
0.18
1.96
0.048
Dec
210-240
0.18
0.48
0.110
0.18
0.87
0.110
0.18
1.96
0.110
Dec
240-270
0.18
0.48
0.083
0.18
0.87
0.083
0.18
1.96
0.083
Dec
270-300
0.18
0.48
0.057
0.18
0.87
0.057
0.18
1.96
0.057
Dec
300-330
0.18
0.48
0.083
0.18
0.87
0.083
0.18
1.96
0.083
Dec
330-360
0.18
0.48
0.044
0.18
0.87
0.044
0.18
1.96
0.044
A-24
-------
REFERENCES
U.S. EPA. (2013). AERSURFACE User's Guide. U.S. Environmental Protection Agency. EPA
454/B-08-001. Revised January 16, 2013.
U.S. EPA. (2015). AERMINUTE User's Guide. U.S. Environmental Protection Agency. EPA
454/B-15-006.
U.S. EPA. (2016a). User's Guide for the AMS/EPA Regulatory Model - AERMOD. U.S.
Environmental Protection Agency. 454/B-16-011.
U.S. EPA. (2016b). User's Guide for the AERMOD Meteorological Processor (AERMET). U.S.
Environmental Protection Agency. EPA-454/B-16-010.
A-25
-------
APPENDIX B
DEVELOPMENT OF HOURLY EMISSIONS PROFILES
Preface: The source type influenced how the hourly emissions profiles were developed. The
methods followed are summarized below separately for EGU and other sources.
B.l EGU Sources
The NEI stores references to the Office of Regulatory Information Systems (ORIS)
identification code for most sources that have Continuous Emissions Monitoring System
(CEMS) data in the CAMD database. For these stacks the relative hourly profiles were derived
from the hourly values in the CAMD database, and the annual emissions totals were taken from
the NEI (Table B-l). EGU emissions came from the NEI for their respective years. Where
CEMS data was available, the CEMS emissions values were used and the emissions in the
annual inventory were adjusted to match the temporal pattern of the year-specific CEMS data.
The EGU units with more than 20 tons of SO2 emissions in at least one year for which CEMS
data are available are listed in Table B-l along with their annual SO2 emissions for 2011, 2012,
and 2013. Sources at the SEMASS Partnership facility (county 25023 and facility ID 8127611)
and IP&L - Harding Street (county 18097 and facility ID 7255211) are designated as EGUs but
are not matched to sources in the CAMD database. These sources were temporalized to hourly
values using average temporal profiles that were derived based on other EGU units in their
respective regions.
B-l
-------
Table B-l. SO2 emissions each year for EGUs included in the air quality modeling.
FIPS
Facility Name
Facility ID
Unit ID
2011
2012
2013
25005
BRAYTON POINT ENERGY LLC
5058411
87339613
3,535
1,228
1,625
25005
BRAYTON POINT ENERGY LLC
5058411
87339713
45
12
118
25005
BRAYTON POINT ENERGY LLC
5058411
87340713
4,298
1,859
1,383
25005
BRAYTON POINT ENERGY LLC
5058411
87340813
10,769
6,033
4,479
18097
IP&L - HARDING STREET
7255211
91188613
8,634
10,531
13,324
18097
IP&L - HARDING STREET
7255211
91188713
7,941
10,270
12,603
18097
IP&L - HARDING STREET
7255211
91188813
681
632
1,846
18097
IP&L - HARDING STREET
7255211
91188813
1,739
109
200
40131
PSO NORTHEASTERN PWR STA
8212411
6698813
8,039
9,008
40131
PSO NORTHEASTERN PWR STA
8212411
6698813
8,879
40131
PSO NORTHEASTERN PWR STA
8212411
6698813
20
38
40131
PSO NORTHEASTERN PWR STA
8212411
6698813
26
40131
PSO NORTHEASTERN PWR STA
8212411
6698313
7,402
9,337
40131
PSO NORTHEASTERN PWR STA
8212411
6698313
9,008
40131
PSO NORTHEASTERN PWR STA
8212411
6698313
27
22
40131
PSO NORTHEASTERN PWR STA
8212411
6698313
26
B.2 Non-EGU Sources
For non-EGU sources that did not have hourly SO2 data in the CAMD database, SCC-
specific temporal profiles from EPA's 201 lv6.3 emissions modeling platform were used to
prepare the hourly factors. Stacks with emissions greater than 20 tons of SO2 in 2011, 2012, or
2013 for which temporal profiles were used are listed in Table B-2 below. The allocation of the
sources to the hourly factors needed for AERMOD was done using tools available within the
Sparse Matrix Operator Kernel Emissions (SMOKE) modeling system version 4.5 (UNC, 2017).
The tools support the generation of "helper files" from which the AERMOD input files can be
derived. The temporal values output from SMOKE were renormalized from scalars to factors
that sum to 1 to aid with quality assurance and usability of the factors.
B-2
-------
Table B-2. SO2 emissions each year for non-EGU release points included in the air quality
modeling1.
FIPS
Facility Name
Facility ID
Unit ID
Release
2011
2012
2013
18097
Citizens Thermal
4885311
100805413
30985212
2,094
1,849
1,575
18097
Citizens Thermal
4885311
100805713
30985012
1,029
853
855
18097
Citizens Thermal
4885311
100805813
30985012
1,225
1,150
1,375
40037
SAPULPA
7320611
72251213
66374812
79
79
98
40037
SAPULPA
7320611
8331413
8217312
33
33
34
40037
SAPULPA
7320611
8331213
8217212
100
100
108
18097
VERTELLUS AGRICULTURE &
7972111
65408713
60023412
20
17
20
18097
QUEMETCO, INC.
8235411
65358713
5022512
49
49
16
18097
QUEMETCO, INC.
8235411
65358713
5022612
71
71
69
40143
TULSA RFNRY WEST
8402711
654613
655312
103
42
26
40143
TULSA RFNRY WEST
8402711
654413
660012
45
20
9
40143
TULSA RFNRY WEST
8402711
654313
659912
380
237
169
40143
TULSA RFNRY WEST
8402711
654113
663512
36
18
11
40143
TULSA RFNRY WEST
8402711
651713
655012
59
65
24
40143
TULSA RFNRY WEST
8402711
651413
661212
270
210
125
40143
TULSA RFNRY WEST
8402711
651313
658812
43
41
11
40143
TULSA RFNRY WEST
8402711
651113
662812
39
40143
TULSA RFNRY WEST
8402711
651113
662812
43
17
40143
TULSA RFNRY WEST
8402711
651013
658912
157
150
37
40143
TULSA RFNRY WEST
8402711
650913
654912
74
55
34
40143
TULSA RFNRY WEST
8402711
650813
656012
38
46
8
40143
TULSA RFNRY WEST
8402711
663113
651512
866
688
360
40143
TULSA RFNRY WEST
8402711
658713
651412
460
370
211
The emissions factors developed for non-EGU sources were monthly, hour-of-day, or
month-hour-of-day, where day was weekday, Saturday, or Sunday. These emission factors
correspond to the MONTH, HROFDY, and MHRDOW emission factors used in AERMOD
(U.S. EPA, 2016). These emission factors are set to sum to 1 for each source. For example, for a
source using the MONTH emission factors, the 12 monthly factors sum to 1. This means that a
particular month's factor allocates a portion of the annual emissions to that month. Further
processing is needed to create hourly emissions for the sources. For monthly factors, the monthly
factor is divided by the number of hours in the month (number of days x 24 hours) and this ratio
is multiplied by the annual emissions to get an hourly emission rate and this rate is then
converted to a g/s rate. This rate is then input into AERMOD as the MONTH emission factor,
and the reference emission rate in AERMOD (emission rate on the SRCPARAM line in the
1 Based on units emitting over 20 tons of SO2.
B-3
-------
AERMOD input file) is set to 1.0. This method creates an hourly emission rate while conserving
the annual emissions.
Consider a source with the following monthly factors (Table B-3) output from SMOKE
for 2011 and annual emissions of 100.32 tons. The factors divide the emissions equally across
the months, resulting in the monthly emissions (in tons) shown for each month. To convert the
monthly emissions for a given month, to g/s, the following equation is used:
Where Ehour is the hourly emission rate in g/s, Eannuai are the annual emissions in tons, Daysm0nth
are the number of days in the month (31 days for January, etc.), 1/24 is the reciprocal of the
number of hours in a day, and 251.9957778 is the conversion factor to convert from tons/hour to
g/s. The resulting hourly emissions rates are also shown in Table B-3. Figure B-l shows how the
hourly emissions are input into AERMOD using the SRCPARAM and EMISFACT keywords.
Equation 1 is also used to calculate the MHRDOW emissions and a similar form of Equation 1 is
used for HROFDY emissions, with the exception that 1/DaySmonth is 1/365 (number of days in the
year).
Table B-3. Example calculation of hourly emissions using the SMOKE MONTH temporal
factors for 2011.
x 251.9957778
annual
*'month J V
Equation B-l
Month
SMOKE factor
DaySmonth Ehour (fl/s)
January
February
0.083333
0.083333
0.083333
0.083333
0.083333
0.083333
0.083333
0.083333
0.083333
0.083333
0.083333
0.083333
31 2.831565
28 3.134947
31 2.831565
30 2.925951
31 2.831565
30 2.925951
31 2.831565
31 2.831565
30 2.925951
31 2.831565
30 2.925951
31 2.831565
March
April
May
June
July
August
September
October
November
December
B-4
-------
SO SRCPARAM SAP SNl
SO EMISFACT SAP SNl
SO EMISFACT SAP SNl
SO EMISFACT SAP SNl
SO EMISFACT SAP SNl
SO EMISFACT SAP SNl
SO EMISFACT SAP SN1
1.000000E+00 28.35000 530.37000 9.60000 1.86000
MONTH 2.831565E+00 3.134947E+00
MONTH 2.831565E+00 2.925951E+00
MONTH 2.831565E+00 2.925951E+00
MONTH 2.831565E+00 2.831565E+00
MONTH 2.925951E+00 2.831565E+00
MONTH 2.925951E+00 2.831565E+00
Figure B-l. Example AERMOD input emission lines for monthly emissions.
B.3 AERMOD inputs
Tables B-4 through B-41 list the cross walks between facility unit identifiers and
AERMOD source identifiers and the 2011-2013 AERMOD inputs for each of the three study
areas. Note that the AERMOD source identifiers are unique to each year. In some cases, a
particular emission release point may not have an AERMOD source identifier for one year but
may have an identifier for other years. Years in which a release point does not have an
AERMOD identifier are left as blanks.
B-5
-------
Table B-4. Fall River 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
BRAYTON POINT ENERGY LLC
87339613
83612912
118371314
BRAY SE1
BRAY SE1
BRAY SE1
BRAYTON POINT ENERGY LLC
87339613
83612912
118371714
BRAY SE2
BRAY SE2
BRAY SE2
BRAYTON POINT ENERGY LLC
87339713
83613312
118371814
BRAY SE3
BRAY SE3
BRAY SE3
BRAYTON POINT ENERGY LLC
87339713
83613312
118371914
BRAY SE4
BRAY SE4
BRAY SE4
BRAYTON POINT ENERGY LLC
87339713
83613312
118372014
BRAY SE5
BRAYTON POINT ENERGY LLC
87339713
83613312
118372114
BRAY SE5
BRAY SE6
BRAY SE5
BRAYTON POINT ENERGY LLC
87340713
83612812
118373214
BRAY SE6
BRAY SE7
BRAY SE6
BRAYTON POINT ENERGY LLC
87340713
83612812
118373514
BRAY SE7
BRAY SE8
BRAY SE7
BRAYTON POINT ENERGY LLC
87340813
83612612
118373614
BRAY SE8
BRAY SE9
BRAY SE8
BRAYTON POINT ENERGY LLC
87340813
83612612
118373714
BRAY SE9
BRAY SE10
BRAY SE9
BRAYTON POINT ENERGY LLC
90543213
83613612
122762214
BRAY SN1
BRAY SN1
BRAY SN1
BRAYTON POINT ENERGY LLC
90543413
83613612
122762414
BRAY SN1
BRAY SN1
BRAY SN1
BRAYTON POINT ENERGY LLC
87341513
83613212
118374814
BRAY SN2
BRAY SN2
BRAY SN2
BRAYTON POINT ENERGY LLC
87341613
83612512
118374914
BRAY_SN2
BRAY_SN2
BRAY_SN2
B-6
-------
Table B-5. 2011 Fall River point source emissions, locations, and stack parameters.
Facility Name
AERMDOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
BRAYTON POINT
ENERGY LLC
BRAY SE1
3534.90
HOURLY
317613.67
4620047.98
5.07
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE2
0.01
HOURLY
317613.67
4620047.98
5.07
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE3
45.24
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE4
0.77
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE5
0.11
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE6
4298.40
HOURLY
317639.35
4620024.01
5.56
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE7
0.01
HOURLY
317639.35
4620024.01
5.56
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE8
0.02
HOURLY
317577.42
4620064.54
4.81
107.29
405.37
24.93
5.94
BRAYTON POINT
ENERGY LLC
BRAY SE9
10769.00
HOURLY
317577.42
4620064.54
4.81
107.29
405.37
24.93
5.94
BRAYTON POINT
ENERGY LLC
BRAY_SN2
0.0004
MONTH
317600.47
4619900.00
8.20
3.66
783.15
24.66
0.30
Table B-6. 2011 Fall River area source emissions, locations, and stack parameters.
Facility Name
AERMDOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height
(m)
X-
dimension
(m)
Y-
dimensio
n(m)
Angle
-------
Table B-7. 2012 Fall River point source emissions, locations, and stack parameters.
Facility Name
AERMDOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
BRAYTON POINT
ENERGY LLC
BRAY SE1
1228.40
HOURLY
317613.67
4620047.98
5.07
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE2
0.08
HOURLY
317613.67
4620047.98
5.07
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE3
12.14
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE4
1.81
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE5
1.53
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE6
0.33
HOURLY
317639.35
4620024.01
5.56
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE7
1859.40
HOURLY
317639.35
4620024.01
5.56
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE8
0.17
HOURLY
317577.42
4620064.54
4.81
107.29
405.37
24.93
5.94
BRAYTON POINT
ENERGY LLC
BRAY SE9
0.13
HOURLY
317577.42
4620064.54
4.81
107.29
405.37
24.93
5.94
BRAYTON POINT
ENERGY LLC
BRAY SE10
6033.0
HOURLY
317577.42
4620064.54
4.81
107.29
405.37
24.93
5.94
BRAYTON POINT
ENERGY LLC
BRAY SN2
0.0014
MONTH
317600.47
4619900.00
8.20
3.66
783.15
24.66
0.30
Table B-8. 2012 Fall River area source emissions, locations, and stack parameters.
Facility Name
AERMDOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height
(m)
X-
dimension
(m)
Y-
dimensio
n(m)
Angle
-------
Table B-9. 2013 Fall River point source emissions, locations, and stack parameters.
Facility Name
AERMDOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
BRAYTON POINT
ENERGY LLC
BRAY SE1
1625.20
HOURLY
317613.67
4620047.98
5.07
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE2
0.01
HOURLY
317613.67
4620047.98
5.07
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE3
118.06
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE4
0.77
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE5
0.11
HOURLY
317536.89
4620117.91
4.72
152.40
432.04
21.73
5.64
BRAYTON POINT
ENERGY LLC
BRAY SE6
1383.00
HOURLY
317639.35
4620024.01
5.56
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE7
0.01
HOURLY
317639.35
4620024.01
5.56
107.29
383.15
20.45
4.42
BRAYTON POINT
ENERGY LLC
BRAY SE8
0.02
HOURLY
317577.42
4620064.54
4.81
107.29
405.37
24.93
5.94
BRAYTON POINT
ENERGY LLC
BRAY SE9
4479.30
HOURLY
317577.42
4620064.54
4.81
107.29
405.37
24.93
5.94
BRAYTON POINT
ENERGY LLC
BRAY_SN2
0.0004
MONTH
317600.47
4619900.00
8.20
3.66
783.15
24.66
0.30
Table B-10. 2013 Fall River area source emissions, locations, and stack parameters.
Facility Name
AERMDOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height
(m)
X-
dimension
(m)
Y-
dimensio
n(m)
Angle
-------
Table B-ll. Indianapolis, IN Indianapolis Belmont WWTP 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
INDIANAPOLIS BELMONT WWTP
68272413
64154812
124267014
BELL SN1
BELL SN1
INDIANAPOLIS BELMONT WWTP
68272613
64155012
124267214
BELL SN1
BELL SN1
INDIANAPOLIS BELMONT WWTP
32403713
30985312
123964514
BELL SN1
BELL SN1
BELL SN1
INDIANAPOLIS BELMONT WWTP
32403813
30985312
123964614
BELL SN1
BELL SN1
BELL SN1
INDIANAPOLIS BELMONT WWTP
32403913
30985312
123964814
BELL SN1
BELL SN1
BELL SN1
INDIANAPOLIS BELMONT WWTP
32404013
30985312
123964714
BELL SN1
BELL SN1
BELL SN1
Table B-12. Indianapolis, IN Citizens Thermal 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
Citizens Thermal
100805713
30985012
141379114
CIT SN1
CIT SN1
CIT SN1
Citizens Thermal
100805813
30985012
141379414
CIT SN1
CIT SN1
CIT SN1
Citizens Thermal
100805413
30985212
141378314
CIT SN2
CIT SN2
CIT SN2
Citizens Thermal
100805713
30985012
141379214
CIT SN3
CIT SN3
Citizens Thermal
100805813
30985012
141379514
CIT SN3
CIT SN3
Citizens Thermal
100805313
30984812
141378114
CIT SN3
CIT SN4
CIT SN4
Citizens Thermal
100805413
30984812
141378414
CIT SN3
CIT SN4
CIT SN4
Citizens Thermal
100805513
30985212
141378714
CIT SN4
CIT SN5
CIT SN5
Citizens Thermal
100805613
30985212
141378914
CIT SN4
CIT SN5
CIT SN5
Citizens Thermal
100805913
30984812
141379714
CIT SN5
CIT SN6
CIT SN6
Citizens Thermal
100806013
30984812
141379914
CIT SN5
CIT SN6
CIT SN6
B-10
-------
Table B-13. Indianapolis, IN IP&L Harding Street 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
IP&L-HARDING STREET
91608313
87281612
124834714
IPL SE1
IPL SE1
IPL SE1
IP&L-HARDING STREET
91608413
87281712
124834814
IPL SE1
IPL SE1
IPL SE1
IP&L-HARDING STREET
91608213
87281512
124834614
IPL SE2
IPL SE2
IPL SE2
IP&L-HARDING STREET
91188213
87281812
123965914
IPL SE3
IPL SE3
IP&L-HARDING STREET
91188313
87281912
123966114
IPL SE4
IP&L-HARDING STREET
91188613
87281212
123966614
IPL SE4
IPL SE3
IPL SE5
IP&L-HARDING STREET
91188713
87281312
123966814
IPL SE5
IPL SE4
IPL SE6
IP&L-HARDING STREET
91188813
87281412
123966914
IPL SE6
IPL SE5
IPL SE7
IP&L-HARDING STREET
91188813
101276612
123967114
IPL SE7
IPL SE6
IPL SE8
IP&L-HARDING STREET
91608513
88573012
124834914
IPL SE8
IPL SE7
IPL SE9
B-ll
-------
Table B-14. Indianapolis, IN Rolls Royce 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
ROLLS ROYCE CORPORATION
68294413
64180912
124304714
RR SN1
RR SN1
RR SN1
ROLLS ROYCE CORPORATION
68294313
64180812
124304614
RR SN2
ROLLS ROYCE CORPORATION
2995413
2866112
124164914
RR SN2
ROLLS ROYCE CORPORATION
2996413
2865012
124166414
RR SN2
RR SN3
RR SN3
ROLLS ROYCE CORPORATION
2996513
2865912
124166514
RR SN3
RR SN4
RR SN4
ROLLS ROYCE CORPORATION
2996613
2865112
124166614
RR SN3
RR SN4
RR SN4
ROLLS ROYCE CORPORATION
2995313
2864812
124164814
RR SN4
RR SN5
RR SN5
ROLLS ROYCE CORPORATION
2995813
64180712
124165414
RR SN5
RR SN6
RR SN6
ROLLS ROYCE CORPORATION
2995413
2866112
124165114
RR SN6
RR SN7
RR SN7
ROLLS ROYCE CORPORATION
2995313
2864812
124164714
RR SN7
ROLLS ROYCE CORPORATION
2995413
2866112
124165014
RR SN8
RR SN8
RR SN8
ROLLS ROYCE CORPORATION
2997413
2865812
41165514
RR SN9
RR SN9
RR SN9
ROLLS ROYCE CORPORATION
2994913
2866312
124166214
RR SN10
RR SN10
RR SN10
ROLLS ROYCE CORPORATION
2996113
2866812
124166014
RR SN10
RR SN10
RR SN10
ROLLS ROYCE CORPORATION
2994913
2866312
124166114
RR SN11
RR SN11
RR SN11
ROLLS ROYCE CORPORATION
2996113
2866812
124165914
RR SN11
RR SN11
RR SN11
ROLLS ROYCE CORPORATION
2997513
2865612
124165814
RR SN12
RR SN12
RR SN12
ROLLS ROYCE CORPORATION
2997613
2864712
124165714
RR SN12
RR SN12
RR SN12
ROLLS ROYCE CORPORATION
2995913
2866412
124165614
RR SN13
RR SN13
RR SN13
ROLLS ROYCE CORPORATION
2996213
2865412
124165514
RR SN13
RR SN13
RR SN13
ROLLS ROYCE CORPORATION
2997413
2865812
124167114
RR SN14
RR SN14
RR SN14
ROLLS ROYCE CORPORATION
2996713
2866912
124166714
RR SN15
RR SN15
RR SN15
B-12
-------
Table B-15. Indianapolis, IN Vertellus 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408713
60023312
90663014
VERT SN1
VERT SN1
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408713
60023312
141512314
VERT SN1
VERT SN1
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408713
60023412
90662914
VERT SN2
VERT SN2
VERT SN1
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408713
60023412
90663314
VERT SN2
VERT SN2
VERT SN1
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408713
60023412
90663414
VERT SN2
VERT SN2
VERT SN1
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408713
101303012
90662214
VERT SN4
VERT SN4
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408613
2863012
90661214
VERT SN12
VERT SN12
VERT SN10
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408613
2861612
141511014
VERT SN13
VERT SN13
VERT SN11
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408613
2864112
90660614
VERT SN14
VERT SN14
VERT SN12
VERTELLUS AGRICULTURE &
NUTRITION SPECIALTIES LLC
65408613
2863312
90660214
VERT_SN15
VERT_SN15
VERT_SN13
B-13
-------
Table B-16. Indianapolis, IN Quemetco 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
QUEMETCO, INC.
65358713
5022612
90566814
QUE SN1
QUE SN1
QUE SN1
QUEMETCO, INC.
65358913
5022612
90567014
QUE SN1
QUE SN1
QUE SN1
QUEMETCO, INC.
65358713
5022612
90566714
QUE SN1
QUEMETCO, INC.
109197013
112719612
154715314
QUE SN2
QUEMETCO, INC.
65358713
5022512
90566614
QUE SN2
QUE SN2
QUE SN3
QUEMETCO, INC.
65359113
5022512
90567214
QUE SN2
QUE SN2
QUE SN3
B-14
-------
Table B-17. 2011 Indianapolis Belmont WWTP, Citizens Thermal, and IP&L Harding Street point source emissions, locations,
and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height (m)
Stack
temperature (K)
Stack
velocity
(m S"1)
Stack
diameter (m)
INDIANAPOLIS
BELMONT WWTP
BELL SN1
24.90
MONTH
568970.00
4397879.00
208.61
45.72
297.59
0.64
3.20
Citizens Thermal
CIT SN1
2254.90
MONTH
571351.00
4401766.00
216.45
82.91
566.48
4.60
4.42
Citizens Thermal
CIT SN2
2093.70
MONTH
571396.00
4401766.00
217.46
82.91
463.71
4.72
4.64
Citizens Thermal
CIT SN3
0.16
MONTH
571380.00
4401766.00
217.35
82.91
488.71
5.33
4.63
Citizens Thermal
CIT SN4
0.08
MONTH
571396.00
4401766.00
217.46
82.91
463.71
4.72
4.64
Citizens Thermal
CIT SN5
0.0005
MONTH
571380.00
4401766.00
217.35
82.91
488.71
5.33
4.63
IP&L-HARDING
STREET
I PL SE1
0.11
HOURLY
569200.00
4396339.00
208.02
9.45
791.48
7.16
3.81
IP&L-HARDING
STREET
I PL SE2
0.10
HOURLY
569180.00
4396327.00
207.98
9.75
791.48
7.16
3.81
IP&L-HARDING
STREET
I PL SE3
0.10
HOURLY
568867.00
4396303.00
208.00
20.12
827.59
57.39
4.21
IP&L-HARDING
STREET
I PL SE4
8633.50
HOURLY
568749.00
4396008.00
208.08
79.55
440.93
65.84
1.98
IP&L-HARDING
STREET
I PL SE5
7940.50
HOURLY
568752.00
4395965.00
208.32
79.55
449.82
63.52
1.98
IP&L-HARDING
STREET
I PL SE6
680.70
HOURLY
568984.00
4395792.00
206.56
172.21
329.26
14.33
6.10
IP&L-HARDING
STREET
I PL SE7
1739.00
HOURLY
568984.00
4395792.00
206.56
172.21
414.82
23.44
6.10
IP&L-HARDING
STREET
IPL_SE8
0.20
HOURLY
569050.00
4396339.00
208.26
22.86
810.93
36.58
5.49
B-15
-------
Table B-18. 2011 Rolls Royce, Vertellus, and Quemetco point source emissions, locations, and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
ROLLS ROYCE
RR SN1
0.02
MONTH
567493.00
4398570.00
212.29
4.57
866.48
32.34
0.30
ROLLS ROYCE
RR SN2
0.89
HROFDY
567428.00
4398870.00
212.70
17.37
588.71
33.04
1.22
ROLLS ROYCE
RR SN3
16.17
HROFDY
567402.00
4398886.00
212.70
19.81
755.37
45.51
0.91
ROLLS ROYCE
RR SN9
3.60
MHRDOW
567435.00
4398899.00
212.72
9.14
866.48
21.21
1.52
ROLLS ROYCE
RR SN10
6.19
MONTH
567551.00
4399165.00
212.10
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN11
0.06
MONTH
567551.00
4399165.00
212.10
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN12
23.36
MONTH
567544.50
4399165.00
212.24
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN13
1.56
MONTH
567512.00
4399163.00
212.51
18.29
533.15
6.52
1.22
ROLLS ROYCE
RR SN14
0.04
MHRDOW
567513.00
4399174.00
212.61
9.14
866.48
21.21
1.52
ROLLS ROYCE
RR SN15
0.002
MONTH
567439.00
4398911.00
212.70
15.24
755.37
13.53
1.68
VERTELLUS
VERT SN1
3.98
MONTH
566836.00
4399683.00
214.94
9.14
453.71
6.28
1.22
VERTELLUS
VERT SN2
26.43
MONTH
566981.00
4399746.00
215.16
9.14
504.26
7.53
1.22
VERTELLUS
VERT SN4
0.19
MONTH
566995.00
4399731.00
214.89
10.97
422.04
5.49
0.81
VERTELLUS
VERT SN12
0.04
MONTH
566851.06
4399666.50
214.85
20.42
823.15
5.09
1.07
VERTELLUS
VERT SN13
0.02
MONTH
566901.00
4399710.00
215.15
20.73
823.15
5.09
1.07
VERTELLUS
VERT SN14
0.05
MONTH
566866.94
4399637.00
214.79
21.64
633.15
6.10
1.52
VERTELLUS
VERT SN15
0.03
MONTH
566864.94
4399640.00
214.79
24.69
823.15
6.07
1.07
QUEMETCO
QUE SN1
70.78
MONTH
559977.54
4400993.45
235.78
30.48
327.04
16.86
1.22
QUEMETCO
QUE_SN2
53.59
MONTH
559993.31
4400853.53
235.10
50.29
321.48
14.84
3.35
B-16
-------
Table B-19. 2011 Indianapolis, IN Rolls Royce area source emissions, locations, and release parameters.
Facility
Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height (m)
X-dimension
(m)
Y-dimension
(m)
Angle
-------
Table B-20. 2012 Indianapolis Belmont WWTP, Citizens Thermal, and IP&L Harding Street point source emissions, locations,
and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height (m)
Stack
temperature (K)
Stack
velocity
(m S"1)
Stack
diameter (m)
INDIANAPOLIS
BELMONT WWTP
BELL SN1
24.90
MONTH
568970.00
4397879.00
208.61
45.72
297.59
0.64
3.20
Citizens Thermal
CIT SN1
2002.70
MONTH
571351.00
4401766.00
216.45
82.91
566.48
4.60
4.42
Citizens Thermal
CIT SN2
1849.50
MONTH
571396.00
4401766.00
217.46
82.91
463.71
4.72
4.64
Citizens Thermal
CIT SN3
0.0000004
MONTH
571351.00
4401766.00
216.45
82.91
566.48
4.60
4.42
Citizens Thermal
CIT SN4
0.18
MONTH
571380.00
4401766.00
217.35
82.91
488.71
5.33
4.63
Citizens Thermal
CIT SN5
0.07
MONTH
571396.00
4401766.00
217.46
82.91
463.71
4.72
4.64
Citizens Thermal
CIT SN6
0.001
MONTH
571380.00
4401766.00
217.35
82.91
488.71
5.33
4.63
IP&L-HARDING
STREET
I PL SE1
0.19
HOURLY
569200.00
4396339.00
208.02
9.45
791.48
7.16
3.81
IP&L-HARDING
STREET
I PL SE2
0.16
HOURLY
569180.00
4396327.00
207.98
9.75
791.48
7.16
3.81
IP&L-HARDING
STREET
I PL SE3
10531.00
HOURLY
568749.00
4396008.00
208.08
79.55
440.93
65.84
1.98
IP&L-HARDING
STREET
I PL SE4
10270.00
HOURLY
568752.00
4395965.00
208.32
79.55
449.82
63.52
1.98
IP&L-HARDING
STREET
I PL SE5
632.10
HOURLY
568984.00
4395792.00
206.56
172.21
329.26
14.33
6.10
IP&L-HARDING
STREET
I PL SE6
109.00
HOURLY
568984.00
4395792.00
206.56
172.21
414.82
23.44
6.10
IP&L-HARDING
STREET
IPL_SE7
0.20
HOURLY
569050.00
4396339.00
208.26
22.86
810.93
36.58
5.49
B-18
-------
Table B-21. 2012 Rolls Royce, Vertellus, and Quemetco point source emissions, locations, and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
ROLLS ROYCE
RR SN1
1.74
MONTH
567493.00
4398570.00
212.29
4.57
866.48
32.34
0.30
ROLLS ROYCE
RR SN2
0.23
HROFDY
567402.00
4398886.00
212.70
19.81
755.37
45.51
0.91
ROLLS ROYCE
RR SN3
0.70
HROFDY
567428.00
4398870.00
212.70
17.37
588.71
33.04
1.22
ROLLS ROYCE
RR SN4
13.98
HROFDY
567402.00
4398886.00
212.70
19.81
755.37
45.51
0.91
ROLLS ROYCE
RR SN9
3.49
MHRDOW
567435.00
4398899.00
212.72
9.14
866.48
21.21
1.52
ROLLS ROYCE
RR SN10
6.08
MONTH
567551.00
4399165.00
212.10
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN11
0.03
MONTH
567551.00
4399165.00
212.10
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN12
7.29
MONTH
567544.50
4399165.00
212.24
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN13
1.57
MONTH
567512.00
4399163.00
212.51
18.29
533.15
6.52
1.22
ROLLS ROYCE
RR SN14
0.02
MHRDOW
567513.00
4399174.00
212.61
9.14
866.48
21.21
1.52
ROLLS ROYCE
RR SN15
0.0001
MONTH
567439.00
4398911.00
212.70
15.24
755.37
13.53
1.68
VERTELLUS
VERT SN1
1.38
MONTH
566836.00
4399683.00
214.94
9.14
453.71
6.28
1.22
VERTELLUS
VERT SN2
22.18
MONTH
566981.00
4399746.00
215.16
9.14
504.26
7.53
1.22
VERTELLUS
VERT SN4
0.90
MONTH
566995.00
4399731.00
214.89
10.97
422.04
5.49
0.81
VERTELLUS
VERT SN12
0.06
MONTH
566851.06
4399666.50
214.85
20.42
823.15
5.09
1.07
VERTELLUS
VERT SN13
0.01
MONTH
566901.00
4399710.00
215.15
20.73
823.15
5.09
1.07
VERTELLUS
VERT SN14
0.03
MONTH
566866.94
4399637.00
214.79
21.64
633.15
6.10
1.52
VERTELLUS
VERT SN15
0.02
MONTH
566864.94
4399640.00
214.79
24.69
823.15
6.07
1.07
QUEMETCO
QUE SN1
70.78
MONTH
559977.54
4400993.45
235.78
30.48
327.04
16.86
1.22
QUEMETCO
QUE SN2
53.59
MONTH
559993.31
4400853.53
235.10
50.29
321.48
14.84
3.35
B-19
-------
Table B-22. 2012 Indianapolis, IN Rolls Royce area source emissions, locations, and release parameters.
Facility
Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height (m)
X-dimension
(m)
Y-dimension
(m)
Angle
-------
Table B-23. 2013 Indianapolis Belmont WWTP, Citizens Thermal, and IP&L Harding Street point source emissions, locations,
and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height (m)
Stack
temperature (K)
Stack
velocity
(m S"1)
Stack
diameter (m)
INDIANAPOLIS
BELMONT WWTP
BELL SN1
20.10
MONTH
568970.00
4397879.00
208.61
45.72
297.59
0.64
3.20
Citizens Thermal
CIT SN1
2229.80
MONTH
571351.00
4401766.00
216.45
82.91
566.48
4.60
4.42
Citizens Thermal
CIT SN2
1575.00
MONTH
571396.00
4401766.00
217.46
82.91
463.71
4.72
4.64
Citizens Thermal
CIT SN3
0.0000001
MONTH
571351.00
4401766.00
216.45
82.91
566.48
4.60
4.42
Citizens Thermal
CIT SN4
0.28
MONTH
571380.00
4401766.00
217.35
82.91
488.71
5.33
4.63
Citizens Thermal
CIT SN5
0.24
MONTH
571396.00
4401766.00
217.46
82.91
463.71
4.72
4.64
Citizens Thermal
CIT SN6
0.002
MONTH
571380.00
4401766.00
217.35
82.91
488.71
5.33
4.63
IP&L-HARDING
STREET
I PL SE1
0.02
HOURLY
569200.00
4396339.00
208.02
9.45
791.48
7.16
3.81
IP&L-HARDING
STREET
I PL SE2
0.01
HOURLY
569180.00
4396327.00
207.98
9.75
791.48
7.16
3.81
IP&L-HARDING
STREET
I PL SE3
0.20
HOURLY
568867.00
4396303.00
208.00
20.12
827.59
57.39
4.21
IP&L-HARDING
STREET
I PL SE4
0.20
HOURLY
568910.00
4396306.00
208.01
20.12
822.04
62.15
4.21
IP&L-HARDING
STREET
I PL SE5
13324.00
HOURLY
568749.00
4396008.00
208.08
79.55
440.93
65.84
1.98
IP&L-HARDING
STREET
I PL SE6
12603.00
HOURLY
568752.00
4395965.00
208.32
79.55
449.82
63.52
1.98
IP&L-HARDING
STREET
I PL SE7
1846.10
HOURLY
568984.00
4395792.00
206.56
172.21
329.26
14.33
6.10
IP&L-HARDING
STREET
I PL SE8
200.30
HOURLY
568984.00
4395792.00
206.56
172.21
414.82
23.44
6.10
IP&L-HARDING
STREET
IPL_SE9
0.30
HOURLY
569050.00
4396339.00
208.26
22.86
810.93
36.58
5.49
B-21
-------
Table B-24. 2013 Rolls Royce, Vertellus, and Quemetco point source emissions, locations, and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
ROLLS ROYCE
RR SN1
0.48
MONTH
567493.00
4398570.00
212.29
4.57
866.48
32.34
0.30
ROLLS ROYCE
RR SN3
1.15
HROFDY
567428.00
4398870.00
212.70
17.37
588.71
33.04
1.22
ROLLS ROYCE
RR SN4
12.39
HROFDY
567402.00
4398886.00
212.70
19.81
755.37
45.51
0.91
ROLLS ROYCE
RR SN9
2.78
MHRDOW
567435.00
4398899.00
212.72
9.14
866.48
21.21
1.52
ROLLS ROYCE
RR SN10
7.24
MONTH
567551.00
4399165.00
212.10
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN11
0.05
MONTH
567551.00
4399165.00
212.10
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN12
2.80
MONTH
567544.50
4399165.00
212.24
15.24
677.59
17.47
1.98
ROLLS ROYCE
RR SN13
4.77
MONTH
567512.00
4399163.00
212.51
18.29
533.15
6.52
1.22
ROLLS ROYCE
RR SN14
0.02
MHRDOW
567513.00
4399174.00
212.61
9.14
866.48
21.21
1.52
ROLLS ROYCE
RR SN15
0.001
MONTH
567439.00
4398911.00
212.70
15.24
755.37
13.53
1.68
VERTELLUS
VERT SN1
25.01
MONTH
566981.00
4399746.00
215.16
9.14
504.26
7.53
1.22
VERTELLUS
VERT SN10
0.07
MONTH
566851.06
4399666.50
214.85
20.42
823.15
5.09
1.07
VERTELLUS
VERT SN11
0.02
MONTH
566901.00
4399710.00
215.15
20.73
823.15
5.09
1.07
VERTELLUS
VERT SN12
0.02
MONTH
566866.94
4399637.00
214.79
21.64
633.15
6.10
1.52
VERTELLUS
VERT SN13
0.02
MONTH
566864.94
4399640.00
214.79
24.69
823.15
6.07
1.07
QUEMETCO
QUE SN1
68.77
MONTH
559977.54
4400993.45
235.78
30.48
327.04
16.86
1.22
QUEMETCO
QUE SN2
3.97
MONTH
559993.31
4400853.53
235.10
50.29
297.04
10.70
3.35
QUEMETCO
QUE_SN3
23.82
MONTH
559993.31
4400853.53
235.10
50.29
321.48
14.84
3.35
B-22
-------
Table B-25. 2013 Rolls Royce area source emissions, locations, and release parameters.
Facility
Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height (m)
X-dimension
(m)
Y-dimension
(m)
Angle
-------
Table B-27. Tulsa Refinery-West 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
TULSA RFNRY WEST
72317213
66440812
100094814
REFWEST SN1
REFWEST SN1
TULSA RFNRY WEST
651913
655212
15606514
REFWEST SN2
REFWEST SN2
TULSA RFNRY WEST
652113
657112
15606314
REFWEST SN3
REFWEST SN3
REFWEST SN1
TULSA RFNRY WEST
663913
654212
16298114
REFWEST SN2
TULSA RFNRY WEST
72311713
66439312
100085314
REFWEST SN4
REFWEST SN4
TULSA RFNRY WEST
664013
654712
16298014
REFWEST SN3
TULSA RFNRY WEST
660813
654812
16303714
REFWEST SN4
TULSA RFNRY WEST
107042213
110579312
151543514
REFWEST SN5
REFWEST SN5
TULSA RFNRY WEST
654413
660012
15477714
REFWEST SN5
REFWEST SN6
REFWEST SN6
TULSA RFNRY WEST
654313
659912
15477814
REFWEST SN6
REFWEST SN7
REFWEST SN7
TULSA RFNRY WEST
654613
655312
15477514
REFWEST SN7
REFWEST SN8
REFWEST SN8
TULSA RFNRY WEST
663113
651512
16299414
REFWEST SN8
REFWEST SN9
REFWEST SN9
TULSA RFNRY WEST
651113
662812
15607614
REFWEST SN9
REFWEST SN11
REFWEST SN11
TULSA RFNRY WEST
650813
656012
15607914
REFWEST SN10
REFWEST SN10
REFWEST SN10
TULSA RFNRY WEST
653513
659612
15478614
REFWEST SN11
REFWEST SN12
REFWEST SN12
TULSA RFNRY WEST
654213
662012
15477914
REFWEST SN12
REFWEST SN13
REFWEST SN13
TULSA RFNRY WEST
651013
658912
15607714
REFWEST SN13
REFWEST SN14
REFWEST SN14
TULSA RFNRY WEST
653613
659012
15478514
REFWEST SN14
REFWEST SN15
REFWEST SN15
TULSA RFNRY WEST
651713
655012
15606714
REFWEST SN15
REFWEST SN16
REFWEST SN16
TULSA RFNRY WEST
654113
663512
15478014
REFWEST SN16
REFWEST SN17
REFWEST SN17
TULSA RFNRY WEST
651313
658812
15607314
REFWEST SN17
REFWEST SN18
REFWEST SN18
TULSA RFNRY WEST
650913
654912
15607814
REFWEST SN18
REFWEST SN19
REFWEST SN19
TULSA RFNRY WEST
651413
661212
15607214
REFWEST SN19
REFWEST SN20
REFWEST SN20
TULSA RFNRY WEST
663813
651712
16298214
REFWEST SN21
TULSA RFNRY WEST
654513
656112
15477614
REFWEST SN22
TULSA RFNRY WEST
653713
659112
15478414
REFWEST SN23
TULSA RFNRY WEST
658713
651412
16408914
REFWEST SN20
REFWEST SN21
REFWEST SN24
B-24
-------
Table B-28. PSO Northeastern Power Station and Sapulpa 2011-2013 AERMOD source identifier crosswalk.
Facility Name
Unit ID
Process ID
Release Point ID
AERMOD 2011
AERMOD 2012
AERMOD 2013
PSO NORTHEASTERN
6698313
6664412
15999814
PSO SE1
PSO SE1
PSO SE1
PSO NORTHEASTERN
6698313
6664412
15999914
PSO SE2
PSO SE2
PSO SE2
PSO NORTHEASTERN
6698513
6664212
15999514
PSO SE3
PSO SE3
PSO SE3
PSO NORTHEASTERN
6698813
6664412
15999114
PSO SE4
PSO SE4
PSO SE4
PSO NORTHEASTERN
6698813
6664412
15999214
PSO SE5
PSO SE5
PSO SE5
PSO NORTHEASTERN
6698913
6664012
15999014
PSO SE6
PSO SE6
PSO SE6
PSO NORTHEASTERN
6699113
6664712
15998814
PSO SE7
PSO SE7
PSO SE7
SAPULPA
8331213
8217212
17068814
SAP SN1
SAP SN1
SAP SN2
SAPULPA
8331413
8217312
17068514
SAP SN2
SAP SN2
SAP SN3
SAPULPA
72251213
66374812
100009614
SAP SN3
SAP SN3
SAP SN4
SAPULPA
8331113
66375112
17068914
SAP SN4
SAP SN4
SAP SN5
SAPULPA
8331113
66375212
17068914
SAP SN4
SAP SN4
SAP SN5
SAPULPA
8331113
66375312
17068914
SAP SN4
SAP SN4
SAP SN5
SAPULPA
8331113
66375412
17068914
SAP SN4
SAP SN4
SAP SN5
SAPULPA
8331113
66375512
17068914
SAP SN4
SAP SN4
SAP SN5
SAPULPA
8331113
66375612
17068914
SAP SN4
SAP SN4
SAP SN5
SAPULPA
8331113
66375712
17068914
SAP SN4
SAP SN4
SAP SN5
SAPULPA
8331113
66376612
17068914
SAP SN4
SAP SN4
SAP SN5
SAPULPA
8331113
66375812
17068914
SAP SN5
SAP SN5
SAP SN6
SAPULPA
8331113
66375912
17068914
SAP SN5
SAP SN5
SAP SN6
SAPULPA
8331113
66376012
17068914
SAP SN5
SAP SN5
SAP SN6
SAPULPA
8331113
66376112
17068914
SAP SN5
SAP SN5
SAP SN6
SAPULPA
8331113
66376212
17068914
SAP SN5
SAP SN5
SAP SN6
SAPULPA
8331113
66376312
17068914
SAP SN5
SAP SN5
SAP SN6
SAPULPA
8331113
66376412
17068914
SAP SN5
SAP SN5
SAP SN6
SAPULPA
8331113
66376512
17068914
SAP SN5
SAP SN5
SAP SN6
SAPULPA
72251313
66375012
100009714
SAP SN6
SAP SN6
SAP SN7
SAPULPA
108757113
112230012
153985314
SAP_SN1
B-25
-------
Table B-29. 2011 Tulsa East Refinery point source emissions, locations, and stack parameters.
Facility Name
AERMOD source
ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
TULSA
RFNRY-EAST
REFEAST SN1
2.00
MONTH
230409.02
4000701.87
192.12
73.15
1088.71
43.34
0.49
TULSA
RFNRY-EAST
REFEAST SN2
0.25
MONTH
229761.77
4000607.68
192.00
30.78
317.59
6.68
0.76
TULSA
RFNRY-EAST
REFEAST SN4
0.83
MONTH
229823.09
4000610.90
192.00
60.96
444.26
5.88
0.61
TULSA
RFNRY-EAST
REFEAST SN5
15.21
MONTH
229944.74
4000860.87
194.00
58.22
572.59
22.92
1.52
TULSA
RFNRY-EAST
REFEAST SN6
0.12
MONTH
229658.38
4000653.14
192.00
29.26
313.71
8.23
1.13
TULSA
RFNRY-EAST
REFEAST SN7
0.13
MONTH
229663.74
4000658.82
192.00
30.48
311.48
7.86
1.13
TULSA
RFNRY-EAST
REFEAST SN8
0.04
MONTH
229954.38
4001000.54
192.90
13.72
570.37
16.06
1.07
TULSA
RFNRY-EAST
REFEAST SN9
0.25
MONTH
229946.71
4000617.28
194.83
42.67
583.15
14.60
1.46
TULSA
RFNRY-EAST
REFEAST SN11
0.44
MONTH
229945.36
4000870.85
194.00
46.02
624.82
3.99
1.77
TULSA
RFNRY-EAST
REFEAST SN12
1.83
MONTH
229956.32
4001096.60
194.47
53.34
449.82
3.47
3.51
TULSA
RFNRY-EAST
REFEAST SN14
0.66
MONTH
229971.12
4000687.91
194.77
38.10
466.48
3.84
2.53
TULSA
RFNRY-EAST
REFEAST SN15
0.74
MONTH
229950.17
4000673.17
194.97
37.80
560.93
10.00
1.77
TULSA
RFNRY-EAST
REFEAST SN16
0.16
MONTH
229950.84
4000700.18
195.00
37.80
533.15
7.04
1.37
TULSA
RFNRY-EAST
REFEAST SN17
1.44
MONTH
229912.85
4001441.17
192.39
21.64
449.82
6.25
2.13
TULSA
RFNRY-EAST
REFEAST_SN18
1.43
MONTH
229940.84
4001441.17
192.31
21.64
449.82
6.19
2.13
B-26
-------
Table B-30. 2011 Tulsa West Refinery point source emissions, locations, and stack parameters.
Facility Name
AERMOD source
ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
TULSA RFNRY
WEST
REFWEST SN1
0.03
MONTH
228617.00
4003889.00
195.00
5.49
616.48
5.06
0.15
TULSA RFNRY
WEST
REFWEST SN2
0.005
MONTH
228750.30
4003806.26
195.10
6.71
588.71
13.20
0.15
TULSA RFNRY
WEST
REFWEST SN3
5.73
MONTH
228706.00
4002861.00
195.00
43.89
477.59
11.83
0.30
TULSA RFNRY
WEST
REFWEST SN4
0.007
MONTH
228658.38
4003859.03
195.10
7.62
547.04
7.25
0.21
TULSA RFNRY
WEST
REFWEST SN5
44.78
MONTH
229176.29
4003711.77
195.10
30.48
637.59
1.92
1.62
TULSA RFNRY
WEST
REFWEST SN6
380.27
MONTH
229185.32
4003728.24
195.10
38.10
548.15
5.15
1.62
TULSA RFNRY
WEST
REFWEST SN7
103.02
MONTH
229202.04
4003723.20
195.20
18.90
505.93
2.99
1.07
TULSA RFNRY
WEST
REFWEST SN8
866.22
MONTH
228262.29
4003837.45
194.30
41.15
522.04
4.88
2.26
TULSA RFNRY
WEST
REFWEST SN9
39.26
MONTH
228236.99
4003995.32
194.20
15.24
471.48
4.11
0.85
TULSA RFNRY
WEST
REFWEST SN10
37.86
MONTH
228237.62
4003989.27
194.20
15.24
683.15
2.99
1.37
TULSA RFNRY
WEST
REFWEST SN11
0.006
MONTH
228251.07
4004028.52
193.90
25.91
768.71
4.05
1.52
TULSA RFNRY
WEST
REFWEST SN12
0.01
MONTH
228262.17
4004029.83
193.90
27.43
736.48
2.19
2.13
TULSA RFNRY
WEST
REFWEST SN13
157.00
MONTH
228246.58
4004020.78
193.90
27.74
922.04
4.82
2.13
TULSA RFNRY
WEST
REFWEST SN14
18.64
MONTH
228246.08
4004012.79
193.90
30.78
877.59
2.04
1.13
TULSA RFNRY
WEST
REFWEST SN15
59.37
MONTH
228239.18
4003982.16
194.30
23.47
523.15
26.33
0.61
TULSA RFNRY
WEST
REFWEST SN16
36.35
MONTH
229175.91
4003721.81
195.10
27.43
560.93
3.20
0.91
TULSA RFNRY
WEST
REFWEST SN17
43.23
MONTH
228239.37
4003969.12
194.60
20.12
594.26
2.38
1.37
TULSA RFNRY
WEST
REFWEST SN18
74.03
MONTH
228279.45
4003823.37
194.50
38.10
726.48
2.26
2.13
TULSA RFNRY
WEST
REFWEST SN19
270.43
MONTH
228279.45
4003823.37
194.50
38.10
738.71
4.88
2.26
TULSA RFNRY
WEST
REFWEST SN20
460.16
MONTH
228688.88
4003894.68
195.19
33.53
394.26
3.41
3.20
B-27
-------
Table B-31. 2011 PSO Northeastern and Sapulpa point source emissions, locations, and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
PSO
NORTHEASTERN
PSO SE1
9007.70
HOURLY
258002.59
4034618.88
195.67
182.88
419.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE2
26.14
HOURLY
258002.59
4034618.88
195.67
182.88
419.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE3
2.36
HOURLY
257841.41
4035283.44
195.41
55.78
393.71
16.28
5.49
PSO
NORTHEASTERN
PSO SE4
8879.30
HOURLY
258002.59
4034618.88
195.67
182.88
419.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE5
25.54
HOURLY
258002.59
4034618.88
195.67
182.88
419.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE6
0.18
HOURLY
257850.92
4035160.78
195.23
45.72
366.48
19.69
5.74
PSO
NORTHEASTERN
PSO SE7
0.20
HOURLY
257850.92
4035160.78
195.23
45.72
366.48
21.55
5.49
SAPULPA
SAP SN1
100.32
MONTH
220648.04
3989373.19
215.01
28.35
530.37
9.60
1.86
SAPULPA
SAP SN2
33.08
MONTH
220621.83
3989378.25
215.62
32.31
498.71
19.39
1.29
SAPULPA
SAP SN3
78.85
MONTH
220621.83
3989378.25
215.62
29.87
515.37
10.27
1.71
SAPULPA
SAP SN4
0.02
MONTH
220667.19
3989381.92
214.54
26.52
310.93
2.13
2.29
SAPULPA
SAP SN5
0.03
MONTH
220667.19
3989381.92
214.54
29.26
310.93
2.13
2.29
Table B-32. 2011 Sapulpa area source emissions, locations, and release parameters.
Facility
Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height (m)
X-dimension
(m)
Y-dimension
(m)
Angle
-------
Table B-33. 2012 Tulsa East Refinery point source emissions, locations, and stack parameters.
Facility Name
AERMOD source
ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
TULSA
RFNRY-EAST
REFEAST SN1
3.91
MONTH
230409.02
4000701.87
192.12
73.15
1088.71
43.34
0.49
TULSA
RFNRY-EAST
REFEAST SN2
1.15
MONTH
229761.77
4000607.68
192.00
30.78
317.59
6.68
0.76
TULSA
RFNRY-EAST
REFEAST SN4
0.38
MONTH
229823.09
4000610.90
192.00
60.96
444.26
5.88
0.61
TULSA
RFNRY-EAST
REFEAST SN5
11.19
MONTH
229944.74
4000860.87
194.00
58.22
572.59
22.92
1.52
TULSA
RFNRY-EAST
REFEAST SN6
0.14
MONTH
229658.38
4000653.14
192.00
29.26
313.71
8.23
1.13
TULSA
RFNRY-EAST
REFEAST SN7
0.15
MONTH
229663.74
4000658.82
192.00
30.48
311.48
7.86
1.13
TULSA
RFNRY-EAST
REFEAST SN8
0.04
MONTH
229954.38
4001000.54
192.90
13.72
570.37
16.06
1.07
TULSA
RFNRY-EAST
REFEAST SN9
0.26
MONTH
229946.71
4000617.28
194.83
42.67
583.15
14.60
1.46
TULSA
RFNRY-EAST
REFEAST SN11
0.58
MONTH
229945.36
4000870.85
194.00
46.02
624.82
3.99
1.77
TULSA
RFNRY-EAST
REFEAST SN12
1.35
MONTH
229956.32
4001096.60
194.47
53.34
449.82
3.47
3.51
TULSA
RFNRY-EAST
REFEAST SN14
0.62
MONTH
229971.12
4000687.91
194.77
38.10
466.48
3.84
2.53
TULSA
RFNRY-EAST
REFEAST SN15
0.72
MONTH
229950.17
4000673.17
194.97
37.80
560.93
10.00
1.77
TULSA
RFNRY-EAST
REFEAST SN16
0.16
MONTH
229950.84
4000700.18
195.00
37.80
533.15
7.04
1.37
TULSA
RFNRY-EAST
REFEAST SN18
1.46
MONTH
229912.85
4001441.17
192.39
21.64
449.82
6.25
2.13
TULSA
RFNRY-EAST
REFEAST_SN19
1.20
MONTH
229940.84
4001441.17
192.31
21.64
449.82
6.19
2.13
B-29
-------
Table B-34. 2012 Tulsa West Refinery point source emissions, locations, and stack parameters.
Facility Name
AERMOD source
ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
TULSA RFNRY
WEST
REFWEST SN1
0.007
MONTH
228617.00
4003889.00
195.00
5.49
616.48
5.06
0.15
TULSA RFNRY
WEST
REFWEST SN2
0.005
MONTH
228750.30
4003806.26
195.10
6.71
588.71
13.20
0.15
TULSA RFNRY
WEST
REFWEST SN3
7.66
MONTH
228706.00
4002861.00
195.00
43.89
477.59
11.83
0.30
TULSA RFNRY
WEST
REFWEST SN4
0.007
MONTH
228658.38
4003859.03
195.10
7.62
547.04
7.25
0.21
TULSA RFNRY
WEST
REFWEST SN5
0.017
MONTH
228617.00
4003889.00
195.00
5.49
616.48
5.06
0.15
TULSA RFNRY
WEST
REFWEST SN6
20.44
MONTH
229176.29
4003711.77
195.10
30.48
637.59
1.92
1.62
TULSA RFNRY
WEST
REFWEST SN7
237.06
MONTH
229185.32
4003728.24
195.10
38.10
548.15
5.15
1.62
TULSA RFNRY
WEST
REFWEST SN8
41.63
MONTH
229202.04
4003723.20
195.20
18.90
505.93
2.99
1.07
TULSA RFNRY
WEST
REFWEST SN9
687.65
MONTH
228262.29
4003837.45
194.30
41.15
522.04
4.88
2.26
TULSA RFNRY
WEST
REFWEST SN10
45.53
MONTH
228237.62
4003989.27
194.20
15.24
683.15
2.99
1.37
TULSA RFNRY
WEST
REFWEST SN11
43.48
MONTH
228236.99
4003995.32
194.20
15.24
471.48
4.11
1.52
TULSA RFNRY
WEST
REFWEST SN12
0.004
MONTH
228251.07
4004028.52
193.90
25.91
768.71
4.05
1.52
TULSA RFNRY
WEST
REFWEST SN13
0.007
MONTH
228262.17
4004029.83
193.90
27.43
736.48
2.19
2.13
TULSA RFNRY
WEST
REFWEST SN14
150.00
MONTH
228246.58
4004020.78
193.90
27.74
922.04
4.82
2.13
TULSA RFNRY
WEST
REFWEST SN15
18.25
MONTH
228246.08
4004012.79
193.90
30.78
877.59
2.04
1.13
TULSA RFNRY
WEST
REFWEST SN16
65.03
MONTH
228239.18
4003982.16
194.30
23.47
523.15
26.33
0.61
TULSA RFNRY
WEST
REFWEST SN17
18.27
MONTH
229175.91
4003721.81
195.10
27.43
560.93
3.20
0.91
TULSA RFNRY
WEST
REFWEST SN18
41.40
MONTH
228239.37
4003969.12
194.60
20.12
594.26
2.38
1.37
TULSA RFNRY
WEST
REFWEST SN19
54.57
MONTH
228279.45
4003823.37
194.50
38.10
726.48
2.26
2.13
TULSA RFNRY
WEST
REFWEST SN20
210.11
MONTH
228279.45
4003823.37
194.50
38.10
738.71
4.88
2.26
TULSA RFNRY
WEST
REFWEST SN21
370.21
MONTH
228688.88
4003894.68
195.19
33.53
394.26
3.41
3.20
B-30
-------
Table B-35. 2012 PSO Northeastern and Sapulpa point source emissions, locations, and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
PSO
NORTHEASTERN
PSO SE1
7401.70
HOURLY
258002.59
4034618.88
195.67
182.88
394.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE2
26.69
HOURLY
258002.59
4034618.88
195.67
182.88
394.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE3
3.08
HOURLY
257841.41
4035283.44
195.41
55.78
393.71
16.28
5.49
PSO
NORTHEASTERN
PSO SE4
8038.60
HOURLY
258002.59
4034618.88
195.67
182.88
394.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE5
19.99
HOURLY
258002.59
4034618.88
195.67
182.88
394.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE6
2.27
HOURLY
257850.92
4035160.78
195.23
45.72
366.48
19.69
5.74
PSO
NORTHEASTERN
PSO SE7
2.42
HOURLY
257850.92
4035160.78
195.23
45.72
366.48
21.55
5.49
SAPULPA
SAP SN1
100.32
MONTH
220648.04
3989373.19
215.01
28.35
530.37
9.60
1.86
SAPULPA
SAP SN2
33.08
MONTH
220621.83
3989378.25
215.62
32.31
498.71
19.39
1.29
SAPULPA
SAP SN3
78.85
MONTH
220621.83
3989378.25
215.62
29.87
515.37
10.27
1.71
SAPULPA
SAP SN4
0.02
MONTH
220667.19
3989381.92
214.54
26.52
310.93
2.13
2.29
SAPULPA
SAP SN5
0.03
MONTH
220667.19
3989381.92
214.54
29.26
310.93
2.13
2.29
Table B-36. 2012 Sapulpa area source emissions, locations, and release parameters.
