SEI

EPA

United States
Environmental
Protection Agency

Sulfur Dioxide Health Assessment
Plan: Scope and Methods for
Exposure and Risk Assessment

Draft
November 2007

Office of Air Quality Planning and Standards
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711


-------
DISCLAIMER

This draft scope and methods plan has been prepared by staff from the Ambient
Standards Group, Office of Air Quality Planning and Standards, U.S. Environmental Protection
Agency. Any opinions, findings, conclusions, or recommendations are those of the authors and
do not necessarily reflect the views of the EPA. This document is being circulated to obtain
review and comment from the Clean Air Scientific Advisory Committee (CASAC) and the
general public. Comments on this document should be addressed to Dr. Stephen E. Graham,
U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards, C504-06,
Research Triangle Park, North Carolina 27711 (email: graham.stephen@epa.gov).


-------
Table of Contents

1.	INTRODUCTION	1

1.1	Overview of Scope and Methods Plan	1

1.2	Background on SO2 NAAQS	2

2.	AIR QUALITY CONSIDERATIONS	5

3.	EXPOSURE ASSESSMENT SCOPE AND METHODS	7

3.1	Overview	7

3.2	TIER I: Air Quality Characterization	7

3.2.1	Approach	8

3.2.1.1	1-hr Average Ambient Monitoring Data Analysis	9

3.2.1.2	5-Minute Ambient Monitoring Data Analysis	10

3.2.1.3	Statistical Model Development	12

3.2.1.4	Statistical Model Application	13

3.2.2	Generated Outcomes	14

3.2.3	Variability and Uncertainty	14

3.3	TIER II: Exposure Assessment	17

3.3.1	Dispersion Modeling Approach	18

3.3.2	Approach for Estimating 5-Minute Peak Concentrations	19

3.3.3	Exposure Modeling Approach	21

3.3.4	Populations Modeled	24

3.3.5	Generated Outcomes	24

3.3.6	Variability and Uncertainty	24

3.4	Criteria for Determining Approach	25

4.	RISK ASSESSMENT SCOPE AND METHODS	27

4.1	Overview	27

4.2	TIER I: Health Effects Evaluation	29

4.3	TIER II Assessment	30

4.3.1	Approaches	31

4.3.2	Generated Outcomes	32

4.3.3	Variability and Uncertainty	32

4.4	TIER III Assesment	33

4.4.1	Approach	33

4.4.1.1	Selection of Health Effect Endpoints	34

4.4.1.2	Selection of Concentration-Response Functions	34

4.4.1.3	Baseline Health Effects Incidence Considerations	35

4.4.2	Generated Outcomes	35

4.4.3	Variability and Uncertainty	36

4.4	Criteria for Determining Approach	36

4.5	Broader Health Risk Characterization	37

5.	SCHEDULE AND MILESTONES	38

6.	REFERENCES	39

1


-------
List of Tables

Table 1. Summary of Metrics and Tools Used for each Tier of the SO2 Exposure Assessment... 7
Table 2. Key Milestones for the Exposure and Health Risk Assessment for the SO2 NAAQS

Review	38

List of Figures

Figure 1. Ambient monitoring site average S02 ambient concentrations between 1990 and 2006
(white line), upper and lower shaded regions indicate concentrations for sites within
the 90th and 10l percentiles, respectively. The data were obtained from a mix of

source- and population- oriented ambient monitors	6

Figure 2. Percent of Total SOx Emissions in the United States by Major Source Categories	9

Figure 3. Cumulative density functions (CDF) of 5-minute peak to 1-hr mean ratios for three

locations. Data obtained from Stoeckenius et. al (1990) Table 2-18	11

Figure 4. Illustrative example of a semi-empirical cumulative density function (CDF) of peak-
to-mean ratios (PMRs). Figure modified using data from Thompson (2000) (Table 5)
that were derived from 5-minute ambient monitoring data, years 1990-2000	 12