Facility
Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height (m)
X-dimension
(m)
Y-dimension
(m)
Angle
-------
Table B-37. 2013 Tulsa East Refinery point source emissions, locations, and stack parameters.
Facility Name
AERMOD source
ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
TULSA
RFNRY-EAST
REFEAST SN1
11.85
MONTH
230409.02
4000701.87
192.12
73.15
1088.71
43.34
0.49
TULSA
RFNRY-EAST
REFEAST SN2
0.34
MONTH
229761.77
4000607.68
192.00
30.78
317.59
6.68
0.76
TULSA
RFNRY-EAST
REFEAST SN4
0.26
MONTH
229823.09
4000610.90
192.00
60.96
444.26
5.88
0.61
TULSA
RFNRY-EAST
REFEAST SN5
5.08
MONTH
229944.74
4000860.87
194.00
58.22
572.59
22.92
1.52
TULSA
RFNRY-EAST
REFEAST SN6
0.10
MONTH
229658.38
4000653.14
192.00
29.26
313.71
8.23
1.13
TULSA
RFNRY-EAST
REFEAST SN7
0.10
MONTH
229663.74
4000658.82
192.00
30.48
311.48
7.86
1.13
TULSA
RFNRY-EAST
REFEAST SN8
0.05
MONTH
229954.38
4001000.54
192.90
13.72
570.37
16.06
1.07
TULSA
RFNRY-EAST
REFEAST SN9
0.22
MONTH
229946.71
4000617.28
194.83
42.67
583.15
14.60
1.46
TULSA
RFNRY-EAST
REFEAST SN11
0.45
MONTH
229945.36
4000870.85
194.00
46.02
624.82
3.99
1.77
TULSA
RFNRY-EAST
REFEAST SN12
0.83
MONTH
229956.32
4001096.60
194.47
53.34
449.82
3.47
3.51
TULSA
RFNRY-EAST
REFEAST SN14
0.44
MONTH
229971.12
4000687.91
194.77
38.10
466.48
3.84
2.53
TULSA
RFNRY-EAST
REFEAST SN15
0.57
MONTH
229950.17
4000673.17
194.97
37.80
560.93
10.00
1.77
TULSA
RFNRY-EAST
REFEAST SN16
0.11
MONTH
229950.84
4000700.18
195.00
37.80
533.15
7.04
1.37
TULSA
RFNRY-EAST
REFEAST SN17
0.99
MONTH
229912.85
4001441.17
192.39
21.64
449.82
6.25
2.13
TULSA
RFNRY-EAST
REFEAST_SN18
1.02
MONTH
229940.84
4001441.17
192.31
21.64
449.82
6.19
2.13
B-32
-------
Table B-38. 2013 Tulsa West Refinery point source emissions, locations, and stack parameters.
Facility Name
AERMOD source
ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
TULSA RFNRY
WEST
REFWEST SN1
8.22
MONTH
228706.00
4002861.00
195.00
43.89
477.59
11.83
0.30
TULSA RFNRY
WEST
REFWEST SN2
0.15
MONTH
228659.61
4003895.03
195.10
18.29
433.15
8.23
1.52
TULSA RFNRY
WEST
REFWEST SN3
0.26
MONTH
228660.10
4003903.01
195.10
18.29
440.37
6.49
1.52
TULSA RFNRY
WEST
REFWEST SN4
0.10
MONTH
228658.38
4003859.03
195.10
24.38
425.93
6.25
1.52
TULSA RFNRY
WEST
REFWEST SN5
0.02
MONTH
228617.00
4003889.00
195.00
5.49
616.48
5.06
0.15
TULSA RFNRY
WEST
REFWEST SN6
9.09
MONTH
229176.29
4003711.77
195.10
30.48
637.59
1.92
1.62
TULSA RFNRY
WEST
REFWEST SN7
169.39
MONTH
229185.32
4003728.24
195.10
38.10
548.15
5.15
1.62
TULSA RFNRY
WEST
REFWEST SN8
26.45
MONTH
229202.04
4003723.20
195.20
18.90
505.93
2.99
1.07
TULSA RFNRY
WEST
REFWEST SN9
360.29
MONTH
228262.29
4003837.45
194.30
41.15
522.04
4.88
2.26
TULSA RFNRY
WEST
REFWEST SN10
8.45
MONTH
228237.62
4003989.27
194.20
15.24
683.15
2.99
1.37
TULSA RFNRY
WEST
REFWEST SN11
16.96
MONTH
228236.99
4003995.32
194.20
15.24
471.48
4.11
1.52
TULSA RFNRY
WEST
REFWEST SN12
0.002
MONTH
228251.07
4004028.52
193.90
25.91
768.71
4.05
1.52
TULSA RFNRY
WEST
REFWEST SN13
0.003
MONTH
228262.17
4004029.83
193.90
27.43
736.48
2.19
2.13
TULSA RFNRY
WEST
REFWEST SN14
36.95
MONTH
228246.58
4004020.78
193.90
27.74
922.04
4.82
2.13
TULSA RFNRY
WEST
REFWEST SN15
4.42
MONTH
228246.08
4004012.79
193.90
30.78
877.59
2.04
1.13
TULSA RFNRY
WEST
REFWEST SN16
23.79
MONTH
228239.18
4003982.16
194.30
23.47
523.15
26.33
0.61
TULSA RFNRY
WEST
REFWEST SN17
10.56
MONTH
229175.91
4003721.81
195.10
27.43
560.93
3.20
0.91
TULSA RFNRY
WEST
REFWEST SN18
10.76
MONTH
228239.37
4003969.12
194.60
20.12
594.26
2.38
1.37
TULSA RFNRY
WEST
REFWEST SN19
34.20
MONTH
228279.45
4003823.37
194.50
38.10
726.48
2.26
2.13
TULSA RFNRY
WEST
REFWEST SN20
124.53
MONTH
228279.45
4003823.37
194.50
38.10
738.71
4.88
2.26
TULSA RFNRY
WEST
REFWEST SN21
0.03
MONTH
228524.37
4004105.79
195.40
27.74
555.37
3.02
1.22
B-33
-------
Table B-39. 2013 Tulsa West Refinery point source emissions, locations, and stack parameters.
Facility Name
AERMOD source
ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
TULSA RFNRY
WEST
REFWEST SN22
0.14
MONTH
229194.24
4003726.69
195.20
34.14
478.71
3.20
1.83
TULSA RFNRY
WEST
REFWEST SN23
0.07
MONTH
228527.85
4004113.59
195.10
34.14
610.93
2.47
1.68
TULSA RFNRY
WEST
REFWEST_SN24
211.21
MONTH
228688.88
4003894.68
195.19
33.53
394.26
3.41
3.20
Table B-40. 2013 PSO Northeastern and Sapulpa point source emissions, locations, and stack parameters.
Facility Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x (m)
UTM-y (m)
Elevation
(m)
Stack
height
(m)
Stack
temperature
(K)
Stack
velocity
(m S"1)
Stack
diameter
(m)
PSO
NORTHEASTERN
PSO SE1
9337.20
HOURLY
258002.59
4034618.88
195.67
182.88
394.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE2
22.32
HOURLY
258002.59
4034618.88
195.67
182.88
394.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE3
1.38
HOURLY
257841.41
4035283.44
195.41
55.78
393.71
16.28
5.49
PSO
NORTHEASTERN
PSO SE4
9007.50
HOURLY
258002.59
4034618.88
195.67
182.88
394.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE5
38.16
HOURLY
258002.59
4034618.88
195.67
182.88
394.26
13.81
8.23
PSO
NORTHEASTERN
PSO SE6
2.88
HOURLY
257850.92
4035160.78
195.23
45.72
366.48
19.69
5.74
PSO
NORTHEASTERN
PSO SE7
3.11
HOURLY
257850.92
4035160.78
195.23
45.72
366.48
21.55
5.49
SAPULPA
SAP SN1
0.01
MONTH
220685.88
3989163.75
216.57
2.44
755.37
21.73
0.10
SAPULPA
SAP SN2
108.29
MONTH
220648.04
3989373.19
215.01
28.35
530.37
9.60
1.86
SAPULPA
SAP SN3
33.74
MONTH
220621.83
3989378.25
215.62
32.31
498.71
19.39
1.29
SAPULPA
SAP SN4
98.48
MONTH
220621.83
3989378.25
215.62
29.87
515.37
10.27
1.71
SAPULPA
SAP SN5
0.02
MONTH
220667.19
3989381.92
214.54
26.52
310.93
2.13
2.29
SAPULPA
SAP_SN6
0.03
MONTH
220667.19
3989381.92
214.54
29.26
310.93
2.13
2.29
B-34
-------
Table B-41. 2013 Sapulpa area source emissions, locations, and release parameters.
Facility
Name
AERMOD
source ID
Emissions
(tons year1)
Emission
factor
UTM-x
(m)
UTM-y
(m)
Elevation
(m)
Release
height (m)
X-dimension
(m)
Y-dimension
(m)
Angle
-------
REFERENCES
UNC (University of North Carolina). (2017). Sparse Matrix Operator Kernel Emissions
modeling system User Manual. Available at:
https://www.cmascenter.org/help/documentation.cfm?MODEL=smoke&VERSION=4.5
U.S. EPA. (2016). User's Guide for the AMS/EPA Regulatory Model - AERMOD. EPA-454/B-
16-011. U.S. Environmental Protection Agency, Research Triangle Park, NC 27711.
B-36
-------
APPENDIX C
AIR QUALITY MODELING DOMAINS FOR STUDY AREAS
Preface: The modeling domains, including receptors and modeled sources, for the three study
areas are shown in Figures C-l and C-2, for Fall River, Figures C-3 and C-4 for Indianapolis,
and Figures C-5 and C-6 for Tulsa. Sources are denoted by stars, monitors by triangles, and
gridded receptors by small dots. The blue airport symbol denotes the location of the NWS station
used in the modeling.
C-l
-------
Brayton Point
m WjE^ i sumNM
250051004
jO 1.75 3.5
i Kilometers!
Figure C-l. Fall River study area air quality modeling domain.
C-2
-------
[Elevation (m )|
Brayton Point
250051004
Kilometers
Figure C-2. Detailed view of Fall River study area air quality modeling domain.
C-3
-------
180970078
180970073
CitizensThermal
Marion
¦j* Quemetco, Inc.
Vertellus
180970057
Belmont WWTP
IP & L - Harding St.
Kilometers
Figure C-3. Indianapolis study area air quality modeling domain.
C-4
-------
Citizens Thermal
Vertellus
Quemetco. Inc.
.m 180970057
Rolls Royce Corp
Belmont WWTP
'MmjfoJh,
IP & L - Harding St.
|0 0.45 0.9
Kilometersj
Figure C-4. Detailed view of Indianapolis study area air quality modeling domain.
C-5
-------
WitSWW&frdSi
t _
PSO Northeastern Station
|Elevation (m)|
401431127
West Refinery
401430175
m 401430235
East Refinery
SAPULPA
Kilometers I
Figure C-5. Tulsa study area air quality modeling domain.
C-6
-------
401431127
West Refinery
401430175
401430235
East Refinery
0 0.5 1 2 3 4
Kilometers!
BBaggBMBSflflHaiBBBBMi -
Figure C-6. Detailed view of Tulsa study area air quality modeling domain.
C-7
-------
APPENDIX D
MODELED AIR QUALITY EVALUATION
AERMOD output for the three study areas was evaluated using three methods. First,
comparison of the 99th percentile of daily 1-hour maximum concentrations for each and
subsequent 3-year design values were compared at each monitor. Second, simple QQ-plots were
generated to provide a quick visual performance of the model for 1-hour, 3-hour, and 24-hour
averages. The QQ-plots are comparisons of the observed and modeled concentrations, unpaired
in time and space, consistent with regulatory evaluations of AERMOD (U.S. EPA, 2003;
Venkatram et al., 2001). Third, for a more rigorous comparison, the EPA Protocol for
determining best performing model, or sometimes called the Cox-Tikvart method (U.S. EPA,
1992; Cox and Tikvart, 1990) was used. Normally, this protocol is used to determine which
model or model scenarios among a suite of models or scenario is the better performer for
regulatory application and focuses on the higher concentrations in the concentration distribution
as these are the concentrations of interest in most regulatory applications (State Implementation
Plans and Prevention of Significant Deterioration). For example, U.S. EPA (2016) used the
protocol to determine which was a better performer in terms of meteorological data, observed or
prognostic data. For the study presented here, we are only evaluating one model and one
scenario, i.e., AERMOD for 2011-2013. Therefore, the protocol will not be used to its full
extent, but rather to provide information regarding the performance of the model for these study
areas. An explanation of the protocol follows.
The protocol uses fractional bias (equation D-l) for evaluating model performance.
Where FB is the fractional bias, OB is the average of the highest 25 observed concentrations and
PR is the average of the highest 25 predicted averages.
In the evaluation, air quality models are subjected to a comprehensive statistical
comparison that involves both an operational and scientific component. The operational
component is to measure the model's ability to estimate concentration statistics most directly
used for regulatory purposes and the scientific component evaluates the model's ability to
perform accurately throughout the range of meteorological conditions and the geographic area of
concern (U.S. EPA, 1992). The test statistic used for the comparison is the robust highest
concentration (RHC) statistic and is given by:
|"ob-pr"|
[ob+pr\
Equation D-l
RHC = X(JV) + [X - X(JV)] x In [^]
Equation D-2
D-l
-------
Where X(N) is the Nth largest value, X\s the average of N-l values, and Nis the number of
values exceeding the threshold value, usually 26.
The operational component of the evaluation compares performance in terms of the
largest network-wide RHC test statistic. The RHC is calculated separately for each monitor
within the network for both observed and modeled values. The absolute fractional bias (AFB) is
calculated for both 3 and 24-hour averages using the absolute value of the results of equation 1.
The inputs to the AFB calculation are the highest observed RHC and the highest modeled RHC.
The scientific component of the evaluation is also based on absolute fractional bias but
the bias is calculated using the RHC for each meteorological condition and monitor. The
meteorological conditions are a function of atmospheric stability and wind speed. For the
purposes of these studies, six unique conditions were defined based on two wind speed
categories (below and above 2.0 m/s) and three stability categories: unstable, neutral, and stable.
1 In scientific evaluation, only 1-hour concentrations are used and the AFB is based on RHC
values paired in space and stability/wind speed combination.
A composite performance measure (CPM) is calculated from the 1-hour, 3-hour, and 24-
hour AFB's:
CPM = i X (.AFBtJ) + ^ X pFB3-^24j Equation D-3
Where AFBy is the absolute fractional bias for monitor i and meteorological condition j, AFBij
is the average absolute fractional bias across all monitors and meteorological conditions, AFB3 is
the absolute fractional bias for the 3-hour average, and AFB24 is the absolute fractional bias for
the 24-hour average. The closer the CPM is to zero, the better the performance of the model.
Also, since the absolute fraction biases are calculated using equation 1, which is bounded by 2
(U.S. EPA, 1992), then the maximum value for the CPM is also 2.
Both the QQ-plots and the EPA protocol are applied to the model output in two ways.
First, evaluations were conducted by comparing model output and observations unpaired in time
and space, consistent with regulatory evaluations of AERMOD (U.S. EPA, 2003; Venkatram et
al., 2001). In regulatory applications, the emphasis is not on where potential modeled NAAQS
violations occur, but whether they occur. Second, given the nature of this particular study as an
exposure analysis, where individual receptors are being used on an hourly basis, the QQ-plots
and the EPA protocol were both applied to model output at individual monitors. This would be a
pairing in space but not necessarily time. This would help answer the question, is the model
1 In U.S. EPA (1992), the three stability categories are related to the Pasquill-Gifford categories, unstable being A,
B, and C, neutral being D, and stable being E and F. Since AERMOD does not use the stability categories, the
stability class was determined using Monin-Obukhov length and surface roughness using methodology from
AERMOD subroutine LTOPG.
D-2
-------
performing well at predicting the locations of concentrations of interest. Also, since the monitors
in each of the study areas are located near populations, if the model performs well near these
monitors then reasonable performance in the population areas, or areas of interest for exposure,
can be expected. For all three areas, QQ-plots and the EPA protocol were performed for the
entire three-year period, 2011-2013, and for each year individually to see if individual years were
driving the total period comparisons.
Fall River: Modeled Air Quality Evaluation
Only one monitor (Figure C-l, Figure C-2) was located in the vicinity of Brayton Power
Station. Table D-l shows the monitored and modeled annual 99th percentile daily 1-hour
maximum concentration and the three-year design value. With the exception of 2011, the model
under-predicts the 99th percentile of the daily 1-hour maximum concentration and under-predicts
the 3-year design value.
Table D-l. Fall River monitored and modeled annual 99th percentile daily 1-hour
maximum concentrations (|j,g m3) and 3-year design value (|j.g m3).
Year
Monitor
Observed
Model
2011
250051004
169.8
177.1
2012
250051004
171.1
138.2
2013
250051004
161.9
84.9
Design Value
250051004
167.6
133.4
Figures D-l through D-3 show the QQ-plots for 1-hour, 3-hour, and 24-hour averages
respectively. In each figure, panel a is the ranked comparisons for the entire 3-year period, while
panels b-d are the individual years' ranked pairings. For the 1-hour comparison across all three
years, the model is over predicting at the lower end of the concentration distributions (less than
50 |Lxg m"3), predicts very well at the middle of the distribution (50 -125 |Lxg m"3) and then shifts to
under-prediction from 150 to 250 |Lxg m"3. At the very high end, i.e. the last three observations,
the model over-predicts, under-predicts and is almost equal to the highest monitored
concentration. Analyzing the three individual years, the model appears to perform the best in
2011. The 3-hour QQ-plots exhibit similar patterns as the 1-hour plots. The 24-hour plots exhibit
a pattern of over-prediction at the low to mid-range of the distributions and then under prediction
at the high ends.
D-3
-------
Fall River 1-hr Q-Q Plot
Fall River 1-hr Q-Q Plot - 2011
0 100 290 300 «90
Observed
100 300 900 400
Deserved
Figure D-l, Fall River 1-hour QQ plots.
Fall River 1-hr Q-Q Plot -2012
Fall River 1-hr Q-Q Plot-2013
100 200 300 400
Observed
200
Observed
D-4
-------
Fall River 3-hr Q-Q Plot
Observed
Fall River 3-hr Q-Q Plot - 2011
200
Observed
Fall River 3-hr Q-Q Plot - 2012
Fall River 3-hr Q-Q Plot - 2013
100 190
Observed
200 250
100 159
Observed
200
Figure D-2. Fall River 3-hour QQ plots.
D-5
-------
Fall River 24-hr Q-Q Plot
Fall River 24-hr Q-Q Plot-2011
cr
/ ~
1
/ ~
* X
S 50* /
• / •
yC •
¦y
Cu •• /
/~* x
r /
" V
o- /
9 29 SO 79 tQO
Observed
Observed
Fall River 24-hr Q-Q Plot - 2012
'M' C y/
Fall River 24-hr Q-Q Plot - 2013
« d J
|
^ 0 *
i /
« 50 ¦ /
*
- w-
^ /
• /
°" /
• /
0 25 50 75 TOO
Observed
Observed
Figure D-3. Fall River QQ-plots.
In addition to the QQ-plots, composite performance metrics, CPM, were calculated for
the entire period and each of the individual years.
Table D-2 lists the CPM values for 2011-13 and CPM values for the individual years.
Also shown are the absolute fractional biases for 1-hour, 3-hour, and 24-hours. Overall,
considering impacts from the three averaging periods, 2011 was the better performing year of the
three years and the 2011-2013 CPM shows the influence of 2013.
D-6
-------
Table D-2. Fall River composite performance metrics (CPM) and absolute fractional biases
'or 1-hour, 3-hour, and 24-hour averages.
Period
CPM
AFBl.hr
AFBs-hr
.£
CM
OQ
Ll_
<
2011-2013
0.45
0.68
0.30
0.38
2011
0.29
0.56
0.21
0.10
2012
0.35
0.43
0.22
0.41
2013
0.49
0.75
0.52
0.20
Indianapolis: Modeled Air Quality Evaluation
Three monitors were available for model evaluation in Indianapolis (Figure C-3). Table
D-3 lists the annual 99th percentile daily 1-hour maximum concentration and 3-year design value
for each monitor. The model is over-predicting at monitor 180970057 (the nearest monitor to the
sources) and generally under-predicting each year and the design values at the other monitors.
The modeled design value at 180970057 is within 10% of the monitored design value while at
the other monitors, the modeled design values are within 3% of the monitored design values.
Table D-3. Indianapolis monitored and modeled annual 99th percentile daily 1-hour
maximum concentrations (|j,g m3) and 3-year design value (|j.g m3).
Monitor
Year
Observed
Modeled
2011
164.8
268.7
180970057
2012
239.4
330.8
2013
204.3
367.5
Design Value
202.8
322.4
2011
155.6
122.1
180970073
2012
146.8
129.9
2013
110.7
151.4
Design Value
137.7
134.4
2011
156.2
153.9
180970078
2012
159.9
162.2
2013
182.4
168.7
Design Value
166.1
161.6
One-hour, 3-hour, and 24-hour QQ-plots across all three monitors are shown in Figures
D-4 through D-6, respectively. For 1-hour averages, the 3-year QQ-plot and 2012 and 2013 QQ-
plots show an over-prediction trend except at the higher concentrations for the 3-year period and
2012, where there is under-prediction. Analysis of the 2012 higher 1-hour concentrations (Figure
D-4) showed very high observations for those years which the model did not simulate while
2011 actually shows very good model performance. For the 3-hour averages (Figure D-5), all
three years and the entire period show good model to monitor agreement with some over-
prediction in 2013. The 24-hour averages (Figure D-6), the 3-year period and each individual
year exhibit over-prediction. Overall, 2011 appeared to show the better performance among the
years among all averaging periods. Figures D-7 through D-9 show the 1-hour, 3-hour, and 24-
D-7
-------
hour QQ-plots for the individual monitors for the 3-year period and by year. Results were mixed
among the three monitors. For the 1-hour averages, monitor 180970057, the closest monitor to
the modeled sources (Figure C-3, Figure C-4), the modeled concentrations were higher than
monitored values except at the highest concentrations for 2012. The annual 99th percentile daily
1-hour maxima and design value in Table D-3 reflect the over-prediction. For 2013, the model
overestimated throughout the distribution. For the other two monitors, the modeled values
showed good agreement through most of the concentration distribution and then tended toward
underestimation at the higher end of the distributions. The same general trend was seen with the
3-hour average concentrations (Figure D-8) for monitor 180970073. All three monitors exhibit
over-prediction for the 24-hour period (Figure D-9) which could be a consequence of the
seasonal background included in the Indianapolis modeling as 3-year average background
concentrations are added to each individual hour in the modeling. The inclusion of an average
background could be over-estimated for some individual hours and when calculating a multi-
hour average, e.g. 3 or 24-hour, the overestimates could accumulate over time.
D-8
-------
Marion 1-hr Q-Q Plot
a
Observed
Manon 1-hr Q-Q Plot - 2011
Observed
Manon 1-hr Q-Q Plot - 2012
Observed
Manon 1-hr Q-Q Plot - 2013
d
Observed
Figure D-4. Indianapolis 1-hour QQ-plots.
D-9
-------
Marion 3-hr Q-Q Plot
200
Observed
Marion 3-hr Q-Q Plot - 2011
~MTV M
Marion 3-hr Q-Q Plot - 2012
Manon 3-hr Q-Q Plot - 2013
d
ZJO
Observed
200
Observed
Figure D-5. Indianapolis 3-hour QQ-plots.
D-
10
-------
Marion 24-hr Q-Q Plot
a
Observed
Manon 24-hr Q-Q Plot - 2011
b
so
Observed
Manon 24-hr Q-Q Plot - 2012
Manon 24-hr Q-Q Plot - 2013
Observed
Figure D-6. Indianapolis 24-hour QQ-plots.
Do—nwd
D-ll
-------
Marion 1-N QO Plot Monitor 180970057
a /
OftMTV«d
Manon Uv Q-Q Plot Monitor 180970057 . 2011
" b /
* /
(Ml—tv—
Marion 1*r Q-Q Plot Monrtor 180970057 2012
C //
xr/'
¦/
Olwvtl
Manon 1-hr Q-Q Ptot - Monitor 180970057 2013
~ d
V^.
OMarv«d
Manon 1*r Q-Q Plot Monrtor 180970073
OMftfvad
Manon 1 h* OQ Plot • Monrtor 180970073 2011
OM*r.»3
Manon 1-hr QX3 Plot Monrtor 180970073 2012
g /
y
OtMTMl
Manon 1 hi Q-Q Plot Monitor 180970073 2013
h /
f/
ObMTv*}
Marion 1 fv Q-Q Plot Monitor 180970078
ObMfVM
Manon 1* Q-Q Plot Monitor 180970078 2011
j /
Manon IhrQ-QPW Monitor 180970078 2012
OMawt
Manon HwQ-Q PVot Monitor 180970078 2013
OMarvM
Figure D-7. 1-hour QQ plots for individual monitors in Indianapolis.
D-12
-------
Manon 3hr OQ PW Monitor 180970057
Manon 3-hr Q-Q Plot • Monitor 180970057 2011
Manon 3-hr Q-Q PfeC Monitor 180970057 2012
C
Manon 3 fw Q-Q Pfc* Monitor 180970057 2013
Manon 3-h* Q-Q Ptct Monitor 180970073
Manon 3-hr Q-Q Pl<* Momtcr 180970073 2011
Manon 3-hr Q-Q Ptc* Monitor 180970073 2012
Manon 3-hr Q-Q PtoJ - Monitor 180970073 2013
Manon 3hr OQ PW Monitor 180970070
Manon 3-hr Q-Q PV* Monitor 180970078 2011
Manon 3 hrQ-QPk* Monitor 180970078 2012
Manon 3-hr Q-Q PW Monitor 180970078 2013
Figure D-8. 3-hour QQ-plots for individual monitors in Indianapolis.
D-13
-------
Manon 24-hr O-Q Plot Monrtor 180970057
OBUfrM
Manon 244* Q-Q Plot - Morutor 180970057 • 2011
1 b
j' /
: //
06mrv«rt
Manon 244* Q-Q Plot • Monitor 180970057 • 2012
C * //
.j-* /
v7
ObMTVtd
Manon 244* Q-Q Plot Monitor 180970057 2013
" d /
/* /
Ofttarvea
Man on 24 hr QQ Plot Monitor 180970073
e /
; ^ ^
OftMTVtd
Manon 244* QQ Plot Monrtor 180970073 • 2011
. f /
** /
# /
* ^
OMerved
Marion 244* QQ Plot Monitor 180970073 2012
g
akMNM
Manon 244* Q-Q Plot - Monrtor 180970073 • 2013
h /
* /
//
Observed
Marion 24 4* Q-Q PW Monitor 180970078
i y/
OftMTWd
Manon 24 4* QQ Plot - Monitor 180970078 2011
j /
//"
nHwirK
Marion 244* QQ Plot Monrtor 180970078 2012
1 k
¦//X
Oft—rv«rt
Marion 244* QQ Plot Monrtor 180970078 ¦ 2013
1 /
Otaerved
Figure D-9. 24-hour QQ-plots for individual monitors in Indianapolis.
D-14
-------
CPM values were calculated for 2011, 2012, and 2013 and the entire 3-year period and
are shown in Table D-4 across all monitors and each individual monitor.
Table D-4. Indianapolis composite performance metrics (CPM) and absolute fractional
biases for 1-hour, 3-hour, and 24-hour averages.
Period
Monitor
CPM
AFBl.hr
AFBs-hr
.£
CM
OQ
Ll_
<
2011-2013
All
0.21
0.41
0.03
0.19
180970057
0.33
0.62
0.03
0.34
180970073
0.28
0.28
0.40
0.17
180970078
0.25
0.35
0.37
0.03
2011
All
0.32
0.45
0.14
0.38
180970057
0.48
0.61
0.26
0.57
180970073
0.34
0.44
0.49
0.07
180970078
0.26
0.29
0.20
0.30
2012
All
0.32
0.53
0.01
0.43
180970057
0.37
0.66
0.01
0.43
180970073
0.44
0.60
0.16
0.56
180970078
0.23
0.34
0.21
0.15
2013
All
0.29
0.53
0.17
0.16
180970057
0.45
0.74
0.17
0.43
180970073
0.44
0.44
0.26
0.61
180970078
0.24
0.41
0.29
0.03
The CPM values based on all monitors indicates relatively good model performance, for
each individual year, as well as the entire 3-year period. Monitor 180970057 tends to have higher
CPM values than the other monitors, possibly due to the inclusion of background increasing
concentration while the monitor is impacted by most of the modeled sources as well. The one
outlier in the CPM values is monitor 180970073 for 2011, with a CPM value of 0.74, much
higher than the other monitors in 2011 or the CPM based on all three monitors. The high CPM
appears to be due to the high AFB values for the 3-hour and 24-hour periods for the monitor as
the monitor under-predicts compared to the other monitors for 2011 (Figures 3-1 If and 3-12f).
Tulsa: Modeled Air Quality Evaluation
Three monitors were available for model evaluation in Tulsa (Figures C-5 and C-6).
Table D-5 shows the annual 99th percentile of the daily 1-hour maximum concentrations and
design values for each monitor. The model under-predicts the design value for 401430175 but
does very well at the design value predictions for the other two monitors.
D-15
-------
Table D-5. Tulsa monitored and modeled annual 99th percentile daily 1-hour maximum
trations (|j,g m 3) and 3-year design value (|j.g
m 3).
Monitor
Year
Observed
Modeled
2011
177.9
141.3
401430175
2012
143.9
117.7
2013
109.9
63.9
Design Value
143.9
107.6
2011
88.9
122.8
401430235
2012
2013
62.8
49.8
99.7
52.6
Design Value
67.1
91.7
2011
66.2
63.9
401431127
2012
40.5
56.6
2013
51.8
36.8
Design Value
52.8
52.4
One-hour, 3-hour, and 24-hour average QQ-plots are shown in Figures D-10 through D-
12 respectively across all monitors and QQ-plots by monitor are shown in Figures D-13 through
D-15. For the 1-hour averages (Figure D-10), the model tends to over-predict for much of the
concentration distribution for the total 3-year period as well as 2011 and 2012. 2013 shows a
trend to more of the distribution being under-predicted. The 3-hour averages (Figure D-l 1) also
show a trend of over-prediction and then under-prediction at the high end of the concentration
distributions but perhaps less pronounced over-prediction than for the 1-hour averages. The 24-
hour averages (Figure D-l2) for the 3-year period show slight over-prediction at the lower ends
of the distribution with good agreement in the middle followed by under-prediction but over-
prediction at the very top of the distribution. 2011 shows slight over-prediction for much of the
distribution, followed by under-prediction and over-prediction for the top three concentrations.
2012 and 2013 show mostly under-prediction, except at the lower end of the concentration
distributions.
With regards to individual monitor performance, monitor 401430175 (located just north
of the West Refinery in Figure C-5 and Figure C-6, appeared to have better model performance
for the 1-hour averages based on the 1-hour QQ-plots (Figure D-13a) when considering the
entire 3-year period. Monitor 401430175 under-predicted for 2011, a mix of under-prediction
and slight over-prediction for 2012 and mostly over-prediction 2013. The other two monitors
mostly over-predicted for the 3-year period and each individual year. For the 3-hour averages,
monitor 4011431127 appeared to be the better performer (Figure D-14i-l) while monitor
401430175 tended toward over-prediction at the low end of the concentrations and under-
prediction at the higher end. Monitor 401430235 mostly over-predicted. Similar trends for the
monitors are seen in the 24-hour averages (Figure D-15).
D-16
-------
Tulsa 1-hr Q-Q Plot
Observed
Tulsa 1-hr Q-Q Plot-2011
Observed
:oi 2M
Tulsa 1-hr Q-Q Plot - 2012
c
Tulsa 1-hr Q-Q Plot - 2013
d
Observed
Observed
Figure D-10. Tulsa 1-hour QQ-plots.
D-17
-------
Tulsa 3-hr Q-Q Plot
Tulsa 3-hr Q-Q Plot-2011
209
100
Observed
90 100
Observed
Tulsa 3-hr Q-Q Rot - 2012
Observed
Figure D-l I. Tulsa 3-hour QQ-plots.
Tulsa 3-hr Q-Q Rot - 2013
Observed
D-18
-------
Tulsa 24-hr Q-Q Plot
a
90
Observed
Tulsa 24-hr Q-Q Plot-2011
b
Observed
Tulsa 24-hr Q-Q Plot - 2012
«
Observed
Tulsa 24-hr Q-Q Plot-2013
Observed
Figure D-12. Tulsa 24-hour QQ-plots.
D-19
-------
Tulsa 1-hrQ-Q PV* Motto* 401430175
Tulsa 14* QOPW Monitor 401430175 2011
Tulsa WwOOPW Morator 401430175 • 2012
Tulsa T-hrOOPW Monitor 441430175 2013
Tulsa 14* Q Q Plot - Monitor 401430235
Tulsa 14* OQ PW
401430235 2011
Tulsa 14* OO PM Monitor 401430235 2012
Tulsa 14* O-O PM ¦ Monitor 401430235 2013
Tulsa 14* QO Plot Monitor 401431127 2011
Tulsa 14* Q-0 Plot ¦
401431127 - 2013
Tulsa 14* Q-0 PM Monitor 401431127
Tulsa 14* Q.0 Plot Mowtor 401431127 2012
Figure D-13. 1-hour QQ-plots for individual monitors in Tulsa, OK.
D-20
-------
Tulsa Jhr QQ Ptot • Monitor 40M3C175
Tulsa W»Q-QPW-Mondor 4014J0175 .2011
Tulsa 3*rQQ PW* Monitor 401430175 • 2012
Tulsa WirQ-Q PW Monitor 401430175 - 2013
Tulsa 3-*rQQ Ptot Monrtor 401430235
Tulsa 3h» QCPW Monitor 4014X235 ¦ 2011
Tulsa 3-#* QQ PkH Monitor 401430236 ¦ 2012
Tulsa 3*r QQ PW Monitor 401430235 • 2013
Tulsa 3-hr Q-Q Plot Monitor 401431127
Tulsa 34w QQ Plot Monitor 401431127 - 2011
Tulsa 3-hr Q-Q Plot Monitor 401431127 - 2012
Tulsa 3-hr Q-Q PM • Monitor 401431127 2013
Figure D-14. 3-hour QQ-plots for individual monitors in Tulsa, OK.
D-21
-------
Tulsa 24*r Q-Q Plot Morator 401430175
Tuisa 244v QQ Plot Morator 401430175 - 2011
Tuisa 244w Q-Q PkH Morator 401430175 • 2012
Tuisa 244w QQ Plot Morator 401430175 - 2013
Tuisa 244* Q-Q Plot Morator 401430235
Tuisa 24 tv QQ Plot Morator 401430235 2011
Tuisa 24-ftr QQ Rot Morator 401430235 2012
Tuisa 24 *w QQ Plot Morator 401430235 2013
Tuisa 244u QQ Plot Morator 401431127
Tuisa 244* Q-Q Plot Morator 401431127 2011
Tuisa 24-hr QQ Plot Morator 401431127 2012
Tuisa 244u Q-Q Plot Morator 401431127 • 2013
Figure D-15. 24-hour QQ-plots for individual monitors in Tulsa, OK.
D-22
-------
CPM values were calculated for 2011, 2012, and 2013 and the entire 3-year period (Table
D-6) across all monitors and each individual monitor. The CPM values among the individual
monitors and the CPM based on all monitors tend to be very close to one another. The model
with best agreement is 410431127 which tends to have the lower CPM with the exception of
4010432035 in 2013. Based on the CPM values, the model appears to do reasonably well against
the monitored values, with the exception of 2013, where the high CPM of 401430175 is driving
the overall CPM value across all monitors.
Table D-6. Tulsa composite performance metrics (CPM) and absolute fractional biases for
1-hour, 3-hour, and 24-hour averages.
Period
Monitor
CPM
AFBl.hr
AFBs-hr
.£
CM
OQ
Ll_
<
2011-2013
All
0.29
0.42
0.29
0.16
401430175
0.34
0.57
0.29
0.16
401432035
0.36
0.27
0.34
0.47
410431127
0.31
0.42
0.18
0.33
2011
All
0.28
0.36
0.33
0.17
401430175
0.34
0.52
0.33
0.17
401432035
0.31
0.24
0.24
0.45
410431127
0.29
0.32
0.14
0.41
2012
All
0.43
0.42
0.37
0.51
401430175
0.49
0.59
0.37
0.51
401432035
0.42
0.54
0.30
0.41
410431127
0.34
0.13
0.34
0.55
2013
All
0.72
0.63
0.84
0.68
401430175
0.83
0.97
0.84
0.68
401432035
0.33
0.42
0.18
0.36
410431127
0.37
0.50
0.37
0.24
Overall Model Performance Summary
Overall, for the three modeled areas, given uncertainties in emissions and meteorology
and temporal resolution of the emissions for many of the sources (i.e., monthly, hour-of-day,
month-hour-of-day, not individual hours), AERMOD appears to show adequate model
performance, both from a regulatory evaluation standpoint, and the narrower analysis on a
monitor-by-monitor-basis. When evaluating on an annual basis, 2011 tended to be the better
performing year, which is not surprising given that 2011 is one of the triennial emissions
inventory years. Also, as noted, given the temporal resolution of the most of the emissions, the
model performance is quite good. With some of the sources using a monthly temporal profile,
emissions for each hour for a given month would be the same (See Appendix B of this document
for an example). Given the lack of temporal variability of source emissions in the model and the
fact that a monitor does pick up temporal variability of emissions not seen by the model, the
performance of AERMOD is acceptable for the purposes of this exposure assessment.
D-23
-------
REFERENCES
Cox, W.M. and J. A. Tikvart. (1990). A statistical Procedure for Determining the Best Performing
Air Quality Simulation Model. Atmos. Environ., 24A (9): 2387-2395.
U.S. EPA. (1992). Protocol for Determining the Best Performing Model, EPA-454/R-92-025.
U.S. Environmental Protection Agency, Research Triangle Park, NC.
U.S. EPA. (2003). AERMOD: Latest Features and Evaluation Results. EPA-454/R-03-003. U.S.
Environmental Protection Agency, Research Triangle Park, NC 27711.
U.S. EPA. (2016). Evaluation of Prognostic Meteorological Data in AERMOD Applications,
EPA-454/R-16-004. U.S. Environmental Protection Agency, Research Triangle Park,
North Carolina 27711.
Venkatram, A., R. W. Brode, A. J. Cimorelli, J. T. Lee, R. J. Paine, S. G. Perry, W. D. Peters, J.
C. Weil, and R. B. Wilson. (2001). A complex terrain dispersion model for regulatory
applications. Atmos.Environ., 35, 4211-4221.
D-24
-------
APPENDIX E
ASTHMA PREVALENCE
E.l Overview
This appendix describes the development of the most recent asthma prevalence file used
by EPA's Air Pollution Exposure Model (APEX) to estimate individuals (e.g., children, adults)
having asthma. This development involved three basic steps: 1) processing National Health
Interview Survey (NHIS) asthma prevalence data, 2) processing U.S. Census poverty/income
status data, and 3) combining the two sets considering variables known to influence asthma (e.g.,
age, sex, poverty status, U.S. region) to estimate asthma prevalence stratified by age and sex for
all US Census tracts.
E.2 General History
The current processing approach is based on work originally performed by Cohen and
Rosenbaum (2005) and then revised and extended by U.S. EPA (2014). Briefly for the earlier
APEX asthma prevalence file development, Cohen and Rosenbaum (2005) calculated asthma
prevalence for children aged 0 to 17 years for each age, sex, and four U.S. regions using 2003
NHIS survey data. The regions defined by NHIS were 'Midwest', 'Northeast', 'South', and
'West'. The asthma prevalence was defined as the probability of a 'Yes' response to the question
"EVER been told that [the child] had asthma?"1 among those persons that responded either 'Yes'
or 'No' to this question.2 The responses were weighted to take into account the complex survey
design of the NHIS.3 Standard errors and confidence intervals for the prevalence were calculated
using a logistic model (PROC SURVEY LOGISTIC). A scatterplot technique (LOESS
smoother) was applied to smooth the prevalence curves and compute the standard errors and
confidence intervals for the smoothed prevalence estimates. Logistic analysis of the raw and
smoothed prevalence curves showed statistically significant differences in prevalence by gender
and region, supporting their use as stratification variables in the final data set. These smoothed
prevalence estimates were used as an input to APEX to estimate air pollutant exposure in
children with asthma (U.S. EPA 2007; 2008; 2009).
1 The response was recorded as variable "CASHMEV" in the downloaded dataset. Data and documentation are
available at http://www.cdc.gov/nchs/nhis/anest data related .1.997 forwardLfafm.
2 If there were another response to this variable other than "yes" or "no" (i.e., refused, not ascertained, don't know,
and missing), the surveyed individual was excluded from the analysis data set.
3 In the SURVEY LOGISTIC procedure, the variable "WTF SC" was used for weighting, "PSU" was used for
clustering, and "STRATUM" was used to define the stratum.
E-l
-------
In the revision documented in U.S. EPA (2014), several years of NHIS survey data
(2006-2010) were combined and used to calculate asthma prevalence for that period. Asthma
prevalence for children (by age in years) as was estimated as described above but also included
an estimate of adult asthma prevalence (by age groups). In addition, two sets of asthma
prevalence for each adults and children were estimated. The first data set, as was done
previously, was based on responses to the question "EVER been told that [the child] had
asthma". The second data set was developed using the probability of a 'Yes' response to a
question that followed those that answered 'Yes' to the first question regarding ever having
asthma, specifically, do those persons "STILL have asthma?". And finally, in addition to the
nominal variables region and sex, the asthma prevalence in this new analysis were further
stratified by a family income/poverty ratio (i.e., whether the family income was considered
below or at/above the US Census estimate of poverty level for the given year).
These updated asthma prevalence data were linked to U.S. census tract level poverty
ratios probabilities, also stratified by age. Staff considered the variability in population exposures
to be better represented when accounting for and modeling these newly refined attributes of this
susceptible population. This is because of the 1) significant observed differences in asthma
prevalence by age, sex, region, and poverty status, 2) the variability in the spatial distribution of
poverty status across census tracts, stratified by age, and 3) the potential for spatial variability in
local scale ambient concentrations.
It is in this spirit that staff update the asthma prevalence files used by APEX, using the
most recent data available that reasonably bound the exposure assessment period of interest.
Step 1: NHIS Data Set Description and Processing
The objective of this first processing step was to estimate asthma prevalence for children
and adults considering several influential variables. First, raw 2011-2015 data and associated
documentation were downloaded from the Center for Disease Control (CDC) and Prevention's
NHIS website.4 The 'Sample Child' and 'Sample Adult' files were selected because of the
availability of person-level attributes of interest within these files, i.e., age in years ('age_p'), sex
('sex'), U.S. geographic region ('region'), coupled with the response to questions of whether or
not the surveyed individual ever had and still has asthma. In total, five years of recent survey
data were obtained, comprising over 64,000 children and 170,000 children for years 2011-2015
(Table E-l).
4 See http://www.cdc.gov/nchs/tihis.htni (accessed April 11, 2017).
E-2
-------
Information regarding personal and family income and poverty ranking are also provided
by the NHIS in separate files. Five files ('INCIMPx.dat') are available for each survey year, each
containing either the actual responses (where recorded or provided by survey participant) or
imputed values for the desired financial variable.5 For this current analysis, the ratio of income to
poverty was provided as a continuous variable ('POVRATI3') and used to develop a nominal
variable for this evaluation: either the survey participant was below or above a selected poverty
threshold. This was done in this manner to be consistent with data generated as part of the second
data set processing step, i.e., a table containing census tract level poverty ratio probabilities
stratified by age (step 2).
When considering the number of stratification variables, the level of asthma prevalence,
and poverty distribution among the survey population, sample size was an important issue. For
the adult data, there were insufficient numbers of persons available to stratify the data by single
ages (for some years of age there were no survey persons). Therefore, the adult survey data were
grouped as follows: ages 18-24, 25-34, 35-44, 45-54, 55-64, 65-74, and, >75.6 To increase the
number of persons within the age, gender, and four region groupings of our characterization of
'below poverty' asthmatics persons, the poverty ratio threshold was selected as <1.5, therefore
including persons that were within 50% above the poverty threshold. If the mean of the five
imputed/recorded values were <1.5, the person's family income was categorized 'below' the
poverty threshold, if the mean of the 5 values were >1.5, the person's family income was
categorized 'above' the poverty threshold.
The person-level income files were then merged with the sample adult and child files
using the 'HHX' (a household identifier), 'FMX' (a family identifier), and 'FPX' (an individual
identifier) variables. Note, all persons within the sample adult and child files had corresponding
financial survey data.
Two asthma survey response variables were of interest in this analysis and were used to
develop the two separate prevalence data sets for each children and adults. The response to the
first question "Have you EVER been told by a doctor or other health professional that you [or
your child] had asthma?" was recorded as variable name 'CASHMEV' for children and
' AASMEV' for adults. Only persons having responses of either 'Yes' or 'No' to this question
were retained to estimate the asthma prevalence. This assumes that the exclusion of those
5 Financial information was not collected from all persons; therefore, the NHIS provides imputed data. Details into
the available variables and imputation method are provided with each year's data set. For example, see "Multiple
Imputation of Family Income and Personal Earnings in the National Health Interview Survey: Methods and
Examples" at https://www.cdc.gov/nchs/data/nhis/tecdocl5.pdf.
6 These same age groupings were used to create the companion file containing the census tract level poverty ratio
probabilities (section 2).
E-3
-------
responding otherwise, i.e., those that 'refused' to answer, instances where it was "not
ascertained', or the person 'does not know', does not affect the estimated prevalence rate if either
'Yes' or 'No' answers could actually be given by these persons. There were very few persons
providing an unusable response (Table E-l), thus the above assumption is reasonable. A second
question was asked as a follow to persons responding "Yes" to the first question, specifically,
"Do you STILL have asthma?" and noted as variables 'CASSTILL' and ' AASSTILL' for
children and adults, respectively. Again, while only persons responding 'Yes' and 'No' were
retained for further analysis, the representativeness of the screened data set is assumed
unchanged from the raw survey data given the few persons having unusable data.
Table E-l. Number of total surveyed persons from NHIS (2006-2010) sample adult and
child files and the number of those responding to asthma survey questions.
CHILDREN
2011
2012
2013
2014
2015
TOTAL
All Persons
12,844
13,275
12,860
13,380
12,281
64,640
Yes/No Asthma
12,831
13,263
12,851
13,366
12,269
64,580
Yes/No to Still Have + No Asthma
12,831
13,248
12,844
13,359
12,269
64,551
ADULTS
2011
2012
2013
2014
2015
TOTAL
All Persons
33,014
34,525
34,557
36,697
33,672
172,465
Yes/No Asthma
32,982
34,505
34,525
36,667
33,651
172,330
Yes/No to Still Have + No Asthma
32,953
34,468
34,498
36,615
33,614
172,148
Logistic Models
As described in the previous section, four person-level analytical data sets were created
from the raw NHIS data files, generally containing similar variables: a 'Yes' or 'No' asthma
response variable (either 'EVER' or 'STILL'), an age (or age group for adults), their sex ('male'
or 'female'), US geographic region ('Midwest', 'Northeast', 'South', and 'West'), and poverty
status ('below' or above'). One approach to calculate prevalence rates and their uncertainties for
a given gender, region, poverty status, and age is to calculate the proportion of 'Yes' responses
among the 'Yes' and 'No' responses for that demographic group, appropriately weighting each
response by the survey weight. This simplified approach was initially used to develop 'raw'
asthma prevalence rates however this approach may not be completely appropriate. The two
main issues with such a simplified approach are that the distributions of the estimated prevalence
rates would not be well approximated by normal distributions and that the estimated confidence
intervals based on a normal approximation would often extend outside the [0, 1] interval. A
better approach for such survey data is to use a logistic transformation and fit the model:
Prob (asthma) = exp(beta) / (1 + exp(beta)),
E-4
-------
where beta may depend on the explanatory variables for age, sex, poverty status, or region. This
is equivalent to the model:
Beta = logit {prob (asthma)} = log {prob (asthma) / [1 - prob (asthma)]}.
The distribution of the estimated values of beta is more closely approximated by a normal
distribution than the distribution of the corresponding estimates of prob (asthma). By applying a
logit transformation to the confidence intervals for beta, the corresponding confidence intervals
for prob (asthma) will always be inside [0, 1], Another advantage of the logistic modeling is that
it can be used to compare alternative statistical models, such as models where the prevalence
probability depends upon age, region, poverty status, and sex, or on age, region, poverty status
but not sex.
In previous analyses using the 2006-2010 NHIS asthma prevalence data, a variety of
logistic models and compared them for use in estimating asthma prevalence, where the
transformed probability variable beta is a given function of age, gender, poverty status, and
region (Cohen and Rosenbaum, 2005; U.S. EPA, 2014). The SAS procedure
SURVEYLOGISTIC was used to fit the various logistic models, taking into account the NHIS
survey weights and survey design (using both stratification and clustering options), as well as
considering various combinations of the selected explanatory variables.
As an example, Table E-2 lists the models fit and their log-likelihood goodness-of-fit
measures using the sample child data and for the "EVER" asthma response variable using the
2006-2010 NHIS data. A total of 32 models were fit, depending on the inclusion of selected
explanatory variables and how age was considered in the model. The 'Strata' column lists the
eight possible stratifications: no stratification, stratified by gender, by region, by poverty status,
by region and gender, by region and poverty status, by gender and poverty status, and by region,
gender and poverty status. For example, "5. region, gender" indicates that separate prevalence
estimates were made for each combination of region and gender. As another example, "2.
gender" means that separate prevalence estimates were made for each gender, so that for each
gender, the prevalence is assumed to be the same for each region. Note the prevalence estimates
are independently calculated for each stratum.
The 'Description' column of Table E-2 indicates how beta depends upon the age:
Linear in age Beta = a + (3 x age, where a and (3 vary with strata.
Quadratic in age Beta = a + (3 x age + y x age2 where a (3 and y vary with strata.
E-5
-------
Cubic in age Beta = a + (3 x age + y x age2 + 8 x age3 where a (3, y, and 8 vary
with the strata.
f(age) Beta = arbitrary function of age, with different functions for
different strata
The category f(age) is equivalent to making age one of the stratification variables, and is
also equivalent to making beta a polynomial of degree 17 in age (since the maximum age for
children is 17), with coefficients that may vary with the strata.
The fitted models are listed in order of complexity, where the simplest model (1) is a
non-stratified linear model in age and the most complex model (model 32) has a prevalence that
is an arbitrary function of age, gender, poverty status, and region. Model 32 is equivalent to
calculating independent prevalence estimates for each of the 288 combinations of age, sex,
poverty status, and region.
E-6
-------
Table E-2. Alternative logistic models for estimating child asthma prevalence using the
"EVER" asthma response variable and goodness of fit test results using the
2006-2010 NHIS data.
Model
Description
Strata
- 2 Log Likelihood
DF
1
ogit(prob) = linear in age
1.none
288740115.1
2
2
ogit(prob) = linear in age
2. gender
287062346.4
4
3
ogit(prob) = linear in age
3. region
288120804.1
8
4
ogit(prob) = linear in age
4. poverty
287385013.1
4
5
ogit(prob) = linear in age
5. region, gender
286367652.6
16
6
ogit(prob) = linear in age
6. region, poverty
286283543.6
16
7
ogit(prob) = linear in age
7. gender, poverty
285696164.7
8
8
ogit(prob) = linear in age
8. region, gender, poverty
284477928.1
32
9
2.
ogit(prob) = quadratic in age
1.none
286862135.1
3
10
2.
ogit(prob) = quadratic in age
2. gender
285098650.6
6
11
2.
ogit(prob) = quadratic in age
3. region
286207721.5
12
12
2.
ogit(prob) = quadratic in age
4. poverty
285352164
6
13
2.
ogit(prob) = quadratic in age
5. region, gender
284330346.1
24
14
2.
ogit(prob) = quadratic in age
6. region, poverty
284182547.5
24
15
2.
ogit(prob) = quadratic in age
7. gender, poverty
283587631.7
12
16
2.
ogit(prob) = quadratic in age
8. region, gender, poverty
282241318.6
48
17
3.
ogit(prob) = cubic in age
1.none
286227019.6
4
18
3.
ogit(prob) = cubic in age
2. gender
284470413
8
19
3.
ogit(prob) = cubic in age
3. region
285546716.1
16
20
3.
ogit(prob) = cubic in age
4. poverty
284688169.9
8
21
3.
ogit(prob) = cubic in age
5. region, gender
283662673.5
32
22
3.
ogit(prob) = cubic in age
6. region, poverty
283404487.5
32
23
3.
ogit(prob) = cubic in age
7. gender, poverty
282890785.3
16
24
3.
ogit(prob) = cubic in age
8. region, gender, poverty
281407414.3
64
25
4.
ogit(prob) = f(age)
1.none
285821686.2
18
26
4.
ogit(prob) = f(age)
2. gender
283843266.2
36
27
4.
ogit(prob) = f(age)
3. region
284761522.8
72
28
4.
ogit(prob) = f(age)
4. poverty
284045849.2
36
29
4.
ogit(prob) = f(age)
5. region, gender
282099156.1
144
30
4.
ogit(prob) = f(age)
6. region, poverty
281929968.5
144
31
4.
ogit(prob) = f(age)
7. gender, poverty
281963915.7
72
32
4.
ogit(prob) = f(age)
8. region, gender, poverty
278655423.1
288
E-7
-------
Table E-2 also includes the -2 Log Likelihood statistic, a goodness-of-fit measure, and
the associated degrees of freedom (DF), which is the total number of estimated parameters. Any
two models can be compared using their -2 Log Likelihood values: models having lower values
are preferred. If the first model is a special case of the second model, then the approximate
statistical significance of the first model is estimated by comparing the difference in the -2 Log
Likelihood values with a chi-squared random variable having r degrees of freedom, where r is
the difference in the DF (hence a likelihood ratio test). For all pairs of models from Table E-2, all
the differences in the -2 Log Likelihood statistic are at least 600,000 and thus significant at p-
values well below 1 percent. Based on its having the lowest -2 Log Likelihood value, the last
model fit (model 32: retaining all explanatory variables and using f(age)) was preferred and used
to estimate the asthma prevalence in the prior analyses7 as well as employed for this updated
2011-2015 NHIS data analysis.
The SURVEYLOGISTIC procedure produces estimates of the beta values and their 95%
confidence intervals for each combination of age, region, poverty status, and gender. By
applying the inverse logit transformation,
Prob (asthma) = exp( beta) / (1 + exp(beta)),
one can convert the beta values and associated 95% confidence intervals into predictions and
95%) confidence intervals for the prevalence. The standard error for the prevalence was estimated
as:
Std Error {Prob (asthma)} = Std Error (beta) x exp(- beta) / (1 + exp(beta) )2,
which follows from the delta method (i.e., a first order Taylor series approximation).
Estimated asthma prevalence using this approach and termed here as 'unsmoothed' are
provided in Attachment 1. Graphical representation is provided in a series of figures
incorporating the following variables:
• Region
• Gender
• Age (in years) or Age_group (age categories)
7 Similar results were obtained when estimating prevalence using the 'STILL' have asthma variable as well as when
investigating model fit using the adult data sets. In the Cohen and Rosenbaum (2005) analysis, adult data were not
used and the poverty to income ratio was not a variable in their models. Also, because age was a categorical
variable in the adult data sets in U.S. EPA (2014) and analyses conducted here, it could only be evaluated using
f(age_group).
E-8
-------
• Poverty Status
• Prevalence = predicted prevalence
• SE = standard error of predicted prevalence
• LowerCI = lower bound of 95 % confidence interval for predicted prevalence
• UpperCI = upper bound of 95 % confidence interval for predicted prevalence
A series of 8 plots are provided per figure that vary by region and poverty status (i.e., 4 x
2 = 8). Results for children are given in Figures 1 ('EVER' had Asthma) and 2 ('STILL' have
asthma) while adults are provided in Figures 3 ('EVER' had Asthma) and 4 ('STILL' have
asthma) within Attachment 1. Data used for each figure/plot can be provided upon request.
Loess Smoother
The estimated prevalence curves show that the prevalence is not necessarily a smooth
function of age. The linear, quadratic, and cubic functions of age modeled by
SURVEYLOGISTIC were identified as a potential method for smoothing the curves, but they
did not provide the best fit to the data. One reason for this might be due to the attempt to fit a
global regression curve to all the age groups, which means that the predictions for age A are
affected by data for very different ages. A local regression approach that separately fits a
regression curve to each age A and its neighboring ages was used, giving a regression weight of
1 to the age and lower weights to the neighboring ages using a tri-weight function:
Weight = {1 - [ |age - A| / q ]3}, where | age - A| <= q.
The parameter q defines the number of points in the neighborhood of the age A. Instead
of calling q the smoothing parameter, SAS defines the smoothing parameter as the proportion of
points in each neighborhood. A quadratic function of age to each age neighborhood was fit
separately for each gender and region combination. These local regression curves were fit to the
beta values, the logits of the asthma prevalence estimates, and then converted them back to
estimated prevalence rates by applying the inverse logit function exp(beta) / (1 + exp(beta)). In
addition to the tri-weight variable, each beta value was assigned a weight of
1 / [std error (beta)]2, to account for their uncertainties.
In this application of LOESS, weights of 1 / [std error (beta)] 2 were used such that a2 =
1. The LOESS procedure estimates a2 from the weighted sum of squares. Because it is assumed
a2 = 1, the estimated standard errors are multiplied by 1 / estimated a and adjusted the widths of
the confidence intervals by the same factor.
E-9
-------
There are several potential values that can be selected for the smoothing parameter; the
optimum value was determined by evaluating three regression diagnostics: the residual standard
error, normal probability plots, and studentized residuals. To generate these statistics, the LOESS
procedure was applied to estimated smoothed curves for beta, the logit of the prevalence, as a
function of age, separately for each region, gender, and poverty classification. For the children
data sets, curves were fit using the choices of 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, and 1.0 for the
smoothing parameter. This selected range of values was bounded using the following
observations. With only 18 points (i.e., the number of single year ages for children), a smoothing
parameter of 0.2 cannot be used because the weight function assigns zero weights to all ages
except age A, and a quadratic model cannot be uniquely fit to a single value. A smoothing
parameter of 0.3 also cannot be used because that choice assigns a neighborhood of 5 points only
(0.3 x 18 = 5, rounded down), of which the two outside ages have assigned weight zero, making
the local quadratic model fit exactly at every point except for the end points (ages 0, 1, 16 and
17). Usually one uses a smoothing parameter below 1 so that not all the data are used for the
local regression at a given x value. Note also that a smoothing parameter of 0 can be used to
generate the raw, unsmoothed, prevalence. The selection of the smoothing parameter used for the
adult curves would follow a similar logic, although the lower bound could effectively be
extended only to 0.9 given the number of age groups. This limits the selection of smoothing
parameter applied to the two adult data sets to a value of 0.9, though values of 0.8 - 1.0 were
nevertheless compared for good measure.
The first regression diagnostic used was the residual standard error, which is the LOESS
estimate of a. As discussed above, the true value of a equals 1, so the best choice of smoothing
parameter should have residual standard errors as close to 1 as possible. For children 'EVER'
having asthma and when considering the best models (of the 112 possible, those having
0.95
-------
The second regression diagnostic was developed from an approximate studentized
residual. The residual errors from the LOESS model were divided by standard error (beta) to
make their variances approximately constant. These approximately studentized residuals should
be approximately normally distributed with a mean of zero and a variance of a2 = 1. To test this
assumption, normal probability plots of the residuals were created for each smoothing parameter,
combining all the studentized residuals across genders, regions, poverty status, and ages. The
results for the children data indicate little distinction or affect by the selection of a particular
smoothing parameter (e.g., see Figure E-l), although linearity in the plotted curve is best
expressed with smoothing parameters generally between 0.6 and 0.9. When considering the adult
data sets, the appropriate value would generally be 0.9.
Probability Plot for student
0oo o
25 50 15
Normal Percentiles
90 95
99.9
Figure E-l. Normal probability plot of studentized residuals generated using logistic model,
smoothing set to 0.6, and the children 'STILL' asthmatic data set.
E-l 1
-------
The third regression di agnostic are plots of the studentized residuals against the smoothed
beta values. All the studentized residuals for a given smoothing parameter are plotted together
within the same graph. Also plotted is a LOESS smoothed curve fit to the same set of points,
with SAS's optimal smoothing parameter choice, to indicate the typical pattern. Ideally there
should be no obvious pattern and an average studentized residual close to zero with no regression
slope (e.g., see Figure E-2). For the children data sets, these plots generally indicate no unusual
patterns, and the results for smoothing parameters 0.4 through 0.6 indicate a fit LOESS curve
closest to the studentized residual equals zero line. When considering the adult data sets, 0.9 -
1.0 appear to be appropriate values.
student
5-1
4
3
Studentized residual versus smoothed logits of still prevalence rates by smoothing paiamerer-2011-2015
SmoothinaPaTameter=0 6
T 1 1 1 r-
-4,00000
+< X . O o
dp'&.xO** xo
J§ x °
O X
—1 1 1 1 1 i j 1 z 1 1 r—
1 r"
T 1 1 ! r
• 1.OO000
-1—J—I—t—I
-5 00000
-3 00000
-2 00000
Predicted loeitprev
re2£?endpov v ,J All: LOESS Smoothed
OOO Muhvest-F emal e-B elovvPovertyLev
Michvest-Male-BelowPovertyLevel
000 Northeaat-Female-BelowPovertyL
000 Northeast-Male-BelowPovertyLev
- South-Feinale Belo'.vPovertyLevel
X x x SoutF-Male-BelowPo'.-ert-1-Le'.'ei
xxx West-Feuiale-Be!owPo\ertyLe\el
;< x x West-Male-BelowFovertyLevel
000 Midwest-Female-AbovePovertyLev
OOO Midweat-Male-AbovePovertyLevel
Korthea?,t-Fenaale-Abo'."ePoverlyL
Korthe^st-Male-AbovePovertyLev
+ + + Soutli-Female-AbovePovettvLevel
South-Male-Abo veP cvertyLevel
XXX West-Female-AbovePovertyLevel
xxx West-Male-AbavePovertvLevel
Figure E-2. Studentized residuals versus model predicted betas generated using a logistic
model and using the children 'STILL' asthmatic data set, with smoothing set to 0.6.
When considering both children asthma prevalence responses evaluated, the residual
standard error (estimated values for sigma) suggests the choice of smoothing parameter as
E-12
-------
varied, ranging from 0.4 to 0.7. The normal probability plots of the studentized residuals suggest
preference for smoothing at or above 0.6. The plots of residuals against smoothed predictions
suggest the choices of 0.4 through 0.6. We therefore chose the final value of 0.6 to use for
smoothing the children's asthma prevalence. For the adults, there were small differences in the
statistical metrics used to evaluate the smoothing. A value of 0.9 was selected for smoothing,
consistent with what was used in my prior analysis (U.S. EPA, 2014).
The smoothed asthma prevalence and associated graphical presentation are provided in
Attachment 2 following a similar format to that presented in Attachment 1.
Step 2: U.S. Census Tract Poverty Ratio Data Set Description and Processing
This section briefly describes the approach used to generate census tract level poverty
ratios for all U.S. census tracts, stratified by age and age groups where available. Details
regarding the data processing is provided below in Attachment 3.8 Data used was from 2013 U.S.
Census 5-year American Community Survey (ACS).
First, ACS internal point latitudes and longitudes were obtained from the 2013 Gazetteer
files.9 Next, the individual state level ACS sequence files (SF-56) were downloaded,10 retaining
the number of persons across the variable "B17024" for each state considering the appropriate
logical record number.11 The data provided by the B17024 variable is stratified by age or age
groups (ages <5, 5, 6-11, 12-14, 15, 16-17, 18-24, 25-34, 35-44, 45-54, 55-64, 65-74, and >75)
and income/poverty ratios, given in increments of 0.25. We calculated two new variables for
each age using the number of persons from the B17024 stratifications; the fraction of those
persons having poverty ratios <1.5 and > 1.5 by summing the appropriate B17024 variable and
dividing by the total number of persons in that age/age group. Then, individual state level
geographic data ("geo" files) and their associated documentation were downloaded12 and
8 Code has been adapted from ACS 2012 SAS programs and from ACS 2012 SAS Macros available at
http://www2.census.gov/acs2012 Syr/sumitiaiyf'ile/lJserTools/SF20125YR SAS.zip and
http://www2.census.gov/acs2Q12 5yr/summaryfile/UserTools/SF All Macro.sas
9 Data set and content description is available at: http://www.census.gov/geo/maps-data/data/gazetteer2Q13.html
10 We used the summary tables (B17024), giving census tract populations by poverty income ratio and age group
downloaded from http://www2.census.gov/acs2013 5vr/summarvfile/2009-2013 ACSSF Bv State All Tables/. We
unzipped each state's ACS2013 5-yr table zip, then gathered sequence file 56.
11 Information regarding variable names is available at
https://www2.census.gov/acs2Q13 5vr/summaryfile/ACS 2013 SF_Tech_Doc.pdf. A file for the appropriate logical
record number, "Sequence_Number_and_Table_Number_Lookup.xls", can be found at
https://www2 .census. gov/acs2Q13 Syr/sunimaiyfile/.
12 Geographic data were obtained from obtained from http://www2.census.gov/acs2013 5yr/summaryfile/2009-
2013 ACSSF By State All Tables/b. Unzipped were each state's ACS2013 5-yr table ("g2013" file names).
E-13
-------
screened for tract level information using the "sumlev" variable equal to ' 140'. Also identified
was the US Region for each state, consistent with that used for the NHIS asthma prevalence
data.13
Finally, the poverty ratio data were combined with the above described census tract level
geographic data using the "stusab" and "logrecno" variables. Because APEX requires the input
data files to be complete, additional processing of the poverty probability file was needed. For
where there was missing tract level poverty information,14 we substituted an age-specific value
using the average for the particular county the tract was located within, or the state-wide average.
The percent of tracts substituted using county averaged values varied by age group though, on
average, was approximately 1.7% of the total tracts (Table E-4). Only a handful of tracts in six of
the age groups were substituted using state averaged values.
Table E-4. Percent of tracts substituted with county average or state average poverty
status.
Percent
Substituted
Age Groups
<5
6-11
12-17
18-24
25-34
35- 44
45-54
55-64
65-74
>75
all
Filled with
County Avg.
1.9
2.1
2.0
1.5
1.4
1.4
1.3
1.4
1.7
2.0
1.7
Filled with
State Avg.
0.004
0.003
0.004
0.001
0
0
0
0
0
0.001
0.001
The final output was a single file containing relevant tract level poverty probabilities
(pov_prob) by age groups for all U.S. census tracts.
Step 3: Combining Census Tract Poverty Ratios with the Asthma Prevalence Data
The two data sets were merged considering the region identifier and stratified by age and
sex. The final asthma prevalence was calculated using the following weighting scheme:
Asthma prevalence=round((povjprob *prev_belowpov)+((1-povjprob) *prev_abovepov), 0.0001);
whereas each U.S. census tract value now expresses a tract specific poverty-weighted
asthma prevalence, stratified by ages (children 0-17), age groups (adults), and two sexes. These
13 https://www2.ceiisus.gOv/s:eo/pdfs/iiiaps-data/maps/refereiice/us regdiv.pdf
14 Whether there were no data collected by the Census for the selected poverty status or whether there were simply
no persons in that age group is relatively inconsequential to estimating the asthmatic persons exposed, particularly
considering latter case as no persons in that age group would be modeled by APEX when using the same Census
population data set.
E-14
-------
final asthma prevalence data used for the assessment are found within the APEX
asthmaprevalence.txt file.
E-15
-------
REFERENCES
Cohen J and Rosenbaum A. (2005). Analysis of NHIS Asthma Prevalence Data. Memorandum
to John Langstaff by ICF Incorporated. For US EPA Work Assignment 3-08 under EPA
contract 68D01052. Available in US EPA (2007) Appendix G.
U.S. EPA. (2007). Ozone Population Exposure Analysis for Selected Urban Areas (July 2007).
Office of Air Quality Planning and Standards, Research Triangle Park, NC. EPA-452/R-
07-010. Available at: Mtfi 'Jkm.gov/ttn/naaas/staiidards/ozone/s o3 cr td.html.
U.S. EPA. (2008). Risk and Exposure Assessment to Support the Review of the NO2 Primary
National Ambient Air Quality Standard. Report no. EPA-452/R-08-008a. November
2008. Available at: http://www.epa.gov/ffti/naaas/standards/nox/dafa/20081121 NQ2 R l.pdf.
U.S. EPA. (2009). Risk and Exposure Assessment to Support the Review of the SO2 Primary
National Ambient Air Quality Standard. Report no. EPA-452/R-09-007. August 2009.
Available at:
http://www.epa.gov/ttn/naaqs/standards/so2/data/200908S02REAFinalReport.pdf.
U.S. EPA. (2014). Health Risk and Exposure Assessment for Ozone, Final Report. Chapter 5
Appendices. Report no. EPA-452/R-14-004c. August 2014. Available at
https://nepis.epa.eov/Exe/ZYPDF.cei/P100KCI7.PDF7Dockev •CCI7.PDF.
E-16
-------
Attachment 1 - Non-Smoothed Asthma Prevalence (Figures 1-4)
Figure 1 - Children (Ever Have Asthma)
Figure 1. Raw asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Above Poverty Level
Figure 1. Raw asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Below Poverty Level
10 11 12 13 14 15 16 17 0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
gender
~ Female Male
gender Female
Figure 1. Raw asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Above Poverty Level
Figure 1. Raw asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Below Poverty Level
gender
~ Female Male
gender
E-17
-------
Figure 1, cont. - Children (Ever Have Asthma)
Figure 1. Raw asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2
0.1
0.0
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 1. Raw asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2-
0.1
0.0'
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
E-18
Figure 1. Raw asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Below Poverty Level
o
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 1. Raw asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Below Poverty Level
o
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
-------
Figure 2 - Children (Still Have Asthma)
Figure 2. Raw asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2
0.1
0.0
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 2. Raw asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2
0.1
0.0
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
E-19
Figure 2. Raw asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Below Poverty Level
o
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 2. Raw asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Below Poverty Level
o
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
-------
Figure 2, cont. - Children (Still Have Asthma)
Figure 2. Raw asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Above Poverty Level
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
age
gender
~ Female Male
Figure 2. Raw asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Above Poverty Level
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
age
gender Female Male
E-20
Figure 2. Raw asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Below Poverty Level
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 2. Raw asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Below Poverty Level
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
-------
Figure
3
- Adults (Ever Have Asthma)
Figure 3. Raw adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Above Poverty Level
Figure 3. Raw adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Below Poverty Level
gender
45-54
age^grp
~ Female
75+ 18-24
gender
45-54
age^grp
~ Female
Figure 3. Raw adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Above Poverty Level
Figure 3. Raw adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Below Poverty Level
gender
45-54
age^grp
~ Female
75+ 18-24
gender
45-54
age^grp
~ Female
E-21
-------
Figure 3, cont. - Adults (Ever Have Asthma)
Figure 3. Raw adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Above Poverty Level
Figure 3. Raw adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Below Poverty Level
x T
^ i 1
r^T—^
J-— 1
gender
age^grp
~ Female Male
75+ 18-24
gender
age^grp
~ Female
Figure 3. Raw adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Above Poverty Level
Figure 3. Raw adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Below Poverty Level
35-44
45-54
65-74
gender Female
gender
E-22
-------
Figure 4 - Adults (Still Have Asthma)
Figure 4. Raw adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Above Poverty Level
Figure 4. Raw adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Below Poverty Level
75+ 18-24
gender
age^gip
~ Female
gender
age^grp
Figure 4. Raw adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Above Poverty Level
Figure 4. Raw adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Below Poverty Level
75+ 18-24
gender
age^grp
~ Female
gender
age^grp
~ Female
E-23
-------
Figure 4, cont. - Adults (Still Have Asthma)
Figure 4. Raw adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Above Poverty Level
Figure 4. Raw adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Below Poverty Level
gender Female Male
Figure 4. Raw adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Below Poverty Level
75+ 18-24
-34 35-44 45-54 55-64
age^grp
gender Female Male
Figure 4. Raw adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Above Poverty Level
35-44 45-54 55-64
age^grp
gender Female Male
75+ 18-24
gender
45-54
age^grp
~ Female
E-24
-------
Attachment 2 -Smoothed Asthma Prevalence (Figures 1-4)
Figure 1 - Children (Ever Have Asthma)
Figure 1. Smoothed asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Above Poverty Level
Figure 1. Smoothed asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Below Poverty Level
10 11 12 13 14 15 16 17
gender Female
gender Female
1. Smoothed asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Above Poverty Level
Figure 1. Smoothed asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Below Poverty Level
2 3 4 5 6 7
9 10 11 12 13 14 15 16 17 0
gender
~ Female Male
gender
E-25
-------
Figure 1, cont.
- Children (Ever Have Asthma)
Figure 1. Smoothed asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2-
0.1
0.0;
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 1. Smoothed asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2
0.1
0.0
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
E-26
Figure 1. Smoothed asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Below Poverty Level
o
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 1. Smoothed asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Below Poverty Level
Female
-------
Figure 2 - Children (Still Have Asthma)
Figure 2. Smoothed asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2-
0.1
0.0:
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 2. Smoothed asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2
0.1
0.0
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
E-27
Figure 2. Smoothed asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Below Poverty Level
o
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 2. Smoothed asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Below Poverty Level
o
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
-------
Figure 2, cont. - Children (Still Have Asthma)
Figure 2. Smoothed asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2-
0.1
0.0;
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 2. Smoothed asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Above Poverty Level
prev
0.6-
0.5
0.4
0.3
0.2-
0.1
0.0*
0
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
E-28
Figure 2. Smoothed asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Below Poverty Level
o
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
age
gender Female Male
Figure 2. Smoothed asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Below Poverty Level
Female
-------
Figure 3 - Adults (Ever Have Asthma)
Figure 3. Smoothed adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Above Poverty Level
Figure 3. Smoothed adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Below Poverty Level
75+ 18-24
gender
age^rp
gender
age^grp
Figure 3. Smoothed adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Northeastpov_rat=Above Poverty Level
Figure 3. Smoothed adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Below Poverty Level
75+ 18-24
gender
agej?rp
gender
age^grp
- Female
E-29
-------
Figure 3, cont. - Adults (Ever Have Asthma)
Figure 3. Smoothed adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Above Poverty Level
Figure 3. Smoothed adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Below Poverty Level
75+ 18-24
gender
age^rp
gender
age^grp
Figure 3. Smoothed adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Above Poverty Level
Figure 3. Smoothed adult asthma 'EVER' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Below Poverty Level
75+ 18-24
gender
age^grp
- Female
gender
agegrp
- Female
E-30
-------
Figure 4 - Adults (Still Have Asthma)
Figure 4. Smoothed adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Above Poverty Level
Figure 4. Smoothed adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Midwest pov_rat=Below Poverty Level
75+ 18-24
gender
age^gip
gender
age^grp
" Female
Figure 4. Smoothed adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Above Poverty Level
Figure 4. Smoothed adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=Northeast pov_rat=Below Poverty Level
75+ 18-24
gender
age^grp
- Female
gender
age^grp
- Female
E-31
-------
Figure 4, cont. - Adults (Still Have Asthma)
Figure 4. Smoothed adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Above Poverty Level
Figure 4. Smoothed adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=South pov_rat=Below Poverty Level
4 f"
41
75+ 18-24
gender
age^grp
~ Female Male
gender
age^grp
~ Female
Figure 4. Smoothed adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Above Poverty Level
Figure 4. Smoothed adult asthma 'STILL' prevalence rates and confidence intervals-2011-2015
region=West pov_rat=Below Poverty Level
75+ 18-24
gender
age^grp
- Female
gender
age^grp
- Female
E-32
-------
Attachment 3 - Processing Code for US Census Poverty Status Data from 2013 ACS
options m logic;
LIBNAME sas 'F:\SGRAHAM\NHIS\NHIS_1115_Process'; run; location of sas data library;
*imports ACS2013_5yr internal point Latitude and Longitude;
PROC IMPORT OUT= acs2013_5yr_tract_lat_long
DATAFILE= "F:\SGRAHAM\NHIS\NHIS_1115_Process\2013_Gaz_tracts_national.txt"
DBMS=TAB REPLACE;
GETNAMES=YES;
DATAROW=2;
RUN;
*formats a new variable GEOIDjnerge using LAT LONs GEOID in order to merge LAT and LON to geography dataset by GEOIDjnerge;
data sas.acs2013_5yr_tract_lat Jong (keep = GEOIDjnerge LAT LON);
set work.acs2013_5yr_tract_lat_long(rename=(GEOID=GEOID_char INTPTLAT=LAT INTPTLONG=LON));
length GEOID_merge $12.;
GEOIDjnerge = put(GEOID_char,Bestl2.); *STATE COUNTY and TRACT from ACS2013 Sequence File data make up GEOID in Lat Lon file;
run;
%macro Read_poverty(geo); *lmports ACS2013_5yr sequence file 56, income/poverty data (Table B17024) by state (geo);
DATA work.SFe0056&geo;
LENGTH FILEID $6
FILETYPE $6
STUSAB $2
CHARITER $3
SEQUENCE $4
LOGRECNO $7;
INFILE "F:\SGRAHAM\NHIS\NHIS_1115_Process\e20135&geo.0056000.txt" DSD TRUNCOVER DELIMITER =7 LRECL=3000;
FILEID ='File Identification'
FILETYPE='File Type'
STUSAB = 'State/U.S.-Abbreviation (USPS)'
CHARITER='Character Iteration'
SEQUENCE='Sequence Number'
LOGRECNO='Logical Record Number'
/*AGE BY RATIO OF INCOME TO POVERTY LEVEL IN THE PAST 12 MONTHS */
/*Universe: Population for whom poverty status is determined */
B17024el='Total:'
B17024e2='Under 6 years:'
B17024e3='Under .50'
B17024e4='.50to .74'
B17024e5='.75to .99'
B17024e6='1.00 to 1.24'
B17024e7='1.25to 1.49'
B17024e8='1.50 to 1.74'
B17024e9='1.75to 1.84'
B17024el0='1.85 to 1.99'
B17024ell='2.00 to 2.99'
B17024el2='3.00 to 3.99'
B17024el3='4.00 to 4.99'
B17024el4='5.00 and over'
B17024el5='6 to 11 years:'
B17024el6='Under .50'
B17024el7='.50 to .74'
B17024el8='.75to .99'
B17024el9='1.00 to 1.24'
B17024e20='1.25 to 1.49'
B17024e21='1.50 to 1.74'
B17024e22='1.75 to 1.84'
B17024e23='1.85 to 1.99'
B17024e24='2.00 to 2.99'
B17024e25='3.00 to 3.99'
B17024e26='4.00 to 4.99'
B17024e27='5.00 and over'
B17024e28='12 to 17 years:'
B17024e29='Under .50'
B17024e30='.50 to .74'
B17024e31='.75to .99'
B17024e32='1.00 to 1.24'
B17024e33='1.25 to 1.49'
B17024e34='1.50 to 1.74'
B17024e35='1.75 to 1.84'
B17024e36='1.85 to 1.99'
B17024e37='2.00 to 2.99'
B17024e38='3.00 to 3.99'
B17024e39='4.00 to 4.99'
B17024e40='5.00 and over'
B17024e41='18 to 24 years:'
E-33
-------
B17024e42=
B17024e43=
B17024e44=
B17024e45=
B17024e46=
B17024e47=
B17024e48=
B17024e49=
B17024e50=
B17024e51=
B17024e52=
B17024e53=
B17024e54=
B17024e55=
B17024e56=
B17024e57=
B17024e58=
B17024e59=
B17024e60=
B17024e61=
B17024e62=
B17024e63=
B17024e64=
B17024e65=
B17024e66=
B17024e67=
B17024e68=
B17024e69=
B17024e70=
B17024e71=
B17024e72=
B17024e73=
B17024e74=
B17024e75=
B17024e76=
B17024e77=
B17024e78=
B17024e79=
B17024e80=
B17024e81=
B17024e82=
B17024e83=
B17024e84=
B17024e85=
B17024e86=
B17024e87=
B17024e88=
B17024e89=
B17024e90=
B17024e91=
B17024e92=
B17024e93=
B17024e94=
B17024e95=
B17024e96=
B17024e97=
B17024e98=
B17024e99=
B17024el00=
B17024el01=
B17024el02=
B17024el03=
B17024el04=
B17024el05=
B17024el06=
B17024el07=
B17024el08=
B17024el09=
B17024ell0=
B17024elll=
B17024ell2=
B17024ell3=
B17024ell4=
B17024ell5=
B17024ell6=
B17024ell7=
B17024ell8=
B17024ell9=
B17024el20=
B17024el21=
B17024el22=
B17024el23=
B17024el24=
B17024el25=
B17024el26=
B17024el27=
B17024el28=
B17024el29=
'Under .50'
'.50 to .74'
'.75 to .99'
'1.00 to 1.24'
'1.25 to 1.49'
'1.50 to 1.74'
'1.75 to 1.84'
'1.85 to 1.99'
'2.00 to 2.99'
'3.00 to 3.99'
'4.00 to 4.99'
'5.00 and over'
'25 to 34 years:'
'Under .50'
'.50 to .74'
'.75 to .99'
'1.00 to 1.24'
'1.25 to 1.49'
'1.50 to 1.74'
'1.75 to 1.84'
'1.85 to 1.99'
'2.00 to 2.99'
'3.00 to 3.99'
'4.00 to 4.99'
'5.00 and over'
'35 to 44 years:'
'Under .50'
'.50 to .74'
'.75 to .99'
'1.00 to 1.24'
'1.25 to 1.49'
'1.50 to 1.74'
'1.75 to 1.84'
'1.85 to 1.99'
'2.00 to 2.99'
'3.00 to 3.99'
'4.00 to 4.99'
'5.00 and over'
'45 to 54 years:'
'Under .50'
'.50 to .74'
'.75 to .99'
'1.00 to 1.24'
'1.25 to 1.49'
'1.50 to 1.74'
'1.75 to 1.84'
'1.85 to 1.99'
'2.00 to 2.99'
'3.00 to 3.99'
'4.00 to 4.99'
'5.00 and over'
'55 to 64 years:'
'Under .50'
'.50 to .74'
'.75 to .99'
'1.00 to 1.24'
'1.25 to 1.49'
'1.50 to 1.74'
'1.75 to 1.84'
'1.85 to 1.99'
'2.00 to 2.99'
'3.00 to 3.99'
'4.00 to 4.99'
'5.00 and over'
'65 to 74 years:'
'Under .50'
'.50 to .74'
'.75 to .99'
'1.00 to 1.24'
'1.25 to 1.49'
'1.50 to 1.74'
'1.75 to 1.84'
'1.85 to 1.99'
'2.00 to 2.99'
'3.00 to 3.99'
'4.00 to 4.99'
'5.00 and over'
'75 years and ove
'Under .50'
'.50 to .74'
'.75 to .99'
'1.00 to 1.24'
'1.25 to 1.49'
'1.50 to 1.74'
'1.75 to 1.84'
'1.85 to 1.99'
'2.00 to 2.99'
'3.00 to 3.99'
-------
B17024el30='4.00 to 4.99'
B17024el31='5.00 and over1
RUN;
%mend;
FILEID $
FILETYPE $
STUSAB $
CHARITER $
SEQUENCE $
LOGRECNO $
B17024el-B17024el31
if B17024el >=0;
%macro AnyGeo(geo); *lmports geo data file, assigns a census region, limits to 2013ACS_5yr census tracts by state ('geo'), assigns lat Ion;
data work.g20135&geo (drop = AIANHH AIANHHFP AIHHTLI AITS
CDCURR CNECTA
DIVISION
PLACE
SUBMCD
ZCTA3
FILEID MAC<
PUMA1
SUM LEVEL TAZ
ZCTA5
);
MEMI
PUMA5
METDIV
REGION
UA
NAME
SDELM
SDSEC
UACP
AITSCE
COMPONENT CONCIT
NECTA
SDUNI
SLDL
UGA
US
BLKGRP
COUSUB
PCI
CBSA
CSA
SLDU
UR
/*Location of geo data file for import*/
INFILE "F:\SGRAHAM\NHIS\NHIS_H15_Process\g20135&geo..txt" MISSOVERTRUNCOVER LRECL=500; /*change directory*/
LABEL FILEID ='File Identification'
STUSAB ='State Postal Abbreviation'
INPUT
FILEID
$1-6
SUMLEVEL='Summary Level'
LOGRECNO='Logical Record Number'
REGION ='Region'
STATECE ='State (Census Code)'
COUNTY ='County'
PLACE ='Place (FIPS Code)'
BLKGRP ='Block Group'
CSA ='Combined Statistical Area'
UA ='Urban Area'
VTD ='Voting District'
SUBMCD ='Subbarrio (FIPS)'
SDSEC ='School District (Secondary)'
UR ='Urban/Rural'
TAZ ='Traffic Analysis Zone'
GEOID ='geographic Identifier'
COMPONENT='geographic Component'
US ='US'
DIVISION ='Division'
STATE ='State (FIPS Code)'
COUSUB ='County Subdivision (FIPS)'
TRACT ='Census Tract'
CONCIT ='Consolidated City'
METDIV ='Metropolitan Division'
UACP ='Urban Area Central Place'
ZCTA3 ='ZIP Code Tabulation Area (3-digit)'
SDELM ='School District (Elementary)'
SDUNI ='School District (Unified)'
PCI ='Principal City Indicator'
UGA ='Urban Growth Area'
NAME -Area Name'
AIANHH ='American Indian Area/Alaska Native Area/Hawaiian Home Land (Census)'
AIANHHFP='American Indian Area/Alaska Native Area/Hawaiian Home Land (FIPS)'
AIHHTLI ='American Indian Trust Land/Hawaiian Home Land Indicator'
AITSCE ='American Indian Tribal Subdivision (Census)'
AITS ='American Indian Tribal Subdivision (FIPS)'
ANRC ='Alaska Native Regional Corporation (FIPS)'
CBSA ='Metropolitan and Micropolitan Statistical Area'
MACC ='Metropolitan Area Central City'
MEMI ='Metropolitan/Micropolitan Indicator Flag'
NECTA ='New England City and Town Combined Statistical Area'
CNECTA ='New England City and Town Area'
NECTADIV='New England City and Town Area Division'
CDCURR ='Current Congressional District'
SLDU ='State Legislative District Upper'
SLDL ='State Legislative District Lower'
ZCTA5 ='ZIP Code Tabulation Area (5-digit)'
PUMA5 ='Public Use Microdata Area - 5% File'
PUMA1 ='Public Use Microdata Area -1% File'
STUSAB $ 7-8
COMPONENT $
REGION $ 22-22
STATE $ 26-27
PLACE $ 36-40
CONCIT $48-52
AIHHTLI $62-62
ANRC $ 71-75
METDIV $84-88
NECTA $ 91-95
UA $ 104-108
SLDU $ 116-118
ZCTA3 $ 128-130
LOGRECNO $ 14-20
DIVISION $23-23
COUNTY $28-30
TRACT $ 41-46
AIANHH $ 53-56
AITSCE $ 63-65
CBSA $ 76-80
MACC $ 89-89
SUMLEVEL $ 9-11
US $ 21-21
STATECE $24-25
COUSUB $ 31-35
BLKGRP $47-47
AIANHHFP $57-61
AITS $ 66-70
CSA $ 81-83
CNECTA $96-98
UACP $ 109-113
SLDL $ 119-121
ZCTA5 $ 131-135
MEMI $90-90
NECTADIV $ 99-103
CDCURR $114-115
VTD $ 122-127
SUBMCD $ 136-140
E-35
-------
SDELM $ 141-145 SDSEC $ 146-150 SDUNI $ 151-155
UR $ 156-156 PCI $ 157-157 TAZ $ 158-163
UGA $ 164-168 PUMA5 $ 169-173 PUMA1 $ 174-178
GEOID $ 179-218 /* GEOID is 40 char in length */
NAME $ 219-418
IF sumlevel='140'; *imports data for tracts only, similar to WHERE tract IS NOT NULL;
run;
data work.g20135&geo (keep = STUSAB CENSUS_REGION LOGRECNO GEOID_merge STATE COUNTY TRACT);
set work.g20135&geo;
length CENSUS_REGION $12.;
if STUSAB = 'CT' OR STUSAB = 'ME' OR STUSAB = 'MA' OR STUSAB = 'NH' OR STUSAB = "Rl"
OR STUSAB = VP OR STUSAB = W OR STUSAB = 'NY' OR STUSAB = 'PA'
then do;
CENSUS_REGION = 'Northeast'; *assign census region-
end;
else if STUSAB = 'IN' OR STUSAB = ML' OR STUSAB = 'Ml' OR STUSAB = 'OH' OR STUSAB = 'Wl'
OR STUSAB = 'IA' OR STUSAB = 'KS' OR STUSAB = 'MN' OR STUSAB = 'MO' OR STUSAB = 'NE'
OR STUSAB = 'ND' OR STUSAB = 'SD'
then do;
CENSUS_REGION = 'Midwest-
end;
else if STUSAB = 'DE' OR STUSAB = 'DC' OR STUSAB = 'FL' OR STUSAB = 'GA' OR STUSAB = 'MD'
OR STUSAB = 'NC' OR STUSAB = 'SC' OR STUSAB = 'VA' OR STUSAB = 'WV' OR STUSAB = 'AL'
OR STUSAB = 'KY' OR STUSAB = 'MS' OR STUSAB = 'TN' OR STUSAB = 'AR' OR STUSAB = 'LA'
OR STUSAB = 'OK' OR STUSAB = 'TX'
then do;
CENSUS_REGION = 'South';
end;
else if STUSAB = 'AZ' OR STUSAB = 'CO' OR STUSAB = 'ID' OR STUSAB = 'NM' OR STUSAB = 'MT'
OR STUSAB = 'UT' OR STUSAB = 'NV' OR STUSAB = 'WY' OR STUSAB = 'AK' OR STUSAB = 'CA'
OR STUSAB = 'HI' OR STUSAB = 'OR' OR STUSAB = 'WA'
then do;
CENSUS_REGION = 'West';
end;
else CENSUS_REGION = 'Other';
where tract ne ";*limit to 2013ACS_5yr census tracts only;
length GEOID_char $12.;
GEOID_char = CATS (STAT E,COUNTY,TRACT); *format GEOID_merge to match LAT LONs GEOID_merge;
GEOID_merge = put(input(GEOID_char,12.),12.);
run;
proc sort data=sas.Acs2013_5yr_tract_lat_long;
by GEOID_merge;
run;
proc sort data=work.g20135&geo;
by GEOID_merge;
run;
data work.g20135&geo.coord (keep = STUSAB CENSUS_REGION LOGRECNO STATE COUNTY TRACT GEOID_merge LAT LON); *adds internal point lat Ion;
merge work.g20135&geo(in=a) sas.Acs2013_5yr_tract_lat_long;
by GEOID_merge;
if a;
run;
%mend;
%macro pov_ratio_caIc(geo);*calculates ratios above or below 1.5 income/poverty ratio by age group by tract. *f ills tracts with 0 persons in an age class with the county-level ratio;
proc means data=work.SFe^g_0056&geo noprint;*creates a sum by county of each census poverty/in come variable (for the entire county);
class counts-
output out = work.pov_ratio_county_sum_&geo
sum = CountySum_B17024el-CountySum_B17024el31
run;
proc sort data =work.pov_ratio_county_sum_&geo;
by county;
run;
proc sort data =work.SFe_g_0056&geo;
by county;
run;
data work.SFe^g_filled_co_0056&geo (drop = _TYPE FREO^);
merge work.SFe_g_0056&geo (in=a) work.pov_ratio_county_sum_&geo;
by county;
if a;
run;
proc means data=work.SFe^g_0056&geo noprint;*creates a sum by state of each census poverty/income variable (for the entire stated-
class state;
output out = work.pov_ratio_state_sum_&geo
sum = StateSum_B17024el-StateSum_B17024el31
E-36
-------
proc sort data =work.pov_ratio_state_sum_&geo;
by state;
run;
proc sort data =work.SFe_g_filled_co_0056&geo;
by state;
run;
data work.SFe^g_filled_st_co_0056&geo (drop = _TYPE FREQJ;
merge work.SFe^g_filled_co_0056&geo (in=a) work.pov_ratio_state_sum_&geo;
by state;
if a;
data work.pov_pct_&geo;
set work.SFe_g_filled_st_co_0056&geo;
length filled_e2 $26 filled_el5 $26 filled_e28 $26 filled_e41 $26 filled_e54 $26 filled_e67 $26 filled_eSO $26 filled_e93 $26
filled_el06 $26 filled_ell9 $26;
IF B17024e2 A= Othen do;
*where age group population in a tract is not equal to zero, calculate below/above poverty ratio based on income/poverty variables for the tract using tract-level data;
filled_e2 = 'Tract Values Used';
pctB17024e3=B17024e3/B17024e2;
pctB17024e4=B17024e4/B17024e2;
pctB17024e5=B17024e5/B17024e2;
pctB17024e6=B17024e6/B17024e2;
pctB17024e7=B17024e7/B17024e2;
pctB17024e8=B17024e8/B17024e2;
pctB17024e9=B17024e9/B17024e2;
pctB17024el0=B17024el0/B17024e2;
pet B17024e11=B17024e11/B17024e2;
pctB17024el2=B17024el2/B17024e2;
pctB17024el3=B17024el3/B17024e2;
pctB17024el4=B17024el4/B17024e2;end;
ELSE IF CountySum_B17024e2 A= Othen do;
*where age group population in a tract is zero, but the county is not equal to zero, calculate below/above poverty ratio based on income/poverty variables using county-level
data;
filled_e2 = 'Filled with County Values';
pctB17024e3=CountySum_B17024e3/CountySum_B17024e2;
pet B17024e4=C o u nty Su m_B17024e 4/C o u nty Su m _B 17024e2;
pet B17024e 5=C o u nty Su m_B17024e 5/C o u nty Su m _B 17024e2;
pet B17024e6=C o u nty Su m_B17024e 6/C o u nty Su m _B 17024e2;
pctB17024e7=CountySum_B17024e7/CountySum_B17024e2;
pctB17024e8=CountySum_B17024e8/CountySum_B17024e2;
pctB17024e9=CountySum_B17024e9/CountySum_B17024e2;
pet B17024e 10=Co u ntyS u m_B 17024e 10/C o u ntyS u m_B17024e 2;
pctB17024ell=CountySum_B17024ell/CountySum_B17024e2;
pctB17024el2=CountySum_B17024el2/CountySum_B17024e2;
pctB17024el3=CountySum_B17024el3/CountySum_B17024e2;
pctB17024el4=CountySum_B17024el4/CountySum_B17024e2;end;
ELSE IF CountySum_B17024e2 = Othen do;
* where age group population in a county and tract are both zero, calculate below/above poverty ratio based on income/poverty variables using state-level data for children 17
and under;
filled_e2 = 'Filled with State Values';
pet B17024e3=s u m (Stat eS u m_B 17024e 3,Stat eS u m_B17024e 16 ,Stat eS u m_B 17024e2 9 )/s u m (Stat eS u m_B 17024e 2 ,St ate Su m_B17024e 15,Stat eS u m_B 17024e 28);
pet B17024e4=s u m (St at eS u m_B 17024e 4,Stat eS u m_B17024e 17 ,Stat eS u m_B 17024e3 0 )/s u m (Stat eS u m_B 17024e2 ,St ate Su m_B17024e 15,Stat eS u m_B 17024e 28);
pctB17024e5=sum (Stat eSum_B17024e5,StateSum_B17024el8,StateSum_B17024e31)/sum (Stat eSum_B 17024e2,StateSum_B17024el5,StateSum_B17024e28);
pet B17024e6=s u m (Stat eS u m_B 17024e 6,Stat eS u m_B17024e 19 ,Stat eS u m_B 17024e3 2 )/s u m (Stat eS u m_B 17024e 2 ,St ate Su m_B17024e 15,Stat eS u m_B 17024e 28);
pet B17024e7=s u m (Stat eS u m_B 17024e 7,Stat eS u m_B17024e 20,Stat eS u m_B 17024e3 3 )/s u m (Stat eS u m_B 17024e 2 ,St ate Su m_B 17024e 15,Stat eS u m_B 17024e 28);
pet B17024e8=s u m (Stat eS u m_B 17024e 8,Stat eS u m_B17024e 21 ,Stat eS u m_B 17024e3 4 )/s u m (Stat eS u m_B 17024e 2 ,St ate Su m_B17024e 15,Stat eS u m_B 17024e 28);
pctB17024e9=sum(StateSum_B17024e9,StateSum_B17024e22,StateSum_B17024e35)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024el0=sum(StateSum_B17024el0,StateSum_B17024e23,StateSum_B17024e36)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024ell=sum(StateSum_B17024ell,StateSum_B17024e24,StateSum_B17024e37)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024el2=sum(StateSum_B17024el2,StateSum_B17024e25,StateSum_B17024e38)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024el3=sum(StateSum_B17024el3,StateSum_B17024e26,StateSum_B17024e39)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024el4=sum(StateSum_B17024el4,StateSum_B17024e27,StateSum_B17024e40)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);end;
IF B17024el5 A= Othen do;
filled_el5 = 'Tract Values Used';
pctB17024el6=B17024el6/B17024el5;
pctB17024el7=B17024el7/B17024el5;
pctB17024el8=B17024el8/B17024el5;
pctB17024el9=B17024el9/B17024el5;
pctB17024e20=B17024e20/B17024el5;
pctB17024e21=B17024e21/B17024el5;
pctB17024e22=B17024e22/B17024el5;
pctB17024e23=B17024e23/B17024el5;
pctB17024e24=B17024e24/B17024el5;
pctB17024e25=B17024e25/B17024el5;
pctB17024e26=B17024e26/B17024el5;
pctB17024e27=B17024e27/B17024el5;end;
ELSE IF CountySum_B17024el5 A= 0 then do;
E-37
-------
filled_el5= 'Filled with County Va I lies';
pet B17024e 16=Co u ntyS u m_B 17024e 16/C o u ntyS u m_B17024e 15;
pet B17024e 17=Co u ntyS u m_B 17024e 17/C o u ntyS u m_B17024e 15;
pctB17024el8=CountySum_B17024el8/CountySum_B17024el5;
pet B17024e 19=Co u ntyS u m_B 17024e 19/Co u nty Su m_B 17024e 15;
pctB17024e20=CountySum_B17024e20/CountySum_B17024el5;
pet B17024e21=Co u ntyS u m_B 17024e 21/C o u ntyS u m_B17024e 15;
pet B17024e2 2=Co u ntyS u m_B 17024e 22/C o u ntyS u m_B17024e 15;
pet B17024e2 3=Co u ntyS u m_B 17024e 23/C o u ntyS u m_B17024e 15;
pctB17024e24=Cou ntyS u m_B 17024e 24/C o u ntyS u m_B17024e 15;
pctB17024e25=CountySum_B17024e25/CountySum_B17024el5;
pet B17024e2 6=Co u ntyS u m_B 17024e 26/C o u ntyS u m_B17024e 15;
pctB17024e27=CountySum_B17024e27/CountySum_B17024el5;end;
ELSE IF CountySum_B17024el5 = 0 then do;
filled_el5 = 'Filled with State Values';
pctB17024el6=sum(StateSum_B17024e3,StateSum_B17024el6,StateSum_B17024e29)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B 17024e28)
pctB17024el7=sum (Stat eSum_B17024e4,StateSum_B17024el7,StateSum_B17024e30)/sum (Stat eSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28)
pctB17024el8=sum(StateSum_B17024e5,StateSum_B17024el8,StateSum_B17024e31)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B 17024e28)
pctB17024el9=sum(StateSum_B17024e6,StateSum_B17024el9,StateSum_B17024e32)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28)
pctB17024e20=sum(StateSum_B17024e7,StateSum_B17024e20,StateSum_B17024e33)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B 17024e28)
pctB17024e21=sum(StateSum_B17024e8,StateSum_B17024e21,StateSum_B17024e34)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28)
pctB17024e22=sum(StateSum_B17024e9,StateSum_B17024e22,StateSum_B17024e35)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B 17024e28)
pctB17024e23=sum(StateSum_B17024el0,StateSum_B17024e23,StateSum_B17024e36)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024e24=sum(StateSum_B17024ell,StateSum_B17024e24,StateSum_B17024e37)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024e25=sum(StateSum_B17024el2,StateSum_B17024e25,StateSum_B17024e38)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024e26=sum(StateSum_B17024el3,StateSum_B17024e26,StateSum_B17024e39)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024e27=sum(StateSum_B17024el4,StateSum_B17024e27,StateSum_B17024e40)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);end;
IF B17024e28 A=0thendo;
filled_e28 = 'Tract Values Used';
pctB17024e29=B17024e29/B17024e28;
pctB17024e30=B17024e30/B17024e28;
pctB17024e31=B17024e31/B17024e28;
pctB17024e32=B17024e32/B17024e28;
pctB17024e33=B17024e33/B17024e28;
pctB17024e34=B17024e34/B17024e28;
pctB17024e35=B17024e35/B17024e28;
pctB17024e36=B17024e36/B17024e28;
pctB17024e37=B17024e37/B17024e28;
pctB17024e38=B17024e38/B17024e28;
pctB17024e39=B17024e39/B17024e28;
pctB17024e40=B17024e40/B17024e28;end;
ELSE IF CountySum_B17024e28 A= Othen do;
filled_e28= 'Filled with County Values';
pctB17024e29=CountySum_B17024e29/CountySum_B17024e28;
pctB17024e30=CountySum_B17024e30/CountySum_B17024e28;
pctB17024e31=CountySum_B17024e31/CountySum_B17024e28;
pctB17024e32=CountySum_B17024e32/CountySum_B17024e28;
pctB17024e33=CountySum_B17024e33/CountySum_B17024e28;
pctB17024e34=CountySum_B17024e34/CountySum_B17024e28;
pctB17024e35=CountySum_B17024e35/CountySum_B17024e28;
pet B17024e3 6=Co u ntyS u m_B 17024e 36/C o u ntyS u m_B17024e 28;
pctB17024e37=CountySum_B17024e37/CountySum_B17024e28;
pctB17024e38=CountySum_B17024e38/CountySum_B17024e28;
pctB17024e39=CountySum_B17024e39/CountySum_B17024e28;
pctB17024e40=CountySum_B17024e40/CountySum_B17024e28;end;
ELSE IF CountySum_B17024e28 = 0 then do;
filled_e28 = 'Filled with State Values';
pctB17024e29=sum(StateSum_B17024e3,StateSum_B17024el6,StateSum_B17024e29)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B 17024e28)
pctB17024e30=sum(StateSum_B17024e4,StateSum_B17024el7,StateSum_B17024e30)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28)
pctB17024e31=sum(StateSum_B17024e5,StateSum_B17024el8,StateSum_B17024e31)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B 17024e28)
pctB17024e32=sum (Stat eSum_B17024e6,StateSum_B17024el9,StateSum_B17024e32)/sum(StateSum_B17024e2,Stat eSum_B17024el5,StateSum_B17024e28)
pctB17024e33=sum(StateSum_B17024e7,StateSum_B17024e20,StateSum_B17024e33)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B 17024e28)
pctB17024e34=sum (Stat eSum_B17024e8,StateSum_B17024e21,StateSum_B17024e34)/sum (Stat eSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28)
pctB17024e35=sum(StateSum_B17024e9,StateSum_B17024e22,StateSum_B17024e35)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B 17024e28)
pctB17024e36=sum(StateSum_B17024el0,StateSum_B17024e23,StateSum_B17024e36)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024e37=sum(StateSum_B17024ell,StateSum_B17024e24,StateSum_B17024e37)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024e38=sum(StateSum_B17024el2,StateSum_B17024e25,StateSum_B17024e38)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024e39=sum(StateSum_B17024el3,StateSum_B17024e26,StateSum_B17024e39)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);
pctB17024e40=sum(StateSum_B17024el4,StateSum_B17024e27,StateSum_B17024e40)/sum(StateSum_B17024e2,StateSum_B17024el5,StateSum_B17024e28);end;
IF B17024e41 A= Othen do;
filled_e41 = 'Tract Values Used';
pctB17024e42=B17024e42/B17024e41;
pctB17024e43=B17024e43/B17024e41;
pctB17024e44=B17024e44/B17024e41;
pctB17024e45=B17024e45/B17024e41;
pctB17024e46=B17024e46/B17024e41;
pctB17024e47=B17024e47/B17024e41;
pctB17024e48=B17024e48/B17024e41;
E-38
-------
pctB17024e49=B17024e49/B17024e41;
pet B17024e 50=B17024e 50/ B17 024e 41;
pet B17024e51=B17024e 51/B17024e41;
pctB17024e52=B17024e 52/B17024e41;
pctB 17024e53=B 17024e 53/B 17024e41;e nd;
ELSE IF CountySum_B17024e41 A= Othen do;
filled_e41 = 'Filled with County Va I lies';
pet B17024e42=Co u ntyS u m_B 17024e 42/C o u ntyS u m_B17024e41;
pctB17024e43=CountySum_B17024e43/CountySum_B17024e41;
pctB17024e44=CountySum_B17024e44/CountySum_B17024e41;
pctB17024e45=CountySum_B17024e45/CountySum_B17024e41;
pet B17024e46=Co u ntyS u m_B 17024e 46/C o u ntyS u m_B17024e 41;
pctB17024e47=CountySum_B17024e47/CountySu m_B17024e41;
pctB17024e48=CountySum_B17024e48/CountySum_B17024e41;
pctB17024e49=CountySum_B17024e49/CountySum_B17024e41;
pet B17024e 50=Co u ntyS u m_B 17024e 50/C o u ntyS u m_B17024e 41;
pctB17024e51=CountySum_B17024e51/CountySum_B17024e41;
pctB17024e52=CountySum_B17024e52/CountySum_B17024e41;
pet B17024e 53=Co u ntyS u m_B 17024e 53/C o u ntyS u m_B17024e 41; e n d;
ELSE IF CountySum_B17024e41 = 0 then do;
* where age group population in a county and tract are both zero, calculate below/above poverty ratio based on income/poverty variables using state-level data for adults 18
and over;
filled_e41 = 'Filled with State Values';
pctB17024e42=sum (Stat eSum_B17024e42,StateSum_B17024e55,StateSum_B17024e68,StateSum_B 17024e81,StateSum_B17024e94,StateSum_B17024el07,StateSum_B 17024
el20)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e43=sum (Stat eSum_B17024e43,StateSum_B17024e56,StateSum_B17024e69,StateSum_B17024e82,StateSum_B17024e95,StateSum_B 17024el08,StateSum_B17024
el21)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e44=sum(StateSum_B17024e44,StateSum_B17024e57,StateSum_B17024e70,StateSum_B17024e83,StateSum_B17024e96,StateSum_B17024el09,StateSum_B17024
el22)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e45=sum (Stat eSum_B17024e45,StateSum_B17024e58,StateSum_B17024e71,StateSum_B17024e84,StateSum_B17024e97,StateSum_B17024ell0,StateSum_B 17024
el23)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e46=sum (Stat eSum_B17024e46,StateSum_B17024e59,StateSum_B17024e72,StateSum_B 17024e85,StateSum_B17024e98,StateSum_B17024elll,StateSum_B 17024
el24)/sum (StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B 17024e 119);
pctB17024e47=sum (Stat eSum_B17024e47,StateSum_B17024e60,StateSum_B17024e73,StateSum_B 17024e86,StateSum_B17024e99,StateSum_B17024ell2,StateSum_B 17024
el25)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e48=sum (Stat eSum_B17024e48,StateSum_B17024e61,StateSum_B17024e74,StateSum_B 17024e87,StateSum_B17024el00,StateSum_B17 024ell3,StateSum_B 1702
4el26)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e49=sum (Stat eSum_B17024e49,StateSum_B17024e62,StateSum_B17024e75,StateSum_B 17024e88,StateSum_B17024el01,StateSum_B17024ell4,StateSum_B 1702
4el27)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e50=sum(StateSum_B17024e50,StateSum_B17024e63,StateSum_B17024e76,StateSum_B17024e89,StateSum_B17024el02,StateSum_B17024ell5,StateSum_B1702
4el28)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e51=sum(StateSum_B17024e51,StateSum_B17024e64,StateSum_B17024e77,StateSum_B17024e90,StateSum_B17024el03,StateSum_B17024ell6,StateSum_B1702
4el29)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e52=sum (Stat eSum_B17024e52,StateSum_B17024e65,StateSum_B17024e78,StateSum_B 17024e91,StateSum_B17024el04,StateSum_B17 024ell7,StateSum_B 1702
4el30)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e53=sum (Stat eSum_B17024e53,StateSum_B17024e66,StateSum_B17024e79,StateSum_B 17024e92,StateSum_B17024el05,StateSum_B17 024ell8,StateSum_B 1702
4el31)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);end;
IF B17024e54 A= Othen do;
filled_e54 = 'Tract Values Used';
pctB17024e55=B17024e55/B17024e54;
pctB17024e56=B17024e56/B17024e54;
pctB17024e57=B17024e57/B17024e54;
pctB17024e58=B17024e58/B17024e54;
pctB17024e59=B17024e59/B17024e54;
pctB17024e60=B17024e60/B17024e54;
pctB17024e61=B17024e61/B17024e54;
pctB17024e62=B17024e62/B17024e54;
pctB17024e63=B17024e63/B17024e54;
pctB17024e64=B17024e64/B17024e54;
pctB 17024e65=B 17024e65/B 17024e54;
pctB 17024e66=B 17024e66/B17024e54;end;
ELSE IF CountySum_B17024e54 A= Othen do;
filled_e54= 'Filled with County Va I ues';
pctB17024e55=CountySum_B17024e55/CountySum_B17024e54;
pctB17024e56=Cou ntyS u m_B 17024e 56/C o u ntyS u m_B17024e 54;
pctB17024e57=CountySum_B17024e57/CountySum_B17024e54;
pctB17024e58=CountySum_B17024e58/CountySum_B17024e54;
pctB17024e59=CountySum_B17024e59/CountySum_B17024e54;
pctB17024e60=CountySum_B17024e60/CountySum_B17024e54;
pctB17024e61=CountySum_B17024e61/CountySum_B17024e54;
pctB17024e62=CountySum_B17024e62/CountySum_B17024e54;
pctB17024e63=CountySum_B17024e63/CountySum_B17024e54;
pctB17024e64=CountySum_B17024e64/CountySum_B17024e54;
pctB17024e65=CountySum_B17024e65/CountySum_B17024e54;
pctB17024e66=CountySum_B17024e66/CountySum_B17024e54;end;
ELSE IF CountySum_B17024e54 = 0 then do;
E-39
-------
filled_e54 = 'Filled with State Values';
pctB17024e55=sum (Stat eSum_B17024e42,StateSum_B17024e55,StateSum_B17024e68,StateSum_B17024e81,StateSum_B 17024e94,StateSum_B 17024el07,StateSum_B17024
el20)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e56=sum (Stat eSum_B 1702 4e43,StateSum_B17024e56,StateSum_B17024e69,StateSum_B17024e82,StateSum_B17024e95,StateSum_B 17024el08,StateSum_B 17024
el21)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e57=sum (Stat eSum_B17024e44,StateSum_B17024e57,StateSum_B17024e70,StateSum_B 17024e83,StateSum_B17024e96,StateSum_B17024el09,StateSum_B 17024
el22)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e58=sum (Stat eSum_B17024e45,StateSum_B17024e58,StateSum_B17024e71,StateSum_B 17024e84,StateSum_B17024e97,StateSum_B17024ell0,StateSum_B 17024
el23)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B 17024e 119);
pctB17024e59=sum (Stat eSum_B17024e46,StateSum_B17024e59,StateSum_B17024e72,StateSum_B 17024e85,StateSum_B17024e98,StateSum_B17024elll,StateSum_B 17024
el24)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B 17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024e60=sum (Stat eSum_B17024e47,StateSum_B17024e60,StateSum_B17024e73,StateSum_B 17024e86,StateSum_B17024e99,StateSum_B17024ell2,StateSum_B 17024
el25)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e61=sum (Stat eSum_B17024e48,StateSum_B17024e61,StateSum_B17024e74,StateSum_B17024e87,StateSum_B17024el00,StateSum_B 17024ell3,StateSum_B1702
4e 126 )/s u m (St ateS u m_B 17024e41 ,St ate Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e80,St at eS u m_B 17024e 93 ,St ate Su m_B17024e 106 ,Stat eS u m_B 17024e 119);
pctB17024e62=sum (Stat eSum_B17024e49,StateSum_B17024e62,StateSum_B17024e75,StateSum_B17024e88,StateSum_B17024el01,StateSum_B17024ell4,StateSum_B 1702
4el27)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e63=sum (Stat eSum_B17024e50,StateSum_B17024e63,StateSum_B17024e76,StateSum_B17024e89,StateSum_B17024el02,StateSum_B17024ell5,StateSum_B 1702
4el28)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e64=sum (Stat eSum_B17024e51,StateSum_B17024e64,StateSum_B17024e77,StateSum_B 17024e90,StateSum_B17024el03,StateSum_B17 024ell6,StateSum_B 1702
4el29)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e65=sum (Stat eSum_B17024e52,StateSum_B17024e65,StateSum_B17024e78,StateSum_B 17024e91,StateSum_B17024el04,StateSum_B17 024ell7,StateSum_B 1702
4el30)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSu m_B 17024el06,StateSum_B17024ell9);
pctB17024e66=sum (Stat eSum_B17024e53,StateSum_B17024e66,StateSum_B17024e79,StateSum_B 17024e92,StateSum_B17024el05,StateSum_B17 024ell8,StateSum_B 1702
4el31)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);end;
IF B17024e67 A=0thendo;
filled_e67 = 'Tract Values Used';
pctB17024e68=B17024e68/B17024e67;
pctB17024e69=B17024e69/B17024e67;
pctB17024e70=B17024e70/B17024e67;
pctB17024e71=B17024e71/B17024e67;
pctB17024e72=B17024e72/B17024e67;
pctB17024e73=B17024e73/B17024e67;
pctB17024e74=B17024e74/B17024e67;
pctB17024e75=B17024e75/B17024e67;
pctB17024e76=B17024e76/B17024e67;
pctB17024e77=B17024e77/B17024e67;
pctB17024e78=B17024e78/B17024e67;
pctB17024e79=B17024e79/B17024e67;end;
ELSE IF CountySum_B17024e67 A= Othen do;
filled_e67 = 'Filled with County Values';
pctB17024e68=CountySum_B17024e68/CountySum_B17024e67;
pctB17024e69=CountySum_B17024e69/CountySum_B17024e67;
pctB17024e70=CountySum_B17024e70/CountySum_B17024e67;
pctB17024e71=CountySum_B17024e71/CountySum_B17024e67;
pctB17024e72=CountySum_B17024e72/CountySum_B17024e67;
pctB17024e73=CountySum_B17024e73/CountySum_B17024e67;
pctB17024e74=CountySum_B17024e74/CountySum_B17024e67;
pet B17024e7 5=Co u ntyS u m_B 17024e 7 5/C o u ntyS u m_B17024e 67;
pctB17024e76=CountySum_B17024e76/CountySum_B17024e67;
pctB17024e77=CountySum_B17024e77/CountySu m_B17024e67;
pctB17024e78=CountySum_B17024e78/CountySum_B17024e67;
pctB17024e79=CountySum_B17024e79/CountySum_B17024e67;end;
ELSE IF CountySum_B17024e67 = 0 then do;
filled_e67 = 'Filled with State Values';
pctB17024e68=sum(StateSum_B17024e42,StateSum_B17024e55,StateSum_B17024e68,StateSum_B17024e81,StateSum_B17024e94,StateSum_B17024el07,StateSum_B 17024
el20)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e69=sum (Stat eSum_B17024e43,StateSum_B17024e56,StateSum_B17024e69,StateSum_B 17024e82,StateSum_B17024e95,StateSum_B17024el08,StateSum_B 17024
e 12 l)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B 17024e 106,Stat eSum_B 17024e 119);
pctB17024e70=sum(StateSum_B17024e44,StateSum_B17024e57,StateSum_B17024e70,StateSum_B17024e83,StateSum_B17024e96,StateSum_B17024el09,StateSum_B17024
el22)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B 17024e 119);
pctB17024e71=sum (Stat eSum_B17024e45,StateSum_B17024e58,StateSum_B17024e71,StateSum_B 17024e84,StateSum_B17024e97,StateSum_B17024ell0,StateSum_B 17024
el23)/sum (StateSu m_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024e72=sum (Stat eSum_B17024e46,StateSum_B17024e59,StateSum_B17024e72,StateSum_B 17024e85,StateSum_B17024e98,StateSum_B17024elll,StateSum_B 17024
el24)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06, StateSu m_B17024e 119);
pctB17024e73=sum (Stat eSum_B17024e47,StateSum_B17024e60,StateSum_B17024e73,StateSum_B17024e86,StateSum_B 17024e99,StateSum_B 17024ell2,StateSum_B17024
e 12 5)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06, StateSu m_B17024e 119);
pctB17024e74=sum (Stat eSum_B17024e48,StateSum_B17024e61,StateSum_B17024e74,StateSum_B17024e87,StateSum_B17024el00,StateSum_B17024ell3,StateSum_B 1702
4el26)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024el 19);
E-40
-------
pctB17024e75=sum (Stat eSum_B17024e49,StateSum_B17024e62,StateSum_B17024e75,StateSum_B 17024e88,StateSum_B17024el01,StateSum_B17 024ell4,StateSum_B 1702
4e 127 )/s u m (St ateS u m_B 17024e41 ,St ate Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e80,St at eS u m_B 17024e 93 ,St ate Su m_B17024e 106,Stat eS u m_B 17024e 119);