11


-------
List of Acronyms/Abbreviations

AERMOD

American Meteorological Society (AMS)/EPA Regulatory Model

ALA

American Lung Association

APEX

EPA's Air Pollutants Exposure model, version 4

AQS

EPA's Air Quality System

CASAC

Clean Air Scientific Advisory Committee

CEM

Continuous Emission Monitoring (CEM) data

CHAD

EPA's Consolidated Human Activity Database

CDF

Cumulative Density Function

COV

Coefficient of Variation

C-R

Concentration-Response relationship

CRSTER

Single Source Dispersion Model

EPA

United States Environmental Protection Agency

EOC

Exposure of Concern

FEVi

decreased Forced Expiratory Volume in one second

HEM

EPA's Human Exposure Model

hr

Hour

ISA

Integrated Science Assessment

ISCST

EPA's Industrial Source Complex Short-Term model

km

Kilometer

ME

Microenvironment

min

Minute(s)

MW

Megawatt

NAAQS

National Ambient Air Quality Standards

NCEA

National Center for Environmental Assessment

NEI

National Emissions Inventory

NEM

NAAQS Exposure Model

NCDC

National Climatic Data Center

NWS

National Weather Service

03

Ozone

OAQPS

Office of Air Quality Planning and Standards

ORD

Office of Research and Development

PM

Particulate Matter

PMR

Peak-to-Mean Ratio

ppb

parts per billion

ppm

parts per million

PRB

Policy-Relevant Background

SD

Standard Deviation

so2

Sulfur Dioxide

sox

Sulfur Oxides

SRaw

Specific Airway Resistance

TRIM

EPA's Total Risk Integrated Methodology

UARG

Utility Air Regulatory Group

in


-------
1. INTRODUCTION

The U.S. Environmental Protection Agency (EPA) is presently conducting a review of
the sulfur dioxide (SO2) national ambient air quality standards (NAAQS). Sections 108 and 109
of the Clean Air Act (Act) govern the establishment and periodic review of the NAAQS. These
standards are established for pollutants that may reasonably be anticipated to endanger public
health and welfare, and whose presence in the ambient air results from numerous or diverse
mobile or stationary sources. The NAAQS are to be based on air quality criteria, which are
meant to accurately reflect the latest scientific knowledge useful in characterizing and assessing
identifiable effects on public health or welfare that may be expected from the presence of the
pollutant in ambient air. The EPA Administrator is to promulgate and periodically review, at
five-year intervals, primary (health-based) and secondary (welfare-based) NAAQS for such
pollutants.1 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
appropriate. The Act also requires that an independent scientific review committee advise the
Administrator as part of this NAAQS review process, a function now performed by the Clean Air
Scientific Advisory Committee (CASAC).

EPA's plan and schedule for this SO2 NAAQS review is presented in the Plan for Review
of the Primary National Ambient Air Quality Standardfor Sulfur Dioxide (US EPA, 2007a).

That plan discusses the preparation of two key components in the NAAQS review process: an
Integrated Science Assessment (ISA) and risk/exposure assessments. The ISA (US EPA, 2007b)
critically evaluates and integrates scientific information on the health effects associated with
exposure to oxides of sulfur (SOx) in the ambient air. The risk/exposure assessments will
develop, as appropriate, quantitative estimates of human exposure and health risk and related
variability and uncertainties, drawing upon the information summarized in the ISA. This draft
document describes the scope and methods planned to conduct these assessments.

1.1 OVERVIEW OF SCOPE AND METHODS PLAN

This plan is designed to outline the scope and approaches and highlight key issues in the
estimation of population exposures and health risks posed by S02 under existing air quality
levels, upon just meeting the current SO2 primary NAAQS, and upon just meeting potential
alternative standards that may be under consideration. The risk/exposure assessments will draw
upon the information presented in the Integrated Science Assessment (ISA) and related Annexes.
This includes information on atmospheric chemistry, air quality, human exposure, the impact of
local source emissions, and health effects of concern.