pctB17024e76=sum (Stat eSum_B17024e50,StateSum_B17024e63,StateSum_B17024e76,StateSum_B 17024e89,StateSum_B17024el02,StateSum_B17 024ell5,StateSum_B 1702
4el28)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e77=sum (Stat eSum_B17024e51,StateSum_B17024e64,StateSum_B17024e77,StateSum_B 17024e90,StateSum_B17024el03,StateSum_B17 024ell6,StateSum_B 1702
4e 129 )/s u m (St ateS u m_B 17024e 4 l,State Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e 80,St ate Su m_B 17024e 9 3,St ate Su m_B17024e 106,Stat eS u m_B 17024e 119);
pctB17024e78=sum (Stat eSum_B17024e52,StateSum_B17024e65,StateSum_B17024e78,StateSum_B 17024e91,StateSum_B17024el04,StateSum_B17 024ell7,StateSum_B 1702
4e 130 )/s u m (St ateS u m_B 17024e41 ,St ate Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e80,St at eS u m_B 17024e 93 ,St ate Su m_B17024e 106 ,Stat eS u m_B 17024e 119);
pctB17024e79=sum(StateSum_B17024e53,StateSum_B17024e66,StateSum_B17024e79,StateSum_B17024e92,StateSum_B17024el05,StateSum_B17024ell8,StateSum_B1702
4e 131 )/s u m (St ateS u m_B 17024e41 ,St ate Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e80,St at eS u m_B 17024e 93 ,St ate Su m_B17024e 106 ,Stat eS u m_B 17024e 119); e n d;
IF B17024e80 A= Othen do;
filled_e80 = 'Tract Values Used';
pet B17024e81=B17024e81/B17024e80;
pctB17024e82=B17024e82/B17024e80;
pctB17024e83=B17024e83/B17024e80;
pet B17024e84=B17024e84/B17024e80;
pctB17024e85=B17024e85/B17024e80;
pctB17024e86=B17024e86/B17024e80;
pctB17024e87=B17024e87/B17024e80;
pctB17024e88=B17024e88/B17024e80;
pctB17024e89=B17024e89/B17024e80;
pctB17024e90=B17024e90/B17024e80;
pctB17024e91=B17024e91/B17024e80;
pctB17024e92=B17024e92/B17024e80;end;
ELSE IF CountySum_B17024e80 A= Othen do;
filled_e80= 'Filled with County Va I lies';
pctB17024e81=CountySum_B17024e81/CountySum_B17024e80;
pctB17024e82=CountySum_B17024e82/CountySum_B17024e80;
pctB17024e83=CountySum_B17024e83/CountySum_B17024e80;
pctB17024e84=CountySum_B17024e84/CountySum_B17024e80;
pctB17024e85=CountySum_B17024e85/CountySum_B17024e80;
pctB17024e86=CountySum_B17024e86/CountySum_B17024e80;
pctB17024e87=CountySum_B17024e87/CountySum_B17024e80;
pctB17024e88=CountySum_B17024e88/CountySum_B17024e80;
pctB17024e89=CountySum_B17024e89/CountySum_B17024e80;
pctB17024e90=CountySum_B17024e90/CountySum_B17024e80;
pctB17024e91=CountySum_B17024e91/CountySum_B17024e80;
pctB17024e92=CountySum_B17024e92/CountySum_B17024e80;end;
ELSE IF CountySum_B17024e80 = 0 then do;
filled_e80 = 'Filled with State Values';
pctB17024e81=sum (Stat eSum_B17024e42,StateSum_B17024e55,StateSum_B17024e68,StateSum_B 17024e81,StateSum_B17024e94,StateSum_B17024el07,StateSum_B 17024
el20)/sum (StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B 17024e 119);
pctB17024e82=sum (Stat eSum_B17024e43,StateSum_B17024e56,StateSum_B17024e69,StateSum_B 17024e82,StateSum_B17024e95,StateSum_B17024el08,StateSum_B 17024
el21)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e83=sum (Stat eSum_B17024e44,StateSum_B17024e57,StateSum_B17024e70,StateSum_B 17024e83,StateSum_B17024e96,StateSum_B17024el09,StateSum_B 17024
el22)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e84=sum (Stat eSum_B17024e45,StateSum_B17024e58,StateSum_B17024e71,StateSum_B 17024e84,StateSum_B17024e97,StateSum_B17024ell0,StateSum_B 17024
el23)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e85=sum (Stat eSum_B17024e46,StateSum_B17024e59,StateSum_B17024e72,StateSum_B17024e85,StateSum_B17024e98,StateSum_B 17024elll,StateSum_B17024
el24)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e86=sum (Stat eSum_B17024e47,StateSum_B17024e60,StateSum_B17024e73,StateSum_B17024e86,StateSum_B 17024e99,StateSum_B17024ell2,StateSum_B 17024
el25)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e87=sum (Stat eSum_B17024e48,StateSum_B17024e61,StateSum_B17024e74,StateSum_B 17024e87,StateSum_B17024el00,StateSum_B17 024ell3,StateSum_B 1702
4el26)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B 17024e 119);
pctB17024e88=sum (Stat eSum_B17024e49,StateSum_B17024e62,StateSum_B17024e75,StateSum_B 17024e88,StateSum_B17024el01,StateSum_B17 024ell4,StateSum_B 1702
4el27)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024e89=sum (Stat eSum_B17024e50,StateSum_B17024e63,StateSum_B17024e76,StateSum_B 17024e89,StateSum_B17024el02,StateSum_B17 024ell5,StateSum_B 1702
4el28)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024el 19);
pctB17024e90=sum (Stat eSum_B17024e51,StateSum_B17024e64,StateSum_B17024e77,StateSum_B 17024e90,StateSum_B17024el03,StateSum_B17024ell6,StateSum_B 1702
4el29)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024el 19);
pctB17024e91=sum (Stat eSum_B17024e52,StateSum_B17024e65,StateSum_B17024e78,StateSum_B17024e91,StateSum_B 17024el04,StateSum_B17024ell7,StateSum_B1702
4el30)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024el 19);
pctB17024e92=sum (Stat eSum_B17024e53,StateSum_B17024e66,StateSum_B17024e79,StateSum_B17024e92,StateSum_B17024el05,StateSum_B17024ell8,StateSum_B 1702
4el31)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);end;
IF B17024e93 A= Othen do;
filled_e93 = 'Tract Values Used';
pctB17024e94=B17024e94/B17024e93;
pctB17024e95=B17024e95/B17024e93;
pctB17024e96=B17024e96/B17024e93;
pctB17024e97=B17024e97/B17024e93;
pctB17024e98=B17024e98/B17024e93;
E-41
-------
pet B17024e99=B17024e99/B17024e93;
pctB17024el00=B17024el00/B17024e93;
pctB17024el01=B17024el01/B17024e93;
pctB17024el02=B17024el02/B17024e93;
pctB17024el03=B17024el03/B17024e93;
pctB17024el04=B17024el04/B17024e93;
pctB17024el05=B17024el05/B17024e93;end;
ELSE IF CountySum_B17024e93 A= Othen do;
filled_e93 = 'Filled with County Values';
pctB17024e94=CountySum_B17024e94/CountySum_B17024e93;
pctB17024e95=CountySum_B17024e95/CountySum_B17024e93;
pet B17024e9 6=Co u ntyS u m_B 17024e 96/C o u ntyS u m _B 17024e9 3;
pctB17024e97=CountySum_B17024e97/CountySum_B17024e93;
pctB17024e98=CountySum_B17024e98/CountySum_B17024e93;
pctB17024e99=CountySum_B17024e99/CountySum_B17024e93;
pctB17024el00=CountySum_B17024el00/CountySum_B17024e93;
pctB17024el01=CountySum_B17024el01/CountySum_B17024e93;
pctB17024el02=CountySum_B17024el02/CountySum_B17024e93;
pctB17024el03=CountySum_B17024el03/CountySum_B17024e93;
pctB17024el04=CountySum_B17024el04/CountySum_B17024e93;
pctB17024el05=CountySum_B17024el05/CountySum_B17024e93;end;
ELSE IF CountySum_B17024e93 = 0 then do;
filled_e93 = 'Filled with State Values';
pctB17024e94=sum (Stat eSum_B17024e42,StateSum_B17024e55,StateSum_B17024e68,StateSum_B 17024e81,StateSum_B17024e94,StateSum_B17024el07,StateSum_B 17024
el20)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e95=sum (Stat eSum_B17024e43,StateSum_B17024e56,StateSum_B17024e69,StateSum_B 17024e82,StateSum_B17024e95,StateSum_B17024el08,StateSum_B 17024
el21)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e96=sum(StateSum_B17024e44,StateSum_B17024e57,StateSum_B17024e70,StateSum_B17024e83,StateSum_B17024e96,StateSum_B17024el09,StateSum_B17024
el22)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e97=sum (Stat eSum_B17024e45,StateSum_B17024e58,StateSum_B17024e71,StateSum_B17024e84,StateSum_B 17024e97,StateSum_B17024ell0,StateSum_B 17024
el23)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024e98=sum (Stat eSum_B17024e46,StateSum_B17024e59,StateSum_B17024e72,StateSum_B 17024e85,StateSum_B17024e98,StateSum_B17024elll,StateSum_B 17024
el24)/sum (StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B 17024e 119);
pctB17024e99=sum (Stat eSum_B17024e47,StateSum_B17024e60,StateSum_B17024e73,StateSum_B 17024e86,StateSum_B17024e99,StateSum_B17024ell2,StateSum_B 17024
el25)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024el00=sum(StateSum_B17024e48,StateSum_B17024e61,StateSum_B17024e74,StateSum_B17024e87,StateSum_B17024el00,StateSum_B17024ell3,StateSum_B170
24e 126 )/sum (Stat eSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el01=sum(StateSum_B17024e49,StateSum_B17024e62,StateSum_B17024e75,StateSum_B17024e88,StateSum_B17024el01,StateSu m_B17024ell4,StateSum_B170
24e 127 )/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el02=sum(StateSum_B17024e50,StateSum_B17024e63,StateSum_B17024e76,StateSum_B17024e89,StateSum_B17024el02,StateSum_B17024ell5,StateSum_B170
24el28)/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el03=sum(StateSum_B17024e51,StateSum_B17024e64,StateSum_B17024e77,StateSum_B17024e90,StateSum_B17024el03,StateSum_B17024ell6,StateSum_B170
24e 129 )/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el04=sum(StateSum_B17024e52,StateSum_B17024e65,StateSum_B17024e78,StateSum_B17024e91,StateSum_B17024el04,StateSum_B17024ell7,StateSum_B170
24el30)/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B 17024e 119);
pctB17024el05=sum(StateSum_B17024e53,StateSum_B17024e66,StateSum_B17024e79,StateSum_B17024e92,StateSum_B17024el05,StateSum_B17024ell8,StateSum_B170
24e 13 l)/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119 );end;
IF B17024el06 A= Othen do;
filled_el06 = 'Tract Values Used';
pctB17024el07=B17024el07/B17024el06;
pctB17024el08=B17024el08/B17024el06;
pctB17024el09=B17024el09/B17024el06;
pctB17024ell0=B17024ell0/B17024el06;
pctB17024elll=B17024elll/B17024el06;
pctB17024ell2=B17024ell2/B17024el06;
pctB17024ell3=B17024ell3/B17024el06;
pctB17024ell4=B17024ell4/B17024el06;
pctB17024ell5=B17024ell5/B17024el06;
pctB17024ell6=B17024ell6/B17024el06;
pctB17024ell7=B17024ell7/B17024el06;
pctB17024ell8=B17024ell8/B17024el06;end;
ELSE IF CountySum_B17024el06 A= Othen do;
filled_el06 = 'Filled with County Values';
pctB17024el07=CountySum_B17024el07/CountySum_B17024el06;
pctB17024el08=CountySum_B17024el08/CountySum_B17024el06;
pctB17024el09=CountySum_B17024el09/CountySum_B17024el06;
pctB17024ell0=CountySum_B17024ell0/CountySum_B17024el06;
pctB17024elll=CountySum_B17024elll/CountySum_B17024el06;
pctB17024ell2=CountySum_B17024ell2/CountySum_B17024el06;
pctB17024ell3=CountySum_B17024ell3/CountySum_B17024el06;
pctB17024ell4=CountySum_B17024ell4/CountySum_B17024el06;
pctB17024ell5=CountySum_B17024ell5/CountySum_B17024el06;
pctB17024ell6=CountySum_B17024ell6/CountySum_B17024el06;
pctB17024ell7=CountySum_B17024ell7/CountySum_B17024el06;
pctB17024ell8=CountySum_B17024ell8/CountySum_B17024el06;end;
ELSE IF CountySum_B17024el06 = Othen do;
E-42
-------
filled_el06 = 'Filled with State Values';
pctB17024el07=sum (Stat eSum_B17024e42,StateSum_B17024e55,StateSum_B17024e68,StateSum_B17024e81,StateSum_B17024e94,StateSum_B17024el07,StateSum_B 1702
4e 120 )/s u m (St ateS u m_B 17024e41 ,St ate Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e80,St at eS u m_B 17024e 93 ,St ate Su m_B1702 4e 106,Stat eS u m_B 17024e 119);
pctB17024el08=sum (StateSu m_B17024e43,StateSum_B17024e56,StateSum_B17024e69,StateSum_B 17024e82,StateSum_B17024e95,StateSum_B17 024el08,StateSum_B 1702
4el21)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024el09=sum(StateSum_B17024e44,StateSum_B17024e57,StateSum_B17024e70,StateSum_B17024e83,StateSum_B17024e96,StateSum_B17024el09,StateSum_B1702
4e 122 )/s u m (St ateS u m_B 17024e41 ,St ate Su m _B 17024e 54,St at eS u m_B 17024e 67 ,St ate Su m_B17024e 80,St ate Su m_B17024e 9 3,Stat eS u m_B 17024e 106,Stat eS u m_B 17024e 119);
pctB17024ell0=sum (StateSu m_B17024e45,StateSum_B17024e58,StateSum_B17024e71,StateSum_B 17024e84,StateSum_B17024e97,StateSum_B17 024ellO,StateSum_B 1702
4e 123 )/s u m (St ateS u m_B 17024e41 ,St ate Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e80,St at eS u m_B 17024e 93 ,St ate Su m_B17024e 106 ,Stat eS u m_B 17024e 119);
pctB17024elll=sum (StateSu m_B17024e46,StateSum_B17024e59,StateSum_B17024e72,StateSum_B 17024e85,StateSum_B17024e98,StateSum_B17024elll,StateSum_B 1702
4e 124 )/s u m (St ateS u m_B 17024e41 ,St ate Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e80,St at eS u m_B 17024e 93 ,St ate Su m_B17024e 106 ,Stat eS u m_B 17024e 119);
pctB17024ell2=sum(StateSum_B17024e47,StateSum_B17024e60,StateSum_B17024e73,StateSum_B17024e86,StateSum_B17024e99,StateSum_B17024ell2,StateSum_B1702
4e 12 5 )/s u m (St ateS u m_B 17024e41 ,St ate Su m_B17024e 54,Stat eS u m_B 17024e 67 ,St ate Su m_B17024e80,St at eS u m_B 17024e 93 ,St ate Su m_B17024e 106 ,Stat eS u m_B 17024e 119);
pctB17024ell3=sum(StateSum_B17024e48,StateSum_B17024e61,StateSum_B17024e74,StateSum_B17024e87,StateSum_B17024el00,StateSum_B17024ell3,StateSum_B170
24e 126 )/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024ell4=sum(StateSum_B17024e49,StateSum_B17024e62,StateSum_B17024e75,StateSum_B17024e88,StateSum_B17024el01,StateSum_B17024ell4,StateSum_B170
24e 127 )/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024ell5=sum(StateSum_B17024e50,StateSum_B17024e63,StateSum_B17024e76,StateSum_B17024e89,StateSum_B17024el02,StateSum_B17024ell5,StateSum_B170
24el28)/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024ell6=sum(StateSum_B17024e51,StateSum_B17024e64,StateSum_B17024e77,StateSum_B17024e90,StateSum_B17024el03,StateSum_B17024ell6,StateSum_B170
24e 129 )/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024ell7=sum(StateSum_B17024e52,StateSum_B17024e65,StateSum_B17024e78,StateSum_B 17024e91,StateSum_B17024el04,StateSum_B17024ell7,StateSum_B 170
24el30)/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024ell8=sum(StateSum_B17024e53,StateSum_B17024e66,StateSum_B 17024e79,StateSum_B17024e92,StateSum_B17024el05,StateSum_B17024ell8,StateSum_B 170
24el31)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);end;
IF B17024ell9 A= Othen do;
filled_ell9 = 'Tract Values Used';
pctB17024el20=B17024el20/B17024ell9;
pctB17024el21=B17024el21/B17024ell9;
pctB17024el22=B17024el22/B17024ell9;
pctB17024el23=B17024el23/B17024ell9;
pctB17024el24=B17024el24/B17024ell9;
pctB17024el25=B17024el25/B17024ell9;
pctB17024el26=B17024el26/B17024ell9;
pctB17024el27=B17024el27/B17024ell9;
pctB17024el28=B17024el28/B17024ell9;
pctB17024el29=B17024el29/B17024ell9;
pctB17024el30=B17024el30/B17024ell9;
pctB17024el31=B17024el31/B17024ell9;end;
ELSE IF CountySum_B17024ell9 A= Othen do;
filled_ell9 = 'Filled with County Values';
pctB17024el20=CountySum_B17024el20/CountySum_B17024ell9;
pctB17024el21=CountySum_B17024el21/CountySum_B17024ell9;
pctB17024el22=CountySum_B17024el22/CountySum_B17024ell9;
pctB17024el23=CountySum_B17024el23/CountySum_B17024ell9;
pctB17024el24=CountySum_B17024el24/CountySum_B17024ell9;
pctB17024el25=CountySum_B17024el25/CountySum_B17024ell9;
pctB17024el26=CountySum_B17024el26/CountySum_B17024ell9;
pctB17024el27=CountySum_B17024el27/CountySum_B17024ell9;
pctB17024el28=CountySum_B17024el28/CountySum_B17024ell9;
pctB17024el29=CountySum_B17024el29/CountySum_B17024ell9;
pctB17024el30=CountySum_B17024el30/CountySum_B17024ell9;
pctB17024el31=CountySum_B17024el31/CountySum_B17024ell9;end;
ELSE IF CountySum_B17024ell9 = Othen do;
filled_ell9 = 'Filled with State Values';
pctB17024el20=sum(StateSum_B17024e42,StateSum_B17024e55,StateSum_B17024e68,StateSum_B 17024e81,StateSum_B17024e94,StateSum_B17 024el07,StateSum_B 1702
4el20)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024el21=sum(StateSum_B17024e43,StateSum_B17024e56,StateSum_B17024e69,StateSum_B17024e82,StateSum_B17024e95,StateSum_B17024el08,StateSum_B1702
4el21)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024el22=sum(StateSum_B17024e44,StateSum_B17024e57,StateSum_B17024e70,StateSum_B17024e83,StateSum_B17024e96,StateSum_B17024el09,StateSum_B1702
4el22)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024el23=sum(StateSum_B17024e45,StateSum_B17024e58,StateSum_B17024e71,StateSum_B 17024e84,StateSum_B17024e97,StateSum_B17024ell0,StateSum_B 1702
4el23)/sum(StateSum_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024ell9);
pctB17024el24=sum(StateSum_B17024e46,StateSum_B17024e59,StateSum_B17024e72,StateSum_B 17024e85,StateSum_B17024e98,StateSum_B17 024elll,StateSum_B 1702
4el24)/sum (StateSu m_B17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B 17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el25=sum(StateSum_B17024e47,StateSum_B17024e60,StateSum_B17024e73,StateSum_B 17024e86,StateSum_B17024e99,StateSum_B17 024ell2,StateSum_B 1702
4el25)/sum (StateSu m_B17024e41,StateSum_B 17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024el 19);
pctB17024el26=sum (StateSu m_B17024e48,StateSum_B17024e61,StateSum_B17024e74,StateSum_B 17024e87,StateSum_B17024el00,StateSum_B17024ell3,StateSum_B170
24e 126 )/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
E-43
-------
pctB17024el27=sum(StateSum_B17024e49,StateSum_B17024e62,StateSum_B17024e75,StateSum_B17024e88,StateSum_B17024el01,StateSum_B17024ell4,StateSum_B170
24e 127 )/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el28=sum(StateSum_B17024e50,StateSum_B17024e63,StateSum_B17024e76,StateSum_B17024e89,StateSum_B17024el02,StateSum_B17024ell5,StateSum_B170
24el28)/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el29=sum(StateSum_B17024e51,StateSum_B17024e64,StateSum_B17024e77,StateSum_B17024e90,StateSum_B17024el03,StateSum_B17024ell6,StateSum_B170
24e 129 )/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el30=sum(StateSum_B17024e52,StateSum_B17024e65,StateSum_B17024e78,StateSum_B17024e91,StateSum_B17024el04,StateSum_B17024ell7,StateSum_B170
24el30)/sum (Stat eSum_B 17024e41,StateSum_B17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024e 119);
pctB17024el31=sum(StateSum_B17024e53,StateSum_B17024e66,StateSum_B17024e79,StateSum_B17024e92,StateSum_B17024el05,StateSum_B17024ell8,StateSum_B170
24e 13 l)/sum (Stat eSum_B 17024e41,StateSum_B 17024e54,StateSum_B17024e67,StateSum_B17024e80,StateSum_B17024e93,StateSum_B17024el06,StateSum_B17024el 19 );end;
run;
data work.pov_ratio_&geo /*calculates percents at or above poverty defined as +/-1.5 in come/ poverty ratio */
(keep= STUSAB CENSUS_REGION LOGRECNO STATE COUNTY TRACT GEO I D_merge LAT LON
/*B17024e2 B17024el5 B17024e28 B17024e41 B17024e54 B17024e67 B17024e80 B17024e93
B17024el06 B17024ell9*/
filled_e2 filled_el5filled_e28 filled_e41 filled_e54filled_e67 filled_eSOfilled_e93 filled_el06
filled_ell9
p0-pl7 np0-npl7
p5u p6toll pl2tol7 pl8to24 p25to34 p35to44 p45to54 p55to64 p65to74 p75plus
np5u np6toll npl2tol7 npl8to24 np25to34 np35to44 np45to54 np55to64 np65to74 np75plus);
set work.pov_pct_&geo;
/*The first sum provides the prob < 1.5 pov ratio with 'p' meaning poverty. The second sum is > 1.5 pov with 'np' meaning not poverty. */
p5u=sum(pctB17024e3,pctB17024e4,pctB17024e5,pctB17024e6,pctB17024e7);
np5u=sum(pctB17024e8,pctB17024e9,pctB17024el0,pctB17024ell,pctB17024el2,pctB17024el3,pctB17024el4);
p6toll=sum(pctB17024el6,pctB17024el7,pctB17024el8,pctB17024el9,pctB17024e20);
np6toll=sum(pctB17024e21,pctB17024e22,pctB17024e23,pctB17024e24,pctB17024e25,pctB17024e26,pctB17024e27);
pl2tol7=sum(pctB17024e29,pctB17024e30,pctB17024e31,pctB17024e32,pctB17024e33);
npl2tol7=sum(pctB17024e34,pctB17024e35,pctB17024e36,pctB17024e37,pctB17024e38,pctB17024e39,pctB17024e40);
pl8to24=sum(pctB17024e42,pctB17024e43,pctB17024e44,pctB17024e45,pctB17024e46);
npl8to24=sum(pctB17024e47,pctB17024e48,pctB17024e49,pctB17024e50,pctB17024e51,pctB17024e52,pctB17024e53);
p25to34=sum(pctB17024e55,pctB17024e56,pctB17024e57,pctB17024e58,pctB17024e59);
np25to34=sum(pctB17024e60,pctB17024e61,pctB17024e62,pctB17024e63,pctB17024e64,pctB17024e65,pctB17024e66);
p35to44=sum(pctB17024e68,pctB17024e69,pctB17024e70,pctB17024e71,pctB17024e72);
np35to44=sum(pctB17024e73,pctB17024e74,pctB17024e75,pctB17024e76,pctB17024e77,pctB17024e78,pctB17024e79);
p45to54=sum(pctB17024e81,pctB17024e82,pctB17024e83,pctB17024e84,pctB17024e85);
np45to54=sum(pctB17024e86,pctB17024e87,pctB17024e88,pctB17024e89,pctB17024e90,pctB17024e91,pctB17024e92);
p55to64=sum(pctB17024e94,pctB17024e95,pctB17024e96,pctB17024e97,pctB17024e98);
np55to64=sum(pctB17024e99,pctB17024el00,pctB17024el01,pctB17024el02,pctB17024el03,pctB17024el04,pctB17024el05);
p65to74=sum(pctB17024el07,pctB17024el08,pctB17024el09,pctB17024ell0,pctB17024elll);
np65to74=sum(pctB17024ell2,pctB17024ell3,pctB17024ell4,pctB17024ell5,pctB17024ell6,pctB17024ell7,pctB17024ell8);
p75plus=sum(pctB17024el20,pctB17024el21,pctB17024el22,pctB17024el23,pctB17024el24);
np75plus=sum(pctB17024el25,pctB17024el26,pctB17024el27,pctB17024el28,pctB17024el29,pctB17024el30,pctB17024el31);
/*copy the percents +/-1.5 in come/ poverty ratio for ages 5 and under, 6toll, and 12tol7 to separate ages 1-17 for which asthma
prevalence data are available*/
p0=p5u; pl=p5u; p2=p5u; p3=p5u; p4=p5u; p5=p5u;
np0=np5u; npl=np5u; npl=np5u; np2=np5u; np3=np5u; np4=np5u; np5=np5u;
p6=p6toll; p7=p6toll; p8=p6toll; p9=p6toll; pl0=p6toll; pll=p6toll;
np6=p6toll; np7=p6toll; np8=p6toll; np9=np6toll; npl0=np6toll; npll=np6toll;
pl2=pl2tol7; pl3=pl2tol7; pl4=pl2tol7; pl5=pl2tol7; pl6=pl2tol7; pl7=pl2tol7;
npl2=npl2tol7; npl3=npl2tol7; npl4=npl2tol7; npl5=npl2tol7; npl6=npl2tol7; npl7=npl2tol7;
run;
data work.QA_pov_ratio_&geo /* checks that all calculated percents sum to 1 where they exist*/
(keep= STUSAB CENSUS_REGION LOGRECNO STATE COUNTY TRACT GEOID_merge LAT LON
filled e2 filled el5filled e28 filled e41 filled e54filled e67 filled eSOfilled e93 filled el06
/*B17024e2 B17024el5 B17024e28 B17024e41 B17024e54 B17024e67 B17024e80 B17024e93
sum5u sum6toll suml2tol7 suml8to24 sum25to34 sum35to44 sum45to54 sum55to64 sum65to74
filled_ell9
B17024el06 B17024ell9*/
sum75plus);
set work.pov_ratio_&geo;
sum5u=p5u+np5u;
sum6toll=p6toll+n p6toll;
suml2tol7=pl2tol7+npl2tol7;
suml8to24=pl8to24+npl8to24;
sum25to34=p25to34+np25to34;
sum35to44=p35to44+np35to44;
sum45to54=p45to54+np45to54;
sum55to64=p55to64+np55to64;
sum65to74=p65to74+np65to74;
sum75plus=p75plus+np75plus;
run;
data work.pov_ratio_&geo; *changes order of columns (variables);
retain STUSAB CENSUS_REGION LOGRECNO STATE COUNTY TRACT GEO I D_merge LAT LON
p0-pl7
p5u p6toll pl2tol7 pl8to24 p25to34 p35to44 p45to54 p55to64 p65to74 p75plus
np0-npl7
np5u np6toll npl2tol7 npl8to24 np25to34 np35to44 np45to54 np55to64 np65to74 np75plus
E-44
-------
filled_e2 filled_el5 filled_e28 filled_e41 filled_e54 filled_e67 filled_e80 filled_e93 filled_el06
filled_ell9;
set work.pov_ratio_&geo;
run;
%mend;
%macro lmport_Pov_Calc_Ratio(geo); *Runs macros that imports and merges income/poverty data with geographic data (by state) then calculates ratios above or below 1.5 income/poverty
ratio (by age group);
%AnyGeo(&.geo);
%Read_poverty{&. geo);
proc sort data=work.SFe0056&geo; *sort estimate data;
by logrecno;
run;
proc sort data=work.g20135&geo.coord; *sort geo data;
by logrecno;
run;
data work.SFe^g_0056&geo; * merges estimate and geo data;
merge work.SFe0056&geo(in=a) work.g20135&geo.coord;
by logrecno;
retain STUSAB STATE COUNTY TRACT LAT LON;
if a;
run;
%pov_ratio_ co/c(&ge o );
proc append base=sas.pov_acs2013_5yr data=work.pov_ratio_&geo; run;
proc append base=sas.QA_pov_acs2013_5yr data=work.QA_pov_ratio_&geo; run;
%mend;
*runs macro for 50 United States, District of Columbia, and Puerto Rico;
%lmport_ Pov_Calc_Ratio(a I );
%lmport_ Pov_Calc_Ratio (a k );
%lmport_ Pov_Calc_Ratio (a z );
%lmport_ Pov_Calc_Ratio (a r );
%lmport_ Pov_Calc_Ratio (ca );
%lmport_ Pov_Calc_Ratio (co );
%lmport_ Pov_Calc_Ratio (ct );
%lmport_ Pov_Calc_Ratio (d e );
%lmport_ Pov_Calc_Ratio (d c );
%lmport_ Pov_Calc_ Ratio( f I );
%lmport_ Pov_Calc_Ratio (ga );
%lmport_Pov_Calc_Ratio( h i);
%lmport_ Pov_Calc_Ratio (id);
%lmport_ Pov_Calc_Ratio (i I );
%lmport_ Pov_Calc_Ratio (in);
%lmport_ Pov_Calc_Ratio (i a );
%lmport_ Pov_Calc_Ratio (ks );
%lmport_ Pov_Calc_Ratio (ky );
%lmport_ Pov_Calc_Ratio( I a );
%lmport_ Pov_Calc_Ratio (me);
%lmport_ Pov_Calc_Ratio (m d );
%lmport_ Pov_Calc_Ratio (ma);
%lmport_ Pov_Calc_Ratio (m i );
%lmport_ Pov_Calc_Ratio (m n );
%lmport_ Pov_Calc_Ratio (m s );
%lmport_ Pov_Calc_Ratio (mo);
%lmport_ Pov_Calc_Ratio (mt );
%lmport_ Pov_Calc_Ratio (n e );
%lmport_ Pov_Calc_Ratio (n v );
%lmport_Pov_Calc_Ratio( n h );
%lmport_ Pov_Calc_Ratio (nj );
%lmport_ Pov_Calc_Ratio (n m );
%lmport_ Pov_Calc_Ratio (ny );
%lmport_ Pov_Calc_Ratio (n c );
%lmport_ Pov_Calc_Ratio (n d );
%lmport_ Pov_Calc_Ratio (o h );
%lmport_ Pov_Calc_Ratio (ok);
%lmport_ Pov_Calc_Ratio (o r);
%lmport_ Pov_Calc_Ratio (pa );
%lmport_ Pov_Calc_Ratio (r i );
%lmport_ Pov_Calc_Ratio (sc );
%lmport_ Pov_Calc_Ratio (sd );
%lmport_ Pov_Calc_Ratio (t n );
%lmport_ Pov_Calc_Ratio (tx );
%lmport_ Pov_Calc_Ratio (ut );
%lmport_ Pov_Calc_Ratio (vt );
%lmport_ Pov_Calc_Ratio (va );
%lmport_ Pov_Calc_Ratio (wa );
%lmport_ Pov_Calc_Ratio (w v );
%lmport_ Pov_Calc_Ratio (w i );
%lmport_ Pov_Calc_Ratio (wy );
%lmport_ Pov_Calc_Ratio (p r);
E-45
-------
Attachment 4 - Proc Survey Logistic Model Results - Evaluating Influence of
Personal Attributes and their Interaction
Table 1. Logistic model parameter estimates, coefficient variation, and statistical
significance for personal attributes that influence asthma prevalence in adults.
US REGION
Statistical
(degrees of
Parameter
Significance
freedom)
Variable
Estimate
Std Err
(ProbT)
Intercept
-2.610
0.114
<0.001
family income
0.045
0.188
0.810
Black African American
-0.184
0.232
0.432
BMI
0.367
0.136
0.009
sex
0.418
0.091
<0.001
age
-0.006
0.002
0.002
family income*sex
0.345
0.236
0.150
Northeast
(51)
Black African American*sex
-0.038
0.314
0.905
BMI*sex
0.296
0.181
0.107
family income*BMI
0.567
0.276
0.045
Black African American*BMI
-0.163
0.290
0.577
family income*Black African American
0.675
0.499
0.183
family income*Black African American*BMI
-0.071
0.667
0.915
family income*BMI*sex
-0.401
0.375
0.290
Black African American*BMI*sex
0.257
0.389
0.512
family income*Black African American*sex
-0.298
0.648
0.647
family income * Black African American*BMI*sex
-0.349
0.862
0.688
Intercept
-2.849
0.100
<0.001
family income
0.543
0.137
<0.001
Black African American
0.347
0.265
0.195
BMI
0.280
0.118
0.020
sex
0.420
0.072
<0.001
age
-0.004
0.001
0.012
family income*sex
-0.219
0.141
0.125
Midwest
(66)
Black African American*sex
0.109
0.315
0.732
BMI*sex
0.380
0.130
0.005
family income*BMI
0.122
0.237
0.609
Black African American*BMI
-0.223
0.357
0.534
family income*Black African American
-0.295
0.374
0.434
family income*Black African American*BMI
0.133
0.481
0.784
family income*BMI*sex
0.080
0.277
0.774
Black African American*BMI*sex
-0.551
0.427
0.202
family income*Black African American*sex
0.093
0.428
0.829
family income * Black African American*BMI*sex
0.390
0.533
0.466
E-46
-------
US REGION
Statistical
(degrees of
Parameter
Significance
freedom)
Variable
Estimate
Std Err
(ProbT)
Intercept
-3.147
0.092
<0.001
family income
0.325
0.123
0.010
Black African American
0.217
0.150
0.151
BMI
0.341
0.102
0.001
sex
0.540
0.076
<0.001
age
-0.001
0.001
0.492
family income*sex
-0.039
0.152
0.797
South
(115)
Black African American*sex
-0.220
0.205
0.286
BMI*sex
0.262
0.130
0.045
family income*BMI
0.123
0.202
0.546
Black African American*BMI
-0.084
0.235
0.721
family income*Black African American
0.076
0.218
0.728
family income*Black African American*BMI
-0.345
0.404
0.394
family income*BMI*sex
-0.050
0.236
0.832
Black African American*BMI*sex
0.119
0.273
0.664
family income*Black African American*sex
0.057
0.286
0.841
family income * Black African American*BMI*sex
0.268
0.460
0.561
Intercept
-3.042
0.086
<0.001
family income
0.206
0.111
0.068
Black African American
0.432
0.295
0.147
BMI
0.467
0.090
<0.001
sex
0.532
0.069
<0.001
age
0.001
0.001
0.652
family income*sex
-0.150
0.138
0.282
West
(70)
Black African American*sex
-0.812
0.314
0.012
BMI*sex
0.235
0.113
0.040
family income*BMI
-0.039
0.193
0.841
Black African American*BMI
0.107
0.420
0.799
family income*Black African American
0.164
0.441
0.711
family income*Black African American*BMI
-0.451
0.730
0.539
family income*BMI*sex
-0.022
0.257
0.932
Black African American*BMI*sex
0.383
0.495
0.442
family income*Black African American*sex
0.998
0.491
0.046
family income * Black African American*BMI*sex
0.025
0.937
0.979
E-47
-------
Table 2. Logistic model parameter estimates, coefficient variation, and statistical
significance for personal attributes that influence asthma prevalence in children.
US REGION
Statistical
(degrees of
Parameter
Significance
freedom)
Variable
Estimate
Std Err
(ProbT)
Intercept
-1.141
0.734
0.127
family income
-0.378
0.298
0.210
Black African American
0.176
0.289
0.544
BMI
0.633
0.406
0.125
sex
-0.314
0.202
0.127
age
-0.063
0.047
0.188
family income*sex
0.716
0.492
0.152
Northeast
(51)
Black African American*sex
-0.032
0.441
0.943
BMI*sex
0.237
0.692
0.733
family income*BMI
0.310
0.856
0.718
Black African American*BMI
0.620
0.833
0.460
family income*Black African American
0.300
0.459
0.517
family income*Black African American*BMI
-0.895
1.583
0.574
family income*BMI*sex
-0.416
1.264
0.743
Black African American*BMI*sex
-0.647
1.320
0.626
family income*Black African American*sex
-0.047
0.878
0.958
family income * Black African American*BMI*sex
0.689
2.715
0.801
Intercept
-3.116
0.539
<0.001
family income
0.626
0.275
0.026
Black African American
0.648
0.308
0.039
BMI
-0.047
0.338
0.889
sex
-0.195
0.173
0.263
age
0.052
0.032
0.114
family income*sex
-0.313
0.352
0.378
Midwest
(66)
Black African American*sex
0.410
0.533
0.445
BMI*sex
0.397
0.533
0.459
family income*BMI
-0.038
0.554
0.945
Black African American*BMI
0.740
0.827
0.374
family income*Black African American
-0.351
0.577
0.545
family income*Black African American*BMI
-0.679
1.303
0.604
family income*BMI*sex
0.416
0.820
0.614
Black African American*BMI*sex
-1.402
1.114
0.213
family income*Black African American*sex
-0.010
0.919
0.991
family income * Black African American*BMI*sex
1.394
1.607
0.389
Intercept
-1.786
0.350
<0.001
South
family income
0.358
0.158
0.025
(115)
Black African American
1.078
0.181
<0.001
BMI
0.735
0.241
0.003
E-48
-------
US REGION
Statistical
(degrees of
Parameter
Significance
freedom)
Variable
Estimate
Std Err
(ProbT)
sex
0.119
0.141
0.398
age
-0.052
0.025
0.039
family income*sex
0.210
0.243
0.390
Black African American*sex
-1.027
0.299
0.001
BMI*sex
0.086
0.384
0.823
family income*BMI
-0.740
0.434
0.091
Black African American*BMI
-0.634
0.427
0.140
family income*Black African American
-0.515
0.285
0.074
family income*Black African American*BMI
1.153
0.637
0.073
family income*BMI*sex
0.330
0.616
0.593
Black African American*BMI*sex
0.767
0.806
0.343
family income*Black African American*sex
0.346
0.446
0.439
family income * Black African American*BMI*sex
-1.095
1.035
0.292
Intercept
-1.786
0.448
<0.001
family income
0.252
0.189
0.186
Black African American
0.315
0.283
0.270
BMI
0.123
0.335
0.714
sex
-0.362
0.180
0.049
age
-0.022
0.029
0.456
family income*sex
-0.290
0.270
0.287
West
(70)
Black African American*sex
-0.406
0.569
0.478
BMI*sex
0.682
0.458
0.141
family income*BMI
-0.528
0.588
0.372
Black African American*BMI
-0.166
0.958
0.863
family income*Black African American
0.612
0.439
0.167
family income*Black African American*BMI
-10.865
1.411
<0.001
family income*BMI*sex
-0.137
0.795
0.864
Black African American*BMI*sex
-1.933
1.603
0.232
family income*Black African American*sex
0.486
0.812
0.551
family income *Black African American*BMI*sex
12.284
2.211
<0.001
E-49
-------
APPENDIX F
DESCRIPTION OF THE AIR POLLUTANTS EXPOSURE MODEL (APEX)
Purpose: This Appendix briefly describes the EPA's Air Pollutants Exposure (APEX) model.
Table of Contents
F.l Overview F-2
F.2 Model Inputs F-4
F.3 Demographic Characteristics F-5
F.4 Attributes of Individuals F-5
F.5 Construction of Longitudinal Diary Sequence F-6
F.6 Key Physiological Processes Modeled F-8
F.7 Estimating Microenvironmental Concentrations F-9
F.7.1 Mass Balance Model F-10
F.7.2 Factors Model F-16
F.8 Exposure and Dose Time Series Calculations F-17
F.9 Model Output F-19
F-l
-------
F.l Overview
APEX is the human inhalation exposure model within the Total Risk Integrated
Methodology (TRIM) framework (U.S. EPA, 2017a, b). APEX is conceptually based on the
probabilistic NAAQS Exposure Model (pNEM) that was used to estimate population exposures
for the 1996 O3 NAAQS review (Johnson et al., 1996a, b, c). Since that time the model has been
restructured, improved, and expanded to reflect conceptual advances in the science of exposure
modeling and newer input data available for the model. Key improvements to algorithms include
replacement of the cohort approach with a probabilistic sampling approach focused on
individuals, accounting for fatigue and oxygen debt after exercise in the calculation of ventilation
rates (Isaacs et al., 2008), new approaches for construction of longitudinal activity patterns for
simulated persons (Glen et al., 2008; Rosenbaum et al., 2008), and new equations for estimating
resting metabolic rate (RMR) and ventilation rate (see Appendix H). Major improvements to data
input to the model include updated air exchange rates (AERs), population census and commuting
data, distributions of body mass and height (Appendix G), and the daily time-location-activities
database (Appendix I).
APEX estimates human exposure to criteria and toxic air pollutants at local, urban, or
regional scales using a stochastic, microenvironmental approach. That is, the model randomly
selects data on a sample of hypothetical individuals in an actual population database and
simulates each individual's movements through time and space (e.g., at home, in vehicles) to
estimate their exposure to the pollutant. APEX can assume people live and work in the same
general area (i.e., that the ambient air quality is the same at home and at work) or optionally can
model commuting and thus exposure at the work location for individuals who work.
The APEX model is a microenvironmental, longitudinal human exposure model for
airborne pollutants. It is applied to a specified study area, which is typically a metropolitan area.
The time period of the simulation is typically one year, but can easily be made either longer or
shorter. APEX uses census data, such as gender and age, to generate the demographic
characteristics of simulated individuals. It then assembles a composite activity diary to represent
the sequence of activities and microenvironments that the individual experiences. Each
microenvironment has a user-specified method for determining air quality. The inhalation
exposure in each microenvironment is simply equal to the air concentration in that
microenvironment. When coupled with breathing rate information and a physiological model,
various measures of dose can also be calculated.
The term microenvironment is intended to represent the immediate surroundings of an
individual, in which the pollutant of interest is assumed to be well-mixed. Time is modeled as a
sequence of discrete time steps called events. In APEX, the concentration in a microenvironment
may change between events. For each microenvironment, the user specifies the method of
F-2
-------
concentration calculation (either mass balance or regression factors, described later in this
paper), the relationship of the microenvironment to the ambient air, and the strength of any
pollutant sources specific to that microenvironment. Because the microenvironments that are
relevant to exposure depend on the nature of the target chemical and APEX is designed to be
applied to a wide range of chemicals, both the total number of microenvironments and the
properties of each are free to be specified by the user.
The ambient air data are provided as input to the model in the form of time series at a list
of specified locations. Typically, hourly air concentrations are used, although temporal
resolutions as small as one minute may be used. The spatial range of applicability of a given
ambient location is called an air district. Any number of air districts can be accommodated in a
model run, subject only to computer hardware limitations. In principle, any microenvironment
could be found within a given air district. Therefore, to estimate exposures as an individual
engages in activities throughout the period it is necessary to determine both the
microenvironment and the air district that apply for each event.
An exposure event is determined by the time reported in the activity diary; during any
event the district, microenvironment, ambient air quality, and breathing rate are assumed to
remain fixed. Since the ambient air data change every hour, the maximum duration of an event is
limited to one hour. The event duration may be less than this (as short as one minute) if the
activity diary indicates that the individual changes microenvironments or activities performed
within the hour.
An APEX simulation includes the following steps:
(1) Characterize the study area - APEX selects sectors (e.g., census tracts) within a study
area based on user-defined criteria and thus identifies the potentially exposed population
and defines the air quality and weather input data required for the area.
(2) Generate simulated individuals - APEX stochastically generates a sample of simulated
individuals based on the census data for the study area and human profile distribution
data (such as age-specific employment probabilities). The user must specify the size of
the sample. The larger the sample, the more representative it is of the population in the
study area and the more stable the model results are (but also the longer the computing
time).
(3) Construct a long-term sequence of activity events and determine breathing rates - APEX
constructs an event sequence (activity pattern) spanning the period of simulation for each
simulated person. The model then stochastically assigns breathing rates to each event,
based on the type of activity and the physical characteristics of the simulated person.
(4) Calculate pollutant concentrations in microenvironments - APEX enables the user to
define any microenvironment that individuals in a study area would visit. The model then
calculates concentrations of each pollutant in each of the microenvironments.
F-3
-------
(5) Calculate pollutant exposures for each simulated individual - Microenvironmental
concentrations are time weighted based on individuals' events (i.e., time spent in the
microenvironment) to produce a sequence of time-averaged exposures (or minute by
minute time series) spanning the simulation period.
(6) Estimate dose - APEX can also calculate the dose time series for each of the simulated
individuals based on the exposures and breathing rates for each event. However, dose is
not needed for the SO2 assessment and thus will not be discussed further.
(7) Estimate a health response - APEX can link an exposure-response (E-R) function
generated from controlled human exposure study data with the modeled exposures to
estimate the fraction of the population that could experience and adverse health outcome
(e.g., lung function decrements).
The model simulation continues until exposures are determined for the user-specified
number of simulated individuals. APEX then calculates population exposure statistics (such as
the number of exposures exceeding user-specified levels) for the entire simulation and writes out
tables of distributions of these statistics.
F.2 Model Inputs
APEX requires certain inputs from the user. The user specifies the geographic area and
the range of ages and age groups to be used for the simulation. Hourly (or shorter) ambient air
quality and hourly temperature data must be furnished for the entire simulation period. Other
hourly meteorological data (humidity, wind speed, wind direction, precipitation) can be used by
the model to estimate microenvironmental concentrations, but are optional.
In addition, most variables used in the model algorithms are represented by user-specified
probability distributions which capture population variability. APEX provides great flexibility in
defining model inputs and parameters, including options for the frequency of selecting new
values from the probability distributions. The model also allows different distributions to be used
at different times of day or on different days, and the distribution can depend conditionally on
values of other parameters. The probability distributions available in APEX include beta, binary,
Cauchy, discrete, exponential, extreme value, gamma, logistic, lognormal, loguniform, normal,
off/on, Pareto, point (constant), triangle, uniform, Weibull, and nonparametric distributions.
Minimum and maximum bounds can be specified for each distribution if a truncated distribution
is appropriate. There are two options for handling truncation. The generated samples outside the
truncation points can be set to the truncation limit; in this case, samples "stack up" at the
truncation points. Alternatively, new random values can be selected, in which case the
probability outside the limits is spread over the specified range, and thus the probabilities inside
the truncation limits will be higher than the theoretical untruncated distribution.
F-4
-------
F.3 Demographic Characteristics
The starting point for constructing a simulated individual is the population census
database; this contains population counts for each combination of age, gender, race, and sector.
The user may decide what spatial area is represented by a sector, but the default input file defines
a sector as a census tract. Census tracts are variable in both geographic size and population
number, though usually have between 1,500 and 8,000 persons. Currently, the default file
contains population counts from the 2010 census for every census tract in the United States, thus
the default file should be sufficient for most exposure modeling purposes. The combination of
age, gender, race, and sector are selected first. The sector becomes the home sector for the
individual, and the corresponding air district becomes the home district. The probabilistic
selection of individuals is based on the sector population and demographic composition, and
taken collectively, the set of simulated individuals constitutes a random sample from the study
area.
The second step in constructing a simulated individual is to determine their employment
status. This is determined by a probability which is a function of age, gender, and home sector.
An input file is provided which contains employment probabilities from the 2010 census for
every combination of age (16 and over), gender, and census tract. APEX assumes that persons
under age 16 do not commute. For persons who are determined to be workers, APEX then
randomly selects a work sector, based on probabilities determined from the commuting matrix.
The work sector is used to assign a work district for the individual that may differ from the home
district, and thus different ambient air quality may be used when the individual is at work.
The commuting matrix contains data on flows (number of individuals) traveling from a
given home sector to a given work sector. Based on commuting data from the 2000 census, a
commuting data base for the entire United States has been prepared. This permits the entire list
of non-zero flows to be specified on one input file. Given a home sector, the number of
destinations to which people commute varies anywhere from one to several hundred other tracts.
F.4 Attributes of Individuals
In addition to the above demographic information, each individual is assigned status and
physiological attributes. The status variables are factors deemed important in estimating
microenvironmental concentrations, and are specified by the user. Status variables can include,
but are not limited to, people's housing type, whether their home has air conditioning, whether
they use a gas stove at home, whether the stove has a gas pilot light, and whether their car has air
conditioning. Physiological variables are important when estimating pollutant specific dose.
These variables could include height, weight, blood volume, pulmonary diffusion rate, resting
F-5
-------
metabolic rate, energy conversion factor (liters of oxygen per kilocalorie energy expended),
hemoglobin density in blood, maximum limit on metabolic equivalents of work (MET) ratios
(see below), and endogenous CO production rate. All of these variables are treated
probabilistically taking into account interdependencies where possible, and reflecting variability
in the population.
Two key personal attributes determined for each individual in this assessment are body
mass (BM) and body surface area (BSA). Each simulated individual's body mass was randomly
sampled from age- and gender-specific body mass distributions generated from National Health
and Nutrition Examination Survey (NHANES) data for the years 2009-2014.1 Details in their
development and the parameter values are provided in Appendix G. Then age- and gender-
specific body surface area can be estimated for each simulated individual. Briefly, the BSA
calculation is based on logarithmic relationships developed by Burmaster (1998) that use body
mass as an independent variable as follows:
BSA=e-2-2781£Ma6821 Equation F-l
where,
BSA = body surface area (m2)
BM = body mass (kg)
F.5 Construction of Longitudinal Diary Sequence
The activity diary determines the sequence of microenvironments visited by the
simulated person. A longitudinal sequence of daily diaries must be constructed for each
simulated individual to cover the entire simulation period. The default activity diaries in APEX
are derived from those in the EPA's Consolidated Human Activity Database (CHAD) (McCurdy
et al., 2000; U.S. EPA 2002; 2017c), although the user could provide area specific diaries if
available. There are over 55,000 CHAD diaries used for the current SO2 assessment, each
covering a 24-hour period, that have been compiled from several studies. CHAD is essentially a
cross-sectional database that, for the most part, only has one diary per person. Therefore, APEX
must assemble each longitudinal diary sequence for a simulated individual from many single-day
diaries selected from a pool of similar people.
1 Demographic (Demo) and Body Measurement (BMX) datasets for each of the NHANES studies were obtained
from http://www.cdc.gov/nchs/nhanes/nhanes_questionnaires.htm.
F-6
-------
APEX selects diaries from CHAD by matching gender and employment status, and by
requiring that age falls within a user-specified range on either side of the age of the simulated
individual. For example, if the user specifies plus or minus 20%, then for a 40-year old simulated
individual, the available CHAD diaries are those from persons aged 32 to 48. Each simulated
individual therefore has an age window of acceptable diaries; these windows can partially
overlap those for other simulated individuals. This differs from a cohort-based approach, where
the age windows are fixed and non-overlapping. The user may optionally request that APEX
allow a decreased probability for selecting diaries from ages outside the primary age window,
and also for selecting diaries from persons of missing gender, age, or employment status. These
options allow the model to continue the simulation when diaries are not available within the
primary window.
The available CHAD diaries are classified into diary pools, based on the temperature and
day of the week. The model will select diaries from the appropriate pool for days in the
simulation having matching temperature and day type characteristics. The rules for defining
these pools are specified by the user. For example, the user could request that all diaries from
Monday to Friday be classified together, and Saturday and Sunday diaries in another class.
Alternatively, the user could instead create more than two classes of weekdays, combine all
seven days into one class, or split all seven days into separate classes.
The temperature classification can be based either on daily maximum temperature, daily
average temperature, or both. The user specifies both the ranges and numbers of temperatures
classes. For example, the user might wish to create four temperature classes and set their ranges
to below 50 °F, 50-69 °F, 70-84 °F, and above a daily maximum of 84 °F. Then day type and
temperature classes are combined to create the diary pools. For example, if there are four
temperature classes and two-day type classes, then there will be eight diary pools.
APEX then determines the day-type and the applicable temperature for each person's
simulated day. APEX allows multiple temperature stations to be used; the sectors are
automatically mapped to the nearest temperature station. This may be important for study areas
such as the greater Los Angeles area, where the inland desert sectors may have very different
temperatures from the coastal sectors. For selected diaries, the temperature in the home sector of
the simulated person is used. For each day of the simulation, the appropriate diary pool is
identified and a CHAD dairy is randomly drawn. When a diary for every day in the simulation
period has been selected, they are concatenated into a single longitudinal diary covering the
entire simulation for that individual. APEX contains three algorithms for stochastically selecting
diaries from the pools to create the longitudinal diary. The first method selects diaries at random
after stratification by age, gender, and diary pool; the second method selects diaries based on
metrics related to exposure (e.g., time spent outdoors) with the goal of creating longitudinal
F-7
-------
diaries with variance properties designated by the user (Glen et al., 2008); and the third method
uses a clustering algorithm to obtain more realistic recurring behavioral patterns (Rosenbaum
2008).
The final step in processing the activity diary is to map the CHAD location codes into the
set of APEX microenvironments, supplied by the user as an input file. The user may define the
number of microenvironments, from one up to the number of different CHAD location codes.
F.6 Key Physiological Processes Modeled
Ventilation is a general term describing the movement of air into and out of the lungs.
The rate of ventilation is determined by the type of activity an individual performs which in turn
is related to the amount of oxygen required to perform the activity. Minute or total ventilation
rate is used to describe the volume of air moved in or out of the lungs per minute. Quantitatively,
the volume of air breathed in per minute (Vf ) is slightly greater than the volume expired per
minute (VE ). Clinically, however, this difference is not important, and by convention, the
ventilation rate is always measured by the expired volume.
The rate of oxygen consumption (V 02) is related to the rate of energy usage in
performing activities as follows:
V02 =EE x ECF Equation (F-2)
where,
V02 = Oxygen consumption rate (liters Ch/minute)
EE = Energy expenditure (kcal/minute)
ECF = Energy conversion factor (liters Ch/kcal).
The ECF shows little variation and typically, commonly a value between 0.20 and 0.21 is
used to represent the conversion from energy units to oxygen consumption. APEX can randomly
sample from a uniform distribution defined by these lower and upper bounds to estimate an ECF
for each simulated individual. The activity-specific energy expenditure is highly variable and can
be estimated using metabolic equivalents (METs), or the ratios of the rate of energy consumption
for non-rest activities to the resting rate of energy consumption, as follows
EE =MET x RMR Equation F-3
F-8
-------
where,
EE = Energy expenditure (kcal/minute)
MET = Metabolic equivalent of work (unitless)
RMR = Resting metabolic rate (kcal/minute)
APEX contains distributions of METs for all activities that might be performed by
simulated individuals. APEX randomly samples from the various METs distributions to obtain
values for every activity performed by each individual. Age- and sex-specific RMR are estimated
once for each simulated individual using a linear regression model developed based on use BM,
age, and the natural logarithms of BM and (age+1) (Equation F-4).2 Details regarding the model
derivation, ergression coefficient values, and performance evaluation are provided in Appendix
H.
RMR = /?0 + PiBM + /?2 log(BM) + P3Age + (34\og(Age) + Et Equation F-4
APEX also contains an algorithm that accounts for variability in ventilation rate (VE)
due to variation in oxygen consumption (V02). The approach indirectly considers influential
variables such as age, sex, and body mass by use of an individual's maximum MET (or,
equivalently, by VChm), thus the variability within age groups, and both inter- and intra-personal
and variability are also accounted for. Appendix H describes this new algorithm, derived using
the same clinical study data used in developing the former APEX algorithm (Graham and
McCurdy, 2005), though as
Yj? — g(3-300 + o.8i28xin_vo2+ 0.5126 x (yo2^vo2Tn)4+N(o,eb)+N(o,ew)) Equation F-5
F.7 Estimating Microenvironmental Concentrations
The user provides rules for determining the pollutant concentration in each
microenvironment. There are two available models for calculating microenvironmental
concentrations: mass balance and regression factors. Any indoor microenvironment may use
2 The "+1" modifier allows APEX to round age upwards instead of downwards to whole years, which is necessary to
avoid undefined log(0) values.
F-9
-------
either model; for each microenvironment, the user specifies whether the mass balance or factors
model will be used.
F.7.1 Mass Balance Model
The mass balance method assumes that an enclosed microenvironment (e.g., a room
within a home) is a single well-mixed volume in which the air concentration is approximately
spatially uniform. The concentration of an air pollutant in such a microenvironment is estimated
using the following four processes (and illustrated in Figure F-l):
• Inflow of air into the microenvironment;
• Outflow of air from the microenvironment;
• Removal of a pollutant from the microenvironment due to deposition, filtration, and
chemical degradation; and
• Emissions from sources of a pollutant inside the microenvironment.
Microenvironment
Figure F-l. Illustration of the mass balance model used by APEX.
Considering the microenvironment as a well-mixed fixed volume of air, the mass balance
equation for a pollutant in the microenvironment can be written in terms of concentration:
Air
outflow
Air
inflow
Removal due to:
•Chemical reactions
•Deposition
•Filtration
Indoor sources
removal
+c,
source
Equation F-6
F-10
-------
where,
C(t) = Concentration in the microenvironment at time t
C m = Rate of change in C(t) due to air entering the microenvironment
C out = Rate of change in C(t) due to air leaving the microenvironment
C removal Rate of change in C(t) due to all internal removal processes
C source = Rate of change in C(t) due to all internal source terms
Concentrations are calculated in the same units as the ambient air quality data, e.g., ppm,
ppb, ppt, or |ig/m3. In the following equations concentration is shown only in |ig/m3 for brevity.
The change in microenvironmental concentration due to influx of air, C m, is given by:
^in = Coutdoor x ^penetration x ^airexchange Equation F-7
where,
Coutdoor = Ambient concentration at an outdoor microenvironment or outside an
indoor microenvironment (|ig/m3)
/penetration = Penetration factor (unitless)
Rair exchange = Air exchange rate (hr"1)
Because the air pressure is approximately constant in microenvironments that are
modeled in practice, the flow of outside air into the microenvironment is equal to that flowing
out of the microenvironment, and this flow rate is given by the air exchange rate. The air
exchange rate (hr"1) can be loosely interpreted as the number of times per hour the entire volume
of air in the microenvironment is replaced. For some pollutants (especially particulate matter),
the process of infiltration may remove a fraction of the pollutant from the outside air. The
fraction that is retained in the air is given by the penetration factor jpenetration.
A proximity factor (/proximity) and a local outdoor source term are used to account for
differences in ambient concentrations between the geographic location represented by the
ambient air quality data (e.g., a regional fixed-site monitor) and the geographic location of the
microenvironment. That is, the outdoor air at a particular location may differ systematically from
the concentration input to the model representing the air quality district. For example, a
playground or house might be located next to a busy road in which case the air at the playground
or outside the house would have elevated levels for mobile source pollutants such as carbon
monoxide and benzene. The concentration in the air at an outdoor location or directly outside an
indoor microenvironment (Coutdoor) is calculated as:
F-ll
-------
Coutdoor ~ fproximty ^arrbient + ^Loc^Outdo orSources Equation F-8
where,
CamHent = Ambient air district concentration (|ig/m3)
/proximity = Proximity factor (unitless)
CLocaiOutdoorSources = the contribution to the concentration at this location from local
sources not represented by the ambient air district concentration (|ig/m3)
During exploratory analyses, the user may examine how a microenvironment affects
overall exposure by setting the microenvironment's proximity or penetration factor to zero, thus
effectively eliminating the specified microenvironment. Change in microenvironmental
concentration due to outflux of air is calculated as the concentration in the microenvironment
C(t) multiplied by the air exchange rate:
COLt =Rairexcha,ve *C(f) Equation F-9
The third term (C removal) in the mass balance calculation (Equation F-6) represents
removal processes within the microenvironment. There are three such processes in general:
chemical reaction, deposition, and filtration. Removal can be important for pollutants such as O3
and SO2, for example, but not for carbon monoxide. The amount lost to chemical reactions will
generally be proportional to the amount present, which in the absence of any other factors would
result in an exponential decay in the concentration with time. Similarly, deposition rates are
usually given by the product of a (constant) deposition velocity and a (time-varying)
concentration, also resulting in an exponential decay. The third removal process is filtration,
usually as part of a forced air circulation or HVAC system. Filtration will normally be more
effective at removing particles than gases. In any case, filtration rates are also approximately
proportional to concentration. Change in concentration due to deposition, filtration, and chemical
degradation in a microenvironment is simulated based on the first-order equation:
^rerm/al ~ deposition ^filtration ^cherrical )*^(0 Equation F 10
~ Rrerm/al x ^(0
where,
F-12
-------
C removal Change in microenvironmental concentration due to removal processes
(|ig/m3/hr)
^-deposition = Removal rate of a pollutant from a microenvironment due to deposition
(hr"1)
RfUtration = Removal rate of a pollutant from a microenvironment due to filtration
(hr1)
Rchemicai = Removal rate of a pollutant from a microenvironment due to chemical
degradation (hr-1)
Rremovai = Removal rate of a pollutant from a microenvironment due to the
combined effects of deposition, filtration, and chemical degradation (hr-1)
The fourth term in the mass balance calculation represents pollutant sources within the
microenvironment. This is the most complicated term, in part because several sources may be
present. APEX allows two methods of specifying source strengths: emission sources and
concentration sources. Either may be used for mass balance microenvironments, and both can be
used within the same microenvironment. The source strength values are used to calculate the
term C SOUrce (|ig/m3/hr).
Emission sources are expressed as emission rates in units of |ig/hr, irrespective of the
units of concentration. To determine the rate of change of concentration associated with an
emission source Se, it is divided by the volume of the microenvironment:
a SE
Csource£E =~^~ Equation F-ll
where,
C source,SE = Rate of change in C(t) due to the emission source Se (|ig/m3/hr)
Se = The emission rate (|ig/hr)
V = The volume of the microenvironment (m3)
Concentration sources (Sc) however, are expressed in units of concentration. These must
be the same units as used for the ambient concentration (e.g., |ig/m3). Concentration sources are
normally used as additive terms for microenvironments using the factors model. Strictly
speaking, they are somewhat inconsistent with the mass balance method, since concentrations
should not be inputs but should be consequences of the dynamics of the system. Nevertheless, a
suitable meaning can be found by determining the rate of change of concentration (C source) that
F-13
-------
would result in a mean increase of Sc in the concentration, given constant parameters and
equilibrium conditions, in this way:
Assume that a microenvironment is always in contact with clean air (ambient = zero), and
it contains one constant concentration source. Then the mean concentration over time in this
microenvironment from this source should be equal to Sc. The mean source strength expressed in
ppm/hr or |ig/m3/hr is the rate of change in concentration (C source,sc). In equilibrium,
Cs = C source,sc Equation F-12
Rair exchange + Rremoval
where, Cs is the mean increase in concentration over time in the microenvironment due to
the source C source,SC • Thus, C source,sc can be expressed as
Csource,sc = Cs x Rmean Equation F-13
where Rmean is the chemical removal rate. From Equation (F-13), Rmean is the sum of the
air exchange rate and the removal rate {Rair exchange + Rremovd) under equilibrium conditions. In
general, however, the microenvironment will not be in equilibrium, but in such conditions there
is no clear meaning to attach to C source,sc since there is no fixed emission rate that will lead to a
fixed increase in concentration. The simplest solution is to use R mean Rair exchange Rremoval-
However, the user is given the option of specifically specifying Rmean (see discussion below).
This may be used to generate a truly constant source strength C source,sc by making Sc and Rmean
both constant in time. If this is not done, then Rmean is simply set to the sum of {Rair exchange +
Rremoval). If these parameters change over time, then C source,sc also changes. Physically, the
reason for this is that in order to maintain a fixed elevation of concentration over the base
conditions, then the source emission rate would have to rise if the air exchange rate were to rise.
Multiple emission and concentration sources within a single microenvironment are
combined into the final total source term by combining Equations (F-l 1) and (F-13):
1 ne nc
Csource ~ ^source,SE + Csource,SC ~ 77 ^i + ^mean / Equation F-l4
V i=1 i=1
where,
F-14
-------
Sei = Emission source strength for emission source i (|ig/hr, irrespective
of the concentration units)
Sa = Emission source strength for concentration source i (|ig/m3)
ne = Number of emission sources in the microenvironment
nc = Number of concentration sources in the microenvironment
In Equations (F-11) and (F-14), if the units of air quality are ppm rather than |ig/m3, I V
is replaced by f/V, where/= ppm / |ig/m3 = gram molecular weight / 24.45 (i.e., 24.45 being the
volume (liters) of a mole of the gas at 25°C and 1 atmosphere pressure). Equations (F-7), (F-9),
(F-10), and (F-14) can now be combined with Equation (F-6) to form the differential equation for
the microenvironmental concentration C(t). Within the time period of a time step (at most 1
hour), C source and C i„ are assumed to be constant. Using C combined = C source + C m leads to:
_p _o C(t)-R f(f)
.. ^corrbined air exchange V / retwval V / ^ i r
at Equation F-15
— contined ^ mean ^(0
Solving this differential equation leads to:
C(t) = -
'corrbined
R.
'confined
Rn
, (f-*o )
Equation F-16
where,
C(to) = Concentration of a pollutant in a microenvironment at the
beginning of a time step (|ig/m3)
C(t) = Concentration of a pollutant in a microenvironment at time t within
the time step (|ig/m3).
Based on Equation (F-16), the following three concentrations in a microenvironment are
calculated:
Cequil =C(t^r:)=Ccombmed = C source+ Cin EquationF-17
Rmean ^airexchange + ^removal
C(t0 + T) = Cequil + (C(t0)-Cequil)e RmsanT Equation F-18
F-15
-------
1 ta+T m -R T
1 » / Q "mean1
Cn^ = - j C(t)dt = C^, + (C(t0)~ C^,)— — Equation F-19
7 j \/ — equi \\u / equal / r-\ -j-
mean 1
where,
Cequii = Concentration in a microenvironment (|ig/m3) if t > x
(equilibrium state).
C(to) = Concentration in a microenvironment at the beginning of the time
step (|ig/m3)
C(to+T) = Concentration in a microenvironment at the end of the time step
(|ig/m3)
Cmean = Mean concentration over the time step in a microenvironment
(|ig/m3)
Rmean Rair exchange Rremoval (hr )
At each time step of the simulation period, APEX uses Equations (F-17), (F-18), and
(5A-19) to calculate the equilibrium, ending, and mean concentrations, respectively. The
calculation continues to the next time step by using C(to+T) for the previous hour as C(to).
F.7.2 Factors Model
The factors model is simpler than the mass balance model. In this method, the value of
the concentration in a microenvironment is not dependent on the concentration during the
previous time step. Rather, this model uses the following equation to calculate the concentration
in a microenvironment from the user-provided hourly air quality data:
"c
^mean = ^arrtient ^proximty fpenetratio n + ^ ^Ci Equation F-20
i=1
where,
Cmean = Mean concentration over the time step in a microenvironment (|ig/m3)
Cambient= The concentration in the ambient (outdoor) environment (|ig/m3)
/proximity = Proximity factor (unitless)
/penetration = Penetration factor (unitless)
Sa = Mean air concentration resulting from source i (|ig/m3)
nc = Number of concentration sources in the microenvironment
F-16
-------
The user may specify distributions for proximity, penetration, and any concentration
source terms. All of the parameters in Equation (F-20) are evaluated for each time step, although
these values might remain constant for several time steps or even for the entire simulation.
The ambient air quality data are supplied as time series over the simulation period at
several locations across the modeled region. The other variables in the factors and mass balance
equations are randomly drawn from user-specified distributions. The user also controls the
frequency and pattern of these random draws. Within a single day, the user selects the number of
random draws to be made and the hours to which they apply. Over the simulation, the same set
of 24 hourly values may either be reused on a regular basis (for example, each winter weekday),
or a new set of values may be drawn. The usage patterns may depend on day of the week, on
month, or both. It is also possible to define different distributions that apply if specific conditions
are met. The air exchange rate is typically modeled with one set of distributions for buildings
with air conditioning and another set of distributions for those which do not. The choice of a
distribution within a set typically depends on the outdoor temperature and possibly other
variables. In total there are eleven such conditional variables which can be used to select the
appropriate distributions for the variables in the mass balance or factors equations.
For example, the hourly emissions of CO from a gas stove may be given by the product
of three random variables: a binary on/off variable that indicates if the stove is used at all during
that hour, a usage duration sampled from a continuous distribution, and an emission rate per
minute of usage. The binary on/off variable may have a probability for on that varies by time of
day and season of the year. The usage duration could be taken from a truncated normal or
lognormal distribution that is resampled for each cooking event, while the emission rate could be
sampled just once per stove.
F.8 Exposure and Dose Time Series Calculations
The activity diaries provide the time sequence of microenvironments visited by the
simulated individual and the activities performed by each individual. The pollutant concentration
in the air in each microenvironment is assumed to be spatially uniform throughout the
microenvironment and unchanging within each diary event and is calculated by either the factors
or the mass balance method, as specified by the user. The exposure of the individual is given by
the time sequence of airborne pollutant concentrations that are encountered in the
microenvironments visited. Figure F-2 illustrates the exposures for one simulated 12-year old
child over a 2-day period. On both days the child travels to and from school in an automobile,
goes outside to a playground in the afternoon while at school, and spends time outside at home in
the evening.
F-17
-------
ppm
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0.00
00:00 06:00 12:00 18:00 00:00 06:00 12:00 18:00 00:00
time of day
Figure F-2. Example of microenvironmental and exposure concentrations for a simulated
individual over a 48-hour duration. (H: home, A: automobile, S: school, P: playground, O:
outdoors at home).
In addition to exposure, APEX models breathing rates based on the physiology of each
individual and the exertion levels associated with the activities performed. For each activity type
in CHAD, a distribution is provided for a corresponding normalized metabolic equivalent of
work or METs (McCurdy, 2000). METs are derived by dividing the metabolic energy
requirements for the specific activity by a person's resting, or basal, metabolic rate. The MET
ratios have less interpersonal variation than do the absolute energy expenditures. Based on age
and sex, the resting metabolic rate, along with other physiological variables is determined for
each individual as part of their anthropometric characteristics. Because the MET ratios are
sampled independently from distributions for each diary event, it would be possible to produce
time-series of MET ratios that are physiologically unrealistic. APEX employs a MET adjustment
algorithm based on a modeled oxygen deficit to prevent such overestimation of MET and
breathing rates (Isaacs et al., 2008). The relationship between the oxygen deficit and the applied
limits on MET ratios are nonlinear and are derived from published data on work capacity and
oxygen consumption. The resulting combination of microenvironmental concentration and
breathing ventilation rates provides a time series of inhalation intake dose for most pollutants.
hh
0
o
p
p 0
A A O
O
—i 1 1 ¦ 1 ¦ 1 1 1 1 1 ¦ 1 • r
F-18
-------
F.9 Model Output
APEX calculates the exposure and dose time series based on the events as listed on the
activity diary with a minimum of one event per hour but usually more during waking hours.
APEX can aggregate the event level exposure and dose time series to output hourly, daily,
monthly, and annual averages. The types of output files are selected by the user, and can be as
detailed as event-level data for each simulated individual (note, Figure F-3 was produced from
an APEX event output file). A set of summary tables are produced for a variety of exposure and
dose measures. These could include tables of person-minutes at various exposure levels, by
microenvironment, a table of person-days at or above each average daily exposure level, and
tables describing the distributions of exposures for different groups. An example of how APEX
results can be depicted is given Figure F-3 which shows the percent of children with at least one
5-minute maximum exposure at or above different exposure levels, concomitant with moderate
or greater exertion. These are results from a simulation of SO2 exposures for Fall River, MA
during 2011. From this graph it can be observed, for example, that APEX estimates 15 percent of
the children in this area experienced a daily maximum 5-minute SO2 exposure above 100 ppb
while exercising, at least once during the year.
\
\
\
>
-
-
-
-
—1—1—1—1—
—'—'—'—'—
! ! ! !
' * * ' * ' <
50 100 150 200
Daily Maximum 5-minute S02 Exposure (ppb) for 2011
250
Figure F-3. The percent of simulated children (ages 5-18) experiencing at least one daily
maximum 5-minute SO2 exposure during 2011, while at moderate or greater exertion.
F-19
-------
REFERENCES
Burmaster, D.E. (1998). LogNormal distributions for skin area as a function of body weight. Risk
Analysis. 18(l):27-32.
Glen, G., Smith, L., Isaacs, K., McCurdy, T., Langstaff, J. (2008). A new method of longitudinal
diary assembly for human exposure modeling. J Expos Sci Environ Epidem. 18:299-311.
Graham, S.E., McCurdy, T. (2005). Revised ventilation rate (VE) equations for use in inhalation-
oriented exposure models. Report no. EPA/600/X-05/008 is Appendix A of US EPA
(2009). Metabolically Derived Human Ventilation Rates: A Revised Approach Based
Upon Oxygen Consumption Rates (Final Report). Report no. EPA/600/R-06/129F.
Available at: http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=202543.
Isaacs, K., Glen, G., McCurdy, T., Smith, L. (2008). Modeling energy expenditure and oxygen
consumption in human exposure models: accounting for fatigue and EPOC. J Expos Sci
Environ Epidemiol. 18:289-298.
Johnson, T., Capel, J., McCoy, M. (1996a). Estimation of Ozone Exposures Experienced by
Urban Residents Using a Probabilistic Version of NEM and 1990 Population Data.
Prepared by IT Air Quality Services for the Office of Air Quality Planning and
Standards, U.S. Environmental Protection Agency, Research Triangle Park, North
Carolina, September.
Johnson, T., Capel, J., Mozier, J., McCoy, M. (1996b). Estimation of Ozone Exposures
Experienced by Outdoor Children in Nine Urban Areas Using a Probabilistic Version of
NEM. Prepared for the Air Quality Management Division under Contract No. 68-DO-
30094, April.
Johnson, T., Capel, J., McCoy, M., Mozier, J. (1996c). Estimation of Ozone Exposures
Experienced by Outdoor Workers in Nine Urban Areas Using a Probabilistic Version of
NEM. Prepared for the Air Quality Management Division under Contract No. 68-DO-
30094, April.
McCurdy, T. (2000). Conceptual basis for multi-route intake dose modeling using an energy
expenditure approach. J Expo Anal Environ Epidemiol. 10:1-12.
McCurdy T, Glen G, Smith L, Lakkadi Y. (2000). The National Exposure Research Laboratory's
Consolidated Human Activity Database. J Expos Anal Environ Epidemiol. 10:566-578.
Rosenbaum, A. S. (2008). The Cluster-Markov algorithm in APEX. Memorandum prepared for
Stephen Graham, John Langstaff. USEPA OAQPS by ICF International.
U.S. EPA. (2002). Consolidated Human Activities Database (CHAD) Users Guide. Database and
documentation available at: http://www.epa.gov/chadnetl/.
U.S. EPA. (2017a). Air Pollutants Exposure Model Documentation (APEX, Version 5)
Volume I: User's Guide. Office of Air Quality Planning and Standards, U.S.
F-20
-------
Environmental Protection Agency, Research Triangle Park, NC, 27711. EPA-452/R-17-
001a. Available at: https://www.epa.gov/fera/apex-user-aiides
U.S. EPA. (2017b). Air Pollutants Exposure Model Documentation (APEX, Version 5)
Volume II: Technical Support Document. Office of Air Quality Planning and Standards,
U.S. Environmental Protection Agency, Research Triangle Park, NC, 27711. EPA-452/R-
17-00lb. Available at: https://www.epa.eov/fera/apex-iiser-euides
U.S. EPA. (2017c). The Consolidated Human Activity Database - Master Version (CHAD-
Master). Technical Memorandum. U.S. Environmental Protection Agency, National
Exposure Research Laboratory, Research Triangle Park, NC, 27711. In preparation.
Previous version (09/15/2014) available at:
https://www.epa.eov/healthresearch/consolidated-human-activitv-database-chad-use-
human-exposure-and-health-studies-and
F-21
-------
APPENDIX G
ICF FINAL MEMO: JOINT DISTRIBUTIONS OF BODY WEIGHT
AND HEIGHT FOR USE IN APEX
G-l
-------
f
'ICF
Draft Memorandum
To:
Re:
From:
Date:
John Langstaff, Stephen Graham, Kristin Isaacs, U.S. Environmental Protection
Agency
Jonathan Cohen, Graham Glen, John Hader, Chris Holder, ICF
April 20, 2017
Joint Distributions of Body Weight and Height for use in APEX (Revised from October
26, 2016 version to add Section 6).
1. Introduction and Summary
The current version of APEX uses fitted distributions for body weight (BW; also referred to as
body mass) based on an analysis of the data from the National Health and Nutrition
Examination Survey (NHANES) for the years 1999-2004. These distributions were developed in
2005.1 The current version of APEX also uses fitted distributions for height (HT) based on fitted
regressions for HT against age for children under 18 years of age and fitted regressions for HT
against the logarithm of BW for adults 18 years and older. The regression coefficients for
children depend upon the age group and gender.2 ICF was tasked with updating these BW
fitted distributions to use more recent NHANES data and to compute parameters for the
joint distribution of BW and HT.
We downloaded and analyzed BW and HT data from NHANES for the years 2003-2014. We
fitted distributions for the entire period 2003-2014 and also for the more recent period 2009-
2014. As shown in Section 5, the final fitted models were very similar for the 2003-2014 and
2009-2014 periods. In this memorandum, we present detailed results for the 2009-2014
analysis. We provide the final parameter estimates for both groups of years in accompanying
Excel spreadsheets. We can provide the detailed analyses for 2003-2014 upon request.
In Section 2, we present histograms and summary tables for the marginal distributions of BW
and HT for each gender and single year of age. We compared fitted normal and log-normal
distributions using the histograms and log-likelihoods and determined that the best overall
choice was a log-normal distribution for BW and a normal distribution for HT. To allow a
smooth set of parameters for different ages, we chose the same distributional forms (but
different parameters) for each combination of gender and age.
In Section 3, we model the joint distribution of BW and HT as a bivariate normal
distribution for the HT and the logarithm of the BW, with different parameters for each
age and gender. We present scatter plots for selected single years of age.
1 Kristin Isaacs and Luther Smith, Alion Science and Technology, "New Values for Physiological Parameters for the
Exposure Model Input File Physiology.txt". Memorandum to Tom McCurdy, EPA. December 20, 2005.
2 Johnson T, Mihlan G, LaPointe J, Fletcher K, Capel J, Rosenbaum A, Cohen J, Stiefer P. 2000. Estimation of
carbon monoxide exposures and associated carboxyhemoglobin levels for residents of Denver and Los Angeles
using pNEM/CO. Appendices. EPA contract 68-D6-0064.
2635 Meridian Pkwy., Suite 200 ¦ Durham, NC 27713 - 919.293.1620 ¦ 919.293.1645 fax ¦ icf.com
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 2
As shown in Section 4, the estimated parameters for each age do not vary smoothly across the
ages. Therefore, we used a natural cubic spline model to smooth each of the five
parameters across the different ages for each gender. This approach also allowed us to
smoothly extrapolate the parameters for ages 80 to 100, since the NHANES data for recent
periods combines all ages 80 and above into a single age group.
In Section 5 we compare the fitted parameters between the NHANES periods 2009-2014 and
2003-2014 and show that, after smoothing the parameters, the maximum unsigned percentage
difference is 11 percent for the correlation coefficient and less than 1 percent for the means.
Finally, in Section Error! Reference source not found, we compare summaries of the HT, BW,
and body mass index from the Personal Summary files generated by running APEX with the old
and updated method for calculating height and weight. There is now a better correlation
between HT, WT, and age for young children and older adults. Average BW values tend to be
larger with the new method, likely reflecting ongoing trends in BW of the U.S. population, and
simulated body mass indices are roughly in line with NHANES data.
2. Marginal Distributions of BW and HT
V' ; ¦ v. r.
For each of the NHANES cycles (2-year periods), we downloaded the age, HT, BW, and survey
weights for each sampled person by merging the demographic file with the body-measurements
file. We selected the variables discussed below.
Age
For 2003-2004 and 2005-2006, RIDAGEEX is the age in months at the time of examination for
individuals of ages 0-84 years, and RIDAGEYR is the age in years at the time of screening for
all individuals. We used RIDAGEEX to calculate the age in years for individuals under 84
(integer part of RIDAGEEX/12) and RIDAGEYR for individuals 85 and over. We assigned the
age group code "1000" to all individuals 80 and over.
For 2007-2008 and 2009-2010, RIDAGEEX is the age in months at the time of examination for
individuals of ages 0-79 years, and RIDAGEYR is the age in years at the time of screening for
all individuals. We used RIDAGEEX to calculate the age in years for individuals under 80
(integer part of RIDAGEEX/12) and RIDAGEYR for individuals 80 and over. We assigned the
age group code "1000" to all individuals 80 and over.
For 2009-2010 and 2011-2012, RIDEXAGM is the age in months at the time of examination for
individuals of ages 0-19 years, and RIDAGEYR is the age in years at the time of screening for
all individuals. We used RIDEXAGM to calculate the age in years for individuals under 20
(integer part of RIDEXAGM/12) and RIDAGEYR for individuals 20 and over. We assigned the
age group code "1000" to all individuals 80 and over.
Gender
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 3
NHANES codes gender using Males = 1 and Females = 2.
HT
For individuals of ages 2 years and older, we used the NHANES variable BMXHT, which is the
standing HT (cm). For children of ages 0 or 1 years, we used the NHANES variable
BMXRECUM, which is the recumbent HT (cm); for programming convenience we renamed this
variable as BMXHT.
BW
For all individuals, we used the NHANES variable BMXWT, which is the BW (kg).
Survey Weight
The NHANES survey weight variable for each 2-year period is WTMEC2YR, which estimates
the number of people in the U.S. population at the mid-year of the survey period represented by
the sampled individual. Since the NHANES survey was designed to over-sample certain
demographic groups (e.g., Mexican-Americans from 2003-2006 and Hispanics from 2006-
2014), the survey weights are needed to adjust the data to represent the U.S. population.
With two exceptions, all of the analyses in this memorandum used the survey weights to adjust
the data. One of these exceptions is for the histogram plots in the next sub-section, which used
the survey weights rounded to the nearest integer because SAS does not allow fractional
weights for those plots. A second exception is for the natural cubic spline smoothing of the
parameter estimates described in Section 4; the survey weights were used in the calculations of
the unsmoothed parameters but it would not have been appropriate to use them for the final
smoothing step.
tog ranis
Figure 2-1 and Figure 2-2 below are histograms of the BW (kg) and HT (cm; standing HT for
ages 2 and over, recumbent HT for ages 0 and 1), respectively, for each gender and selected
single years of age (the selected ages shown are 1, 5, 10, 15, 20, 25, 30, 40, 60, 70, and 79
years). Superimposed on each histogram are fitted normal and log-normal distributions. The
calculations use the survey weights rounded to the nearest integer (making a negligible error,
since the survey weights are usually several thousand). For BW (Figure 2-1), the distributions
are generally right-skewed and the log-normal distribution appears to fit the data better
than the normal distribution. For HT (Figure 2-2), the distributions are almost symmetric
and it is hard to distinguish the two fitted distributions on the plots. We provide larger
versions of the histograms in Figure 2-1 and Figure 2-2 in Attachment A and Attachment B,
respectively.
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 4
Gender = 1
Gender = 2
a
JV
0 5 10 15 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55
fa
fa
J"
^1—
J
/
K,
60 -
50 -
40 -
30 -
20 -
10 -
60 -
50 -
40 -
30 -
20 -
10 -
120 160 200
120 160 200
A
y\
20 60 100 140 180 220 20 60 100 140 180 220
60 -
50 -
40 -
30 -
20 -
10 -
60 -
50 -
40 -
30 -
20 -
10 -
J-
FK
50 -
40 "
30 -
20 -
10 -
50 -
40 "
30 -
20 -
10 -
0 40 80 120 160 200 0 40 80 120 160 200
J
/;
-A
h
-edi
n\
20 60 100 140 180 220 20 60 100 140 180 220
60 "
40 -
20 -
60 "
40 -
20 -
Pvl
0 40 80 120 160 200 0 40 80 120 160 200
Weight (kg)
| Curves Noimal Lognoimal(Theta=0 Sigma=EST Zeta=EST)~|
50 70 90 110 130 150
A
ft
A
50 70 90 110 130 150
J
ft
110 130 150_ 170 _ 190 210 110 130 150 170_ 190 210
ft
rJ
IV
7
TV
\
X.
J
L
h,
140 150 160 170 180 190 140 150 160 170 180 190
ft
J-
k
ft
x
r
J-
jt|
I
PL
j
J
fL
/
J
K
J\
\
f
J
V
/
\
120 140 160 180 200 220 120 140 160 180 200 220
Heieht (cm)
Figure 2-1. Distributions of BW
Figure 2-2. Distributions of HT
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 5
2.3. Summary Statistics
Table 1 below contains the estimated means ("Mean") and standard deviations ("Std Dev") for
the BW and its natural logarithm ("Log") for each age group and gender. The row for age "1000"
corresponds to ages 80 and older; the summary statistics for this group are shown for
comparison purposes but are not used for the final set of distributions which are only based on
the data for ages 0-79 years. Distributions are fitted separately to each combination of gender
and either a single year of age from 0 to 79 years or the age group 80 years and older. We
weighted the means and standard deviations across the sampled individuals using the exact
survey weights.
To compare the fit of the normal and log-normal distributions, we tabulated the likelihood values.
If f(x) is the probability density function for x (either a log-normal or normal distribution), then
—2LL = -2 x I SWi x log{f(xi)}, where SWi and x are the survey weight and BW, respectively,
for the i'th individual of the given age group and gender. (We omitted the constant term from
f(x)). The value -2LL estimates the corresponding value of minus twice the log-likelihood for the
population. Based on the likelihood method, the better of the two models (normal or log-normal)
will have a lower value of -2LL; this determination is shown in the column "Best."
For the vast majority of cases, the log-normal model is preferred for BW. This pattern is
consistent with the histograms shown above. Since the results of the APEX simulations should
not be too sensitive to the exact ages of the modeled population, it is better to use the same
distribution for all ages and genders, which suggests that BW should be modeled as a log-
normal distribution for all demographic groups.
Table 1. Summary Statistics for BW
Mean
Mean Log
Std Dev
Std Dev
-2LL
-2LL
Age Gender
BW
BW
BW
Log BW
Best
Normal
Log-Normal
0 1
7.815
2.024
1.933
0.261
Normal
1.30E+07
1.32E+07
1 1
11.443
2.429
1.451
0.126
Lognormal
1.02E+07
9.99E+06
2 1
14.130
2.640
1.850
0.126
Lognormal
1.32E+07
1.26E+07
3 1
16.162
2.773
2.436
0.139
Lognormal
1.94E+07
1.81 E+07
4 1
18.693
2.915
3.152
0.157
Lognormal
2.24E+07
2.13E+07
5 1
21.347
3.045
4.002
0.172
Lognormal
2.14E+07
2.02E+07
6 1
23.789
3.149
5.344
0.191
Lognormal
2.81 E+07
2.57E+07
7 1
27.870
3.298
7.526
0.234
Lognormal
3.34E+07
3.11 E+07
8 1
31.112
3.407
8.244
0.241
Lognormal
3.62E+07
3.45E+07
9 1
34.679
3.513
9.531
0.249
Lognormal
3.38E+07
3.22E+07
10 1
40.133
3.656
11.645
0.263
Lognormal
3.49E+07
3.33E+07
11 1
48.057
3.832
14.351
0.280
Lognormal
3.48E+07
3.36E+07
12 1
50.746
3.894
13.498
0.252
Lognormal
3.99E+07
3.88E+07
13 1
60.002
4.060
16.631
0.256
Lognormal
4.08E+07
3.94E+07
14 1
65.258
4.143
18.467
0.259
Lognormal
5.00E+07
4.82E+07
15 1
71.356
4.234
19.846
0.255
Lognormal
4.14E+07
4.00E+07
16 1
74.894
4.289
18.367
0.226
Lognormal
4.57E+07
4.43E+07
17 1
77.237
4.317
20.101
0.235
Lognormal
4.23E+07
4.07E+07
18 1
81.164
4.363
23.222
0.248
Lognormal
4.39E+07
4.18E+07
19 1
79.636
4.350
19.629
0.229
Lognormal
4.71 E+07
4.57E+07
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 6
Mean
Mean Log
Std Dev
Std Dev
-2LL
-2LL
Age Gender
BW
BW
BW
Log BW
Best
Normal
Log-Normal
20 1
79.206
4.341
20.898
0.246
Lognormal
5.21 E+07
5.07E+07
21 1
79.075
4.342
20.585
0.231
Lognormal
4.61 E+07
4.41 E+07
22 1
81.032
4.368
20.166
0.224
Lognormal
4.43E+07
4.26E+07
23 1
86.142
4.418
25.256
0.269
Lognormal
4.57E+07
4.41 E+07
24 1
82.705
4.396
16.561
0.192
Lognormal
4.27E+07
4.19E+07
25 1
85.955
4.422
22.691
0.248
Lognormal
4.40E+07
4.29E+07
26 1
86.496
4.437
19.619
0.213
Lognormal
3.59E+07
3.50E+07
27 1
86.016
4.433
18.552
0.207
Lognormal
3.80E+07
3.73E+07
28 1
88.812
4.459
21.574
0.230
Lognormal
4.74E+07
4.62E+07
29 1
89.171
4.467
20.015
0.215
Lognormal
4.63E+07
4.54E+07
30 1
88.645
4.458
21.090
0.233
Lognormal
4.87E+07
4.80E+07
31 1
88.916
4.465
19.163
0.211
Lognormal
3.86E+07
3.81 E+07
32 1
91.226
4.486
22.585
0.230
Lognormal
4.54E+07
4.41 E+07
33 1
92.027
4.500
19.719
0.208
Lognormal
3.85E+07
3.79E+07
34 1
87.439
4.451
17.985
0.194
Lognormal
3.33E+07
3.26E+07
35 1
88.897
4.461
21.560
0.228
Lognormal
3.94E+07
3.84E+07
36 1
92.644
4.498
25.114
0.240
Lognormal
4.54E+07
4.36E+07
37 1
93.184
4.512
21.813
0.204
Lognormal
4.11 E+07
3.92E+07
38 1
93.366
4.514
20.963
0.210
Lognormal
3.89E+07
3.79E+07
39 1
90.726
4.483
20.780
0.219
Lognormal
4.24E+07
4.16E+07
40 1
92.532
4.504
20.717
0.212
Lognormal
4.58E+07
4.48E+07
41 1
94.364
4.522
22.769
0.218
Lognormal
4.73E+07
4.56E+07
42 1
90.804
4.491
17.670
0.189
Lognormal
3.59E+07
3.54E+07
43 1
92.679
4.510
19.518
0.192
Lognormal
4.57E+07
4.43E+07
44 1
93.069
4.512
20.205
0.202
Lognormal
4.53E+07
4.41 E+07
45 1
88.197
4.463
16.018
0.182
Lognormal
3.79E+07
3.77E+07
46 1
90.498
4.485
18.381
0.200
Lognormal
4.43E+07
4.38E+07
47 1
90.870
4.493
17.327
0.180
Lognormal
4.41 E+07
4.31 E+07
48 1
90.708
4.482
21.347
0.221
Lognormal
4.08E+07
3.98E+07
49 1
90.907
4.488
19.250
0.208
Lognormal
4.00E+07
3.95E+07
50 1
94.131
4.524
20.593
0.199
Lognormal
4.70E+07
4.55E+07
51 1
86.258
4.432
20.135
0.221
Lognormal
3.66E+07
3.57E+07
52 1
92.086
4.501
19.609
0.205
Lognormal
4.26E+07
4.19E+07
53 1
90.250
4.479
19.589
0.215
Lognormal
4.25E+07
4.21 E+07
54 1
93.833
4.521
19.125
0.204
Lognormal
4.32E+07
4.30E+07
55 1
90.353
4.483
18.593
0.203
Lognormal
4.56E+07
4.52E+07
56 1
90.006
4.481
17.833
0.192
Lognormal
4.20E+07
4.13E+07
57 1
89.277
4.474
17.028
0.190
Lognormal
3.66E+07
3.63E+07
58 1
89.392
4.474
18.265
0.195
Lognormal
3.85E+07
3.78E+07
59 1
91.403
4.491
20.709
0.217
Lognormal
4.75E+07
4.66E+07
60 1
90.917
4.488
20.306
0.206
Lognormal
3.96E+07
3.85E+07
61 1
93.150
4.506
22.700
0.233
Lognormal
3.16E+07
3.10E+07
62 1
90.499
4.487
18.053
0.192
Lognormal
3.11 E+07
3.06E+07
63 1
91.326
4.486
23.270
0.234
Lognormal
3.80E+07
3.68E+07
64 1
89.615
4.467
23.395
0.230
Lognormal
3.13E+07
3.00E+07
65 1
91.754
4.493
20.739
0.229
Lognormal
3.88E+07
3.86E+07
66 1
89.407
4.471
18.910
0.210
Lognormal
2.57E+07
2.55E+07
67 1
90.274
4.482
18.677
0.207
Lognormal
1.96E+07
1.95E+07
68 1
88.174
4.447
22.562
0.256
Lognormal
2.67E+07
2.64E+07
69 1
88.345
4.461
17.487
0.204
Normal
2.12E+07
2.13E+07
-------
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 7
Mean
Mean Log
Std Dev
Std Dev
-2LL
BW
BW
BW
Log BW
Best
Normal
88.508
4.465
16.451
0.190
Normal
2.32E+07
86.951
4.442
19.122
0.218
Lognormal
1.23E+07
85.011
4.427
14.707
0.184
Normal
2.07E+07
82.985
4.401
16.298
0.189
Lognormal
1.48E+07
87.057
4.452
15.113
0.172
Lognormal
1.71E+07
84.965
4.418
18.599
0.219
Lognormal
1.54E+07
84.242
4.418
15.364
0.173
Lognormal
1.45E+07
87.413
4.457
14.289
0.166
Normal
1.19E+07
86.227
4.437
17.646
0.199
Lognormal
1.08E+07
79.399
4.361
13.595
0.160
Lognormal
7.74E+06
79.526
4.360
14.305
0.182
Lognormal
7.06E+07
7.370
1.963
1.848
0.270
Normal
1.19E+07
11.090
2.394
1.754
0.152
Lognormal
1.14E+07
13.219
2.573
1.838
0.133
Lognormal
1.43E+07
15.640
2.739
2.510
0.145
Lognormal
1.70E+07
18.059
2.879
3.247
0.168
Lognormal
2.17E+07
20.679
3.012
4.027
0.181
Lognormal
2.12E+07
23.793
3.147
5.253
0.205
Lognormal
2.36E+07
26.881
3.261
7.211
0.238
Lognormal
2.92E+07
32.029
3.433
9.019
0.253
Lognormal
2.99E+07
36.699
3.566
10.701
0.264
Lognormal
3.46E+07
41.050
3.681
11.396
0.256
Lognormal
3.30E+07
47.362
3.818
13.982
0.278
Lognormal
4.43E+07
54.672
3.963
15.597
0.273
Lognormal
4.31 E+07
56.288
4.000
14.933
0.242
Lognormal
3.57E+07
59.807
4.069
13.215
0.209
Lognormal
4.03E+07
63.838
4.126
16.980
0.240
Lognormal
4.48E+07
64.978
4.140
18.345
0.251
Lognormal
4.31 E+07
65.573
4.151
18.055
0.244
Lognormal
4.11 E+07
67.681
4.177
20.459
0.263
Lognormal
4.15E+07
68.713
4.193
20.005
0.266
Lognormal
3.53E+07
67.242
4.175
18.889
0.250
Lognormal
4.92E+07
68.518
4.194
18.688
0.253
Lognormal
4.11 E+07
73.589
4.263
21.062
0.257
Lognormal
4.77E+07
73.890
4.269
19.737
0.258
Lognormal
4.70E+07
74.087
4.270
20.804
0.259
Lognormal
3.92E+07
71.664
4.235
22.042
0.261
Lognormal
4.91 E+07
74.947
4.278
22.693
0.268
Lognormal
4.46E+07
76.495
4.300
21.714
0.272
Lognormal
4.47E+07
76.115
4.293
22.452
0.274
Lognormal
4.54E+07
76.079
4.305
17.674
0.234
Lognormal
3.79E+07
77.839
4.318
22.534
0.262
Lognormal
4.31 E+07
77.715
4.316
22.610
0.264
Lognormal
5.03E+07
79.498
4.331
26.226
0.289
Lognormal
3.77E+07
80.160
4.353
21.345
0.243
Lognormal
4.10E+07
79.954
4.341
24.352
0.278
Lognormal
4.26E+07
76.240
4.309
17.070
0.221
Lognormal
3.04E+07
76.700
4.304
22.247
0.259
Lognormal
5.09E+07
79.289
4.333
23.794
0.276
Lognormal
4.06E+07
79.992
4.354
19.236
0.236
Lognormal
4.41 E+07
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 8
Age
Gender
Mean
BW
Mean Log
BW
Std Dev
BW
Std Dev
Log BW
Best
-2LL
Normal
-2LL
Log-Normal
39
2
76.566
4.305
21.337
0.251
Lognormal
4.81 E+07
4.62E+07
40
2
76.974
4.303
23.274
0.279
Lognormal
4.61 E+07
4.46E+07
41
2
76.441
4.301
21.868
0.260
Lognormal
4.68E+07
4.51 E+07
42
2
76.145
4.298
20.347
0.264
Lognormal
4.63E+07
4.57E+07
43
2
76.903
4.311
20.853
0.243
Lognormal
4.84E+07
4.65E+07
44
2
75.614
4.290
22.250
0.260
Lognormal
4.77E+07
4.55E+07
45
2
75.209
4.290
20.478
0.238
Lognormal
4.98E+07
4.74E+07
46
2
79.677
4.348
21.220
0.240
Lognormal
3.92E+07
3.77E+07
47
2
80.825
4.360
21.865
0.249
Lognormal
4.76E+07
4.60E+07
48
2
78.180
4.324
21.616
0.260
Lognormal
4.68E+07
4.56E+07
49
2
78.804
4.338
19.602
0.240
Lognormal
4.61 E+07
4.53E+07
50
2
79.090
4.345
18.574
0.221
Lognormal
5.30E+07
5.17E+07
51
2
77.540
4.320
20.179
0.244
Lognormal
4.67E+07
4.54E+07
52
2
73.712
4.267
20.579
0.252
Lognormal
5.12E+07
4.93E+07
53
2
77.885
4.325
19.474
0.243
Lognormal
3.77E+07
3.70E+07
54
2
81.799
4.368
23.266
0.262
Lognormal
4.49E+07
4.35E+07
55
2
81.660
4.364
23.736
0.270
Lognormal
4.30E+07
4.17E+07
56
2
78.463
4.332
19.938
0.245
Lognormal
5.21 E+07
5.11 E+07
57
2
77.206
4.320
19.414
0.225
Lognormal
4.11 E+07
3.95E+07
58
2
82.906
4.372
25.218
0.306
Lognormal
3.27E+07
3.24E+07
59
2
75.924
4.305
17.461
0.223
Lognormal
4.32E+07
4.25E+07
60
2
80.438
4.349
23.023
0.276
Lognormal
4.03E+07
3.95E+07
61
2
81.177
4.374
17.290
0.215
Lognormal
4.17E+07
4.15E+07
62
2
81.189
4.373
18.224
0.216
Lognormal
3.11 E+07
3.05E+07
63
2
74.279
4.282
17.151
0.229
Lognormal
3.96E+07
3.92E+07
64
2
78.502
4.333
20.131
0.243
Lognormal
4.00E+07
3.91 E+07
65
2
74.259
4.284
16.038
0.219
Lognormal
3.21 E+07
3.20E+07
66
2
76.788
4.320
15.800
0.207
Lognormal
2.52E+07
2.51 E+07
67
2
77.607
4.318
20.286
0.259
Lognormal
2.64E+07
2.61 E+07
68
2
71.134
4.237
17.438
0.232
Lognormal
2.51 E+07
2.45E+07
69
2
74.826
4.288
16.942
0.237
Normal
2.21 E+07
2.22E+07
70
2
80.651
4.361
19.520
0.243
Lognormal
2.93E+07
2.91 E+07
71
2
77.613
4.318
20.636
0.259
Lognormal
2.55E+07
2.51 E+07
72
2
75.780
4.295
19.888
0.254
Lognormal
2.38E+07
2.33E+07
73
2
76.332
4.307
18.416
0.234
Lognormal
2.22E+07
2.18E+07
74
2
73.923
4.280
16.136
0.216
Lognormal
2.19E+07
2.16E+07
75
2
73.693
4.276
15.862
0.222
Normal
1.45E+07
1.45E+07
76
2
77.133
4.324
16.505
0.209
Lognormal
1.55E+07
1.53E+07
77
2
73.587
4.270
18.167
0.238
Lognormal
1.30E+07
1.27E+07
78
2
72.360
4.258
16.423
0.216
Lognormal
1.29E+07
1.26E+07
79
2
69.868
4.224
15.927
0.208
Lognormal
1.45E+07
1.40E+07
1000
2
64.634
4.148
13.273
0.205
Lognormal
1.13E+08
1.12E+08
Note: Age 1000 = 80 years or older.
Table 2 below is the same as Table 1 above but for HT. In this case, the preferred distribution is
less consistent since 64 percent of the HT cases have "Normal" for the "Best" distribution and
36 percent of the cases have "Lognormal." The histograms also did not show a strong
preference for one of those two distributions. Since the results of the APEX simulations should
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 9
not be too sensitive to the exact ages of the modeled population, it is better to use the same
distribution for all ages and genders, which suggests that HT should be modeled as a normal
distribution for all demographic groups.
Table 2. Summary Statistics for HT
Mean
Mean Log
Std Dev
Std Dev
-2LL
-2LL Log-
Age Gender
HT
HT
HT
Log HT
Best
Normal
Normal
0 1
66.348
4.190
6.538
0.101
Normal
2.66E+07
2.68E+07
1 1
81.551
4.400
4.495
0.055
Lognormal
2.33E+07
2.32E+07
2 1
91.720
4.518
4.508
0.049
Normal
2.32E+07
2.32E+07
3 1
98.932
4.593
4.763
0.048
Normal
2.86E+07
2.86E+07
4 1
106.749
4.669
4.795
0.045
Lognormal
2.81 E+07
2.81 E+07
5 1
114.047
4.735
5.750
0.050
Lognormal
2.55E+07
2.54E+07
6 1
119.584
4.783
5.647
0.047
Normal
2.87E+07
2.88E+07
7 1
126.274
4.837
6.172
0.049
Normal
3.08E+07
3.08E+07
8 1
131.387
4.877
6.487
0.050
Normal
3.28E+07
3.28E+07
9 1
137.145
4.920
6.989
0.051
Lognormal
3.00E+07
2.99E+07
10 1
142.600
4.959
6.965
0.049
Normal
2.88E+07
2.89E+07
11 1
150.274
5.011
8.441
0.056
Lognormal
2.89E+07
2.88E+07
12 1
155.594
5.046
7.455
0.048
Lognormal
3.23E+07
3.23E+07
13 1
163.822
5.097
8.320
0.051
Normal
3.23E+07
3.24E+07
14 1
168.833
5.128
7.825
0.047
Normal
3.74E+07
3.75E+07
15 1
173.395
5.155
7.224
0.042
Normal
2.94E+07
2.95E+07
16 1
174.662
5.162
6.608
0.038
Normal
3.20E+07
3.21 E+07
17 1
175.483
5.166
8.067
0.046
Normal
3.13E+07
3.13E+07
18 1
175.871
5.169
7.309
0.042
Normal
3.00E+07
3.00E+07
19 1
176.655
5.173
7.524
0.043
Lognormal
3.41 E+07
3.41 E+07
20 1
175.034
5.164
7.566
0.044
Normal
3.72E+07
3.73E+07
21 1
176.763
5.174
8.403
0.048
Normal
3.49E+07
3.50E+07
22 1
176.195
5.171
6.516
0.037
Lognormal
3.00E+07
3.00E+07
23 1
174.777
5.162
8.261
0.047
Lognormal
3.20E+07
3.19E+07
24 1
176.734
5.174
7.498
0.042
Lognormal
3.25E+07
3.24E+07
25 1
176.400
5.172
6.713
0.038
Normal
2.92E+07
2.93E+07
26 1
176.482
5.172
6.841
0.039
Normal
2.50E+07
2.51 E+07
27 1
176.625
5.173
6.835
0.039
Normal
2.70E+07
2.70E+07
28 1
177.668
5.179
7.591
0.043
Normal
3.35E+07
3.35E+07
29 1
176.629
5.173
7.984
0.045
Lognormal
3.41 E+07
3.40E+07
30 1
177.154
5.176
7.644
0.044
Normal
3.48E+07
3.49E+07
31 1
176.424
5.172
6.393
0.036
Normal
2.63E+07
2.63E+07
32 1
176.506
5.172
8.069
0.046
Normal
3.25E+07
3.26E+07
33 1
177.685
5.179
7.686
0.043
Lognormal
2.81 E+07
2.81 E+07
34 1
176.909
5.175
7.629
0.043
Normal
2.49E+07
2.49E+07
35 1
175.465
5.166
8.162
0.047
Normal
2.87E+07
2.88E+07
36 1
175.886
5.169
7.555
0.043
Normal
3.08E+07
3.08E+07
37 1
176.134
5.170
7.465
0.043
Normal
2.88E+07
2.88E+07
38 1
176.737
5.174
7.627
0.043
Normal
2.78E+07
2.78E+07
39 1
176.688
5.173
8.195
0.047
Normal
3.13E+07
3.14E+07
40 1
177.188
5.176
8.246
0.046
Lognormal
3.41 E+07
3.40E+07
41 1
177.129
5.176
8.370
0.047
Normal
3.42E+07
3.43E+07
42 1
175.377
5.166
7.477
0.043
Lognormal
2.67E+07
2.67E+07
43 1
177.690
5.179
7.330
0.041
Lognormal
3.28E+07
3.28E+07
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 10
Mean
Mean Log
Std Dev
Std Dev
-2LL
-2LL Log-
Age
Gender
HT
HT
HT
Log HT
Best
Normal
Normal
44
1
176.112
5.170
7.903
0.045
Lognormal
3.32E+07
3.31 E+07
45
1
174.981
5.164
7.396
0.042
Normal
2.89E+07
2.90E+07
46
1
176.634
5.173
6.562
0.038
Normal
3.09E+07
3.10E+07
47
1
175.600
5.167
6.753
0.038
Lognormal
3.17E+07
3.17E+07
48
1
176.122
5.170
7.434
0.043
Normal
2.87E+07
2.88E+07
49
1
177.033
5.176
6.807
0.039
Normal
2.78E+07
2.79E+07
50
1
176.496
5.172
7.690
0.043
Lognormal
3.39E+07
3.38E+07
51
1
174.912
5.163
7.901
0.045
Lognormal
2.69E+07
2.69E+07
52
1
176.530
5.173
6.804
0.039
Normal
2.96E+07
2.96E+07
53
1
176.744
5.174
7.201
0.041
Lognormal
3.02E+07
3.02E+07
54
1
176.288
5.171
7.453
0.042
Normal
3.16E+07
3.16E+07
55
1
175.405
5.166
6.225
0.035
Lognormal
3.10E+07
3.10E+07
56
1
176.729
5.174
7.468
0.043
Normal
3.09E+07
3.10E+07
57
1
175.733
5.168
8.368
0.048
Normal
2.88E+07
2.89E+07
58
1
176.871
5.174
8.038
0.046
Normal
2.93E+07
2.93E+07
59
1
176.603
5.173
6.358
0.036
Normal
3.16E+07
3.17E+07
60
1
175.322
5.166
7.743
0.044
Lognormal
2.90E+07
2.89E+07
61
1
175.231
5.165
7.553
0.044
Normal
2.20E+07
2.20E+07
62
1
174.979
5.164
7.231
0.042
Normal
2.27E+07
2.28E+07
63
1
177.680
5.179
8.229
0.046
Lognormal
2.69E+07
2.69E+07
64
1
173.887
5.158
7.268
0.042
Normal
2.13E+07
2.14E+07
65
1
175.770
5.168
7.209
0.042
Normal
2.72E+07
2.73E+07
66
1
175.376
5.166
8.807
0.051
Normal
2.00E+07
2.01 E+07
67
1
173.978
5.158
6.767
0.039
Lognormal
1.38E+07
1.38E+07
68
1
174.040
5.159
6.660
0.039
Normal
1.81E+07
1.82E+07
69
1
173.767
5.157
8.313
0.048
Normal
1.66E+07
1.66E+07
70
1
173.764
5.157
6.780
0.039
Normal
1.69E+07
1.69E+07
71
1
171.952
5.146
7.098
0.041
Lognormal
8.79E+06
8.75E+06
72
1
173.617
5.156
7.523
0.044
Normal
1.64E+07
1.64E+07
73
1
171.815
5.145
7.548
0.044
Normal
1.14E+07
1.14E+07
74
1
173.762
5.157
6.224
0.036
Lognormal
1.23E+07
1.22E+07
75
1
172.609
5.150
7.212
0.042
Lognormal
1.12E+07
1.12E+07
76
1
172.734
5.151
6.328
0.037
Lognormal
1.05E+07
1.05E+07
77
1
172.442
5.149
7.440
0.043
Normal
9.47E+06
9.48E+06
78
1
174.156
5.159
7.499
0.043
Normal
7.98E+06
7.98E+06
79
1
172.635
5.150
6.417
0.037
Lognormal
5.87E+06
5.86E+06
1000
1
171.292
5.143
6.915
0.041
Normal
5.32E+07
5.32E+07
0
2
64.997
4.169
6.275
0.100
Normal
2.50E+07
2.52E+07
1
2
80.615
4.388
4.947
0.062
Normal
2.25E+07
2.25E+07
2
2
89.528
4.493
4.204
0.046
Lognormal
2.50E+07
2.49E+07
3
2
98.281
4.587
4.248
0.044
Normal
2.29E+07
2.30E+07
4
2
105.404
4.657
4.857
0.046
Normal
2.69E+07
2.70E+07
5
2
112.415
4.721
5.787
0.052
Lognormal
2.53E+07
2.53E+07
6
2
118.957
4.778
5.654
0.048
Normal
2.44E+07
2.44E+07
7
2
124.658
4.824
5.843
0.047
Lognormal
2.68E+07
2.67E+07
8
2
131.786
4.880
6.950
0.052
Lognormal
2.70E+07
2.69E+07
9
2
137.722
4.924
6.500
0.047
Lognormal
2.86E+07
2.86E+07
10
2
144.426
4.971
7.298
0.050
Lognormal
2.80E+07
2.79E+07
11
2
150.574
5.013
7.670
0.052
Normal
3.58E+07
3.60E+07
12
2
156.583
5.052
7.295
0.047
Normal
3.30E+07
3.31 E+07
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 11
Age
Gender
Mean
HT
Mean Log
HT
Std Dev
HT
Std Dev
Log HT
Best
-2LL
Normal
-2LL Log-
Normal
13
2
158.923
5.068
6.149
0.039
Lognormal
2.58E+07
2.58E+07
14
2
160.849
5.080
6.429
0.040
Normal
3.09E+07
3.09E+07
15
2
161.704
5.085
6.674
0.042
Normal
3.22E+07
3.23E+07
16
2
162.002
5.087
6.219
0.038
Lognormal
2.94E+07
2.94E+07
17
2
162.805
5.092
6.661
0.041
Normal
2.95E+07
2.95E+07
18
2
162.208
5.088
6.344
0.039
Lognormal
2.77E+07
2.77E+07
19
2
163.320
5.095
6.174
0.038
Normal
2.35E+07
2.35E+07
20
2
163.411
5.095
7.485
0.046
Normal
3.59E+07
3.60E+07
21
2
161.858
5.086
6.643
0.041
Lognormal
2.87E+07
2.86E+07
22
2
162.038
5.087
6.058
0.037
Lognormal
3.09E+07
3.09E+07
23
2
161.916
5.086
7.447
0.046
Normal
3.38E+07
3.39E+07
24
2
162.774
5.091
7.195
0.044
Lognormal
2.74E+07
2.73E+07
25
2
162.763
5.092
6.405
0.039
Lognormal
3.22E+07
3.21 E+07
26
2
163.198
5.094
6.312
0.039
Normal
2.90E+07
2.91 E+07
27
2
163.593
5.096
7.471
0.046
Normal
3.14E+07
3.14E+07
28
2
163.380
5.095
6.569
0.040
Normal
2.99E+07
3.00E+07
29
2
162.909
5.093
5.527
0.034
Normal
2.49E+07
2.49E+07
30
2
163.515
5.096
7.695
0.047
Normal
3.03E+07
3.03E+07
31
2
164.013
5.099
6.712
0.041
Normal
3.34E+07
3.34E+07
32
2
163.674
5.097
7.194
0.044
Normal
2.48E+07
2.48E+07
33
2
163.856
5.098
6.710
0.041
Normal
2.77E+07
2.78E+07
34
2
163.344
5.095
7.496
0.046
Lognormal
2.90E+07
2.90E+07
35
2
163.531
5.096
6.544
0.041
Normal
2.17E+07
2.18E+07
36
2
163.211
5.094
7.656
0.047
Normal
3.58E+07
3.58E+07
37
2
164.099
5.100
6.902
0.043
Normal
2.69E+07
2.70E+07
38
2
162.956
5.092
7.860
0.048
Lognormal
3.27E+07
3.26E+07
39
2
162.702
5.091
7.675
0.047
Normal
3.44E+07
3.44E+07
40
2
162.678
5.091
7.397
0.045
Lognormal
3.16E+07
3.16E+07
41
2
161.638
5.085
6.643
0.041
Lognormal
3.13E+07
3.12E+07
42
2
163.154
5.094
7.131
0.043
Lognormal
3.28E+07
3.27E+07
43
2
162.756
5.091
6.773
0.042
Normal
3.30E+07
3.30E+07
44
2
162.821
5.092
6.921
0.043
Normal
3.22E+07
3.23E+07
45
2
162.737
5.092
5.720
0.035
Normal
3.19E+07
3.19E+07
46
2
162.146
5.087
7.539
0.047
Normal
2.79E+07
2.80E+07
47
2
163.495
5.096
7.326
0.045
Normal
3.31 E+07
3.31 E+07
48
2
163.566
5.096
6.311
0.039
Normal
3.07E+07
3.08E+07
49
2
162.858
5.092
6.338
0.039
Normal
3.11 E+07
3.13E+07
50
2
162.498
5.090
6.919
0.043
Normal
3.76E+07
3.77E+07
51
2
162.610
5.091
5.990
0.037
Normal
3.06E+07
3.07E+07
52
2
161.654
5.084
7.879
0.051
Normal
3.73E+07
3.80E+07
53
2
163.379
5.095
6.657
0.041
Normal
2.60E+07
2.61 E+07
54
2
162.049
5.087
7.027
0.043
Lognormal
3.02E+07
3.01 E+07
55
2
162.694
5.091
6.633
0.041
Normal
2.81 E+07
2.81 E+07
56
2
162.638
5.091
6.787
0.041
Lognormal
3.60E+07
3.59E+07
57
2
160.512
5.077
7.084
0.044
Lognormal
2.92E+07
2.91 E+07
58
2
160.963
5.080
7.017
0.044
Normal
2.15E+07
2.15E+07
59
2
160.849
5.080
6.991
0.043
Lognormal
3.15E+07
3.14E+07
60
2
161.262
5.082
6.422
0.040
Normal
2.62E+07
2.63E+07
61
2
163.010
5.093
7.148
0.044
Lognormal
3.07E+07
3.07E+07
62
2
160.395
5.077
6.512
0.041
Lognormal
2.17E+07
2.17E+07
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 12
Mean
Mean Log
Std Dev
Std Dev
-2LL
-2LL Log-
Age
Gender
HT
HT
HT
Log HT
Best
Normal
Normal
63
2
161.629
5.084
6.589
0.041
Lognormal
2.83E+07
2.82E+07
64
2
160.269
5.076
6.028
0.038
Normal
2.69E+07
2.70E+07
65
2
161.070
5.081
6.539
0.040
Lognormal
2.33E+07
2.32E+07
66
2
159.425
5.071
5.689
0.036
Normal
1.74E+07
1.75E+07
67
2
160.241
5.076
6.903
0.043
Lognormal
1.83E+07
1.83E+07
68
2
158.931
5.067
7.056
0.045
Normal
1.82E+07
1.83E+07
69
2
159.863
5.073
6.687
0.043
Normal
1.59E+07
1.60E+07
70
2
160.263
5.076
6.986
0.044
Normal
2.07E+07
2.07E+07
71
2
159.678
5.072
7.340
0.046
Normal
1.80E+07
1.80E+07
72
2
158.699
5.066
6.225
0.039
Lognormal
1.59E+07
1.59E+07
73
2
159.618
5.072
7.187
0.045
Normal
1.61E+07
1.61E+07
74
2
159.042
5.068
6.425
0.040
Lognormal
1.57E+07
1.57E+07
75
2
158.332
5.064
7.461
0.047
Normal
1.11E+07
1.11E+07
76
2
159.769
5.073
5.740
0.036
Normal
1.05E+07
1.05E+07
77
2
158.186
5.063
5.841
0.037
Normal
8.57E+06
8.58E+06
78
2
158.001
5.062
7.098
0.045
Normal
9.55E+06
9.57E+06
79
2
158.586
5.065
7.097
0.045
Normal
1.12E+07
1.12E+07
1000
2
155.746
5.047
6.564
0.042
Normal
8.63E+07
8.64E+07
Note: Age 1000 = 80 years or older.