S02 is one of a group of compounds known as sulfur oxides (SOx), which include
multiple chemicals (e.g., SO2, SO, SO3). However only SO2 is present at concentrations

1 Section 109(b)(1) [42 U.S.C. 7409] of the Act 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."

1


-------
significant for human exposures and the ISA indicates there is limited adverse health effect data
for the other gaseous compounds. Therefore, as in past NAAQS reviews, S02 will be considered
as a surrogate for gaseous SOx species in this assessment, with the secondarily formed particulate
species (i.e., sulfate or S04) addressed as part of the particulate matter (PM) NAAQS review.

The planned SO2 exposure and health risk assessments are designed to estimate very
short-term exposures to S02 (i.e., 5-minute) and the associated risk of adverse health effects for
persons in close proximity to local source emissions. Air quality will be characterized in urban
areas along with the associated risk of adverse health effects for 1- and 24-hr averaging times.
Risk and exposures will be assessed using a tiered approach where progression to a more
sophisticated level of analysis will depend on the availability of data and on the anticipated
utility of the results. For example, ambient air quality will initially be used as a surrogate for
exposure, while subsequent analyses may involve incorporating human activity data,
microenvironmental concentrations, and possibly the development of individual exposure
profiles. The exposure assessment will generate ambient concentrations and exposure metrics
that are most relevant for addressing concerns about health effects associated with SO2 exposure.

Health risk will initially be assessed through the identification of exposure concentration
levels associated with adverse health effects, termed potential health effect benchmarks. These
potential health effect benchmarks, based on observed effects from several short-term (i.e., 5-
minute) controlled human exposure studies evaluated in the ISA, will be used to determine how
often air quality concentrations or estimated exposures exceed exposure concentrations
associated with adverse health effects. Thus, the exposure estimates generated will be used to
estimate the number of individuals experiencing potential exposures of concern (EOC).

However, most of the recent supporting evidence for SO2 health effects is from epidemiological
studies, resulting in uncertainties regarding whether the variation in observed health effects is
caused by ambient SO2 concentrations, is from exposure to one or more correlated air pollutants,
or the result of differences in individual activity level (i.e., personal activities performed that
result in increased ventilation rates). An additional characterization of risk may involve use of
concentration-response functions, if and where sufficient and relevant data are identified in the
ISA to support development of functions that are related to ambient SO2 concentrations. These
functions typically have averaging times of 1- and 24-hrs and as a result, the ambient
concentrations used in this type of risk analysis would have the same averaging times.

This plan is intended to facilitate consultation with the CASAC, as well as for public
review, and to obtain advice on the overall scope, approaches, and key issues in advance of the
conduct of such analyses and presentation of results in the first draft of the risk/exposure
assessments. The risk/exposure assessments together with other information contained in the
SOx ISA, are intended to help inform the Administrator's judgments as to whether the current
primary standards are requisite to protect public health with an adequate margin of safety, or
whether revisions to the standards are appropriate.

1.2 BACKGROUND ON S02 NAAQS

As a first step in formulating the scope and methods plan, a point of reference was
developed by extracting key supporting results from the previous review of the NAAQS for SO2.
The most recent review of the SO2 NAAQS, completed in 1996, evaluated both the existing

2


-------
standards of 0.14 ppm daily average and 0.030 ppm annual average and whether an additional
short-term standard (e.g., 5-minute) was necessary to protect against short-term peak exposures.
Based on the health evidence at that time, it was determined that repeated exposures to 5-minute
peak S02 levels (0.60 ppm, noted here as 0.60 ppm-5min) could pose a risk of significant health
effects for asthmatic individuals at elevated ventilation rates (e.g., while exercising). Therefore,
the air quality and exposure analyses conducted for the previous review focused on the likely
frequency of such events using two approaches (US EPA, 1982a; 1982b; 1986a; 1986b; 1994a;
1994b; SAI, 1996).