For an overall comparison, we calculated the values of -2LL for the entire population ages 0-79
years by summing the values of -2LL across all ages and genders. For BW, the -2LL totals were
5.91 x109 for the normal distribution and 5.75* 109 for the log-normal distribution—again
supporting the log-normal distribution. For HT, the -2LL totals were 4.42* 109 for the normal
distribution and 4.43* 109 for the log-normal distribution, which provides some small support for
the normal distribution. The unrounded summary statistics from Table 1 and Table 2 above are
shown in the tabs "Mean", "Weights", and "HTs" of the accompanying Excel file "means.2009 to
2014.102016.xlsx"; the tab "Read Me" gives the content and formats for each tab.
To summarize these results, the recommended distributions are a normal distribution for HTs
and a log-normal distribution for BWs. The parameters vary by age (in years) and gender.
The same conclusion was reached by Brainard and Burmaster (1992)3. Note that in 2002,
the CDC developed growth charts for children by fitting more complicated Box-Cox models to
earlier NHANES data.4 The Box-Cox model uses a power of the normal distribution, which tends
to a log-normal distribution when the power tends to zero. Those approaches would be harder
to implement for APEX, particularly when developing joint distributions for BW and HT.
3. Joint Distributions for BW and HT
The conclusion from Section 2 was that, for each age and gender, we should model BW by a
log-normal distribution and HT by a normal distribution. To fit a joint distribution, it is important to
3 Brainard, J., Burmaster, D.E. "Bivariate distributions for height and weight of men and women in the United States".
Risk Analysis 1992, 12(2)267-275.
4 http://www.cdc.aov/arowthcharts/cdc charts.htm
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 13
realize that HT and BW are not independent. Therefore, we fit the joint distribution of HT and
BW by assuming that the HT and the logarithm of the BW have a bivariate normal
distribution. Table 1 and Table 2 above contain the means and standard deviations of the HT
and the logarithm of the BW. Table 3 below contains the correlations between the HT and the
logarithm of the BW, calculated using the survey weights. The "Mean" tab of the accompanying
Excel file "means.2009 to 2014.102016.xlsx" contains the unrounded values of the correlation
coefficient.
Table 3. Correlation Between Log BW and HT
Age
Correlation Between Log 1
Gender BW and HT I
1 Age
Gender
Correlation Between Log
BW and HT
0
1 0.934
0
2
0.933
1
1 0.804
1
2
0.789
2
1 0.751
2
2
0.765
3
1 0.742
3
2
0.733
4
1 0.755
4
2
0.761
5
1 0.741
5
2
0.744
6
1 0.758
6
2
0.734
7
1 0.706
7
2
0.753
8
1 0.768
8
2
0.720
9
1 0.721
9
2
0.676
10
1 0.685
10
2
0.729
11
1 0.697
11
2
0.606
12
1 0.671
12
2
0.558
13
1 0.563
13
2
0.391
14
1 0.585
14
2
0.344
15
1 0.485
15
2
0.461
16
1 0.430
16
2
0.364
17
1 0.416
17
2
0.359
18
1 0.451
18
2
0.228
19
1 0.312
19
2
0.227
20
1 0.504
20
2
0.294
21
1 0.426
21
2
0.397
22
1 0.299
22
2
0.086
23
1 0.423
23
2
0.294
24
1 0.391
24
2
0.236
25
1 0.388
25
2
0.288
26
1 0.396
26
2
0.325
27
1 0.515
27
2
0.356
28
1 0.337
28
2
0.354
29
1 0.174
29
2
0.269
30
1 0.597
30
2
0.269
31
1 0.298
31
2
0.212
32
1 0.482
32
2
0.248
33
1 0.528
33
2
0.269
34
1 0.292
34
2
0.283
35
1 0.279
35
2
0.200
36
1 0.519
36
2
0.362
37
1 0.434
37
2
0.391
38
1 0.453
38
2
0.328
39
1 0.373
39
2
0.396
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 14
Age
Correlation Between Log 1
Gender BW and HT 1
1 Age
Gender
Correlation Between Log
BW and HT
40
1 0.546
40
2
0.302
41
1 0.357
41
2
0.367
42
1 0.339
42
2
0.300
43
1 0.367
43
2
0.233
44
1 0.470
44
2
0.301
45
1 0.453
45
2
0.240
46
1 0.227
46
2
0.245
47
1 0.405
47
2
0.254
48
1 0.357
48
2
0.042
49
1 0.496
49
2
0.262
50
1 0.590
50
2
0.248
51
1 0.534
51
2
0.167
52
1 0.338
52
2
0.347
53
1 0.510
53
2
0.260
54
1 0.441
54
2
0.235
55
1 0.363
55
2
0.178
56
1 0.292
56
2
0.115
57
1 0.437
57
2
0.301
58
1 0.324
58
2
0.287
59
1 0.472
59
2
0.266
60
1 0.380
60
2
0.414
61
1 0.387
61
2
0.380
62
1 0.475
62
2
0.266
63
1 0.520
63
2
0.310
64
1 0.534
64
2
0.248
65
1 0.372
65
2
0.240
66
1 0.408
66
2
0.331
67
1 0.627
67
2
0.351
68
1 0.490
68
2
0.300
69
1 0.510
69
2
0.287
70
1 0.434
70
2
0.257
71
1 0.413
71
2
0.275
72
1 0.527
72
2
0.262
73
1 0.578
73
2
0.302
74
1 0.220
74
2
0.237
75
1 0.503
75
2
0.083
76
1 0.161
76
2
0.297
77
1 0.400
77
2
0.248
78
1 0.524
78
2
0.292
79
1 0.195
79
2
0.461
1000
1 0.491
1000
2
0.419
Note: Age 1000 = 80 years or older.
Figure 3-1 below illustrates the fitted joint distributions for selected ages (5, 15, 25, 40, 60, and
79 years) and both genders. Each data point shows the HT and the logarithm of the BW for a
single NHANES subject. The red prediction ellipse includes 95 percent of the fitted joint
distribution (which is not necessarily 95 percent of the sampled data). The blue prediction ellipse
includes 80 percent of the fitted joint distribution (which is not necessarily 80 percent of the
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 15
sampled data). The ellipses and correlations were computed using the survey weights, even
though there is only a single point shown for each NHANES subject. The elliptical shapes of the
scatter plot data support the use of a bivariate normal distribution with a non-zero correlation. A
zero correlation would imply that HT and BW are independent. We provide larger versions of the
plots in Figure 3-1 in Attachment C.
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 16
gender -1
gender - 2
I Observations 2951
| Correlation 0.74081
°
°0 8 ^ °
° 8 ° °
^t5° o® °^° S1*"*5r '
00 110 120 130
[Observations 2371 °
| Con-elation 0.4851 ° D 0
0 o° )
9- ^ O 8o0 \
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 17
4. Smoothing the Parameters
4.1. Smooth Parameters Using Natural Cubic Spline
The last step for fitting the joint distributions of BW and HT is to smooth the parameter values
to make them continuous functions of the age rather than varying discontinuously.
Otherwise, a small change in the age of one of the simulated persons can lead to a large
change in the simulated distribution of that person's HT and BW and thus other exposure
parameters. The five parameters for each age and gender are
mean log BW,
standard deviation log BW,
mean HT,
standard deviation HT, and
correlation.
Figure 4-1 below illustrates how the five parameters vary by age for the same gender. Also
shown are the smoothed curves created with a natural cubic spline, without applying any
weighting. For each parameter, we chose the same set of eight knots for the spline function: 0,
10, 20, 30, 40, 50, 60, and 70. Between each two consecutive knots, we fitted a cubic
polynomial so that the curve and its first two derivatives are continuous at the knot. For values
above 70, we fitted a straight line so that the curve and its first derivative are continuous at 70.
(A similar linear curve applies below zero but those values are not needed since age cannot be
negative). The straight line fitted to ages 70 and above is used to extrapolate the
parameter values up to age 100. We provide larger versions of the plots in Figure 4-1 in
Attachment D.
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 18
Cubic spline of mean of log body weight (kg) versus age
Gender=
Gender= 2
20 40 60 80
100 0
AGE
| O the mean, logweight — Linear Predictor |
20 40 60 80
Cubic spline of standard deviation of log body weight (kg) versus age
5 0.25 •
J3
I
0.20 •
Gender = 1
Gender= 2
°
o
°
/ °° \
° °° $ ° 0°° ° °
o o°\
* \
-
0 20 40 60 80 100 0 20 40 60
AGE
| O the standard deviation, logweight ¦ Linear Predictor |
80 100
Cubic spline of mean of height (cm) versus age
Cubic spline of standard deviation of height (cm) versus age
40 60 80 100 0
AGE
|p the mean, BMXHT Linear Predictor |
40 60 80 100
Gender= 1
Gender= 2
»
° ° 0»0# o °
oV s
r v/ o% V
°%° °o° ^
T ^ qO a go
/ ° o °
o L ° o ° ° \
° —
f ' ^
° / »o °° * ° o°
r
/
•
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 19
For simulating the joint distribution of BW and HT in APEX, we propose the following
approach.
First, simulate the values of log BW from a normal distribution. We show the mean and
standard deviation of the log BW for each age and gender in the "SMOOTHED" columns of
Table 4. Truncate the distribution at the lower and upper bounds as shown in the "BOUNDS
FOR LOG BW" columns, which we calculated as
BOUNDS FOR LOG BW = Mean Log BW ± (zo.99 x Std Dev Log BW).
zo.99 is the 99th percentile of a standard normal distribution. Resampling should be done, so
that a new value should be selected if the simulated value is outside these bounds. Thus, the
probability of being outside these two bounds is 0.02. Let w be the simulated value of log BW.
Second, simulate the values of HT from the conditional distribution of HT given that the
log of the BW is w. The simulated value of HT is
Simulated HT = mh + (sh x corr x + (sh x Vl — corr2 x z),
where
mh = Mean HT,
sh = Std Dev HT,
corr = Correlation coefficient (between log BW and HT),
w = Simulated log BW,
mw = Mean Log BW,
sw = Std Dev Log BW, and
z = Simulated and truncated standard normal variate.
The z-score "z" is randomly generated from a standard normal distribution. Analogously to the
truncation of the BW distribution, z should be resampled if its absolute value is greater than
zo.99.
-------
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 20
Table 4. Unsmoothed and Smoothed Parameter Values
BOUNDS FOR
UNSMOOTHED
SMOOTHED
LOG BW
Std Dev Correl-
Std Dev Correl-
Lower Upper
Mean HT Mean Log BW Std Dev HT Log BW ation
Mean HT Mean Log BW Std Dev HT Log BW ation
Bound Bound
66.348
2.024
6.538
0.261
0.934
71.149
2.189
4.541
0.144
0.827
1.855
2.524
81.551
2.429
4.495
0.126
0.804
79.700
2.362
4.830
0.156
0.817
1.999
2.725
91.720
2.640
4.508
0.126
0.751
88.191
2.533
5.117
0.168
0.807
2.141
2.924
98.932
2.773
4.763
0.139
0.742
96.564
2.701
5.399
0.180
0.795
2.283
3.120
106.749
2.915
4.795
0.157
0.755
104.761
2.867
5.673
0.191
0.783
2.421
3.312
114.047
3.045
5.750
0.172
0.741
112.722
3.027
5.937
0.202
0.770
2.557
3.498
119.584
3.149
5.647
0.191
0.758
120.388
3.182
6.188
0.212
0.755
2.689
3.676
126.274
3.298
6.172
0.234
0.706
127.701
3.330
6.424
0.221
0.738
2.816
3.845
131.387
3.407
6.487
0.241
0.768
134.601
3.470
6.642
0.229
0.719
2.937
4.004
137.145
3.513
6.989
0.249
0.721
141.030
3.601
6.840
0.236
0.698
3.052
4.150
142.600
3.656
6.965
0.263
0.685
146.928
3.721
7.014
0.241
0.673
3.160
4.283
150.274
3.832
8.441
0.280
0.697
152.251
3.831
7.164
0.245
0.646
3.260
4.401
155.594
3.894
7.455
0.252
0.671
157.006
3.929
7.290
0.248
0.616
3.352
4.505
163.822
4.060
8.320
0.256
0.563
161.217
4.016
7.393
0.249
0.585
3.437
4.596
168.833
4.143
7.825
0.259
0.585
164.906
4.094
7.474
0.250
0.553
3.514
4.675
173.395
4.234
7.224
0.255
0.485
168.094
4.162
7.535
0.249
0.521
3.583
4.741
174.662
4.289
6.608
0.226
0.430
170.804
4.222
7.578
0.248
0.491
3.646
4.798
175.483
4.317
8.067
0.235
0.416
173.059
4.272
7.604
0.246
0.462
3.701
4.844
175.871
4.363
7.309
0.248
0.451
174.881
4.315
7.613
0.243
0.437
3.749
4.882
176.655
4.350
7.524
0.229
0.312
176.292
4.350
7.608
0.241
0.415
3.790
4.911
175.034
4.341
7.566
0.246
0.504
177.314
4.379
7.590
0.238
0.398
3.824
4.933
176.763
4.342
8.403
0.231
0.426
177.974
4.401
7.561
0.236
0.385
3.852
4.950
176.195
4.368
6.516
0.224
0.299
178.320
4.417
7.523
0.233
0.378
3.874
4.960
174.777
4.418
8.261
0.269
0.423
178.401
4.429
7.481
0.231
0.375
3.891
4.967
176.734
4.396
7.498
0.192
0.391
178.270
4.437
7.437
0.229
0.375
3.904
4.969
176.400
4.422
6.713
0.248
0.388
177.977
4.441
7.395
0.227
0.377
3.914
4.969
176.482
4.437
6.841
0.213
0.396
177.575
4.444
7.359
0.225
0.382
3.921
4.967
176.625
4.433
6.835
0.207
0.515
177.113
4.445
7.333
0.223
0.388
3.926
4.964
177.668
4.459
7.591
0.230
0.337
176.643
4.446
7.319
0.222
0.394
3.930
4.961
176.629
4.467
7.984
0.215
0.174
176.217
4.446
7.322
0.220
0.401
3.934
4.959
177.154
4.458
7.644
0.233
0.597
175.885
4.449
7.344
0.219
0.407
3.939
4.958
176.424
4.465
6.393
0.211
0.298
175.688
4.452
7.388
0.218
0.411
3.946
4.959
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 21
UNSMOOTHED
SMOOTHED
BOUNDS FOR
LOG BW
32 1
176.506
4.486
8.069
0.230
0.482
175.614
4.458
7.450
0.217
0.414
3.953
4.963
33 1
177.685
4.500
7.686
0.208
0.528
175.643
4.465
7.523
0.216
0.416
3.962
4.967
34 1
176.909
4.451
7.629
0.194
0.292
175.752
4.472
7.603
0.215
0.418
3.971
4.973
35 1
175.465
4.461
8.162
0.228
0.279
175.920
4.480
7.683
0.215
0.418
3.980
4.979
36 1
175.886
4.498
7.555
0.240
0.519
176.124
4.487
7.757
0.214
0.417
3.990
4.985
37 1
176.134
4.512
7.465
0.204
0.434
176.344
4.495
7.821
0.213
0.416
3.999
4.990
38 1
176.737
4.514
7.627
0.210
0.453
176.556
4.501
7.867
0.212
0.415
4.008
4.994
39 1
176.688
4.483
8.195
0.219
0.373
176.740
4.506
7.891
0.211
0.413
4.015
4.996
40 1
177.188
4.504
8.246
0.212
0.546
176.874
4.509
7.886
0.209
0.411
4.022
4.996
41 1
177.129
4.522
8.370
0.218
0.357
176.941
4.510
7.850
0.208
0.410
4.027
4.993
42 1
175.377
4.491
7.477
0.189
0.339
176.945
4.509
7.786
0.206
0.408
4.030
4.988
43 1
177.690
4.510
7.330
0.192
0.367
176.899
4.507
7.701
0.204
0.407
4.033
4.982
44 1
176.112
4.512
7.903
0.202
0.470
176.813
4.504
7.602
0.202
0.406
4.034
4.974
45 1
174.981
4.463
7.396
0.182
0.453
176.697
4.500
7.496
0.200
0.405
4.034
4.966
46 1
176.634
4.485
6.562
0.200
0.227
176.561
4.495
7.389
0.199
0.405
4.033
4.958
47 1
175.600
4.493
6.753
0.180
0.405
176.417
4.491
7.288
0.198
0.406
4.031
4.951
48 1
176.122
4.482
7.434
0.221
0.357
176.276
4.487
7.199
0.197
0.407
4.029
4.945
49 1
177.033
4.488
6.807
0.208
0.496
176.147
4.483
7.130
0.197
0.409
4.025
4.941
50 1
176.496
4.524
7.690
0.199
0.590
176.042
4.481
7.087
0.197
0.412
4.022
4.940
51 1
174.912
4.432
7.901
0.221
0.534
175.968
4.480
7.074
0.199
0.416
4.018
4.942
52 1
176.530
4.501
6.804
0.205
0.338
175.922
4.480
7.089
0.200
0.421
4.013
4.946
53 1
176.744
4.479
7.201
0.215
0.510
175.897
4.480
7.126
0.203
0.427
4.009
4.952
54 1
176.288
4.521
7.453
0.204
0.441
175.887
4.482
7.180
0.205
0.433
4.004
4.959
55 1
175.405
4.483
6.225
0.203
0.363
175.885
4.484
7.246
0.208
0.439
4.000
4.967
56 1
176.729
4.481
7.468
0.192
0.292
175.885
4.486
7.319
0.211
0.444
3.996
4.976
57 1
175.733
4.474
8.368
0.190
0.437
175.880
4.488
7.393
0.213
0.449
3.992
4.984
58 1
176.871
4.474
8.038
0.195
0.324
175.865
4.489
7.463
0.216
0.454
3.988
4.991
59 1
176.603
4.491
6.358
0.217
0.472
175.831
4.490
7.524
0.217
0.457
3.985
4.996
60 1
175.322
4.488
7.743
0.206
0.380
175.774
4.491
7.571
0.219
0.460
3.982
4.999
61 1
175.231
4.506
7.553
0.233
0.387
175.688
4.490
7.600
0.219
0.461
3.980
5.000
62 1
174.979
4.487
7.231
0.192
0.475
175.574
4.488
7.612
0.219
0.460
3.979
4.998
63 1
177.680
4.486
8.229
0.234
0.520
175.436
4.486
7.608
0.218
0.459
3.978
4.993
64 1
173.887
4.467
7.268
0.230
0.534
175.277
4.482
7.591
0.217
0.456
3.977
4.987
65 1
175.770
4.493
7.209
0.229
0.372
175.099
4.478
7.561
0.215
0.453
3.978
4.979
66 1
175.376
4.471
8.807
0.210
0.408
174.906
4.474
7.523
0.213
0.448
3.978
4.970
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 22
UNSMOOTHED
SMOOTHED
BOUNDS FOR
LOG BW
67 1
173.978
4.482
6.767
0.207
0.627
174.701
4.469
7.476
0.211
0.444
3.979
4.959
68 1
174.040
4.447
6.660
0.256
0.490
174.487
4.464
7.424
0.208
0.438
3.980
4.948
69 1
173.767
4.461
8.313
0.204
0.510
174.267
4.458
7.368
0.205
0.433
3.981
4.936
70 1
173.764
4.465
6.780
0.190
0.434
174.043
4.453
7.310
0.203
0.427
3.982
4.924
71 1
171.952
4.442
7.098
0.218
0.413
173.819
4.447
7.252
0.200
0.421
3.983
4.912
72 1
173.617
4.427
7.523
0.184
0.527
173.595
4.442
7.193
0.197
0.416
3.984
4.900
73 1
171.815
4.401
7.548
0.189
0.578
173.371
4.436
7.135
0.194
0.410
3.985
4.888
74 1
173.762
4.452
6.224
0.172
0.220
173.148
4.431
7.076
0.191
0.404
3.986
4.875
75 1
172.609
4.418
7.212
0.219
0.503
172.924
4.425
7.018
0.188
0.399
3.987
4.863
76 1
172.734
4.418
6.328
0.173
0.161
172.700
4.420
6.960
0.185
0.393
3.989
4.851
77 1
172.442
4.457
7.440
0.166
0.400
172.476
4.414
6.901
0.183
0.387
3.990
4.839
78 1
174.156
4.437
7.499
0.199
0.524
172.252
4.409
6.843
0.180
0.381
3.991
4.827
79 1
172.635
4.361
6.417
0.160
0.195
172.028
4.403
6.785
0.177
0.376
3.992
4.814
80 1
171.804
4.398
6.726
0.174
0.370
3.993
4.802
81 1
171.580
4.392
6.668
0.171
0.364
3.994
4.790
82 1
171.357
4.387
6.610
0.168
0.359
3.995
4.778
83 1
171.133
4.381
6.551
0.165
0.353
3.996
4.766
84 1
170.909
4.376
6.493
0.162
0.347
3.998
4.754
85 1
170.685
4.370
6.434
0.160
0.341
3.999
4.741
86 1
170.461
4.365
6.376
0.157
0.336
4.000
4.729
87 1
170.237
4.359
6.318
0.154
0.330
4.001
4.717
88 1
170.013
4.353
6.259
0.151
0.324
4.002
4.705
89 1
169.789
4.348
6.201
0.148
0.319
4.003
4.693
90 1
169.565
4.342
6.143
0.145
0.313
4.004
4.680
91 1
169.342
4.337
6.084
0.142
0.307
4.006
4.668
92 1
169.118
4.331
6.026
0.140
0.301
4.007
4.656
93 1
168.894
4.326
5.968
0.137
0.296
4.008
4.644
94 1
168.670
4.320
5.909
0.134
0.290
4.009
4.632
95 1
168.446
4.315
5.851
0.131
0.284
4.010
4.620
96 1
168.222
4.309
5.792
0.128
0.279
4.011
4.607
97 1
167.998
4.304
5.734
0.125
0.273
4.012
4.595
98 1
167.774
4.298
5.676
0.122
0.267
4.013
4.583
99 1
167.550
4.293
5.617
0.120
0.262
4.015
4.571
100 1
167.327
4.287
5.559
0.117
0.256
4.016
4.559
0 2 64.997 1.963 6.275 0.270 0.933 68.702 2.113 4.597 0.164 0.848 1.731 2.495
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 23
UNSMOOTHED
SMOOTHED
BOUNDS FOR
LOG BW
1
2
80.615
2.394
4.947
0.152
0.789
77.867
2.301
4.849
0.173
0.831
1.898
2.705
2
2
89.528
2.573
4.204
0.133
0.765
86.943
2.488
5.097
0.183
0.813
2.063
2.912
3
2
98.281
2.739
4.248
0.145
0.733
95.842
2.671
5.339
0.192
0.794
2.225
3.116
4
2
105.404
2.879
4.857
0.168
0.761
104.476
2.849
5.571
0.200
0.775
2.383
3.314
5
2
112.415
3.012
5.787
0.181
0.744
112.756
3.020
5.790
0.208
0.754
2.535
3.505
6
2
118.957
3.147
5.654
0.205
0.734
120.594
3.184
5.993
0.216
0.731
2.681
3.686
7
2
124.658
3.261
5.843
0.238
0.753
127.901
3.337
6.177
0.223
0.706
2.818
3.857
8
2
131.786
3.433
6.950
0.253
0.720
134.589
3.479
6.338
0.230
0.679
2.944
4.014
9
2
137.722
3.566
6.500
0.264
0.676
140.569
3.608
6.474
0.236
0.649
3.060
4.156
10
2
144.426
3.681
7.298
0.256
0.729
145.754
3.722
6.581
0.240
0.616
3.163
4.281
11
2
150.574
3.818
7.670
0.278
0.606
150.083
3.820
6.657
0.244
0.580
3.252
4.389
12
2
156.583
3.963
7.295
0.273
0.558
153.611
3.904
6.705
0.247
0.541
3.328
4.479
13
2
158.923
4.000
6.149
0.242
0.391
156.424
3.974
6.730
0.250
0.501
3.393
4.555
14
2
160.849
4.069
6.429
0.209
0.344
158.606
4.032
6.737
0.251
0.460
3.447
4.617
15
2
161.704
4.126
6.674
0.240
0.461
160.241
4.079
6.728
0.253
0.420
3.491
4.667
16
2
162.002
4.140
6.219
0.251
0.364
161.413
4.118
6.710
0.254
0.382
3.528
4.708
17
2
162.805
4.151
6.661
0.244
0.359
162.208
4.149
6.687
0.254
0.347
3.558
4.740
18
2
162.208
4.177
6.344
0.263
0.228
162.709
4.174
6.662
0.255
0.315
3.582
4.766
19
2
163.320
4.193
6.174
0.266
0.227
163.000
4.195
6.640
0.255
0.288
3.602
4.788
20
2
163.411
4.175
7.485
0.250
0.294
163.167
4.213
6.626
0.255
0.266
3.619
4.807
21
2
161.858
4.194
6.643
0.253
0.397
163.281
4.229
6.624
0.256
0.252
3.634
4.825
22
2
162.038
4.263
6.058
0.257
0.086
163.358
4.244
6.632
0.257
0.243
3.647
4.842
23
2
161.916
4.269
7.447
0.258
0.294
163.405
4.258
6.649
0.258
0.239
3.658
4.857
24
2
162.774
4.270
7.195
0.259
0.236
163.425
4.270
6.675
0.259
0.240
3.667
4.872
25
2
162.763
4.235
6.405
0.261
0.288
163.423
4.280
6.707
0.260
0.244
3.676
4.885
26
2
163.198
4.278
6.312
0.268
0.325
163.404
4.289
6.744
0.261
0.251
3.683
4.896
27
2
163.593
4.300
7.471
0.272
0.356
163.372
4.297
6.786
0.262
0.260
3.689
4.906
28
2
163.380
4.293
6.569
0.274
0.354
163.332
4.304
6.829
0.262
0.271
3.694
4.914
29
2
162.909
4.305
5.527
0.234
0.269
163.288
4.309
6.874
0.263
0.281
3.698
4.920
30
2
163.515
4.318
7.695
0.262
0.269
163.246
4.314
6.919
0.263
0.292
3.702
4.925
31
2
164.013
4.316
6.712
0.264
0.212
163.208
4.316
6.962
0.263
0.301
3.705
4.927
32
2
163.674
4.331
7.194
0.289
0.248
163.176
4.318
7.002
0.262
0.309
3.708
4.928
33
2
163.856
4.353
6.710
0.243
0.269
163.148
4.319
7.039
0.262
0.315
3.711
4.928
34
2
163.344
4.341
7.496
0.278
0.283
163.124
4.319
7.072
0.261
0.320
3.713
4.926
35
2
163.531
4.309
6.544
0.221
0.200
163.103
4.319
7.100
0.260
0.323
3.715
4.923
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 24
UNSMOOTHED
SMOOTHED
BOUNDS FOR
LOG BW
36
2
163.211
4.304
7.656
0.259
0.362
163.085
4.318
7.122
0.259
0.325
3.717
4.920
37
2
164.099
4.333
6.902
0.276
0.391
163.070
4.317
7.137
0.257
0.324
3.719
4.916
38
2
162.956
4.354
7.860
0.236
0.328
163.056
4.316
7.145
0.256
0.322
3.720
4.913
39
2
162.702
4.305
7.675
0.251
0.396
163.043
4.316
7.144
0.255
0.318
3.722
4.909
40
2
162.678
4.303
7.397
0.279
0.302
163.031
4.315
7.134
0.254
0.311
3.724
4.906
41
2
161.638
4.301
6.643
0.260
0.367
163.018
4.315
7.114
0.253
0.302
3.726
4.904
42
2
163.154
4.298
7.131
0.264
0.300
163.004
4.315
7.085
0.252
0.291
3.729
4.902
43
2
162.756
4.311
6.773
0.243
0.233
162.987
4.316
7.050
0.251
0.280
3.731
4.901
44
2
162.821
4.290
6.921
0.260
0.301
162.965
4.317
7.010
0.251
0.267
3.734
4.901
45
2
162.737
4.290
5.720
0.238
0.240
162.937
4.319
6.967
0.250
0.255
3.736
4.901
46
2
162.146
4.348
7.539
0.240
0.245
162.902
4.320
6.923
0.250
0.243
3.739
4.901
47
2
163.495
4.360
7.326
0.249
0.254
162.858
4.322
6.879
0.249
0.233
3.742
4.902
48
2
163.566
4.324
6.311
0.260
0.042
162.803
4.324
6.837
0.249
0.224
3.745
4.903
49
2
162.858
4.338
6.338
0.240
0.262
162.737
4.326
6.800
0.249
0.218
3.748
4.904
50
2
162.498
4.345
6.919
0.221
0.248
162.657
4.328
6.768
0.248
0.215
3.750
4.906
51
2
162.610
4.320
5.990
0.244
0.167
162.563
4.330
6.743
0.248
0.215
3.753
4.907
52
2
161.654
4.267
7.879
0.252
0.347
162.456
4.332
6.725
0.248
0.218
3.756
4.908
53
2
163.379
4.325
6.657
0.243
0.260
162.336
4.334
6.713
0.247
0.223
3.758
4.909
54
2
162.049
4.368
7.027
0.262
0.235
162.207
4.335
6.706
0.247
0.231
3.761
4.910
55
2
162.694
4.364
6.633
0.270
0.178
162.068
4.337
6.703
0.247
0.240
3.763
4.911
56
2
162.638
4.332
6.787
0.245
0.115
161.922
4.338
6.703
0.246
0.249
3.764
4.911
57
2
160.512
4.320
7.084
0.225
0.301
161.770
4.338
6.705
0.246
0.259
3.766
4.910
58
2
160.963
4.372
7.017
0.306
0.287
161.613
4.338
6.708
0.245
0.269
3.767
4.909
59
2
160.849
4.305
6.991
0.223
0.266
161.454
4.338
6.712
0.245
0.278
3.768
4.907
60
2
161.262
4.349
6.422
0.276
0.414
161.293
4.337
6.716
0.244
0.286
3.769
4.905
61
2
163.010
4.374
7.148
0.215
0.380
161.131
4.335
6.718
0.243
0.292
3.769
4.901
62
2
160.395
4.373
6.512
0.216
0.266
160.970
4.333
6.719
0.242
0.296
3.769
4.897
63
2
161.629
4.282
6.589
0.229
0.310
160.808
4.330
6.719
0.241
0.299
3.769
4.892
64
2
160.269
4.333
6.028
0.243
0.248
160.647
4.327
6.718
0.240
0.300
3.768
4.886
65
2
161.070
4.284
6.539
0.219
0.240
160.485
4.324
6.716
0.239
0.301
3.768
4.880
66
2
159.425
4.320
5.689
0.207
0.331
160.324
4.320
6.713
0.238
0.300
3.766
4.873
67
2
160.241
4.318
6.903
0.259
0.351
160.163
4.316
6.710
0.237
0.299
3.765
4.866
68
2
158.931
4.237
7.056
0.232
0.300
160.001
4.311
6.707
0.235
0.297
3.764
4.858
69
2
159.863
4.288
6.687
0.237
0.287
159.839
4.307
6.703
0.234
0.295
3.763
4.851
70
2
160.263
4.361
6.986
0.243
0.257
159.678
4.302
6.699
0.233
0.292
3.761
4.843
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 25
BOUNDS FOR
UNSMOOTHED
SMOOTHED
LOG BW
71
2
159.678
4.318
7.340
0.259
0.275
159.516
4.298
6.695
0.231
0.290
3.760
4.836
72
2
158.699
4.295
6.225
0.254
0.262
159.355
4.293
6.691
0.230
0.287
3.758
4.828
73
2
159.618
4.307
7.187
0.234
0.302
159.193
4.288
6.687
0.229
0.285
3.757
4.820
74
2
159.042
4.280
6.425
0.216
0.237
159.032
4.284
6.683
0.227
0.282
3.755
4.812
75
2
158.332
4.276
7.461
0.222
0.083
158.870
4.279
6.679
0.226
0.280
3.754
4.805
76
2
159.769
4.324
5.740
0.209
0.297
158.709
4.275
6.675
0.225
0.277
3.752
4.797
77
2
158.186
4.270
5.841
0.238
0.248
158.547
4.270
6.671
0.223
0.275
3.751
4.789
78
2
158.001
4.258
7.098
0.216
0.292
158.386
4.266
6.667
0.222
0.272
3.750
4.782
79
2
158.586
4.224
7.097
0.208
0.461
158.224
4.261
6.663
0.220
0.270
3.748
4.774
80
2
158.063
4.257
6.659
0.219
0.267
3.747
4.766
81
2
157.901
4.252
6.655
0.218
0.265
3.745
4.759
82
2
157.740
4.247
6.651
0.216
0.262
3.744
4.751
83
2
157.578
4.243
6.648
0.215
0.260
3.742
4.743
84
2
157.417
4.238
6.644
0.214
0.257
3.741
4.736
85
2
157.255
4.234
6.640
0.212
0.255
3.739
4.728
86
2
157.094
4.229
6.636
0.211
0.252
3.738
4.720
87
2
156.932
4.225
6.632
0.210
0.250
3.737
4.712
88
2
156.771
4.220
6.628
0.208
0.247
3.735
4.705
89
2
156.609
4.215
6.624
0.207
0.245
3.734
4.697
90
2
156.448
4.211
6.620
0.206
0.242
3.732
4.689
91
2
156.286
4.206
6.616
0.204
0.240
3.731
4.682
92
2
156.125
4.202
6.612
0.203
0.237
3.729
4.674
93
2
155.963
4.197
6.608
0.202
0.235
3.728
4.666
94
2
155.802
4.193
6.604
0.200
0.232
3.727
4.659
95
2
155.640
4.188
6.600
0.199
0.230
3.725
4.651
96
2
155.479
4.183
6.596
0.198
0.227
3.724
4.643
97
2
155.317
4.179
6.592
0.196
0.225
3.722
4.636
98
2
155.156
4.174
6.588
0.195
0.222
3.721
4.628
99
2
154.994
4.170
6.584
0.194
0.220
3.719
4.620
100
2
154.833
4.165
6.580
0.192
0.217
3.718
4.613
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 26
5. Comparison between 2009-2014 and 2003-2014
The fitted models for 2009-2014 are contained in Table 4 and in the tab "Parameters" of the
accompanying Excel file "means.2009 to 2014.102016.xlsx". We give unsmoothed and
smoothed parameters for each age and gender. Using the same approach, the fitted
parameters for 2003-2014 are contained in the tab "Parameters" of the accompanying Excel file
"means.2003 to 2014.102016.xlsx".
The following Table 5 contains a comparison of the parameters between the two sets of years.
The differences and percentage differences are relative to the baseline of 2003-2014:
Difference = Value for 2009-2014 - Value for 2003-2014
Percentage Difference = Difference / Value for 2003-2014 x 100
The tabulated means and maxima are for each gender across all ages 0-79 years, for both the
unsmoothed and smoothed parameters.
The mean differences are between -0.14 and 0.07 across all parameters, so there is only a
small trend in the parameters. (Note that the two periods overlap, but any difference between
the overlapping periods implies a difference between 2003-2008 and 2009-2014.)
The differences are small for the mean parameters: the maximum unsigned percentage
differences are at most 1.7 percent for the unsmoothed mean parameters and at most 0.6
percent for the smoothed mean parameters.
The differences are much higher for the standard deviations and the correlations. For the
unsmoothed data, the maximum unsigned percentage difference is 17 percent for the standard
deviation of the HT and 69 percent for the correlation. For the smoothed data, the differences
are much smaller: the maximum unsigned percentage difference is 5.4 percent for the standard
deviation of the HT and 10.7 percent for the correlation.
The mean unsigned percentage difference is at most 13.7 percent across all unsmoothed
parameters and at most 3.4 percent across all smoothed parameters.
The lack of a large trend between the two time periods, and the small percentage differences for
the smoothed parameters, suggest that it will not make very much difference which set of years
is used for the APEX model inputs. We recommend using the more recent data from 2009-
2014.
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 27
Table 5. Differences between Parameters for 2009-2014 and 2003-2014 (Baseline)
Maximum
Mean
Mean Unsigned
Unsigned
Percentage
Percentage
Percentage
Statistic
Gender
Mean Difference
Difference
Difference
Difference
Mean HT
1
-0.12
-0.07
0.22
0.86
Mean HT
2
-0.14
-0.09
0.23
0.67
Mean Log BW
1
0.00
0.05
0.27
0.91
Mean Log BW
2
0.01
0.16
0.34
1.65
| Std Dev HT
1
-0.04
-0.57
4.19
17.42
I Std Dev HT
2
0.07
0.96
4.47
10.08
c/5 Std Dev Log
1
0.00
0.49
3.82
11.59
3 BW
Std Dev Log
2
0.00
1.04
4.09
13.49
BW
Correlation
1
-0.01
-1.67
10.65
51.40
Correlation
2
0.00
0.22
13.71
68.67
Mean HT
1
-0.12
-0.07
0.12
0.32
Mean HT
2
-0.14
-0.09
0.12
0.40
Mean Log BW
1
0.00
0.05
0.08
0.21
Mean Log BW
2
0.01
0.17
0.19
0.61
"§ Std Dev HT
1
-0.04
-0.58
1.69
5.41
B Std Dev HT
Q
2
0.07
1.00
1.48
4.42
E Std Dev Log
1
0.00
0.50
1.27
2.98
W BW
Std Dev Log
2
0.00
1.06
1.20
4.28
BW
Correlation
1
-0.01
-1.16
2.19
7.00
Correlation
2
0.00
-1.21
3.37
10.71
6. Effect on HT and WT in APEX using Updated
Algorithm
6.1. Description of APEX Runs and Analysis
To summarize the effect of the new algorithm on simulated HT and WT values, we conducted
two separate APEX runs: one employing the HT and BW calculations based on the 1999-2004
NHANES data (referred to as the "old method" in this section) and one employing the HT and
BW calculation method based on the 2009-2014 NHANES data as proposed in this
memorandum (the "new method"). Apart from this difference, the two APEX runs were identical.
Both APEX runs employed 100,000 profiles and modeled ages 0-99 years old. This produced a
set of 100,000 HT, WT, and body mass index (BMI) values (one of each for each profile).
We analyzed statistics of the HT, WT, and BMI of the profiles generated in APEX for each of 14
age bins. We created the age bins so that they each (except for the oldest bin) contained a
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 28
roughly equal number of profiles: 5-year bins ages 0-55 years, then single bins for 55-62 years,
62-75 years, and 75-99 years. We present in Figure 6-1 the number of profiles in each age bin.
c
zs
o
O
20
40
60
80
100
Age Mid-points (years)
Figure 6-1. Number of Profiles in each Age Bin from APEX Runs (100,000 profiles)
6.2. Comparison of HT, WT, and BMI Results
Table 6 presents a statistical summary and comparison of the HT, WT, and BMI values
generated in the two APEX runs employing the old and new methods. These statistics were
calculated only on the basis of gender and not on the basis of age bin.
We also compared the outputs of the two methods on the basis of age bin. Figure 6-2 through
Figure 6-7 present the mean and standard deviation of HT, WT, and BMI values from the old
and new methods in each age bin for the 100,000 profiles generated in APEX.
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 29
Table 6. Statistical Summary of HT, WT, and BMI in APEX using Old and New Methods
%
Difference
Variable
Gender
N
Mean
St. Dev
Min
Max
in Mean
Old
M
49,473
164.948
25.582
63.058
205.788
-0.108
Height
New
49,473
164.770
26.038
58.240
205.776
(cm)
Old
F
50,527
154.176
20.525
63.251
187.350
0.126
New
50,527
154.371
21.230
54.668
190.061
Old
M
49,473
73.943
28.745
3.600
199.198
2.085
Weight
New
49,473
75.484
29.782
6.392
148.412
(kg)
Old
F
50,527
65.056
24.744
3.700
165.998
2.373
New
50,527
66.600
25.885
5.646
138.102
Old
M
49,473
25.611
6.374
5.385
59.404
2.075
BMI
New
49,473
26.143
6.637
10.162
54.052
CM
E
O)
Old
F
50,527
26.189
7.440
5.491
63.184
1.824
New
50,527
26.667
7.690
10.155
61.574
E
o
O)
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 30
o
oo
o
to
o
o
fM
o
o
o
CO
0
HIHHH
20
* Old method
* New method
40
60
80
100
Age bin centers (years)
Figure 6-3. Mean ± Standard Deviation of HT for Females
o
CM
O
o
o
oo
o
(O
o
^r
o
C\l
o -
0
" ?!
* Old method
• New method
20
40
60
80
r
100
Age bin centers (years)
Figure 6-4. Mean ± Standard Deviation of WT for Males
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 31
o
CM
O
O
O
oo
o
CD
O
O
OJ
o -
0
II I
* Old method
• New method
20
40
60
l
80
100
Age bin centers (years)
Figure 6-5. Mean ± Standard Deviation of WT for Females
o
ID
CO
O
CO
ID
CN
O
fM
ID
nr
0
o "
<> o
o <1
II
• Old method
* New method
~~r~
20
40
60
80
100
Age bin centers (years)
Figure 6-6. Mean ± Standard Deviation of BMI for Males
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 32
CD
^r
O0
CD
-4—1
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 33
For BMI values, the new method substantially decreased the standard deviation for ages
0-15 years, with generally lower BMI means as well (except in the youngest age group).
For adults, there is a general increase in the means and standard deviations of BMI
values using the new method, especially for males.
For a previous assessment, we generated the distribution of BMI values shown in Figure
6-8, from NHANES 2003-2014 data. The distributions of BMI values in these simulations
are similar to the NHANES BMI distributions. The majority of BMI values from NHANES
are between about 15 and 35 kg/m2, and the mean BMI values simulated here also fall
within that range. BMI values below 15 kg/m2 and above 40 kg/m2 are relatively rare in
the NHANES data, and the same is true of the BMI values simulated here.
0 20 40 60 80 120
BMI
Legend
Female
Male
Both
BMI Classification Breaks:
18.5, 25, 30
Figure 6-8. Distribution of BMI Values (kg/m2) from NHANES 2003-2014
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 34
Attachment A. Distributions of Body Weight
Distribution of BMXWT
Gender = 1
Gender = 2
125 -
100-
75-
50-
25-
0-
150 -
125 -
100-
75-
50-
25-
0-
i iiiii—i—i—r~
10 15 20 25 30 35 40 45 50
i ill r i—i—i—r~
10 15 20 25 30 35 40 45 50
0 5
Weight (kg)
| Curves Normal Lognomial(Theta=0 Sigma=EST Zeta=EST) |
Distribution of BMXWT
Gender = 1
Gender = 2
100-
o
60-
ii
H.
<
40-
20-
S 100-
60-
ii
ci
<
40-
120
20-
0
40
80
160
200
0
40
80
120
160
200
Weight (kg)
| Curves Normal Lognonml(Theta=0 Sigma=EST Zeta=EST) |
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 35
1—1
60-
50-
o
40-
bi
<
30-
20-
"S
O
i i
o c
js
Ph
60-
50-
in
II
ft
40-
30-
<
o o
i i
o-
Distribution of BMXWT
Gender = 1
Gender = 2
J
r
r—— r
60 100 140 180 220
I 1 1—
100 140 180 220
Weight (kg)
¦ Normal Lognormal(Theta=0 Sigma=EST Zeta=EST) |
Curves
Distribution of BMXWT
Gender = 1
Gender = 2
60-
50-
40-
30-
20-
10-
0-
60-
50-
40-
30-
20-
10-
0-
J
.A
T 1-
40 80 120 160 200
Weight (kg)
40 80 120 160 200
¦ Normal Lognormal(Theta=0 Sigma=EST Zeta=EST) |
Curves
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 36
1—1
50-
o
II
40-
30-
<
20-
10-
s=
o
0 -
jp
Ph
50-
40-
*0
II
30-
&
<
0 o
1 i
LJ
o-
Distribution of BMXWT
Gender = 1
Gender = 2
J
fa
1
/;
A.
i —V— I i 1 1 1 1 1 1 r
60 100 140 180 220 20 60 100 140 180 220
Weight (kg)
¦ Normal Lognormal(Theta=0 Sigma=EST Zeta=EST) |
Curves
Distribution of BMXWT
Gender = 1
Gender = 2
60-
40-
20-
0-
80-
60-
40-
20-
-A
K.
120 160
A
200 0 40
Weight (kg)
120 160
200
¦ Normal Lognorroal(Theta=0 Sigma=EST Zeta=EST) |
Curves
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 37
Attachment B. Distributions of Height
Distribution of BMXHT
Gender = 1
Gender = 2
100-
ii
60-
H.
<
40-
20-
P 100-
II
60-
a
<
40-
20-
50
70
90
110
130
150
50
70
90
110
130
150
Standing Height (cm)
| Curves Normal Lognormal(Theta=0 Sigma=EST Zeta=EST) |
Distribution of BMXHT
Gender = 1
Gender = 2
60-
50-
40-
30-
20-
10-
c
o-
(1>
60-
50-
40-
30-
20-
10-
o-
J
110
A
I I
210 110 130
Standing Height (cm)
170
I
190
I
210
Curves
¦ Normal
Lognormal(Theta=0 Sigma=EST Zeta=EST) |
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 38
Distribution of BMXHT
Gender = 1
Gender = 2
40-
30-
20-
10-
a
<)-
40-
30-
20-
10-
o-
N
J
i i
\
150 160 170 180 190 140 150 160 170 180 190
Standing Height (cm)
¦ Normal Lognormal(Theta=0 Sigma=EST Zeta=EST) |
Curves
Distribution of BMXHT
Gender = 1
Gender = 2
60-
50-
40-
30-
20-
10-
G
o-
P-i
60-
50-
40-
30-
20-
10-
o-
ft
t 1 1 —i 1—
130 150 170 190 210 230
-1 r~— 1 1—
130 150 170 190 210 230
Standing Height (cm)
¦ Normal Lognorroal(Theta=0 Sigma=EST Zeta=EST) |
Curves
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 39
Distribution of BMXHT
Gender = 1
Gender = 2
o
l/"l
II
bi
<
80-
60-
40-
20-
f
h
r
\i
o
>o
II
a
<
Perc
to 0\ oo
o o o o o c
1 1 1 1 1
f
\
i
V ,
f
I
h
1 1
120 140 160 180 200 220 120 140 160 180 200 220
Standing Height (cm)
| Curves Normal Lognorroal(Theta=0 Sigma=EST Zeta=EST) |
Distribution of BMXHT
Gender = 1
Gender = 2
60-
40-
20-
0-
nS 80"
60-
40-
20-
J
A
i i r
120 140 160 180 200 220
120 140 160 180 200 220
Standing Height (cm)
¦ Normal Lognorroal(Theta=0 Sigma=EST Zeta=EST) |
Curves
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 40
Attachment C. Scatter Plots of Log BW versus HT
Scatter Plot
3.75 ¦
3.50 -
3.00 -
2.75-
Observations 295
Correlation 0.7408
8ts>cP P° /
(5qa>
l—
100
-1—
110
-1—
120
-1—
130
Standing Height (cm)
| Prediction Ellipses
80%
95%
Years=2009-2014 Age=5 Gender=l
Scatter Plot
3.50 -
3.25 ¦
j? 3.00 -
2.75-
2.50-
Observations 279
Correlation 0.7436
On
isP
100
-1—
110
120
130
Standing Height (cm)
| Prediction Ellipses
95%
Years=2009-2014 Age=5 Gender=2
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 41
Scatter Plot
5.00-
Observations 237
Correlation 0.4851
°0
4.75-
4.50-
oo
4.25-
4.00-
3.75 -
150 160 170 180 190
Standing Height (cm)
| Prediction Ellipses 80°6 95°6 |
Years=2009-2014 Age=15 Gender=l
Scatter Plot
Observations 223
Correlation 0.4608
4.75-
4.50-
ooo
'<9 oo'
-S
o o
60
g 4.25 -
60
O
/ o
4.00-
oo
3.75 -
140
150
160
170
180
Standing Height (cm)
| Prediction Ellipses 80°6 95°6 |
Years=2009-2014 Age=15 Gender=2
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 42
Scatter Plot
Observations 131
Correlation 0.3881
5.00-
oo
4.75-
o o
o o
Cfj
I 4.50-
00
O
oo
<§>0 Oq
Oo
O O
o OO
o°
4.25-
OO
4.00-
160
170
180
190
Standing Height (cm)
| Prediction Ellipses 80% 95°6 |
Years=2009-2014 Age=25 Gender=l
Scatter Plot
Obs ervations 152
Correlation 0.2878
5.0-
O O
/
o O
o
,«p
.O CD
O
4.0-
o ®
150
160
170
180
Standing Height (cm)
| Prediction Ellipses — 80% 95% |
Years=2009-2014 Age=25 Gender=2
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 43
Scatter Plot
5.25-
Observations 159
Correlation 0.5456
5.00-
0°
4.75-
oo,
4.50-
4.25-
o
4.00-
160 170 180 190
Standing Height (cm)
| Prediction Ellipses 80% 95°6 |
Years=2009-2014 Age=40 Gender=l
Scatter Plot
Obs ervations 165
Correlation 0.3016
5.0-
.op
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 44
Scatter Plot
Observations 174
Correlation 0.3803
5.00-
4.75-
'oo
o o
.00
OO
5 4.50 -
S
BO
O
o Oo
0
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 45
Scatter Plot
Obs ervations ^ 1
Correlation 0.1948
4.6-
.op
4.2-
4.0-
160
170
180
190
Standing Height (cm)
| Prediction Ellipses — 80% 95%
Years=2009-2014 Age=79 Gender=l
Scatter Plot
Obs ervations 63
Correlation 0.4614
140 150 160 170
Standing Height (cm)
| Prediction Ellipses — 80% 95% |
Years=2009-2014 Age=79 Gender=2
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 46
Attachment D. Unsmoothed and Smoothed Values for the Five Joint-distribution
Parameters
NHANES 2009 to 2014
Cubic spline of mean of log body weight (kg) versus age
Gender = 1
Gender = 2
0 20 40 60 80 100 0 20 40 60 80 100
AGE
| O the mean, logweight linear Predictor |
NHANES 2009 to 2014
Cubic spline of standard deviation of log body weight (kg) versus age
Gender = 1
Gender = 2
| O the standard deviation, logweight"
Linear Predictor
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 47
NHANES 2009 to 2014
Cubic spline of mean of height (cm) versus age
Gender = 1
Gender = 2
| O the mean, BMXHT
Linear Predictor |
NHANES 2009 to 2014
Cubic spline of standard deviation of height (cm) versus age
Gender = 1
Gender = 2
Q the standard deviation, BMXHT
Linear Predictor
-------
Joint Distributions of Body Weight and Height for use in APEX
April 20, 2017
Page 48
NHANES 2009 to 2014
Cubic spline of correlation between log body weight and height versus age
Gender = 1
Gender = 2
o
1°
0?p
TO
\ °
s
\o °o°°° 0°° '-V
V o % „
Vo O ° °
° O 0 0 \
° °% 0
0 o
o
° 0
o
\ ^ o
o\ ° oo Q
oT ® ° ° o o
\ ° V""S. 0°
V°o /o \°
00 °o ^
o°
o
O 0
o
0 20 40 60 80 100 0 20 40 60 80 100
AGE
| Q corr — Linear Predictor]
-------
APPENDIX H
ICF FINAL MEMO: RESTING METABOLIC RATE (RMR)
AND VENTILATION RATE (VE) ALGORITHM REFINEMENTS
H-l
-------
~ICF
Memorandum
To: John Langstaff and Stephen Graham, U.S. EPA OAQPS
From: Jessica Levasseur, Graham Glen, and Chris Holder, ICF
Date: February 17, 2017
Re: WA 4-52 Task 4: RMR and Ve Algorithm Refinements
1. Introduction
Ventilation rate (Ve) and resting metabolic rate (RMR) are two key variables used to assign
physiological characteristics to individuals in a simulated population in the U.S. Environmental
Protection Agency (EPA) Air Pollutants Exposure model (APEX). These and other simulated
aspects of individuals' physiology, combined with population demographics as well as activity
data drawn from the EPA Comprehensive Human Activity Database, are used to estimate
exposure to air pollutants in APEX (Isaacs, 2008). The current implementation of algorithms
used to estimate RMR and Ve in APEX are based on studies that are 30 and 10 years old,
respectively (Schofield, 1985; Graham and McCurdy, 2005). The algorithm for Ve also leads to
some sharp discontinuities between modeled age groups.
Under this task, ICF ("we") implemented refinements (i.e., technical improvements) to
RMR and Ve calculations to improve the usefulness or accuracy of APEX simulations. To
complete this task, we conducted multiple literature searches to identify literature relevant to
developing appropriate RMR and Ve algorithms. We identified additional sources of data to
augment the RMR dataset provided to us by the EPA. We identified no new data on Ve to add to
the dataset provided by the EPA.
In this memorandum, we describe these literature searches, the datasets used to develop the
updated RMR and Ve algorithms, and the performance in APEX of the updated algorithms for
RMR and Ve compared to the existing algorithms. Using updated datasets, we aimed to improve
the RMR and Ve algorithms.
Note that all references to "log" or "logarithm" refer to the natural logarithm, not the base-10
logarithm.
2. Ve and RMR Literature Search
In McCurdy (2015), titled "Physiological Parameters and Physical Activity Data for Evaluating
Exposure Modeling Performance: a Synthesis," the author expounds upon important factors that
influence physiological parameters and affect exposure and dose modeling. He also provided a
separate document of "unused references" that contained relevant publications he was unable
to fully evaluate in the synthesis.
In focusing on sections containing relevant mentions of Ve and RMR, we identified 321
publications as potentially useful sources of literature that warranted further investigation. We
then scrutinized these publication titles and abstracts for particular relevance to RMR or Ve
-------
WA 4-52 Task 4 RMR and VE Algorithm Refinements
February 17, 2017
Page 2
prediction or to refining the algorithms for RMR or Ve. Of these 321 publications, we identified
53 as potentially relevant for our task.
We identified population gaps within the RMR and Ve datasets initially provided by the EPA,
namely women, children, older adults, and obese people (for the Ve dataset) and men and older
adults (for the RMR dataset). We focused our literature search on publications specifically
relevant to these underrepresented subpopulations.
2.1. RMR
Only 13 publications were relevant to addressing the population gaps present within the RMR
dataset. We conducted a cited-references search on these 13 RMR publications, returning all
the publications that cite or are cited by these 13 publications. From the RMR cited-references
search, we focused on publications that contained each of the following characteristics:
¦ measured RMR (or an equivalent physiological measurement);
¦ contained information on body weight, height, and sex; and
¦ used primary data from at least 200 subjects or defined new predictive equations.
We identified seven publications that had these characteristics. We acquired new RMR data
from one of these publications—the Oxford-Brookes database (Henry, 2005)—adding
more than 13,000 unique data points to an RMR dataset provided by EPA.
2.2. VE
We conducted a separate literature search for Ve, as requested by the EPA, on those articles
published between 2000 and 2010. Conducting a PubMed search on the following search
criteria returned 387 publications:
¦ "Ventilation Rate" OR "VE" AND (Equation/s OR algorithm/s)
¦ Humans only
¦ English only.
Assessing these abstracts for new potential sources of data and new potential equations, 16
articles appeared relevant. After acquiring full articles, we identified two as possible sources of
data but none had relevant algorithms for Ve prediction. We were unable to acquire these new
datasets for Ve.
3. Updated RMR Dataset
" script - "'ataset
The initial RMR dataset provided to us by the EPA is described in the research report Analyses
of Resting Energy Expenditure (REE) data for US residents by Kriti Sharma, Thomas McCurdy,
and Stephen Graham (no date), which describes a database of 763 individuals ages 4 to 89.
-------
WA 4-52 Task 4 RMR and VE Algorithm Refinements
February 17, 2017
Page 3
3.2. Descripl sokes Database
Published in 2005, Dr. Jaya Henry created the Oxford-Brookes (OB) database that combined
data from a variety of sources, resulting in more than 10,000 RMR values. For a detailed
summary of the OB database creation, please see Henry (2005) and IOM (2005).
3.3. Merging Datasets
We removed duplicates between the OB database and the initial RMR dataset (provided by the
EPA). In addition to information on study author and year of study, this dataset contains
information on:
¦ sex,
¦ age,
¦ BM,
¦ height, and
¦ RMR.
We deleted observations missing any of the following values: RMR, BM, age, or sex. The full
dataset contains 16,254 observations (9,377 males and 6,877 females). Of these, 39 males
and 33 females were missing reported heights. Therefore, for analyses requiring height (see
Section 5), we used a smaller dataset of 16,182 observations (9,338 males and 6,844 females).
4. Ve Dataset
scription of Dataset
Dr. William Adams of UC Davis constructed the Ve dataset provided to us by the EPA. Graham
and McCurdy (2005) also used his data. Dr. Adams collected data from 32 panel studies over
25 years. In addition to information on test exercise parameters, this dataset contains
information on:
¦ sex,
¦ age,
¦ BM,
¦ height,
¦ oxygen consumption rate (VO2), and
¦ VE.
EPA recommended the removal of four data points for quality-assurance reasons. The final Ve
dataset, with no new data added (none were identified), contains 6,636 observations, with
4,565 males and 2,071 females.
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 4
5. Updated RMR Algorithms
Using the new RMR dataset, and with a goal of improving the RMR algorithm while reducing
discontinuities in RMR between age groups, we developed new algorithms for estimating RMR
in APEX. The algorithms follow the general format of a multiple linear regression (MLR) model,
which is described as:
y = 0i*i + 02*2 + + a + (!)
Where:
y =
variable of interest
P =
coefficient of input variable
x =
input variable
a =
intercept
£ =
residual
V =
distribution mean
o =
distribution standard deviation
n =
number of independent regression variables
/ =
person-specific index
It is generally known that RMR and BM, as well as RMR and age, are not exactly linearly
related; the algorithms developed here use BM, age, and the natural logarithms of BM and
(age+1). The "+1" modifier allows APEX to round age upwards instead of downwards to whole
years, which is necessary to avoid undefined log(0)1 values.
To place all the RMR data on an equal footing, we first rounded all ages down to integer values.
Instead of dividing the data at preset age boundaries (as was done in the existing APEX
algorithm), we repeatedly altered the age boundaries until the residual sum of squares was
minimized. Five age groups were sufficient to capture the data for both males and females,
though each sex required different age groups. These age groups are shown in Table 1 and
Table 2 below, along with the optimal regression parameters (not including height) for each age
group and sex. Note that all people over age 99 are treated as 99 years old by APEX and
therefore are included in the oldest age groups.
Table 1. Optimal RMR Regression Parameters for Males by Age Group (n = 9,377), Height Not
Included
Age
Group
n
BM
log(BM)
Age
log(Age)
Intercept
St. Dev.
0-5
625
13.19
270.2
-18.34
131.3
-208.5
69.10
6-13
1355
10.21
260.2
13.04
-205.7
333.4
115.3
14-24
4123
0.207
1078.
115.1
-2794.0
3360.6
161.1
25-54
2531
2.845
729.6
3.181
-191.6
-1067.
178.2
55-99
743
9.291
264.8
-5.288
181.5
-705.9
163.6
Units: RMR = kilocalories/day; BM = kilograms; Age = years
1 Note that all references to "log" or "logarithm" refers to the natural logarithm, not base-10 logarithm.
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 5
Table 2. Optimal RMR Regression Parameters for Females by Age Group (n = 6,877), Height Not
Included
Age
Group
n
BM
log(BM)
Age
log(Age)
Intercept
St. Dev.
0-5
625
11.94
261.5
-22.31
120.9
-183.6
64.16
6-13
1618
5.296
409.1
40.37
-524.9
392.7
99.43
14-29
2657
0.968
676.9
40.89
-1002.
772.7
143.1
30-53
1346
4.935
355.4
16.28
-896.0
2225.
145.3
54-99
631
2.254
445.9
5.464
-489.9
944.2
124.5
Units: RMR = kilocalories/day; BM = kilograms; Age = years
Input values should be in units of kilograms (kg) for BM and years for age, with the RMR
estimate in kilocalories/day (kcal/d). For example, using Equation (1) with information from
Table 1, a 20-year-old male weighing 75 kg would be assigned an RMR as follows:
RMR = 0.207 x 75 + 1078 x log(75) + 115.1 x 20 - 2794 x log(21) + 3360.6
+ 161.1 x JV(0,1)
RMR = 1826.4^^ + 161.1 x JV(0,1) (for any 20-year-old male weighing 75 kg)
While the overall r2 values are fairly high (0.820 males, 0.816 females), the r2 for particular age
groups varies from over 0.9 (for boys and girls ages 0-5 years) to less than 0.6. Transforming
RMR, and including height and log(height) as input variables, did not improve overall fit. For
adults in particular, a substantial amount of variation remains in the residual error of the new
RMR algorithms. To reduce this, more modeling variables would be required than are available
in the RMR dataset.
When including height, the optimal regression parameters are as shown in Table 3 and Table 4
for males and females, respectively. The overall r2 values are 0.815 for males and 0.816 for
females when height is included in the regression. These are not appreciably different from the
regressions without height. Therefore, the proposed updates to RMR regressions do not
use height.
Table 3. Optimal RMR Regression Parameters for Males by Age Group (n = 9,338), Height Included
Age
Group
n
BM
log(BM)
Age
log(Age)
HT
log(HT)
Intercept
St.
Dev.
0-5
596
17.61
106.3
-17.93
87.37
-368.9
676.3
607.6
68.60
6-13
1355
12.64
149.3
30.91
-417.0
-1498.
2151.5
2344.9
115.0
14-24
4123
0.0309
1098.6
114.3
-2777.
31.45
-101.2
3250.7
161.1
25-54
2522
4.692
481.5
2.422
-136.3
1590.
-2014.
-1961.3
176.6
55-99
742
12.60
-108.4
-5.151
170.6
-927.2
2405.
982.6
160.7
Units: RMR = kilocalories/day; BM = kilograms; Age = years; Height = meters
Table 4. Optimal RMR Regression Parameters for Females by Age Group (n = 6,844), Height
Included
Age
St.
Group n
BM
log(BM)
Age
log(Age)
HT
log(HT)
Intercept
Dev.
0-5 611
21.78
-16.26
-9.014
39.09
-942.8
1259.9
1443.0
61.89
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 6
Age
St.
Group
n
BM
log(BM)
Age
log(Age)
HT
log(HT)
Intercept
Dev.
6-13
1618
7.540
262.8
43.41
-604.3
-338.0
758.7
1209.3
98.85
14-29
2648
4.194
391.6
41.38
-1010.3
152.5
433.1
1298.2
141.1
30-53
1346
6.239
208.5
14.38
-803.3
2854.4
-4066.
-180.9
143.9
54-99
621
3.840
284.9
4.510
-400.1
1782.8
-2274.
-588.6
123.1
Units: RMR = kilocalories/day; BM = kilograms; Age = years; Height = meters
We tried many variations on the above regressions, including changing the age outpoints, the
number of age groups, the list of independent variables, and the transformation of the
dependent variable RMR. The SAS program provided in Appendix A contains the code that
produces the regressions in Table 1-Table 4 and some of the plots shown below.
Figure 1 presents scatter plots of observed RMR values (top row) and RMR values predicted by
the updated algorithms described above (bottom row), as a function of age. These figures use
"BMR" to mean "RMR." The updated RMR algorithms have a bias of less than 0.5 percent
between observed and predicted values, compared to the existing APEX algorithms
which have a bias of 1-2 percent (10-30 kcal/d; smaller bias for females).
Figure 2 shows the mean RMR values by age: observed (black), predicted by the existing APEX
algorithms (red), predicted by the updated algorithms (blue), and predicted by the updated
algorithms with height included as an input variable (green; height-related regression
parameters not provided in this memorandum). In the red data points (the existing APEX
algorithms), a discontinuity is seen between ages 59 and 60, particularly for males. For adults
ages 59 and under, the red points are generally higher than the black points (the observed
values), whereas the red points are generally below the black points for ages 60 and above.
The same effect is seen in females, but the discontinuity is less pronounced. In the blue data
points (the updated algorithms), no sizeable discontinuities are seen at the age group
boundaries. As discussed earlier, the inclusion of height (the green points) does not have a
dramatic impact on the fit of the new RMR algorithm.
-------
WA 4-52 Task 4: RMR arid VE Algorithm Refinements
February 17, 2017
Page 7
BMR - all males
BMR - all females
(a)
* ++4* ++t : ++*+
l + t x + * +
+,H ~, *
t#£>
Vri+?r^ ++:
+ , ++ ++ + t +
10 20 30 40 50 60 70
100 110
fitted bmr - all males
i+|j4itviv+ +*v~
* $t$++ ± "n* + + +
10 20 30 40 50 60 70
90 100 110
(C)
Units: RMR = kilocalories/day, Age = years
wte+t +tjfc +
+V
i+J t++ s +*
10 20 30 40 50 60 70 80
100 110
(b)
fitted bmr - all females
100 110
(d)
Figure 1. Top Row: Observed RMR Values by Age for (a) Males and (b) Females. Bottom Row: Predicted RMR Values
by Age for (c) Males and (d) Females using the Updated Algorithms (without height).
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 8
mean bmr and old and new fits - males
data=black, old fit=red, new fit=blue, new with ht=green
old
1800
1700
1600
1500
1400
1300
1200
1100
1000
900
800
700
600
500
400
300
p- + + + + +
+ ^±++++++++++ ++t+tih+i+i+^++++t+ *t+++ -+
* ++:++*±++:+ * +
+++
¦+ ^ ++++* + $ j ++ +
+ + + + 4. * + * +
+++++++L* +
t *+ ++ * *+ + i
+ +++±tT +
+ ±t . Jf $ ^
|+ ^ * $ *+ ¥
y
(a)
30 40 50 60 70
age
mean bmr and old and new fits - females
data=black, old fit=red, newfit=blue, new with ht=green
bmr
1500
1400
1300
1200
1100
1000
900
800
700
600
500
400
300-
*** f^V.V .5 .,*(
^ *t*+ +jJ:
\ ft
* + +
+ t
(b) age
Units: RMR = kilocalories/day, Age = years
Figure 2. Mean RMR Values by Age: Observed (Black), Predicted by the
Existing APEX Algorithms (Red), Predicted by the Updated Algorithms
(Blue), and Predicted by the Updated Algorithms with Height Included as
an Input Variable (Green), for (a) Males and (b) Females.
-------
WA 4-52 Task 4 RMR and VE Algorithm Refinements
February 17, 2017
Page 9
8. Updated Ve Algorithm
Using the existing Ve dataset from Graham and McCurdy (2005), we developed updated Ve
algorithms for APEX that reduce discontinuities in predicted Ve between age groups and that
also utilize maximum VO2 (VCbm) as an input. VCbm is included because ongoing related work
on metabolic equivalents of task (MET) values for persons with unusual maximum capacity for
work suggests that their MET distributions are modified in a predicable way by their maximum
MET (or, equivalent^, by VC>2m). One potential limitation of this analysis is that the VCbm
values might not be well characterized for all people in the dataset.
As discussed earlier with Equation 1 above, we aimed to follow the general format of an MLR
model. In considering VE in particular, the available variables for regression are listed in Table 5
below. As discussed later in this section, we only utilized VO2 and VC^m in the updated Ve
algorithms.
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 10
Table 5. Summary of Variables Available in the Ve Dataset.
Description
Step Stage of exercise regimen at a given work level (0.1-13). <1 indicates resting state
where 0.1 =lay, 0.2=sit, and 0.3=stand (these were not used as they appeared
consistently unusual with regard to values observed in the exercising dataset).
Age Age (y)
BM Body mass (kg)
Char Special characteristics of the study subject. 1=trained athlete; 2=trained non-athlete;
3=normally active; 4=sedentary; 5=obese.
ET Cumulative test time at end of step (min). ".'-missing.
Gend 1=females;-1=males
Grd Percent grade while on treadmill, ".'-missing.
HR Heart rate (b/min) measured during the last minute of each step, ".'-missing.
HT Height (cm)
LBM Lean body mass (kg)
Mach Machine used. 1=cycle ergometer; 2=treadmill; ".'-missing.
VO2 Oxygen consumption (L/min, STPD) measured during the last minute of each step
Spd Treadmill speed (m/min). ".'-missing.
STUD Study number
SUBJ Study subject identifier
TT Total time of test (min). ".'-missing.
VE Ventilation (L/min, BTPS) measured during the last minute of each step
VChm Observed V02max (L/min, STPD) for the test
Wk Cycle ergometer setting (W). ".'-missing.
In_ve log(VE)
ln_vo2 log(V02)
VQ Ve-V02
ln_VQ log(VQ)
ln_bm log(BM)
ve_bm Ve-^BM
ln_ve_bm log(ve_bm)
vo2_bm V02-;-BM
ln_vo2_bm log(vo2_bm)
Note: y = years; kg = kilograms; min = minutes; b/min = beats per minute; cm = centimeters; L = liters; m/min
= meters per minute; log = natural logarithm; STPD = standard temperature and pressure, dry; BTPS =
body temperature and pressure, saturated.
Out of a total 6,636 observations, 65 had values of VC>2m that were less than values of VO2. We
found that using VC>2m as-is, versus using the maximum between VC>2m and VO2, made no
appreciable difference in estimates of Ve; we therefore used VCbm as-is.
Each Ve regression took place in two stages. First, all 6,636 data points were used in each
regression. Then, all the points that were more than 3 studentized residuals away from the fitted
line were removed, and the regression was repeated. This was done to prevent a few outlier
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 11
points from having undue influence. In this second step, 43 points were rejected though overall
they had very little effect on the regression. Note that for a random sample of 6,636 points from
a true normal distribution, about 18 would be expected to be more than 3 standard deviations
from the mean. The number of outliers was therefore only modestly above what would be
expected by chance alone.
The Graham and McCurdy (2005) regressions had four separate age groups (<20, 20-33, 34-
60, and 61+) evaluated independently, so discontinuities appear at the age boundaries. Thus, a
given person ageing across a boundary would experience a sudden shift in their VE /VO2
relationship. Our new analysis uses the same regression equation for all ages, eliminating this
issue.
For a given VO2 level, if VCbm decreases, then (VCVVCbm) increases, and thus Ve also
increases. This relationship eliminated the need to regress upon variables such as age, BM,
height, and sex. For example, males on average need less Ve to support a given VO2, which is
captured by their having higher VCbm. The only variables needed for the new Ve algorithm are
VO2 and VCbm, both of which are already calculated in APEX.
The actual values of VO2 and VCbm are less relevant than the fraction of maximum capacity,
represented by fi = VC^A/Cbm. fi may operate non-linearly (for example, fi = 0.9 is likely more
than twice as encumbering as fi = 0.45). A SAS procedure "Proc Transreg" was used to
determine appropriate transformations. This recommended a power of 4 or 5 be used, that is, y
= Ve"° 25 or y = Ve"02, when only the variable ln_vo2 was used as the independent variable.
Table 6. Reported r2 Statistic Based on Transformation of Ve
Transformation w . „ „
tr_r2
ve_r2
Variables
of VE
2 ln_vo2 0.9479 0.7350
3 ln_vo2 0.9566 0.8779
4 ln_vo2 0.9563 0.8873
5 ln_vo2 0.9544 0.8850
6 ln_vo2 0.9523 0.8821
In Ve In vo2 0.9341 0.8561
Note: VO2 = oxygen consumption rate; ln_vo2 = log(VC>2) = natural log of VO2;
transformation of VEis VE"Nwhen N is an integer; ln_VE= log(VE); tr_r2 = r2 of the
transformed response variable,ve_r2 = r2 of Ve
Table 6 demonstrates that the reported r2 for the regression (called tr_r2) of the transformed
variable Y = VE("1/power) is higher than the r2 for Ve itself (called ve_r2), but that reflects how well
the regression captures the variation in the transformed variable. Because the transformation is
intended to "linearize" the data, it is expected that the regression would fit better on the
transformed variable. Note that the set of variables that produce the optimal r2 for the
transformed variable sometimes is not the same set that is optimal for ve_r2.
When ln_vo2 is the only independent variable, the best transformation (in terms of ve_r2) is
power=4, or y = Ve"0 25, as seen in Table 7. Table 7 shows that the addition of age, sex, or height
makes little impact on the prediction of Ve. Of these, height is the most effective, but it adds less
than 0.01 to r2. However, the addition of either V02m orfi = V02/V02m to the set of independent
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 12
variables gives a substantial improvement in both tr_r2 and ve_r2. However, note that using f2
instead of fi did not improve the fit.
Table 7. Reported r2 Statistic for Variables used with Y=Ve"025
Transformation
of VE
4
4
4
4
4
4
4
Variables
ln_vo2
ln_vo2, age
ln_vo2, sex
ln_vo2, height
ln_vo2, VChm
ln_vo2, fi
ln_vo2, f2
0.9563
0.9566
0.9578
0.9596
0.9715
0.9721
0.9712
ve r2
0.8873
0.8900
0.8923
0.8938
0.9213
0.9378
0.9347
Note: VO2 = oxygen consumption rate; VChm = maximum VO2; ln_vo2 = log(V02) = natural log
of VO2; tr r2 = r2 of the transformed variable; ve r2 = r2 of Ve; fi = VO2/VO2IT1; f2 =
(V02A/02m)2; transformation of VEis VE"Nwhen N is an integer
Once fi is added to the list of independent variables, then the optimal transformation of Ve
changes. For example, the first line of
-------
WA 4-52 Tas! R and VE Algorithm Refinements
February 17, 2017
Page 13
Table 8 shows that a power of 5 (that is, y = Ve"02), now outperforms a power of 4 (see the r2
values in the second-to-last line of Table 7), whereas the opposite was true in Table 6. The
optimal transformation of Ve changes and the optimal set of independent variables depend on
each other. Using the ve_r2 statistic as the measure, then for power=5, f2 provides a better fit
that fi, but that f3 is worse than f2. The same is true for power = 6, although all the fits (except for
the one using fi) are better than with power = 5.
Even higher transformation powers can be used, but in practice large powers provide similar
results to a log transformation2. The last five rows of
2 The SAS Proc Transreg uses the symbolism power=0 to explicitly indicate a log transformation for the response
variable, although since the Tables report values of (-1/power), it would be more correct to call this power = °°
-------
WA 4-52 Tas! R and VE Algorithm Refinements
February 17, 2017
Page 14
Table 8 examines using the natural logarithm of Ve as the dependent variable, with the natural
logarithm of VO2 and various powers of (V02/VC>2m) as independent variables. Using fi orf2
provides a worse fit with ln_VE than is obtained with power = 6, but using f4 provides the best
overall fit.
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 15
Table 8. Reported r2 for Combinations of Independent Variables and Transformations of Ve
9ivi ¦ ¦ lullw 11
of VE
Variables
tr_r2
ve_r2
5
ln_vo2, fi
0.9730
0.9402
5
ln_vo2, f2
0.9729
0.9420
5
ln_vo2, f3
0.9723
0.9402
6
ln_vo2, fi
0.9730
0.9397
6
ln_vo2, f2
0.9734
0.9445
6
ln_vo2, f3
0.9731
0.9442
6
ln_vo2, U
0.9723
0.9427
In Ve
ln_vo2, fi
0.9662
0.9244
In Ve
ln_vo2, f2
0.9714
0.9411
In Ve
ln_vo2, f3
0.9724
0.9466
In Ve
ln_vo2, U
0.9719
0.9481
In Ve
ln_vo2, fs
0.9711
0.9479
Note: VO2 = oxygen consumption rate; VChm = maximum VO2; ln_vo2 = log(VC>2) = natural log of VO2; fi =
VO2/VO2IT1; fN = (V02/VC>2m)N; transformation of Ve is VE"Nwhen N is an integer; tr_r2 = r2 of the transformed
variable; ve_r2 = r2 of Ve
Using the log transformation with the independent variables ln_vo2 and f4=(V02/VC>2m)4, Table
9 examines the effects of adding further independent variables; specifically age, gender, and/or
height.
Table 9. Various Sets of Independent Variables used to Predict log(VE)
Transform
Variables
tr_r2
ve_r2
In Ve
ln_vo2, U
0.9719
0.9481
In Ve
ln_vo2, U, age
0.9720
0.9477
In Ve
ln_vo2, U, gender
0.9721
0.9483
In Ve
ln_vo2, U, height
0.9723
0.9481
In Ve
ln_vo2, U, age gender height
0.9726
0.9477
Note: VO2 = oxygen consumption rate; ln_vo2 = log(VC>2) = natural log of VO2; tr_r2 = r2 of
the transformed variable; ve_r2 = r2 of Ve; U = (VO2/VO2IT1)4
In all cases, the ve_r2 is unchanged to three decimal places, being 0.948 in all cases. Hence,
the recommendation is to use the simplest version of these regressions, as seen in Equation (2)
below.
yE2 _ e(3.298 + 0.7935xin_vo2+ 0.53845 x (702-^02m)4+0.1253xN(0,l))
The following two figures show all 6,636 data points from the Ve dataset. Figure 3 shows
measured Ve and measured VO2. Figure 4 shows predicted Ve ("VE2") and measured VO2,
where VE2 is given by Equation (2) (with an r2 of 0.948, as shown in Table 9) which is based on
the Ve dataset with outliers removed (this is not the final updated Ve algorithm, as noted later in
this section).
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 16
veO
220
200
180
160
140
120
100
80
60
40
20
V02
Figure 3. Measured VO2 and Measured Ve, from the Ve
dataset
ve2
220
0 1 2 3 4 5 6
V02
Figure 4. Measured VO2 and Predicted Ve Using the
Updated Algorithm (i.e., VE2 shown in Equation 2).
As can be seen in the figures, predicted and observed values of Ve are very close.
In concordance with a request from the EPA WAM, we developed a mixed-effects regression
(MER) in addition to the above MLR. MER separates residuals into within-person (ew) and
between-person (eb) effects, known as intrapersonal and interpersonal effects, respectively.
This analysis, using the same independent variables and the same Ve dataset discussed above
yields another VE algorithm. This algorithm, shown below, is the final version of the
updated Ve algorithm to be incorporated into APEX.
-------
WA 4-52 Tas! R and VE Algorithm Refinements
February 17, 2017
Page 17
Pred VE = e(3-300 + °-8128xin-vo2+0 5126 x (^°2"^02™)4+N(0,efc)+N(0,ew)) £3)
N(0,eb) is a normal distribution with mean zero and standard deviation eb=0.09866 meant to
capture /'nte/personal variability, which is sampled once per person. N(0,ew) is an /nfrapersonal
residual with standard deviation of ew=0.07852, which is resampled daily due to natural
/'nfrapersonal fluctuations in Ve that occur daily.
Differences between Equations (2) and (3) may be due to the fact that some of the persons in
the dataset had different numbers of observations. The mean, median, and mode were all
seven observations per person, with a range from one to 13. With regard to implementation in
APEX, the cause of the interpersonal variability may not be necessary to determine. It is
sufficient to specify the size of the two error terms, one sampled once per person and the other
sampled once per day.
Ultimately, the EPA WAM chose Equation (3) to implement in APEX due to its increased ability
to account for inter- and intra-personal effects. The resulting r2 for Ve (0.94) is a substantial
improvement over the existing Ve regressions in APEX (where r2 was 0.892-0.925), with a
large reduction in discontinuities of Ve between ages.
7. Effect of Updated Algorithm(s) on Simulated
Exposure
The updated RMR algorithm is based on an MLR with coefficients shown in Table 1 and Table
2. The updated Ve algorithm is shown in Equation (3).
The existing RMR algorithm in APEX (in units of kilocalories/minute [kcal/min]) is:
RMR = 0.166 X [RMRsiope X BM + RMRint + RMRerr]
(4)
Where:
0.166
RMRsiope
RMRint
RMRerr
the conversion factor for converting megajoules (MJ)/d to kcal/min
slope of the regression equation (MJ/(d-kg))
intercept of the regression equation (MJ/d)
variation in the regression equation (MJ/d)
The existing Ve algorithm in APEX (in units of m i 11 i I iters/m i n ute [ml_/min]) is:
VE = (1,000^) xBM x exp(VEinter + VEslope x ln(V02) +Zx VEresid) (5)
Where:
intercept of the regression equation
slope of the regression equation
random number from normal distribution
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 18
VEresid = variation in the regression equation
And where VO2 (in units L/min/kg) is:
T,~ METXECFXRMR
Where:
ECF = energy conversion factor (L CVkcal)
We compared the effects of the existing and updated RMR and Ve algorithms using a sample of
1000 persons, ages 0 to 95, run for one year each (taken from an APEX run for ozone, in 2010,
in the Los Angeles area). Four runs were made: R1V1 is the combination of old RMR and old Ve
algorithms; R2V1 uses the new RMR and old Ve algorithms; R1V2 uses the old RMR and new
Ve algorithms; and finally, R2V2 uses both new algorithms. Each run produced a sample of
1000 RMR values (one per person), and 8,760,000 Ve values (one per hour, per person).
The RMR results did not vary when just the Ve method was changed. This was expected,
because APEX calculates RMR first. The Ve calculation is affected by any change in RMR.
Statistics comparing the old and new RMR algorithms are presented in Table 10. The new RMR
algorithm produces slightly lower values across the board, with larger decreases at the higher
end of the range. Even then, these differences are below 4 percent. There are fewer extreme
values using the new algorithm, resulting in a smaller standard deviation.
Table 10. RMR Value Statistics (kcal/min) for 1000 Persons, Using Old and New RMR Algorithms
I Statistic
Old RMR
New RMR
% Change I
Mean
1.065
1.040
- 2.4 %
Standard deviation
0.292
0.275
- 5.8 %
10th percentile
0.709
0.702
- 1.0 %
Median
1.057
1.034
- 2.2 %
90th percentile
1.443
1.390
- 3.7 %
The Ve data below have been analyzed in two ways. First, statistics on the full set of 8,760,000
Ve values are generated. When comparing the same Ve algorithm and varying RMR algorithms,
the old Ve algorithm had a drop of 2 percent in mean Ve when switching to the new RMR, and
the new Ve algorithm had a similar drop of 1.5 percent (not shown in a table here). These are
somewhat smaller than the drop in mean RMR of 2.4 percent.
Focusing on the new RMR algorithm, a comparison of Ve statistics from the R2V1 and R2V2
runs is shown in Table 11, using all 8,760,000 Ve values. The high-end Ve values changed very
little between the old and new Ve algorithms (by 0.5 percent), but the new algorithm predicts
higher values at lower Ve levels (by 17.6 percent), resulting in an increase by 6 percent in mean
values. These values are effectively time-weighted, so sleeping Ve accounts for about one-third
of the set (that is, at rest or below). By contrast, the Adams dataset was concerned almost
solely with activities above resting levels. Hence, the regression based on the Adams dataset is
being extrapolated to sleeping as an activity. One would therefore expect that the new Ve
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 19
algorithm would be more robust for the higher activity levels. Note that the new Ve algorithm has
a smaller standard deviation than the old method (by 11.6 percent), resulting in fewer extreme
values.
Table 11. Ve Value Statistics (mL/hr) for 8,760,000 Person-hours, Using the New RMR Algorithm
with the Old and New Ve Algorithms
I Statistic
Old VE
New Ve
% Change I
Mean
19581
20763
+ 6.0 %
Standard deviation
10375
9172
- 11.6 %
10th percentile
8778
10319
+ 17.6 %
Median
17422
19391
+ 11.3 %
90th percentile
33042
32887
- 0.5 %
The second type of analysis is to examine the change in mean Ve per person, and the change
in the 90th percentile of each person's Ve values. First, the 1000 personal means (over the year)
and 1000 personal 90th percentiles are calculated. Table 12 shows modest increases (in the
range of 6 percent) in person-mean Ve values when using the new Ve algorithm, with a 1.8-
percent increase in standard deviation. Table 13 shows that the 90th percentile for each person
(that is, the Ve level that one exceeds for 2.4 hours per day, on average) has changed relatively
little between the old and new algorithms. The mean has dropped 2 percent, but the standard
deviation dropped by 9.1 percent because the upper tail does not extend as far as before.
Table 12. Population Statistics on Personal Mean Ve (mL/hr), Using the New RMR Algorithm with
the Old and New Ve Algorithms
Statistic
Old VE
New Ve
% Change
Mean
19581
20763
+ 6.0%
Standard deviation
6187
6296
+ 1.8%
10th percentile
12236
12843
+ 5.0%
Median
18955
20504
+ 8.2%
90th percentile
27822
29164
+ 4.8%
Table 13. Population Statistics on Personal 90th Percentile of Ve (mL/hr), Using the New RMR
Algorithm with the Old and New Ve Algorithms
Statistic
Old VE
New Ve
% Change
Mean
28017
27445
-2.0%
Standard deviation
11094
10087
-9.1%
10th percentile
14205
14415
1.5%
Median
27026
27339
1.2%
90th percentile
42572
40775
-4.2%
In summary, in comparing the updated APEX algorithms for RMR and Ve to the existing
algorithms:
Average RMR decreases with the updated RMR algorithms, though remains within 3
percent of RMR predicted by the existing algorithm.
As expected, the updated Ve algorithm has no effect on predicted RMR.
-------
WA 4-52 Task 4 RMR and VE Algorithm Refinements
February 17, 2017
Page 20
¦ The updated RMR algorithm impacts Ve predictions less when utilizing the updated Ve
algorithm; this impact is greater at the lower end of estimated Ve values.
¦ The upper end (90th percentile) of predicted Ve values are similar between the existing
and updated Ve algorithms. This appears to be due to two partially cancelling effects: the
population 90th percentile of the personal means increased 4.8 percent, but the
population 90th percentile of the personal 90th percentiles decreased 4.2 percent.
¦ The lower end of predicted Ve values is moderately higher with the updated Ve algorithm
than with the existing Ve algorithm (a 17.6-percent change in the 10th percentile, which
corresponds to sleeping Ve)
¦ Both the updated and existing Ve algorithms predict Ve values exceeding 100,000
mL/min for roughly 1 in every 65,000 person-hours, which was the hard-coded maximum
for Ve in APEX. Note that a switch has been added to the APEX Control Options File to
enable or disable the maximum upper limit. This was disabled for the current comparison
runs, because truncation of the two tails at the same point would cause the two
distributions to look more similar than they otherwise would.
8. Summary Discussion and Next Steps
Through extensive literature searches for both RMR and Ve algorithms, as well as through
augmentation of the RMR dataset, ICF has improved upon the RMR and Ve physiological
algorithms within the APEX model. These updated algorithms perform better than the existing
algorithms in APEX, with reduced discontinuities between APEX age groups and better fits to
the measured datasets. ICF has created "switches" within the APEX Control Options File that
allows users to choose between the available RMR or Ve algorithms. The coding required to
completely replace the older algorithms can be done quickly at EPA's request.
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page 21
9. References
Graham, S. and McCurdy, T. (2005) A NERL Internal Research Report: Revised ventilation rate
(VE) equations for use in inhalation-oriented exposure models. EPA/600/X-05/008.
IOM. (2005). Dietary Reference Intakes for Energy, Carbohydrate, Fiber, Fat, Fatty Acids,
Cholesterol, Protein and Amino Acids. Panel on Macronutrients, Panel on the Definition
of Dietary Fiber, Subcommittee on Upper Reference Levels of Nutrients, Subcommittee
on Interpretation and Uses of Dietary Reference Intakes, and the Standing Committee
on the Scientific Evaluation of Dietary Reference Intakes, Food and Nutrition Board. U.S.
Institute of Medicine. National Academies Press. URL:
http://www.nationalacademies.org/hmd/Activities/Nutrition/SummarvDRIs/DRI-
Tables.aspx. Table: Doubly Labeled Water Data Set: 10_DLW_Database.xls
(downloaded 8/24/16)
Henry, CJK. (2005). Basal metabolic rate studies in humans: measurement and development of
new equations. Public Health Nutrition. 8(7A): 1133-1152.
Isaacs, K. (2008). Estimating ventilation in human exposure models: Summary. Internal
Memorandum. July 28, 2008.
McCurdy, T. Physiological Parameters and Physical Activity Data for Evaluating Exposure
Modeling Performance: a Synthesis. U.S. Environmental Protection Agency,
Washington, DC, EPA/600/R-15/175, 2015.
Schofield, WN (1985). Predicting basal metabolic rate, new standards and review of previous
work. Human Nutrition: Clinical Nutrition. 39C(Supp. 1): 5-41.
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-1
* Written by WGG at ICF, last revised on October 21, 2016
data raw;
infile "C:/main/APEX/WA452/exercise/from Jess/newrmr JL 30augl6.csv"
firstobs=2 dsd dlm=',';
length sex $1 author $20 type $6 citation $80 study $40;
input sex author type age bmr ht bm citation year study recno;
yage = floor(age);
run;
data good bad all;
set raw;
logbm = log(bm);
logbmr = log(bmr);
if sex="M" then gender= 1;
if sex="F" then gender=-l;
bad = 0;
if age=. then bad=l;
if bmr=. then bad=l;
if bm=. then bad=l;
if gender=. then bad=l;
if ht=. then bad=l;
ageO = age;
age = floor(age);
if age>99 then age=99;
logage = log(l+age);
logageO = log(l+age0);
invage = 1/(1+age);
bmage = bm*age;
bmcage = bm*(1+age);
bmlage = bm*logage;
loght = log(ht);
if bad=0 then output good; else output bad;
output all;
run;
data males females;
set all;
if gender= 1 then output males;
if gender=-l then output females;
run;
axisl order = 0 to 3200 by 200;
title 'RMR: All males';
proc gplot data=males;
plot bmr*age /VAXIS=axisl;
run; quit;
title 'RMR: All females';
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-2
proc gplot data=females;
plot bmr*age /VAXIS=axisl;
run; quit;
axisl order = 0 to 3200 by 200;
axis2 order = 0 to 180 by 10;
title 'RMR vs BM: All males';
proc gplot data=males;
plot bmr*bm /VAXIS=axisl HAXIS=axis2;
run; quit;
title 'RMR vs BM: All females';
proc gplot data=females;
plot bmr*bm /VAXIS=axisl HAXIS=axis2;
run; quit;
axisl order = 0 to 3200 by 200;
axis3 order = 0 to 2.0 by 0.1;
title 'RMR vs BM: All males';
proc gplot data=males;
plot bmr*ht /VAXIS=axisl HAXIS=axis3;
run; quit;
title 'RMR vs BM: All females';
proc gplot data=females;
plot bmr*ht /VAXIS=axisl HAXIS=axis3;
run; quit;
proc sort data=males; by yage; run;
proc means data=males noprint;
by yage;
var bmr;
output out=ml n=n mean=mean std=std min=min max=max;
run;
proc print data=ml; run;
%macro c(gen,num,test,vars);
%let last=0;
%do i=l %to #
%let j=%scan(&test,&i);
%put i=&i j =&j ;
title "&gen. &last.-&j";
data a;
%if &gen=M %then set males;;
%if &gen=F %then set females;;
lo = symgetn("last") ;
hi = &j ;
if (age>=lo and age
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-3
proc reg data=a;
model bmr=&vars /vif;
output out=z
p =predicted
residual=residual
rstudent=rstudent;
run; quit;
data null ;
set z end=eof;
retain pertot 0;
retain errtot 0;
pertot = pertot + 1;
errtot = errtot + residual**2;
if (eof) then do;
call symput("pertot&i",trim(left(pertot)));
call symput("errtot&i",trim(left(errtot)));
end;
run;
%let last = &j;
%end;
%let pertot = 0;
%let errtot = 0;
%do i=l %to #
%let pertot = %sysevalf(&pertot+&&pertot&i);
%let errtot = %sysevalf(&errtot+&&errtot&i);
%end;
%put test = &test;
%put pertot = Spertot;
%put errtot = Serrtot;
%mend;
%c(F,5,6 14 30 54 100,bm logbm age logage); * err = 11052 best;
%c(M,5,6 14 25 55 100,bm logbm age logage); * err = 22753 best;
%macro d(gen,num,test,vars) ;
%let last=0;
%do i=l %to #
%let j=%scan(&test, &i) ;
%put i=&i j =&j ;
title "&gen. &last.-&j";
data a;
%if &gen=M %then set males;;
%if &gen=F %then set females;;
lo = symgetn("last") ;
hi = &j ;
if (age>=lo and age
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-4
data null ;
set z end=eof;
retain pertot 0;
retain errtot 0;
pertot = pertot + 1;
errtot = errtot + residual**2;
if (eof) then do;
call symput("pertot&i",trim(left(pertot)));
call symput("errtot&i",trim(left(errtot)));
end;
run;
%let last = &j;
%end;
%let pertot = 0;
%let errtot = 0;
%do i=l %to #
%let pertot = %sysevalf(&pertot+&&pertot&i);
%let errtot = %sysevalf(&errtot+&&errtot&i);
%end;
%put test = &test;
%put pertot = Spertot;
%put errtot = Serrtot;
%mend;
%d(F,5,6 14 30 54 100,bm logbm age logage ht loght); * err = 10767;
%d(M,5,6 14 25 55 100,bm logbm age logage ht loght) ; * err = 22488;
proc sort data=males; by age; run;
proc means data=males noprint;
by age;
var bm logbm bmr ht loght;
output out=ml mean=;
run;
data m2;
set ml;
logage = log(l+age);
if (age<=5) then fit = 13.19*bm + 270.2 *logbm - 18.34*age + 131.3*logage -
208.5 ;
if (age>=6 and age<=13) then fit = 10.21*bm + 260.2 *logbm + 13.04*age -
205.7*logage + 333.4 ;
if (age>=14 and age<=24) then fit = 0.207*bm + 1078. *logbm + 115.1*age -
2794.*logage + 3360.6;
if (age>=25 and age<=54) then fit = 2.845*bm + 729.6 *logbm + 3.181*age -
191.6*logage - 1067. ;
if (age>=55) then fit = 9.291*bm + 264.8 *logbm - 5.288*age + 181.5*logage -
705.9 ;
if (fit<50) then fit=50;
if (fit>3000) then fit=3000;
if (age<=5) then fit2 = 17.61*bm + 106.3 *logbm - 17.93*age + 87.37*logage -
368.9*ht + 676.3 *loght + 607.6;
if (age>=6 and age<=13) then fit2 = 12.64*bm + 149.3 *logbm + 30.91*age -
417.0*logage - 1498.*ht + 2151.5*loght + 2344.9;
if (age>=14 and age<=24) then fit2 = .0309*bm + 1098.6*logbm + 114.3*age -
2777.*logage + 31.45*ht - 101.2 *loght + 3250.7;
if (age>=25 and age<=54) then fit2 = 4.692*bm + 481.5 *logbm + 2.422*age -
136.3*logage + 1590.*ht - 2014. *loght - 1961.3;
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-5
if (age>=55) then fit2 = 12.60*bm - 108.4 *logbm - 5.151*age + 170.6*logage
- 927.2*ht + 2405. *loght + 982.6;
if (fit<50) then fit=50;
if (fit>3000) then fit=3000;
if (age<=2) then old = 0.249*bm - 0.127 ;
if (age>=3 and age<=9) then old = 0.095*bm + 2.110 ;
if (age>=10 and age<=17) then old = 0.074*bm + 2.754 ;
if (age>=18 and age<=29) then old = 0.063*bm + 2.896 ;
if (age>=30 and age<=59) then old = 0.048*bm + 3.653 ;
if (age>=60) then old = 0.049*bm + 2.459 ;
old = 238.845 * old;
if (old<144) then old=144;
if (old>2880) then old=2880;
run;
symboll color=black;
symbol2 color=red;
symbol3 color=blue;
symbol4 color=green;
title "mean bmr and old and new fits - males";
title2 "data=black, old fit=red, new fit=blue, new with ht=green";
proc gplot data=m2;
plot old*age=2 fit*age=3 bmr*age=l fit2*age=4 /overlay;
run; quit;
proc gplot data=m2(where=(age>=48 and age<=63));
plot old*age=2 fit*age=3 bmr*age=l fit2*age=4 /overlay;
run; quit;
proc sort data=females; by age; run;
proc means data=females noprint;
by age;
var bm logbm bmr ht loght;
output out=fl mean=;
run;
data f2;
set fl;
logage = log(l+age);
if (age<=5) then fit = 11.94*bm + 261.5 *logbm - 22.31*age + 120.9*logage -
183.6;
if (age>=6 and age<=13) then fit = 5.296*bm + 409.1 *logbm + 40.37*age -
524.9*logage + 392.7;
if (age>=14 and age<=29) then fit = 0.968*bm + 676.9 *logbm + 40.89*age -
1002.*logage + 772.7;
if (age>=30 and age<=53) then fit = 4.935*bm + 355.4 *logbm + 16.28*age -
896.0*logage + 2225.;
if (age>=54) then fit = 2.254*bm + 445.9 *logbm + 5.464*age - 489.9*logage +
944.2;
if (fit<50) then fit=50;
if (fit>3000) then fit=3000;
if (age<=5) then fit2 = 21.78*bm - 16.26 *logbm - 9.014*age + 39.09 *logage
- 942.8 *ht + 1259.9*loght + 1443.0;
if (age>=6 and age<=13) then fit2 = 7.540*bm + 262.8 *logbm + 43.41*age -
604.3 *logage - 338.0 *ht + 758.7 *loght + 1209.3;
if (age>=14 and age<=29) then fit2 = 4.194*bm + 391.6 *logbm + 41.38*age -
1010.3*logage + 152.5 *ht + 433.1 *loght + 1298.2;
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-6
if (age>=30 and age<=53) then fit2 = 6.239*bm + 208.5 *logbm + 14.38*age -
803.3 *logage + 2854.4*ht - 4066. *loght - 180.9;
if (age>=54) then fit2 = 3.840*bm + 284.9 *logbm + 4.510*age - 400.1 *logage
+ 1782.8*ht - 2274. *loght - 588.6;
if (fit<50) then fit=50;
if (fit>3000) then fit=3000;
if (age<=2) then old = 0.244*bm - 0.130 ;
if (age>=3 and age<=9) then old = 0.085*bm + 2.033 ;
if (age>=10 and age<=17) then old = 0.056*bm + 2.898 ;
if (age>=18 and age<=29) then old = 0.062*bm + 2.036 ;
if (age>=30 and age<=59) then old = 0.034*bm + 3.538 ;
if (age>=60) then old = 0.038*bm + 2.755 ;
old = 238.845 * old;
if (old<144) then old=144;
if (old>2880) then old=2880;
run;
symboll color=black;
symbol2 color=red;
symbol3 color=blue;
symbol4 color=green;
title "mean bmr and old and new fits - females";
title2 "data=black, old fit=red, new fit=blue, new with ht=green";
proc gplot data=f2;
plot bmr*age=l old*age=2 fit*age=3 fit2*age=4 /overlay;
run; quit;
proc gplot data=f2(where=(age>=48 and age <=63));
plot bmr*age=l old*age=2 fit*age=3 fit2*age=4 /overlay;
run; quit;
data mall;
set males;
z = rannor(0);
if (age<=5) then fit = 13.19*bm + 270.2 *logbm - 18.34*age + 131.3*logage -
208.5 + 69.10* z;
if (age>=6 and age<=13) then fit = 10.21*bm + 260.2 *logbm + 13.04*age -
205.7*logage + 333.4 + 115.3*z;
if (age>=14 and age<=29) then fit = 0.207*bm + 1078. *logbm + 115.1*age -
2794.*logage + 3360.6 + 161.l*z;
if (age>=30 and age<=53) then fit = 2.845*bm + 729.6 *logbm + 3.181*age -
191.6*logage - 1067. + 178.2*z;
if (age>=54) then fit = 9.291*bm + 264.8 *logbm - 5.288*age + 181.5*logage -
705.9 + 163.6* z;
if (fit<50) then fit=50;
if (fit>3000) then fit=3000;
if (age<=5) then fit2 = 11.59*bm + 215.6 *logbm - 29.69*age + 112.9*logage +
367.l*ht - 332.7 + 68.93*z;
if (age>=6 and age<=13) then fit2 = 10.42*bm + 239.4 *logbm + 11.87*age -
200.3*logage + 42.18*ht + 339.8 + 115.3*z;
if (age>=14 and age<=24) then fit2 = 0.103*bm + 1094. *logbm + 114.4*age -
2781.*logage - 28.7*ht + 3322.1 + 161.l*z;
if (age>=25 and age<=54) then fit2 = 5.022*bm + 457.5 *logbm + 2.370*age -
134.5*logage + 405.3*ht - 939.6 + 176.7*z;
if (age>=55) then fit2 = 11.78*bm - 44.62 *logbm - 3.177*age + 39.95*logage
+ 490.8*ht + 50.55 + 160.9*z;
if (fit2<50) then fit2=50;
if (fit2>3000) then fit2=3000;
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-7
if (age<=5) then fit3 = 17.61*bm + 106.3 *logbm - 17.93*age + 87.37*logage -
368.9*ht + 676.3*loght + 607.6 + 68.60*z;
if (age>=6 and age<=13) then fit3 = 12.64*bm + 149.3 *logbm + 30.92*age -
417.0*logage - 1498.*ht + 2151.*loght + 2344.9 + 115.0*z;
if (age>=14 and age<=24) then fit3 = .0309*bm + 1098.6*logbm + 114.3*age -
2777.*logage + 31.45*ht - 101.2*loght + 3250.7 + 161.l*z;
if (age>=25 and age<=54) then fit3 = 4.692*bm + 481.5 *logbm + 2.422*age -
136.3*logage + 1590.*ht - 2014.*loght - 1961.3 + 176.6*z;
if (age>=55) then fit3 = 12.67*bm - 113.9 *logbm - 3.228*age + 38.95*logage
- 962.2*ht + 2466.*loght + 1453.5 + 160.9*z;
if (fit3<50) then fit3=50;
if (fit3>3000) then fit3=3000;
if (ht=.) then fit3=.;
if (age<=2) then old = 0.249*bm - 0.127 + 0.29*z;
if (age>=3 and age<=9) then old = 0.095*bm + 2.110 + 0.28*z;
if (age>=10 and age<=17) then old = 0.074*bm + 2.754 + 0.44*z;
if (age>=18 and age<=29) then old = 0.063*bm + 2.896 + 0.64*z;
if (age>=30 and age<=59) then old = 0.048*bm + 3.653 + 0.70*z;
if (age>=60) then old = 0.049*bm + 2.459 + 0.69*z;
old = 238.845 * old;
if (old<144) then old=144;
if (old>2880) then old=2880;
err = BMR-fit;
err2 = BMR-fit2;
err3 = BMR-fit3;
errO = BMR-old;
run;
axisl order = 0 to 3000 by 1000;
title "fitted bmr - all males";
proc gplot data=mall;
plot fit*age;
run; quit;
title "fitted bmr with height - all males";
proc gplot data=mall;
plot fit2*age;
run; quit;
title "fitted bmr with ht and loght - all males";
proc gplot data=mall;
plot fit3*age=3;
run; quit;
title "APEX fit for bmr - all males";
proc gplot data=mall;
plot old*age /vaxis=axis1;
run; quit;
title "error statistics - males";
proc means data=mall n mean std var min max;
var bmr errO err err2 err3;
run;
proc sort data=mall; by age; run;
proc means data=mall noprint;
by age;
var bmr fit fit2 fit3 old err err2 err3 errO;
output out=mstats mean=;
run;
symboll color=black;
symbol2 color=red;
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-8
symbol3 color=blue;
title "mean bmr and old and new fits - males";
title2 "data=black, old fit=red, new fit=blue";
proc gplot data=mstats;
plot old*age=2 fit*age=3 bmr*age=l /overlap;
run; quit;
data fall;
set females;
z = rannor(0);
if (age<=5) then fit = 11.94*bm + 261.3 *logbm - 22.14*age + 120.4*logage -
182.9 + 64.62* z;
if (age>=6 and age<=13) then fit = 5.296*bm + 409.1 *logbm + 40.37*age -
524.9*logage + 392.7 + 99.43*z;
if (age>=14 and age<=29) then fit = 1.004*bm + 674.4 *logbm + 41.linage -
1007.*logage + 790.6 + 143.2*z;
if (age>=30 and age<=53) then fit = 4.935*bm + 355.4 *logbm + 16.29*age -
896.0*logage + 2225.3 + 145.3*z;
if (age>=54) then fit = 2.699*bm + 415.7 *logbm + 8.701*age - 711.6*logage +
1756.8 + 124.6* z;
if (fit<50) then fit=50;
if (fit>3000) then fit=3000;
if (age<=5) then fit2 = 11.09*bm + 175.3 *logbm - 35.26*age + 98.50 *logage
+ 449.0*ht - 304.3 + 63.23*z;
if (age>=6 and age<=13) then fit2 = 6.494*bm + 304.9 *logbm + 31.99*age -
483.8 *logage + 209.0*ht + 411.8 + 98.89*z;
if (age>=14 and age<=29) then fit2 = 4.107*bm + 396.9 *logbm + 41.32*age -
1009.3*logage + 423.2*ht + 1049.9 + 141.l*z;
if (age>=30 and age<=53) then fit2 = 6.969*bm + 155.6 *logbm + 14.74*age -
815.2 *logage + 316.4*ht + 2175.2 + 144.0*z;
if (age>=54) then fit2 = 5.038*bm + 198.6 *logbm + 7.630*age - 610.7 *logage
+ 346.l*ht + 1602.5 + 122.6*z;
if (fit2<50) then fit2=50;
if (fit2>3000) then fit2=3000;
if (age<=5) then fit3 = 21.78*bm - 16.26 *logbm - 9.014*age + 39.09 *logage
- 942.8 *ht + 1259.9*loght + 1443.0 + 61.89*z;
if (age>=6 and age<=13) then fit3 = 7.540*bm + 262.8 *logbm + 43.41*age -
604.3 *logage - 338.0 *ht + 758.7 *loght + 1209.3 + 98.85*z;
if (age>=14 and age<=29) then fit3 = 4.194*bm + 391.6 *logbm + 41.38*age -
1010.3*logage + 152.5 *ht + 423.1 *loght + 1298.2 + 141.l*z;
if (age>=30 and age<=53) then fit3 = 6.239*bm + 208.5 *logbm + 14.38*age -
803.3 *logage + 2854.4*ht - 4066. *loght - 180.9 + 143.9*z;
if (age>=54) then fit3 = 4.506*bm + 236.4 *logbm + 7.564*age - 605.8 *logage
+ 1489.9*ht - 1796.6*loght + 475.8 + 122.6*z;
if (fit3<50) then fit3=50;
if (fit3>3000) then fit3=3000;
if (ht=.) then fit3=.;
if (age<=2) then old = 0.244*bm - 0.130 + 0.25*z;
if (age>=3 and age<=9) then old = 0.085*bm + 2.033 + 0.29*z;
if (age>=10 and age<=17) then old = 0.056*bm + 2.898 + 0.47*z;
if (age>=18 and age<=29) then old = 0.062*bm + 2.036 + 0.50*z;
if (age>=30 and age<=59) then old = 0.034*bm + 3.538 + 0.47*z;
if (age>=60) then old = 0.038*bm + 2.755 + 0.45*z;
old = 238.845 * old;
if (old<144) then old=144;
if (old>2880) then old=2880;
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page A-9
err = BMR-fit;
err2 = BMR-fit2;
err3 = BMR-fit3;
errO = BMR-old;
run;
title "fitted bmr - all females";
proc gplot data=fall;
plot fit*age;
run; quit;
title "fitted bmr with height - all females";
proc gplot data=fall;
plot fit2*age;
run; quit;
title "fitted bmr with ht and loght - all females";
proc gplot data=fall;
plot fit3*age=3;
run; quit;
axisl order = 0 to 3000 by 1000;
title 'BMR - all males';
proc gplot data=mall;
plot bmr*age /vaxis=axis1;
run; quit;
title 'BMR - all females';
proc gplot data=fall;
plot bmr*age /vaxis=axis1;
run; quit;
proc means data=fall n mean std var min max;
var bmr errO err err2 err3;
run;
proc sort data=fall; by age; run;
proc means data=fall noprint;
by age;
var bmr fit fit2 fit3 old err err2 err3 errO;
output out=fstats mean=;
run;
symboll color=black;
symbol2 color=red;
symbol3 color=blue;
title "mean bmr and old and new fits - females";
title2 "data=black, old fit=red, new fit=blue";
proc gplot data=fstats;
plot old*age=2 fit*age=3 bmr*age=l /overlap;
run; quit;
proc means data=males(where=(ht NE .)) n mean std var; var bmr; run;
proc means data=females(where=(ht NE .)) n mean std var; var bmr; run;
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page B-1
* August 2, 2016 by WGG, based on program by Jonathan Cohen;
libname apex 'C:\main\APEX\WA342\task4\task4';
data adams4;
set apex.adams4 end=eof;
* The following four obs deleted by JEL email of 3/2/2016;
if STUD = 2 and SUBJ = 32 and step = 1.0 then delete;
if STUD = 2 and SUBJ = 38 and step = 1.0 then delete;
if STUD = 20 and SUBJ = 8 and step = 5.0 then delete;
if STUD = 30 and SUBJ = 114 and step = 0.1 then delete;
if ve=. or In vo2=. or vo2m=. or gend=. or age=. then delete;
* V02 units are L/min;
vo2 = exp(ln vo2);
* V02m is personal maximum V02 in L/min;
retain suml 0;
suml = suml + ve;
if (eof) then do;
meanve= suml/ N ;
call symput ("mean ve",trim(left(meanve)));
end;
* Macro variable mean ve is used later in calculating r2 for ve;
drop suml meanve;
label vo2='V02';
run;
proc sort data=adams4 out=sorted; by stud subj; run;
data persons;
set sorted;
by stud subj;
retain vo2max nobs 0;
keep stud subj nobs vo2m vo2max;
if first.subj then do; nobs=0; vo2max=vo2m; end;
nobs = nobs+1;
if vo2max
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page B-2
g3 = f1**3;
g4 = f1**4;
g5 = f1**5;
* bmi is body mass index;
bmi = bm/(ht/100)**2;
In bmi = log(bmi);
* ht is height in cm;
In ht = log(ht);
* bm is body mass in kg;
In bm = log(bm);
* age in full years - log uses age rounded up to prevent log(0);
In age = log(l+age);
id = N ;
* Gend=-1 are males, gend=l are females;
run;
*Box-cox analysis
time;
proc transreg data
* model boxcoxfve
0. 125 -0.1111 -0.1
* model boxcoxfve
0. 125 -0.1111 -0.1
* model boxcoxfve
0. 125 -0.1111 -0.1
* model boxcoxfve
0. 125 -0.1111 -0.1
model boxcoxfve /
-0.1111 -0.1 0 0.5
* model boxcoxfve
0. 125 -0. 1111 -0.1
run;
to assess y
= base;
/ lambda= -
0 0.5 1 )=
/ lambda= -
0 0.5 1 )=
/ lambda= -
0 0.5 1 )=
/ lambda= -
0 0.5 1 )=
lambda= -1
1 )= identi
/ lambda= -
0 0.5 1 )=
/
transformation. Run one model statement at a
1 -0.5 -0.3333 -0.25 -0.2 -0.1666 -0.14286 -
identityfln vo2); * -0.2;
1 -0.5 -0.3333 -0.25 -0.2 -0.1666 -0.14286 -
identityfln vo2 fl); * -0.125;
1 -0.5 -0.3333 -0.25 -0.2 -0.1666 -0.14286 -
identityfln vo2 f2); * -0.1;
1 -0.5 -0.3333 -0.25 -0.2 -0.1666 -0.14286 -
identityfln vo2 f3) ; * 0;
-0.5 -0.3333 -0.25 -0.2 -0.1666 -0.14286 -0.125
tyfln vo2 f4); * 0;
1 -0.5 -0.3333 -0.25 -0.2 -0.1666 -0.14286 -
identityfln vo2 f5); * 0;
/* With just In vo2, the best transformation is lambda=-0.2. With higher
powers of vo2/vo2m included
this shifts to 0, which is the log transform.
*/
%macro regr(power,x);
data a;
set base end=eof;
if (&power>0) then y = ve**(-1/f&power));
else y = logfve);
run;
*calculate regression coefficients & include VIF;
proc reg data=a noprint;
model y = &x/ vif;
output out=b
p =predicted
residual=residual
rstudent=rstudent;
run; quit;
*remove studentized outliers;
data c;
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page B-3
set b;
if rstudent = . then delete;
if abs(rstudent) > 3 then delete;
run;
* Redo regression without outliers;
proc reg data=c plots(maxpoints=6700);
model y = &x/ vif;
output out=d
p =predicted2
residual=residual2
rstudent=rstudent2;
run; quit;
* Calculate and report r2 on the original variable ve;
data e;
set d end=eof;
if (&power>0) then pred = l/predicted2**(Spower);
else pred = exp(predicted2);
retain sumb suml 0;
db = (ve-Smean ve)**2;
dl = (ve-pred)**2 ;
s umb = s umb + db;
suml = suml + dl;
if (eof) then do;
vb = s umb / N ;
vl = suml / N ;
statl = 1 - vl/vb;
put "vars &x ";
put "stats " N sumb suml vb vl statl;
end;
keep stud subj ve vo2 In vo2 vo2m y fl f2 f3 f4 f5 gend pred;
run;
%mend regr;
jregx
2,
In
1
<
0
tr r2 =
0. 9479
ve r2 = 0.7350
jregx
3,
In"
vo2)
tr r2 =
0.9566
ve~r2 = 0.8779
jregx
4,
In"
vo2)
tr r2 =
0.9563
ve r2 = 0.8873
jregx
5,
In"
vo2)
tr r2 =
0.9544
ve r2 = 0.8850
jregx
6,
In"
vo2)
tr r2 =
0.9523
ve r2 = 0.8821
jregx
0,
In"
vo2)
tr r2 =
0.9341
ve r2 = 0.8561
jregx
4,
In
vo2)
* tr
r2
= 0
. 9563
ve r2
=
0. 8 E
373;
jregx
4,
In"
vo2
age)
tr
r2
= 0. 9581
ve
r2
= 0.8900;
jregx
4,
In"
vo2
gend)
* tr
r2
= 0.957
8
ve
r2
= 0. 8923;
jregx
4,
In"
vo2
ht)
tr
r2
= 0.9596
ve
r2
= 0.8938;
jregx
4,
In"
vo2
vo2m)
tr
r2
= 0.9715
ve
r2
= 0.9213;
jregx
4,
In"
vo2
fl)
tr r2 =
0. 9721
ve r2 = 0.937 8
jregx
4,
In"
vo2
f 2)
tr r2 =
0. 9712
ve r2 = 0.934 7
jregx
5,
In
vo2
fl)
tr
r2
= 0. 9730
ve
r2
= 0.9402;
jregx
5,
In"
vo2
f 2)
tr
r2
= 0. 9729
ve
r2
= 0.9420;
jregx
5,
In"
vo2
f 3)
tr
r2
= 0.9723
ve
r2
= 0. 9402;
jregx
6,
In"
vo2
fl)
tr
r2
= 0. 9730
ve
r2
= 0.9397;
jregx
6,
In"
vo2
f 2)
tr
r2
= 0.9734
ve r2
=
0.9445;
jregx
6,
In"
vo2
f 3)
tr r2 =
0. 9731
ve r2 = 0.94 42
jregx
6,
In"
vo2
f 4)
tr
r2
= 0.9723
ve
r2
= 0.9427;
jregx
0,
In"
vo2
fl)
tr
r2
= 0.9662
ve
r2
= 0.9244;
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page B-4
jregx 0, In vo2 f2)
jregx 0, In vo2 f3)
:regx 0, ln_vo2 f4)
:regx 0, ln_vo2 f5)
* tr_r2 = 0.9714
* tr_r2 = 0.9724
* tr r2 = 0.9719
* tr r2 = 0.9711 ve r2
ve r2 = 0.9411;
ve r2 = 0.9466;
ve r2 = 0.9481; * best;
: 079479;
jregx 0, In vo2 f4 age) *
jregx 0, In vo2 f4 gend)
jregx 0, ln~vo2 f4 ht) * tr_r2
jregx 0, In vo2 f4 gend age ht)
* tr_r2 = 0.9720
* tr r2 = 0.9721
= 0.9723
* tr r2 =
ve_r2 = 0.9481;
0.9726 ve r2 = 0.9477;
ve r2 = 0.9477;
ve r2 = 0.9483;
* For comparison, repeat the near-optimal regression using vo2max instead of
vo2m;
jregx 0, In vo2 g4) * ve r2 = 0.9481;
/* %regr(0, In vo2 f4) seems to be the best choice. While very high powers
(11+) of 1/ve
give marginally better r2, the log is a more usual choice, especially since
the primary
independent variable (vo2) is also log transformed.