First, existing ambient monitoring data were analyzed to estimate surrogate exposure
metrics. These metrics included the frequency of 5-minute peak concentrations above 0.50, 0.60,
and 0.70 ppm, the number of repeated exceedances of these concentrations, and the sequential
occurrences of peak concentrations within given a day (SAI, 1996). The results of this analysis
indicated that several locations in the U.S. had a substantial number of 5-minute peak
concentrations at or above 0.60 ppm, in the vicinity of local sources.

The previous review also included several annual exposure analyses to estimate the
likelihood that an asthmatic individual would be exposed to short-term peak S02 concentrations
while at elevated exertion levels. These analyses generally combined SO2 emission estimates
from targeted utility and non-utility sources with exposure modeling to estimate the probability
of exposure to short-term peak concentrations. The first such analysis conducted by the Agency
estimated the number of 5-minute exposures >0.5 ppm associated with four selected coal-fired
power utilities (US EPA, 1986b). An expanded analysis sponsored by the Utility Air Regulatory
Group (UARG) also considered the frequency of short-term exposure events that might result
from operation of power utility boilers, however the scope of the assessment addressed all power
plants across the nation (Burton et al., 1987). The probability of peak concentrations
surrounding non-utility sources was the focus of an additional study conducted by the Agency
(Stoeckenius et al., 1990). The resultant combined exposure estimates considering these early
analyses indicated that between 0.7 and 1.8 percent of the total asthmatic population potentially
could be exposed one or more times annually, while outdoors at exercise, to SO2 concentrations
>0.50 ppm-5min. It also was noted that the frequency of exposures above the health effect
benchmark of 0.60 ppm-5min, while not part of the analysis, would be anticipated to be lower.

Further supporting analyses considered in the prior review included a more recent
exposure assessment sponsored by the UARG (Rosenbaum et al., 1992) that centered on
emissions from fossil-fueled power plants. That study accounted for the anticipated reductions
in emissions after implementation of the acid deposition provisions (Title IV) of the 1990 Clean
Air Act Amendments. A 42% reduction in the number of 5-minute exposures to 0.50 ppm for
asthmatic individuals was expected (reducing the number of asthmatics exposed from 68,000
sown to 40,000) in comparison with the previous Burton et al. (1987) analysis. In addition, in
response to the request for public comment on the 1994 reproposal, a new exposure analysis was
submitted by the National Mining Association (Sciences International, Inc. 1995) that
reevaluated non-utility sources. Revised exposure estimates were provided for four of the seven
non-utility source categories by incorporating new emissions data and using less conservative
modeling assumptions in comparison with those used for the earlier Stoeckenius et al. (1990)
non-utility analysis. Significantly fewer exposure events (i.e., occurrence of 5-minute 0.50 ppm
or greater exposures) were estimated in this industry-sponsored revised analysis, decreasing the

3


-------
range of estimated exposures for these four sources by an order of magnitude (i.e., from 73,000-
259,000 short-term exposure events in the original analysis to 7,900-23,100 in the revised
analysis).

Based upon the results of each of these analyses, EPA concluded that exposure of
asthmatics to SO2 at concentrations that can elicit adverse health effects is likely to be a rare
event when viewed in the context of the entire population of asthmatics (61 FR 25566).
Therefore, 5-minute peak SO2 concentrations were judged not to pose a broad public health
threat when viewed from a national perspective, and a 5-minute standard was not promulgated.
In addition, the current standards of 0.14 ppm-24hr and 0.03 ppm-annual average were retained
(61 FR 25566).

EPA's decision not to establish a 5-minute standard was challenged and, on January 30,
1998, the Court of Appeals for the District of Columbia found that EPA inadequately explained
its determination. The court remanded the decision back to EPA, requiring additional rationale
to support the Agency judgment that 5-minute peaks of S02 do not pose a public health problem
even though these peaks would likely cause adverse health outcomes in a subset of asthmatics.
In response, EPA has requested and obtained from State air pollution agencies additional 5-
minute ambient SO2 data to support additional analyses that address issues raised in the Court's
remand of the Agency's last decision.

The planned exposure analysis and health risk assessment described in this Scope and
Methods Plan builds upon the methodology, analyses, and lessons learned from the assessments
conducted for the last review. These plans are based on our current understanding of the SO2
scientific literature and are subject to change as findings of the 1st draft S02 ISA are reviewed by
the CAS AC and the general public. EPA's Office of Research and Development (ORD)

National Center for Environmental Assessment (NCEA) has compiled and synthesized the most
policy-relevant science available to produce a 1st draft of the ISA (US EPA, 2007b), portions of
which have been reviewed and used in the development of the approach below. The approach
described in this plan is subject to modification to take into account CAS AC and public
comments following their review of this document as well as be guided by any additional
information contained in the second and final versions of the ISA.

4


-------
2. AIR QUALITY CONSIDERATIONS

The latest years of SO2 air quality data available since the previous review (1997-2006)
have been assembled for use in the exposure and health risk analyses. Air quality data can be
useful in evaluating historic, current, and prospective trends, for the development of statistical
relationships, as input to an exposure model, and as input to a risk model based on concentration-
response functions. The following air quality scenarios will be considered in this review, the
form and use of which is described below:

•	"as is" represents the 5-minute, 1-hr, 24-hr, and annual average ambient monitoring
concentration data reported by US EPA's Air Quality System (AQS). These data will be
used for estimating health risks associated with current air quality and for the
development of mathematical/statistical relationships among different averaging times
(e.g., 5-minute peak to 1-hr mean concentrations).

•	"simulated' concentrations are "as is " air quality data that have been modified by a
mathematical or statistical procedure to just meet a particular concentration level for a
specific averaging time. These could represent the current primary standards (i.e., 0.14
ppm 24-hr and 0.03 ppm annual averages) and/or potential alternative SO2 standards that
may be under consideration (e.g., a concentration level for a 1-hr average). Simulations
of this type would typically use the more recent 1-hr average ambient monitoring data
(years 2004-2006). Simulated air quality data are used to estimate health risks associated
with air concentrations that deviate from "as is" ambient concentrations. For example,
these simulated concentrations would be used in the air quality characterization (Section
3.2) to estimate the number of short-term peak concentrations above a health effect
benchmark level (Section 4). In addition, they may be combined with concentration-
response functions from epidemiological studies to estimate health risks (Section 4).

Two approaches will be investigated to simulate just meeting the current and potential
alternative SO2 standards, recognizing that currently every location across the U.S. meets both
the existing SO2 annual average and the 24-hr average standards (Figure 1) (US EPA, 2007c). A
proportional method is being considered as the primary approach for simulating air quality data.
SO2 concentrations would be adjusted linearly across the entire concentration distribution at the
design value monitor to just meet a potential alternative standard. Other ambient monitoring
data in the area would then be simulated in proportion to the design value monitor. This general
approach (often referred to as a concentration roll-back) has been used for evaluating just
meeting the current and alternative standards where current air quality monitoring data was
above the existing NAAQS (e.g., PM2.5 in Abt Associates, 2005).

There are additional considerations for simulating SO2 concentrations that would just
meet the existing standards where recent air quality levels are below the current standard(s). A
proportional simulation involving a concentration roll-up may be performed to simulate the
current and alternative standards that would yield ambient concentrations greater than current air
quality. However, EPA has not commonly adjusted ambient concentrations higher to simulate
just meeting alternative standards. Therefore, as part of a possible second approach, historical
monitoring data may be useful in representing scenarios that are at or near the current and

5


-------
potential alternative SO2 standards that are above recent air quality, rather than performing the
concentration roll-up given that the historical data contain higher concentrations than recent air
quality. We recognize that it will be important to characterize any analysis involving an upward
adjustment in S02 ambient concentrations as hypothetical scenarios that are very unlikely to
occur given current control programs and emission standards.