Note: ve r2 is based on the no-outlier data set (3 studentized residuals);
On full Adams data set with (0, In vo2 f4), 6636 obs, r2 = 0.9463, which can
be
checked by running %stats(adams4) below.;
Macro %stats examines the optimal choice, examining the effects of truncating
on the predicted points. It does not seem to make much difference whether
the N(0,1)
is truncated or not, or whether the generated ve values are truncated or
not. Note that
%stats may be re-run several times, and the predicted values will change
because new
random numbers are being drawn.
%macro stats(ds);
proc sort data=&ds out=s; by stud subj; run;
data cloud;
set s end=eof;
by stud subj;
retain ss vv vl vlb v2 v2b ql qlb tl tlb 0;
veO = min(ve, 220) ;
z = rannor(0);
retain zb 0;
if first.subj then zb = rannor(O);
pi = exp(3.29821+0.79351*ln_vo2+0.53845*f4);
plb = min(max(pi,4) , 220) ;
vel = exp(3.29821+0.79351*ln_vo2+0.53845*f4+0.12529*z);
velb = min(max(vel, 4) , 220) ;
ve2 = exp(3.300+0.8128*ln_vo2+0.5126*f4+0.09866*zb+0.07852*z) ;
ve2b = min(max(ve2 , 4) , 220) ;
old = 1/(0.163-0.0816*ln_vo2-0.000342*age-0.00348*gend+0.000233*ht)**2;
outliers
-------
WA 4-52 Task 4: RMR and VE Algorithm Refinements
February 17, 2017
Page B-5
oldb = min(max(old, 4) , 220) ;
ss = ss + ve**2;
ql = ql + (pl-ve)**2;
qlb = qlb + (plb-ve)**2;
tl = tl + (old-ve)**2;
tlb = tlb + (oldb-ve)**2;
vv = vv + (ve-Smean ve)**2;
vl = vl + (vel-&mean ve)**2;
vlb = vlb + (velb-Smean ve)**2;
v2 = v2 + (ve2-&mean ve)**2;
v2b = v2b + (ve2b-&mean ve)**2;
if (eof) then do;
put "data set = &ds";
put ss vv vl vlb v2 v2b;
qql = 1-ql/vv;
qqlb = 1-qlb/vv;
ttl = 1-tl/vv;
ttlb = 1-tlb/vv;
put ql qlb qql qqlb ttl ttlb;
end;
run;
%mend;
stats(base)
stats(e)
axisl order = 0 to 220 by 20;
proc gplot data=cloud;
plot ve0*vo2 /VAXIS=axisl;
plot ve2*vo2 /VAXIS=axisl;
run;quit;
proc means data=cloud N min mean median std max;
var ve vel ve2 old;
run;
proc mixed data=e covtest plots(maxpoints=6700);
class stud subj;
model y = In vo2 f4 /solution ddfm=kr;
random subj(stud)/ solution ;
title 'data= random statement & ddfm=kr';
ods output covParms=mixedcovm old;
ods output solutionF=solutions old;
run;
-------
APPENDIX I
CONSOLIDATED HUMAN ACTIVITY DATABASE (CHAD) DATA
A total of 24 Consolidated Human Activity Database (CHAD) studies were included in
CHAD as of November 2015, with 179,912 diary-days entered. The geographic coverages range
from specific cities to collections of metropolitan areas to the entire US, and the respondents tend
to be adults but some studies include (or are limited to) children. CHAD contains human activity
data from these studies, coded into a harmonized set of location and activity codes. Note,
however, that the data collected in the original studies differed in level of detail in terms of
activity, location, and time resolution. In addition, the translation of the original study data into
CHAD format was performed by different individuals or groups. Therefore, the CHAD data
themselves will vary in specificity and resolution across the studies. One of the goals of this
manual is to provide any user with enough information to assess each study within CHAD for
appropriateness for their application. An overview of the studies is provided in Table 1-1 below.
1-1
-------
Table 1-1. Overview of Activity Studies Included in CHAD-Master (as of November 2015)
Study Name
Geographic
Coverage
Dates (as
incorporated
into CHAD)
Respondent Ages
(years; as
incorporated into
CHAD)
Data Gathering
Diary-Days (as
incorporated
into CHAD)
Study References
Baltimore Retirement
Home Study (BAL)
Baltimore County,
MD
01-02/1997
07-08/1998
>65
daily recall data collected by
study staff over a 3-week
period
391
Williams et al., 2000
American Time Use
Survey, Bureau of Labor
Statistics (BLS)
Whole US
2003-2011
>15
24-hour recall data collected
by telephone interview
combining structured
questions and
conversational interviewing
124,517
BLS, 2014
California Activity Pattern
Studies (CAA, CAC, CAY)
California
CAA and CAY:
10/1987-09/1988
CAC: 04/1989-
02/1990
CAA: 18-94
CAY: 12-17
CAC: <11
24-hour recall data collected
by telephone interviews with
structured questions
CAA: 1,579
CAY: 183
CAC: 1,200
Wiley et al., 1991a; 1991b
Cincinnati Activity
Patterns Study (CIN)
Cincinnati, OH
08-09/1985
<86
activity diary and
background questionnaire
2,614
Johnson, 1989
Detroit Exposure and
Aerosol Research Study
(DEA)
Detroit, Ml
06/2004-10/2007
>18
activities recorded via free-
form entry, while location
data were structured
340
Williams et al., 2008
Denver, Colorado
Personal Exposure Study
(DEN)
Denver, CO
11/1982-02/1983
18-70
activity diary and
background questionnaire
805
Johnson, 1984; Johnson et
al., 1986
EPA Longitudinal Studies
(EPA)
Respondents
residing in Central
NC (Raleigh,
Durham, Chapel
Hill)
1999-2000,
2002, 2006-
2008, 2012-2013
0, 35-67
paper diary; free-from
questionnaire
1,786
Isaacs et al., 2012
Population Study of
Income Dynamics PSID 1,
II, III (ISR)
Whole US
I: 02-12/19971
1: 2002-2003
111:09/2007-
05/2005
I: <12 II and II: <18
interviews; time diaries
1:5,61611:4,997
111:2,741
Alion Science and
Technology, 2012; University
of Michigan, 2014
1-2
-------
Study Name
Geographic
Coverage
Dates (as
incorporated
into CHAD)
Respondent Ages
(years; as
incorporated into
CHAD)
Data Gathering
Diary-Days (as
incorporated
into CHAD)
Study References
Los Angeles Ozone
Exposure Study:
Elementary School/High
School (LAE/LAH)
Los Angeles, CA
Fall/1989,
Fall/1990
10-17
real-time diaries
94
Roth Associates, 1988; Spier
et al., 1992
North Carolina State
University Study (NCS)
Mostly NC, 9 other
states also included
09-10/2013, 09-
10/2014
22-58
diaries recorded in real time
662
Hill, 2014
National Human Activity
Pattern Study (NHAPS):
Air/Water (NHA/NHW)
48 states
09/1992-10/1994
<93
telephone interview and
questionnaire
NHA: 4,723
NHW: 4,663
Klepeiset al., 1995; Tsang
and Klepeis, 1996
National-scale Activity
Study (NSA)
7 metropolitan
areas
06-09/2009
35-92
recall activity diary
questionnaire
6,862
Knowledge Networks, 2009
RTI Ozone Averting
Behavior Study (OAB)
35 metropolitan
areas
07-09/2002,
08/2003
2-12
no information provided at
this time
2,907
Mansfield et al., 2009
RTP Particulate Matter
Panel Study (RTP)
Wake and Orange
Counties, NC
06-11/2000,
01-05/2001
55-85
diaries recorded in real time
998
Williams et al., 2001; 2003a,b
Seattle Study (SEA)
Seattle, WA
10/1999-05/2001
6-91
diaries recorded in real time
1,692
Liu et al., 2003
Study of Use of Products
and Exposure-related
Behaviors (SUP)
California
06/2006-03/2010
<88
24-hour recall data,
collected by phone interview
9,446
Bennett et al., 2012
Valdez Air Health Study
(VAL)
Valdez, AK
04-05/1990,
08/1990,
02-03/1991
11-71
information not provided
397
Goldstein et al., 1992
Washington, DC Study
(WAS)
Washington, DC
11/1982-02/1983
18-71
activity diary and
background questionnaire
699
Hartwell et al., 1984;
Johnson et al., 1986;
Settergren et al., 1984
1-3
-------
REFERENCES
Alion Science and Technology. (2012). PSID Integration into CHAD (a description from Alion
on integrating ISR into CHAD).
Bennett DH, Teague CH, Lee K, Cassady DL, Ritz B, and Hertz-Pi cci otto I. (2012). Passive
sampling methods to determine household and personal care product use. Journal of
Exposure Science and Environmental Epidemiology 22(2): 148-160.
BLS (Bureau of Labor Statistics). (2014). American Time Use Survey User's Guide:
Understanding ATUS 2003 to 2013; Bureau of Labor Statistics, Washington, DC;
December 2014. Available at: http://www.bls.eov/tiis/atusiisersguide.pdf.
Goldstein B, Tardiff R, Hoffnagle G, and Kester R. (1992). Valdez Air Health Study: Summary
Report. Prepared for Alyeska Pipeline Service Company, Anchorage, AK.
Hill Z., 2014. Development and Evaluation of Human Longitudinal Time-Location-Activity
Data. Available at: http://www.lib.ncsu.edU/resolver/l840.4/8287.
Hartwell TD, Clayton CA, Michie RM, Whitmore RW, Zelon HS, Jones SM, and Whitehurst
DA. (1984). Study of Carbon Monoxide Exposure of Residents of Washington, D.C. and
Denver, Colorado. Prepared for the U.S. Environmental Protection Agency. Research
Triangle Park, NC.
Isaacs K, McCurdy T, Glen G, Nysewander M, Errickson A, Forbes S, Graham S, McCurdy L,
Smith L, Tulve N, and Vallero, D. (2012). Statistical properties of longitudinal time-
activity data for use in human exposure modeling. Journal of Exposure Science and
Environmental Epidemiology 23(3): 328-336.
Johnson, T. (1984). Study of Personal Exposure to Carbon Monoxide in Denver, Colorado.
Prepared for U.S. Environmental Protection Agency, Environmental Monitoring Systems
Laboratory, Research Triangle Park, NC.
Johnson, T, Capel J, and Wijnberg L. (1986). Selected Data Analyses Relating to Studies of
Personal Carbon Monoxide Exposure in Denver and Washington, DC. Prepared for U.S.
Environmental Protection Agency, Environmental Monitoring Systems Laboratory,
Research Triangle Park, NC.
Johnson, T. (1989). Human Activity Patterns in Cincinnati, Ohio. Final Report. Prepared for
Electric Power Research Institute, Health Studies Program, Palo Alto, CA.
Klepeis N, Tsang A, and Behar J. (1995). Analysis of the National Human Activity Pattern
Survey (NHAPS) Respondents from a Standpoint of Exposure Assessment. Final Report.
Prepared for U.S. Environmental Protection Agency, National Exposure Research
Laboratory, Las Vegas, NV.
Knowledge Networks. (2009). Field Report: National Scale Activity Survey (NSAS). Conducted
for Research Triangle Institute. Submitted to Carol Mansfield November 13, 2009.
1-4
-------
Liu L-JS, Box M, Kalman D, Kaufman J, Koenig J, Larson T, Lumley T, Sheppard L, and
Wallace L. 2003. Exposure assessment of particulate matter for susceptible populations
in Seattle. Environ Health Perspect 111: 909-918.
Mansfield C, Houtven GV, Johnson F R, and Yang J-C. (2009). Environmental Risks and
Behavior: Do children spend less time outdoors when ozone pollution is high? ASSA
annual meeting, January 5, 2009. Update of Houtven et al. (2003) using the OAB CHAD
data set, and related to Mansfield et al. (2006).
Roth Associates. (1988). LA_partl and LA_part2 (A Study of Activity Patterns Among a Group
of Los Angeles Asthmatics). Electric Power Research Institute
Settergren SK, Hartwell TD, and Clayton CA. (1984). Study of Carbon Monoxide Exposure of
Residents of Washington, DC.: Additional Analyses. Prepared for U.S. Environmental
Protection Agency, Environmental Monitoring Systems Laboratory, Research Triangle
Park, NC.
Spier C, Little D, Trim S, Johnson T, Linn W, and Hackney J. (1992). Activity Patterns in
Elementary and High School Students Exposed to Oxidant Pollution. Journal of Exposure
Analysis and Environmental Epidemiology 2: 277-293.
Tsang AM and Klepeis NE. (1996). Descriptive Statistics Tables from a Detailed Analysis of the
National Human Activity Pattern Survey (NHAPS) Data, U.S. Environmental Protection
Agency, Washington, D.C.
University of Michigan. (2014). The Panel Study of Income Dynamics.
http://psidonline.isr.umich.edu/Stiidies.aspx.
Wiley J, Robinson J, Piazza T, Garrett K, Cirksena K, Cheng Y, and Martin G. (1991a). Activity
Patterns of California Residents. Final Report. Prepared for California Air Resources
Board, Research Division, Sacramento, CA.
Wiley J, Robinson J, Cheng Y, Piazza T, Stork L, and Pladsen K. (1991b). Study of Children's
Activity Patterns. Final Report under contract no A733-149. Prepared for California Air
Resources Board, Research Division, Sacramento, CA.
Williams, R, Suggs, J, Creason, J, Rodes, C, Lawless, P, Kwok, R, Zweidinger, R, and Sheldon,
L. (2000). The 1998 Baltimore particulate matter epidemiology-exposure study: Part 2.
Personal exposure associated with an elderly population. J Expo Anal Environ
Epidemiol. 10(6): 533-543.
Williams RW, Wallace LA, Suggs JC, Evans EG, Creason JP, Highsmith VR, Sheldon LS, Rea
AW, Vette AF, Zweidinger RB, Leovic KW, Norris GA, Landis MS, HowardReed C,
Stevens C, Conner TL, Rodes CE, Lawless PA, Thornburg J, Liu LS, Kalman D,
Kaufman J, Koenig JQ, Larson TL, Lumley T, Sheppard L, Brown K, Suh H, Wheeler A,
Gold D, Koutrakis P, and Lippmann M. (2001). Preliminary particulate matter mass
concentrations associated with longitudinal panel studies: assessing human exposures of
1-5
-------
high risk subpopulations to particulate matter. Office of Research and Development.
United States Environmental Protection Agency. EPA/600/R-01/086.
Williams R, Suggs J, Rea A, Leovic K, Vette A, Croghan C, Sheldon L, Rodes C, Thornburg J,
Ejire A, Herbst M, and Sanders W. (2003a). The Research Triangle Park particulate
matter panel study: PM mass concentration relationships. Atmospheric Environment 37
(38): 5349-5363.
Williams R, Suggs J, Rea A, Sheldon L, Rodes C, and Thornburg J. (2003b). The Research
Triangle Park particulate matter panel study: Modeling ambient source contribution to
personal and residential PM mass concentrations. Atmospheric Environment 37 (36):
5365-5378.
Williams R, Rea A, Vette A, Croghan C, Whitaker D, Stevens C, McDow A, Fortmann R,
Sheldon L, Wilson H, Thornburg J, Phillips M, Lawless P, Rodes C, and Daughtrey H.
(2008). The design and field implementation of the Detroit Exposure and Aerosol
Research Study. Journal of Exposure Science and Environmental Epidemiology 19: 643-
659.
1-6
-------
APPENDIX J
DETAILED EXPOSURE AND RISK RESULTS
Table J-l. APEX estimates for percent of children and adults with asthma in Fall River
study area, 2011.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
32.74
12.17
5.49
2.55
1.31
0.62
200 ppb
0.24
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.41
0.14
0.05
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
1.43
0.71
0.47
0.38
0.27
0.22
200%
0.25
0.11
0.05
0
0
0
UB
100%
3.68
2.50
1.95
1.57
1.26
1.13
200%
1.48
0.99
0.77
0.60
0.52
0.52
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
5.08
0.44
0.05
0
0
0
200 ppb
0.02
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.06
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.34
0.09
0.03
0
0
0
200%
0.05
0
0
0
0
0
UB
100%
1.28
0.55
0.33
0.21
0.16
0.11
200%
0.56
0.25
0.15
0.10
0.07
0.05
J-l
-------
Table J-2. APEX estimates for percent of children and adults with asthma in Fall River
study area, 2012.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
13.21
2.76
0.56
0.12
0.03
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.14
0.03
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.77
0.44
0.25
0.14
0.08
0.08
200%
0.14
0.03
0
0
0
0
UB
100%
2.55
1.76
1.29
1.04
0.91
0.77
200%
1.10
0.74
0.55
0.44
0.38
0.30
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
1.86
0.18
0
0
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.02
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.17
0.04
0
0
0
0
200%
0.01
0
0
0
0
0
UB
100%
0.88
0.38
0.22
0.15
0.10
0.08
200%
0.39
0.16
0.10
0.07
0.05
0.03
J-2
-------
Table J-3. APEX estimates for percent of children and adults with asthma in Fall River
study area, 2013.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
12.29
1.60
0.33
0.03
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.11
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.55
0.08
0.05
0.03
0.03
0
200%
0.05
0
0
0
0
0
UB
100%
1.95
1.04
0.77
0.60
0.52
0.44
200%
0.77
0.44
0.33
0.27
0.25
0.19
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
1.32
0.07
0
0
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.07
0
0
0
0
0
200%
0
0
0
0
0
0
UB
100%
0.54
0.24
0.15
0.11
0.08
0.07
200%
0.24
0.11
0.07
0.06
0.04
0.03
J-3
-------
Table J-4. APEX estimates for percent of children and adults with asthma in Indianapolis
study area, 2011.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
27.03
7.97
2.45
0.97
0.41
0.10
200 ppb
1.04
0
0
0
0
0
300 ppb
0.83
0
0
0
0
0
400 ppb
0.31
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.57
0.19
0.11
0.07
0.06
0.05
200%
0.07
0
0
0
0
0
MEAN
100%
1.48
0.77
0.57
0.44
0.38
0.32
200%
0.36
0.17
0.11
0.09
0.06
0.06
UB
100%
3.80
2.60
2.11
1.83
1.61
1.47
200%
1.61
1.13
0.94
0.80
0.72
0.66
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
4.28
0.60
0.10
0
0
0
200 ppb
0.12
0
0
0
0
0
300 ppb
0.09
0
0
0
0
0
400 ppb
0.07
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.11
0.03
0.01
0
0
0
200%
0.01
0
0
0
0
0
MEAN
100%
0.44
0.17
0.10
0.06
0.04
0.03
200%
0.10
0.03
0.02
0.01
0.01
0.00
UB
100%
1.59
0.90
0.64
0.49
0.39
0.33
200%
0.72
0.42
0.31
0.23
0.20
0.17
J-4
-------
Table J-5. APEX estimates for percent of children and adults with asthma in Indianapolis
study area, 2012.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
22.30
7.66
2.55
0.97
0.41
0.21
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.41
0.18
0.11
0.07
0.05
0.04
200%
0.02
0
0
0
0
0
MEAN
100%
1.29
0.74
0.54
0.42
0.37
0.32
200%
0.32
0.17
0.09
0.09
0.06
0.06
UB
100%
3.53
2.51
2.06
1.81
1.62
1.49
200%
1.50
1.09
0.91
0.79
0.74
0.65
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
3.78
0.55
0.14
0.05
0.03
0.02
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.09
0.02
0.01
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.41
0.17
0.09
0.07
0.05
0.03
200%
0.09
0.03
0.02
0.01
0.01
0
UB
100%
1.54
0.90
0.64
0.48
0.39
0.32
200%
0.69
0.43
0.30
0.24
0.20
0.16
J-5
-------
Table J-6. APEX estimates for percent of children and adults with asthma in Indianapolis
study area, 2013.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
17.95
4.69
1.73
0.31
0.14
0.07
200 ppb
0.93
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.35
0.14
0.08
0.06
0.04
0.03
200%
0.02
0
0
0
0
0
MEAN
100%
1.12
0.64
0.46
0.37
0.32
0.29
200%
0.27
0.13
0.08
0.07
0.06
0.06
UB
100%
3.23
2.27
1.91
1.65
1.48
1.36
200%
1.39
0.99
0.84
0.72
0.67
0.62
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
2.90
0.36
0.12
0.03
0
0
200 ppb
0.17
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0.07
0.02
0.01
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.35
0.14
0.09
0.06
0.04
0.03
200%
0.07
0.03
0.01
0.01
0
0
UB
100%
1.41
0.81
0.58
0.45
0.37
0.31
200%
0.65
0.38
0.28
0.22
0.19
0.16
J-6
-------
Table 3-1. APEX estimates for percent of children and adults with asthma in Tulsa study
area, 2011.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
0.24
0
0
0
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0
0
0
0
0
0
200%
0
0
0
0
0
0
UB
100%
0.44
0.24
0.20
0.16
0.11
0.11
200%
0.20
0.11
0.09
0.07
0.05
0.05
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
0.10
0
0
0
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.01
0
0
0
0
0
200%
0
0
0
0
0
0
UB
100%
0.21
0.09
0.06
0.03
0.03
0.01
200%
0.10
0.05
0.03
0.01
0.01
0.01
J-7
-------
Table J-8. APEX estimates for percent of children and adults with asthma in Tulsa study
area, 2012.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
0.15
0
0
0
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.02
0
0
0
0
0
200%
0
0
0
0
0
0
UB
100%
0.53
0.35
0.26
0.22
0.18
0.16
200%
0.22
0.18
0.13
0.11
0.09
0.07
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
0.06
0
0
0
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.01
0
0
0
0
0
200%
0
0
0
0
0
0
UB
100%
0.23
0.11
0.07
0.05
0.03
0.02
200%
0.11
0.06
0.03
0.02
0.02
0.01
J-8
-------
Table J-9. APEX estimates for percent of children and adults with asthma in Tulsa study
area, 2013.
Percent of children with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
0.03
0
0
0
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of children with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.02
0
0
0
0
0
200%
0
0
0
0
0
0
UB
100%
0.44
0.33
0.27
0.22
0.18
0.16
200%
0.20
0.15
0.15
0.09
0.09
0.07
Percent of adults with asthma at elevated ventilation having exposures at or above 5-minute benchmark
concentrations
number of days per year
benchmark
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
100 ppb
0
0
0
0
0
0
200 ppb
0
0
0
0
0
0
300 ppb
0
0
0
0
0
0
400 ppb
0
0
0
0
0
0
Percent of adults with asthma estimated to experience at least one day with an increase in sRaw > 100%
number of days per year
E-R Function
sRaw
at least 1
at least 2
at least 3
at least 4
at least 5
at least 6
LB
100%
0
0
0
0
0
0
200%
0
0
0
0
0
0
MEAN
100%
0.01
0
0
0
0
0
200%
0
0
0
0
0
0
UB
100%
0.19
0.09
0.05
0.04
0.03
0.02
200%
0.09
0.05
0.03
0.02
0.01
0.01
J-9
-------
Table J-10. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Fall River, 2011,
children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
44
84
131
159
208
269
10
103
189
255
334
383
437
20
149
233
309
387
477
491
30
143
269
360
438
482
554
40
190
298
422
481
513
559
50
249
436
465
516
549
521
60
345
428
510
503
450
364
70
346
447
427
346
253
219
80
477
463
337
233
182
121
90
396
334
206
129
72
56
100
379
204
118
57
34
21
110
271
106
42
22
13
1
120
196
65
29
12
0
1
130
149
39
6
1
1
0
140
70
14
1
0
0
0
150
75
11
2
1
0
0
170
36
4
1
0
0
0
190
8
0
0
0
0
0
200
5
0
0
0
0
0
210
3
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-10
-------
Table J-ll. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Fall River, 2011,
adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
846
1960
3040
4178
5175
6026
10
2399
4026
4673
4712
4524
4225
20
2302
2473
2192
1828
1445
1173
30
1690
1417
1080
768
586
435
40
1257
900
550
376
251
167
50
995
604
333
162
115
104
60
740
327
184
97
76
39
70
554
238
82
69
17
11
80
521
167
52
19
9
2
90
327
71
32
4
2
2
100
236
30
6
0
0
0
110
154
15
0
0
0
0
120
87
9
0
0
0
0
130
63
0
0
0
0
0
140
37
0
0
0
0
0
150
22
0
0
0
0
0
170
24
0
0
0
0
0
190
2
0
0
0
0
0
200
2
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-ll
-------
Table J-12. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Fall River, 2012,
children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
56
107
163
213
273
334
10
120
252
338
428
490
543
20
183
310
420
510
564
630
30
266
411
552
610
738
828
40
350
518
636
724
694
651
50
375
546
539
479
423
350
60
522
551
495
386
296
191
70
513
465
281
191
98
67
80
400
219
122
53
34
21
90
366
150
58
26
11
5
100
391
93
19
4
1
0
110
66
5
1
0
0
0
120
13
1
0
0
0
0
130
5
1
0
0
0
0
140
3
0
0
0
0
0
150
2
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-12
-------
Table J-13. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Fall River, 2012,
adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
1181
2616
4018
5093
6194
7068
10
3038
4422
4656
4604
4189
3734
20
2562
2475
1954
1523
1186
956
30
1770
1287
883
591
398
275
40
1225
666
379
249
143
97
50
764
353
203
102
65
32
60
608
216
91
35
17
17
70
411
123
32
22
13
4
80
273
45
15
4
0
0
90
199
17
4
0
0
0
100
214
22
0
0
0
0
110
11
0
0
0
0
0
120
2
0
0
0
0
0
130
2
0
0
0
0
0
140
0
0
0
0
0
0
150
0
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-13
-------
Table J-14. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Fall River, 2013,
children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
38
101
132
176
239
294
10
173
273
403
494
592
671
20
419
644
819
1006
1089
1180
30
453
718
851
885
917
895
40
706
884
878
763
608
466
50
560
513
333
199
132
89
60
365
245
131
67
27
16
70
166
93
38
19
8
3
80
180
63
18
6
3
1
90
125
36
9
3
1
0
100
193
41
11
1
0
0
110
97
14
1
0
0
0
120
149
2
0
0
0
0
130
4
1
0
0
0
0
140
2
0
0
0
0
0
150
1
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-14
-------
Table J-15. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Fall River, 2013,
adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
1190
2540
3819
4922
5889
6649
10
3914
5240
5344
5184
4788
4422
20
3375
2914
2296
1685
1246
943
30
1597
948
508
314
206
134
40
1077
385
195
91
61
35
50
534
117
50
15
11
4
60
208
52
11
9
0
0
70
97
13
2
0
0
0
80
56
17
2
0
0
0
90
43
4
2
0
0
0
100
74
9
0
0
0
0
110
50
0
0
0
0
0
120
35
0
0
0
0
0
130
4
0
0
0
0
0
140
0
0
0
0
0
0
150
0
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-15
-------
Table J-16. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Indianapolis, 2011,
children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
34
56
64
86
105
142
10
116
213
378
502
685
802
20
292
577
764
914
955
1082
30
401
678
794
985
1154
1240
40
667
1067
1446
1700
1985
2202
50
1015
1566
2030
2375
2487
2573
60
1258
1899
2015
1959
1850
1693
70
1345
1584
1472
1157
925
693
80
1708
1517
1105
794
483
292
90
1064
802
476
228
124
60
100
652
315
127
86
37
7
110
1337
434
124
7
7
4
120
558
79
15
11
0
0
130
142
19
0
0
0
0
140
60
11
0
0
0
0
150
7
0
0
0
0
0
170
64
7
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
22
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
56
0
0
0
0
0
400
34
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-16
-------
Table J-17. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Indianapolis, 2011,
adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
398
1014
1749
2701
3796
4860
10
3211
6683
9807
12203
13952
15240
20
5601
8600
9708
9764
9477
8911
30
6198
6939
6578
5675
4705
3802
40
6422
6067
4381
3479
2570
2184
50
4885
3267
2222
1431
996
647
60
3211
1624
902
454
268
149
70
1917
921
336
143
106
68
80
1767
573
218
100
44
37
90
915
143
68
25
6
0
100
467
87
25
0
0
0
110
697
100
12
0
0
0
120
149
12
0
0
0
0
130
100
19
0
0
0
0
140
37
0
0
0
0
0
150
25
0
0
0
0
0
170
25
0
0
0
0
0
190
6
0
0
0
0
0
200
0
0
0
0
0
0
210
12
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
6
0
0
0
0
0
400
25
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-17
-------
Table J-18. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Indianapolis, 2012,
children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
15
56
67
112
124
135
10
127
292
449
622
779
910
20
296
509
757
888
981
1116
30
378
659
832
929
1127
1184
40
607
963
1150
1408
1543
1828
50
1101
1637
2109
2416
2659
2644
60
1626
2274
2453
2375
2210
1963
70
1723
1835
1558
1281
910
719
80
1277
1075
764
442
273
213
90
1258
697
408
247
161
71
100
779
461
165
67
37
19
110
461
202
82
30
4
0
120
607
127
22
4
4
4
130
127
7
4
4
0
0
140
109
19
4
0
0
0
150
337
15
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-18
-------
Table J-19. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Indianapolis, 2012,
adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
429
896
1593
2651
3672
4736
10
3460
7449
10784
13180
14991
16292
20
5881
8202
9154
9135
8805
8084
30
5489
6690
6030
5308
4524
3933
40
5986
5420
4232
3223
2365
1755
50
5196
3790
2732
1649
1108
803
60
4207
2122
915
516
361
199
70
2197
765
292
180
68
50
80
1058
311
137
93
56
37
90
821
218
100
44
6
6
100
554
118
31
6
6
6
110
380
44
12
12
6
0
120
236
37
6
0
0
0
130
56
0
0
0
0
0
140
50
0
0
0
0
0
150
93
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-19
-------
Table J-20. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Indianapolis, 2013,
children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
30
79
94
127
139
161
10
146
228
401
581
742
903
20
243
536
738
895
981
1127
30
472
757
963
1015
1176
1169
40
712
1240
1412
1802
2041
2363
50
1401
2060
2614
2959
3083
2993
60
1749
2468
2382
1989
1704
1476
70
1940
1513
1247
925
689
461
80
1300
933
524
326
154
90
90
884
498
255
150
71
37
100
880
285
135
26
7
0
110
408
146
37
4
7
7
120
131
49
15
4
0
0
130
64
4
0
0
0
0
140
311
26
0
0
0
0
150
45
0
0
0
0
0
170
0
0
0
0
0
0
190
7
0
0
0
0
0
200
52
0
0
0
0
0
210
0
0
0
0
0
0
230
49
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-20
-------
Table J-21. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Indianapolis, 2013,
adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
454
1164
2159
3379
4499
5569
10
3522
7648
10884
13168
14661
15899
20
5924
8301
8830
8662
8507
7835
30
6254
7331
6590
5557
4536
3952
40
6932
5688
4269
3279
2464
1886
50
5345
3342
2116
1388
940
523
60
3105
1369
728
324
187
143
70
1774
616
249
131
87
50
80
1151
317
87
44
37
25
90
554
112
37
19
6
0
100
454
87
25
6
0
0
110
243
19
6
0
0
0
120
106
12
6
6
0
0
130
19
0
0
0
0
0
140
93
6
6
0
0
0
150
68
6
0
0
0
0
170
0
0
0
0
0
0
190
6
0
0
0
0
0
200
25
0
0
0
0
0
210
19
0
0
0
0
0
230
19
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-21
-------
Table J-22. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Tulsa, 2011, children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
224
460
724
930
1168
1397
10
1679
2616
3040
3251
3281
3218
20
1887
1570
1166
881
698
589
30
807
452
302
266
218
181
40
429
228
167
99
66
49
50
223
104
49
23
21
13
60
119
23
8
8
5
5
70
48
8
5
2
0
0
80
20
2
0
0
0
0
90
16
0
0
0
0
0
100
8
0
0
0
0
0
110
2
0
0
0
0
0
120
2
0
0
0
0
0
130
0
0
0
0
0
0
140
2
0
0
0
0
0
150
0
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-22
-------
Table J-23. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Tulsa, 2011, adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
4898
7860
9613
10849
11728
12248
10
6176
5478
4411
3487
2783
2341
20
2272
1052
618
437
306
244
30
772
333
214
134
89
59
40
437
163
86
36
18
15
50
258
74
24
9
6
0
60
92
15
0
0
0
0
70
50
3
0
0
0
0
80
9
0
0
0
0
0
90
12
0
0
0
0
0
100
6
0
0
0
0
0
110
6
0
0
0
0
0
120
0
0
0
0
0
0
130
3
0
0
0
0
0
140
0
0
0
0
0
0
150
0
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-23
-------
Table J-24. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Tulsa, 2012, children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
203
437
670
882
1105
1285
10
967
1547
1940
2209
2352
2444
20
2397
2476
2133
1823
1555
1353
30
965
551
432
351
307
284
40
607
356
238
175
130
81
50
147
76
46
18
7
5
60
92
15
2
0
0
0
70
38
3
0
0
0
0
80
30
0
0
0
0
0
90
15
0
0
0
0
0
100
2
0
0
0
0
0
110
0
0
0
0
0
0
120
2
0
0
0
0
0
130
0
0
0
0
0
0
140
2
0
0
0
0
0
150
3
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-24
-------
Table J-25. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Tulsa, 2012, adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
4474
7260
8998
10201
11057
11633
10
5228
5017
4331
3615
3027
2611
20
3571
2020
1251
867
659
496
30
957
413
258
172
119
107
40
496
214
92
53
33
21
50
140
30
9
6
0
0
60
65
9
0
0
0
0
70
21
0
0
0
0
0
80
18
0
0
0
0
0
90
6
0
0
0
0
0
100
0
0
0
0
0
0
110
0
0
0
0
0
0
120
0
0
0
0
0
0
130
3
0
0
0
0
0
140
3
0
0
0
0
0
150
0
0
0
0
0
0
170
3
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-25
-------
Table J-26. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Tulsa, 2013, children.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
300
582
813
1034
1234
1417
10
815
1183
1465
1646
1823
1951
20
3009
2987
2710
2436
2146
1885
30
691
422
279
233
178
142
40
576
266
183
104
73
53
50
61
25
10
5
3
3
60
13
2
0
0
0
0
70
0
0
0
0
0
0
80
0
0
0
0
0
0
90
0
0
0
0
0
0
100
2
0
0
0
0
0
110
0
0
0
0
0
0
120
0
0
0
0
0
0
130
0
0
0
0
0
0
140
0
0
0
0
0
0
150
0
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-26
-------
Table 3-21. Estimated daily maximum SO2 exposures for air quality adjusted to just meet
existing standard, while at elevated ventilation (binned): Tulsa, 2013, adults.
POPULATION
ADJUSTED EXPOSURE TO B
NS (NUMBER OF PEOPLE)
Level
At least 1
Exposure
At least 2
Exposures
At least 3
Exposures
At least 4
Exposures
At least 5
Exposures
At least 6
Exposures
0
5190
8264
9901
10997
11698
12251
10
4688
4108
3508
2956
2519
2127
20
4180
2326
1438
924
656
487
30
576
220
83
45
36
24
40
333
59
27
18
6
6
50
24
3
3
0
0
0
60
0
0
0
0
0
0
70
0
0
0
0
0
0
80
3
0
0
0
0
0
90
0
0
0
0
0
0
100
0
0
0
0
0
0
110
0
0
0
0
0
0
120
0
0
0
0
0
0
130
0
0
0
0
0
0
140
0
0
0
0
0
0
150
0
0
0
0
0
0
170
0
0
0
0
0
0
190
0
0
0
0
0
0
200
0
0
0
0
0
0
210
0
0
0
0
0
0
230
0
0
0
0
0
0
250
0
0
0
0
0
0
300
0
0
0
0
0
0
350
0
0
0
0
0
0
400
0
0
0
0
0
0
450
0
0
0
0
0
0
500
0
0
0
0
0
0
550
0
0
0
0
0
0
600
0
0
0
0
0
0
J-27
-------
2013 air quality - just meeting standard
¦Children - at least 1 exposure
•Children - at least 2 exposures
Children - at least 3 exposures
¦Children - at least 4 exposures
Children - at least 5 exposures
Children - at least 6 exposures
2011 air quality - just meeting standard
50 100 150 200
S02 Exposure Concentration (ppb)
SO, Exposure Concentration (ppb)
east 1 exposure
east 2 exposures
east 3 exposures
east 4 exposures
east 5 exposures
east 6 exposures
» > 5
100 150 200
S02 Exposure Concentration (ppb)
Children
a
—Children
a
—Children
a
Children
a
—Children
a
Children
a
90
80
70
60
50
40
30
20
10 +
2012 air quality - just meeting standard
at least 1 exposure
at least 2 exposures
at least 3 exposures
at least 4 exposures
at least 5 exposures
at least 6 exposures
^—Children
—Children
Children
Children
Children
—Children
Figure J-l. Estimated percent of children with asthma expected to experience daily
maximum 5-minute SO2 exposures at or above selected levels in Fall River
study area, air quality adjusted to just meet the existing standard, 2011-2013
(top to bottom panels).
J-2 8
-------
2011 air quality - just meeting standard
-Adults - at least 1 exposure
Adults - at least 2 exposures
Adults - at least 3 exposures
¦Adults - at least 4 exposures
Adults - at least 5 exposures
Adults - at least 6 exposures
2013 air quality - just meeting standard
¦Adults - at least 1 exposure
Adults - at least 2 exposures
Adults - at least 3 exposures
•Adults - at least 4 exposures
Adults - at least 5 exposures
Adults - at least 6 exposures
2012 air quality - just meeting standard
40
30
50 100 150 200
S02 Exposure Concentration (ppb)
50 100 150 200
S02 Exposure Concentration (ppb)
50 100 150 200
S02 Exposure Concentration (ppb)
east 1 exposure
east 2 exposures
east 3 exposures
east 4 exposures
east 5 exposures
east 6 exposures
Adults - at
Adults - at
Adults - at
Adults - at
— Adults - at
Adults - at
Figure J-2. Estimated percent of adults with asthma expected to experience daily maximum
5-minute SO2 exposures at or above selected levels in Fall River study area, air
quality adjusted to just meet the existing standard, 2011-2013 (top to bottom
panels).
J-2 9
-------
2013 air quality - just meeting standard
east 1 exposure
east 2 exposures
east 3 exposures
east 4 exposures
east 5 exposures
east 6 exposures
5 -Q
5 «
50 100 150 200
Daily maximum 5-minute exposure level (ppb)
2011 air quality - just meeting standard
east 1 exposure
east 2 exposures
east 3 exposures
east 4 exposures
east 5 exposures
east 6 exposures
2012 air quality - just meeting standard
east 1 exposure
east 2 exposures
east 3 exposures
east 4 exposures
east 5 exposures
east 6 exposures
Children
—Children
— Children
Children
—Children
Children
Children
—Children
—-Children
Children
—Children
—Children
50 100 150 200
Daily maximum 5-minute exposure level (ppb)
Children
—Children
—Children
Children
—Children
—Children
50 100 150 200
Daily maximum 5-minute exposure level (ppb)
Figure J-3. Estimated percent of children with asthma expected to experience daily
maximum 5-minute SO2 exposures at or above selected levels in Indianapolis
study area, air quality adjusted to just meet the existing standard, 2011-2013
(top to bottom panels).
J-30
-------
3 A, ¦»
S I 2
o !-§
Ill
70
E
^ 4->
£ ra ^
•5 ® =
n — o
ETC Q-
1^-2
Tie
I S g
g o o 60
ro ro
J= 1_ «
C ^ x
I ° 01 50
5
§
2011 air quality - just meeting standard
; \\\
-Adults - at least 1 exposure
-Adults - at least 2 exposures
: |\ \
Adults - at least 3 exposures
-Adults - at least 4 exposures
-Adults - at least5 exposures
Adults - at least 6 exposures
X x X 1
———
—,—,—,—,—,—,—, ,
50 100 150 200
Daily maximum 5-minute exposure level (ppb)
E
2 +-
E ™ *
•5 « =
re •- o
E-~ Q.
5 «u
>• _- 5
^ « -2
¦o J I
i S J
5 o o
90
80
70
60
50
* - s
_C
E 03
X ® =
«TJ — o
E"S °-
5 ro
> - ?
W J5 -M
150
50 100
Daily maximum 5-minute exposure level (ppb)
50 100 150 200
Daily maximum 5-minute exposure level (ppb)
2012 a
ir quality - just meeting standard
-Adults - at least 1 exposure
-Adults - at least 2 exposures
;
Adults - at least 3 exposures
-Adults - at least 4 exposures
-Adults - at least 5 exposures
Adults - at least 6 exposures
\\ \ \
\\ \
\ \ \
W\
1—,—,—,—| 1
1 1 1 1 \ J L. 1. 1
- | \ \
: \ \ \
2013 air quality - just meeting standard
-Adults - at least 1 exposure
-Adults - at least 2 exposures
Adults - at least 3 exposures
-Adults - at least 4 exposures
: 1\\
-Adults - at least5 exposures
Adults - at least 6 exposures
i l\\\
: 1\\
: \\\\\
: 1\
Figure J-4. Estimated percent of adults with asthma expected to experience daily maximum
5-minute SO2 exposures at or above selected levels in Indianapolis study area,
air quality adjusted to just meet the existing standard, 2011-2013 (top to bottom
panels).
J-31
-------
2011 air quality - just meeting standard
—Children - at least 1 exposure
—Children - at least 2 exposures
1
—Children - at least 3 exposures
—Children - at least 4 exposures
—Children - at least 5 exposures
Children - at least 6 exposures
50 100 150
S02 Exposure Concentration (ppb)
2012 air quality - just meeting standard
Children - at least 1 exposure
—Children - at least 2 exposures
— Children - at least 3 exposures
Children - at least 4 exposures
—Children - at least 5 exposures
— Children - at least 6 exposures
5 O ao 50 4-
50 100 150
S02 Exposure Concentration (ppb)
| O & 50
2013 air quality - just meeting standard
Children - at least 1 exposure
—Children - at least 2 exposures
Children - at least 3 exposures
Children - at least 4 exposures
—Children - at least 5 exposures
— Children - at least 6 exposures
50 100 150
S02 Exposure Concentration (ppb)
Figure J-5. Estimated percent of children with asthma expected to experience daily
maximum 5-minute SO2 exposures at or above selected levels in Tulsa study
area, air quality adjusted to just meet the existing standard, 2011-2013 (top to
bottom panels).
J-32
-------
2011 air quality - just meeting standard
—Adults - at least 1 exposure
—Adults - at least 2 exposures
— Adults - at least 3 exposures
—Adults - at least 4 exposures
—Adults - at least 5 exposures
Adults - at least 6 exposures
50 100 150
SO, Exposure Concentration (ppb)
| 5 8°
a <*>
Adults - at least 1 exposure
—Adults - at least 2 exposures
Adults - at least 3 exposures
Adults - at least 4 exposures
—Adults - at least 5 exposures
— Adults - at least 6 exposures
50 100 150
S02 Exposure Concentration (ppb)
2012 air quality - just meeting standard
£ "
x ^
ra ¦-
E | g
P
^ .2
£ St
3 S S
£ jj ™
"> 3 O
* o 2
£ a n
3 x Jr.
-a « -£
3 E 20
2013 air quality - just meeting standard
-Adults - at least 1 exposure
-Adults - at least 2 exposures
Adults - at least 3 exposures
-Adults - at least 4 exposures
-Adults - at least 5 exposures
Adults - at least 6 exposures
_¦ \\\
! \\
\ \v
—.—*—1—>—
—'—*—'—'—
—«—¦—¦—'—
50 100 150
S02 Exposure Concentration (ppb)
Figure J-6. Estimated percent of adults with asthma expected to experience daily maximum
5-ininute SO2 exposures at or above selected levels in Tulsa study area, air
quality adjusted to just meet the existing standard, 2011-2013 (top to bottom
panels).
J-3 3
-------
Table J-28. Exposure-Response Function for S02-attributable increases (>100% and
>200%) in sRaw: Mean, lower prediction interval and upper prediction
interval.
E-RsRaw 100%
E-R sRaw 200%
exposure
mean
lower
upper
exposure
mean
lower
upper
5
2.49E-07
2.87E-10
5.74E-05
5
5.77E-08
6.95E-12
6.09E-05
15
4.02E-05
6.70E-07
1.14E-03
15
8.64E-06
3.07E-08
7.35E-04
25
2.92E-04
1.33E-05
3.71 E-03
25
6.38E-05
8.34E-07
2.04E-03
35
9.45E-04
7.74E-05
7.53E-03
35
2.13E-04
5.97E-06
3.81 E-03
45
2.12E-03
2.58E-04
1.24E-02
45
4.93E-04
2.33E-05
5.93E-03
55
3.90E-03
6.33E-04
1.79E-02
55
9.32E-04
6.48E-05
8.32E-03
65
6.28E-03
1.28E-03
2.41 E-02
65
1.55E-03
1.45E-04
1.09E-02
75
9.26E-03
2.25E-03
3.08E-02
75
2.34E-03
2.81 E-04
1.37E-02
85
1.28E-02
3.61 E-03
3.78E-02
85
3.33E-03
4.88E-04
1.66E-02
95
1.69E-02
5.40E-03
4.51 E-02
95
4.50E-03
7.83E-04
1.96E-02
105
2.15E-02
7.64E-03
5.26E-02
105
5.86E-03
1.18E-03
2.27E-02
115
2.66E-02
1.03E-02
6.03E-02
115
7.40E-03
1.69E-03
2.59E-02
125
3.21 E-02
1.35E-02
6.81 E-02
125
9.11 E-03
2.33E-03
2.92E-02
135
3.80E-02
1.71 E-02
7.60E-02
135
1.10E-02
3.10E-03
3.25E-02
145
4.41 E-02
2.12E-02
8.39E-02
145
1.30E-02
4.02E-03
3.58E-02
160
5.40E-02
2.81 E-02
9.59E-02
160
1.63E-02
5.67E-03
4.09E-02
180
6.80E-02
3.87E-02
1.12E-01
180
2.13E-02
8.42E-03
4.79E-02
195
7.90E-02
4.76E-02
1.24E-01
195
2.53E-02
1.09E-02
5.31 E-02
205
8.65E-02
5.39E-02
1.32E-01
205
2.81 E-02
1.27E-02
5.66E-02
220
9.80E-02
6.38E-02
1.44E-01
220
3.25E-02
1.57E-02
6.19E-02
240
1.14E-01
7.79E-02
1.60E-01
240
3.87E-02
2.03E-02
6.91 E-02
275
1.42E-01
1.04E-01
1.87E-01
275
5.04E-02
2.95E-02
8.17E-02
325
1.82E-01
1.44E-01
2.26E-01
325
6.83E-02
4.48E-02
1.00E-01
375
2.22E-01
1.83E-01
2.64E-01
375
8.72E-02
6.20E-02
1.20E-01
425
2.60E-01
2.20E-01
3.03E-01
425
1.07E-01
7.99E-02
1.40E-01
475
2.97E-01
2.55E-01
3.41 E-01
475
1.27E-01
9.77E-02
1.61 E-01
525
3.32E-01
2.87E-01
3.80E-01
525
1.47E-01
1.15E-01
1.84E-01
575
3.65E-01
3.15E-01
4.17E-01
575
1.67E-01
1.31 E-01
2.08E-01
J-34
-------
APPENDIX K
DAYTIME HOURLY CONCENTRATION ESTIMATES AND MEASUREMENTS BY
SEASON
This appendix relates to the evaluation described in section 3.2.5. This evaluation is
intended to inform the extent to which occurrences of the relatively higher daytime concentration
events at ambient air monitors are reflected in the distribution of daytime model predicted
concentrations. The following steps were performed to prepare the ambient air monitor and
modeled concentration data sets for this evaluation.
(1) Selection of the datasets to best represent the distributions of ambient air monitor and
model concentrations. For the monitor data, we used the reported unadjusted (as is)
values with no augmentation for time points where missing. The monitor datasets used
met completeness criteria (section 3.5.1) for each year used (2011-2013), although some
seasons may be relatively more (or less) complete than others, even within the 3-year
pooled dataset. The AERMOD estimates provide a complete time-series of hourly
ambient air concentrations in each year. Similar to the Appendix D evaluation, this
evaluation uses the AERMOD estimates (as is) for each monitor site.
(2) Stratification of the monitor and model distributions of hourly concentrations by time of
day and season. Time-of-day was split into two categories: daytime (hours most likely
associated with population exposure) included the hours of 6AM to 8PM, and nighttime
included all other hours. Seasons were stratified as winter (December, January,
February), spring (March, April, May), summer (June, July, August), and fall
(September, October, November). The result is 8 datasets of monitor concentrations and
of model concentrations (winter daytime, winter nighttime, spring daytime, etc.) for each
monitor location in each study area.
(3) Addressing missing monitor concentrations for some time points (while having complete
model concentrations)} The model receptor concentration distribution was paired with
the monitor concentration distribution based on percentile within each distribution.
The paired model and monitor concentration distributions were plotted for each of the
two times-of-day and four seasons. Figure 3-1 presents the figures for the monitor location in
each study area with the highest design value and for the daytime hours in the three warmer
1 While the air quality model predicts values for every day and hour, an ambient air monitor typically does not
measure for every hour in every year. Therefore, this distribution of values was calculated for each data set to
have an equal reference point rather than compare only concentrations for reported measurements. To maximize
the relative number of percentiles with respect to hours of data points in each season (which generally range from
2,730 to 3,864), the 0 to 100 percentiles were calcuated using every 0.04 percentile, thus 2,500 values were
generated for every season and time of day pair.
K-l
-------
months. The complete set of daytime graphs for all monitor locations and seasons are provided
here.
Fall River 250051004 - As Is (Winter-Day)
E
I 100
t! 25
s
«
. ¦
r<
25 50 75 100 125
Observed Hourly S02 Concentration (ppb)
Fall River 250051004 - As Is (Spring-Day)
1100
I
t3 25
..
25 50 75 100 125
Observed Hourly S02 Concentration (ppb)
Fall River 250051004 - As Is (Summer-Day)
E
1100
tl 25
• •
• •
•
25 50 75 100 125
Observed Hourly SOz Concentration (ppb)
Fall River 250051004 - As Is (Fall-Day)
1100
I
25 50 75 100 125
Observed Hourly SOz Concentration (ppb)
Figure K-l. Comparison of predicted and observed hourly SQi concentrations in ambient
air (2011-2013): Fall River monitor 250051004 having the highest design value
in study area.
K-2
-------
225
XI
&200
| 175
E
1 150
u
S125
° 100
_>
§ 75
X
"g 50
2
"0 25
£
Q.
0
Indianapolis 180970057 -As Is (Winter-Day)
225
XI
a 200
| 175
2
I 150
0
3125
° 100
_>¦
1 75
X
"g 50
tJ
1 "
a.
o
Indianapolis 180970057 - As Is (Spring-Day)
• «
•
•
..i
•
••••
• $••***
•
/ •••
25 50 75 100 125 150 175 200
Observed Hourly S02 Concentration (ppb)
225
25 50 75 100 125 150 175 200 225
Observed Hourly S02 Concentration (ppb)
225
XI
&200
| 175
e
1 150
u
8125
8 100
_>¦
i 75
X
"g 50
1 25
£ n
ndianapolis 180970057 - As Is (Summer-Day)
225
XI
a 200
| 175
2
I 150
0
I1"
° 100
_>¦
1 75
X
"g 50
tJ
1 25
a.
n
Indianapolis 180970057-As Is (Fall-Day)
• ,/
•
.
•
•
_ /
0 25 50 75 100 125 150 175 200 225
Observed Hourly S02 Concentration (ppb)
0 25 50 75 100 125 150 175 200 225
Observed Hourly S02 Concentration (ppb)
Figure K-2. Comparison of predicted and observed hourly SO2 concentrations in ambient
air (2011-2013): Indianapolis monitor 180970057 having the highest design
value in study area.
K-3
-------
Indianapolis 180970073 (Winter-Day)
225
— 200
a. 175
V. 100
Indianapolis 180970073 (Spring-Day)
225
— 200
_D
Q-175
Q.
^150
o
l/l
u
73
HI
" I 11 I 1 "I I ' . ' ' I "
50 75 100 125 150 175 200 225
Observed S02 (ppb)
125
"8 100
75
50 -
25
0
• •
H-
0 25 50 75 100 125 150 175 200 225
Observed S02 (ppb)
Indianapolis 180970073 (Summer-Day)
225
200
_D
a. 175
a.
^150 +
o
LO
125 :
"S 100
0)
u
0)
75 --
50
25
0
—
I
—
25 50 75 100 125 150 175 200
Observed Hourly S02 (ppb)
u
~u
ai
225
Indianapolis 180970073 (Fall-Day)
225
— 200
_Q
Q-175
Q.
"^150 +
o
l/l
125
"5 100
ai
75 -
50
25
0
H-
25 50 75 100 125 150 175 200
Observed Houlry S02 (ppb)
225
Figure K-3. Comparison of predicted and observed hourly SO2 concentrations in ambient
air (2011-2013): Indianapolis monitor 180970073.
K-4
-------
Indianapolis 180970078 (Winter-Day)
90
.a
a
a. so
60
50
40
30
20
10
i*
• •
a* * *. 1 1 1 1 1 . 1 1 1 1 1
11,11 ¦ , 11
1
rill - ...I..
Indianapolis 180970078 (Spring-Day)
no
l/l
-o
01
"U
01
0 10 20 30 40 50 60 70 80 90 100 110
Observed S02 (ppb)
90
-Q
a
a. so
n 70
o
«
•
i""l
""l""
" -I""
10 20 30 40 50 60 70 80 90 100 110
Observed S02 (ppb)
Indianapolis 180970078 (Summer-Day)
Indianapolis 180970078 (Fall-Day)
_100 -
"g. 90
^ 80
n 70
S 60
¦a 50
a>
t; 40
"O 30
£ 20
a- in
Predicted SOz (ppb)
M l-
MNJW^LnfflSlCOlOOh
oooooooooooc
:
:
i
•
•
«• '* *
•
•
•
r
0 10 20 30 40 50 60 70 80 90 100 110
Observed S02 (ppb)
10 20 30 40 50 60 70 80 90 100 110
Observed S02 (ppb)
Figure K-4. Comparison of predicted and observed hourly SO2 concentrations in ambient
air (2011-2013): Indianapolis monitor 180970078.
K-5
-------
Tulsa 401430175 - As Is (Winter-Day)
Tulsa 401430175 - As Is (Spring-Day)
90
90
Q. 80
CL 80
£ 70
2 30
"S 20
"S 20
"8 10
0
90
0
10
20 30 40 50 60 70
Observed Holurly S02 Concentration (ppb)
90
Observed Hourly S02 Concentration (ppb)
Tulsa 401430175 - As Is (Summer-Day)
Tulsa 401430175 - As Is (Fall-Day)
90
a 80
a 80
£ 70
£ 70
£ 60
60
2 30
2 30
"S 20
"8 io
"S io
0
20 30 40 50 60 70
Observed Hourly S02 Concentration (ppb)
0
10
20
30 40 50 60 70
Observed Hourly S02 Concentration (ppb)
90
Figure K-5. Comparison of predicted and observed hourly SO2 concentrations in ambient
air (2011-2013): Tulsa monitor 401430175 having the highest design value in
the study area.
K-6
-------
70
Tulsa 401430235 (Winter-Day)
1 ¦ 1 I 1 1 1 I 1
10 20 30 40 50
Observed S02 (ppb)
60
70
3" 60
Ql
Q-
70
Tulsa 401430235 (Spring-Day)
:
E
E
E
f'''
:
1
• 1
20 30 40 50
Observed S02 (ppb)
60
70
70
2? 60
Q.
350
840
"g 30
4-*
U
20
01
Tulsa 401430235 (Summer-Day)
® ; /
£ 10 J
o '
0 10
!
4-
—i i , . |
20 30 40 50
Observed SOz (ppb)
60
Tulsa 401430235 (Summer-Day)
70
70
2" 60
Q.
350
840
"g 30
u r
^20 ;
-------
40 -r
Tulsa 401431127 (Winter-Day)
10 15 20 25 30
Observed S02 (ppb)
35
40
3-35
a 30
PM
25
o
to
-o 'u
S 15
o
n10
40
Tulsa 401431127 (Spring-Day)
1
¦
, !>
•
E
f1''1"'1
1 1 1 ' 1 ¦' 1 1 1 1 ' ' 1 '
1 1 ' 1 1 1 ¦ 1 1
1 ' ' • 1 1 ' ' 1
10 15 20 25 30
Observed S02 (ppb)
35
40
3"35 r
a 30
20
Tulsa 401431127 (Summer-Day)
-+-
-+-
10 15 20 25 30
Observed S02 (ppb)
35
40
Tulsa 401431127 (Fall-Day)
p 10
—
—
5 10 15 20 25 30
Observed S02 (ppb)
35
40
Figure K-7. Comparison of predicted and observed hourly SO2 concentrations in ambient
air (2011-2013): Tulsa monitor 401431127.
K-8
-------
United States Office of Air Quality Planning and Standards Publication No. EPA-452/R-18-003
Environmental Protection Health and Environmental Impacts Division May 2018
Agency Research Triangle Park, NC
------- |