Another air quality related issue, potentially relevant to characterizing health risks
associated with SO2 ambient standards, is the characterization of policy-relevant background
(PRB) levels in the U.S. Policy-relevant background is defined as the distribution of S02
concentrations that would be observed in the U.S. in the absence of anthropogenic emissions of
S02 in the U.S., Canada, and Mexico. Estimates of PRB have been reported in the draft ISA
(Section 2.4.3) and Annexes (Section AX2.9), and for most of the continental U.S. the PRB is
estimated to be less than 10 parts per trillion (ppt) annual average. In the Ohio River Valley,
where present-day SO2 concentrations are highest (>5 ppb), this amounts to a contribution of less
than 1% percent of the total observed ambient S02 concentration (AX2.9). In the Northwestern
U.S. and Hawaii, where there are geothermal sources of SO2 (e.g., volcanic activity) the
contribution of PRB to total S02 can be as high as 70 to 80%. However, since PRB is well
below concentrations that might cause potential health effects, PRB will not be considered
separately in any estimation of health risk associated with recent air quality or alternative
standards.

A) Annual Average	B) 24-hour Average

19901991 19921993199419951998199719981999 2000200120022003 2004 20052006 19901991199219931994199519901997199819992000200120022003200420052006

Figure 1. Ambient monitoring site average S02 ambient concentrations between 1990 and 2006 (white
line), upper and lower shaded regions indicate concentrations for sites within the 90th and 10th percentiles,
respectively. The data were obtained from a both source- and population- oriented ambient monitors.

6


-------
3. EXPOSURE ASSESSMENT SCOPE AND METHODS

3.1 OVERVIEW

The exposure assessment for SO2 will estimate short-term human exposures (e.g., 5-
minute peak exposures) considering current source emissions and surrogate exposure metrics
associated with current ambient levels of SO2, with ambient levels that just meet the current
standard, and with ambient levels that just meet alternative standards that may be under
consideration. A two-tiered approach to assessing exposure will be employed, beginning with an
air quality characterization and progressing to a more refined analysis, if appropriate. The goals
of the S02 exposure assessment are: (1) to estimate short-term peak exposures to ambient
concentrations through air quality monitoring and modeling analyses, (2) to develop quantitative
relationships between time-averaged (1-hr) and short-term peak (5-minute) concentrations, and
(3) to identify key assumptions and uncertainties in the exposure estimates. The results from the
air quality analysis and exposure assessment will be used to inform the characterization of health
risks, as described in Section 4. The assessment approaches and tools to be used in each tier of
analysis are summarized in Table 1. Specific objectives, tool applications, assessment inputs,
and outcomes of each tier are described in detail below.

At each tier of the exposure assessment, an evaluation of the uncertainties will be
performed and the relative degree of confidence in the exposure estimates will be determined.
Similar to the exposure assessment approach briefly described above, a tiered approach will be
employed that begins with a qualitative uncertainty analysis and progresses to a quantitative
analysis if data are available and there is value added to the decision making process to warrant
such an analysis. The first step in the uncertainty analysis would be to identify the components
of the assessment that do or do not contribute to uncertainty, and provide a rationale for why this
is the case. A qualitative evaluation would follow for the uncertain components of the
assessment, resulting in a matrix describing, for each area of uncertainty, both the magnitude
(minimal, moderate, major) and the direction of influence (under- or over-estimate) on exposure
estimates. If sufficient data are available and if the overall magnitude of uncertainty is estimated
as high and possibly biased, a quantitative assessment of uncertainty would then be performed
for selected components of the assessment.

Table 1. Summary of Metrics and Tools Used for each Tier of the SP2 Exposure Assessment.

Tier

Exposure Metrics and Tools Used

AQ Characterization

AQ, Supplemental Data

Exposure Assessment

AQ, Supplemental Data, S02
Emissions, AERMOD, APEX

Notes

AERMOD American Meteorological Society (AMS)/EPA Regulatory Model
APEX EPA's Air Pollutants Exposure Model, version 4
AQ Air quality monitoring data

3.2 TIER I: AIR QUALITY CHARACTERIZATION

The first step in assessing exposure will be to conduct an air quality analysis relying
largely on ambient monitoring data and the information provided in the ISA and relevant

7


-------
Annexes. In this analysis, the ambient SO2 concentrations will serve as a surrogate for total
human exposure and will be used in developing of statistical relationships among various
averaging times. This analysis would include information on SO2 air quality patterns, historic
trends, local sources, and any potential concentrations of concern based on the ISA's evaluation
of the health effects evidence. The relationships among short-term peak concentrations and
various averaging times (e.g., hourly, daily, and annual) will be evaluated and used to inform
subsequent analyses (if any) of the current standards and any potential alternative standards that
may be under consideration.

The three objectives in this analysis are to: (1) estimate short- and long-term surrogate
exposures using available 1-hr average ambient S02 monitoring data, (2) develop a statistical
model(s) to estimate relevant surrogate short-term peak exposures (5-minute) associated with
various averaging times, and (3) estimate quantitative relationships among 5-minute peak and 1-
hr average ambient monitoring concentrations in proximity to important local emission sources.

All available ambient monitoring data collected since the prior S02 NAAQS review (i.e.,
both the 1-hr average and 5-minute ambient SO2 monitoring data from years 1997-2006) have
been gathered for use in this assessment and will be used as is for the development of the
statistical model(s). Simulation of recent air quality monitoring data will be required to analyze
any alternative standards that may be under consideration. While ambient S02 concentrations
have declined over time and there are no locations that are not meeting the current standard
(Figure 1), historical data may be useful for characterizing ambient concentrations that were near
or at the current standard levels.

3.2.1 Approach

In general, approximately 300 - 400 ambient monitors collected 1-hr average SO2
concentrations over the time-period of 1997-2006. The 5-minute monitoring is limited to at most
94 monitors, most of which report the maximum 5-minute concentrations, with only 15 monitors
containing continuous monitoring.2 In addition, while there is some overlap in the monitor
siting, not all of the 5-minute monitors have been co-located with hourly monitors. Because of
variable coverage in ambient monitors collecting 1-hr average and/or 5-minute SO2
concentrations across the U.S., some of the proposed air quality analyses may involve either the
1-hr average or the 5-minute data set exclusively.

The analysis of the ambient monitoring data will involve characterizing factors thought to
influence ambient SO2 concentrations. For example, there may be a variable impact on ambient
monitoring concentrations from local sources based on the monitor site location. Power
generating utilities and related processes are the most significant outdoor emission source of
SOx. Using emissions data provided in the Chapter 2 ISA Annex, Table AX2-3, electric utility
fuel combustion is noted as the largest single source category contributing to the total estimated
SOx emissions for the U.S. (Figure 2). Other SOx emission sources may be identified as being
important, based on consideration of local emissions estimates.

2 The monitor counts are approximate since the number operating monitors can vary from one year to the next.

8


-------
Off-Highway

Figure 2. Percent of Total SOx Emissions in the United States by Major Source Categories.

The air quality characterization also will involve development of mathematical
relationships to be used in extrapolating exposure surrogate metrics across different
concentration averaging times. This is because the averaging times for the current SO2 NAAQS
(i.e, daily and annual average concentration) and much of the ambient monitoring data (i.e., 1-hr
average concentration) are not comparable to the averaging times relevant for much of the health
effects data (i.e., 5-minutes). In addition to the analyses addressing concerns about very short-
term exposures, all of the hourly monitoring data will be evaluated considering the averaging
times related to health effects reported to be associated with 1-hr and 24-hr SO2 concentrations in
urban areas. Where very short-term data are available, the number of 5-minute peak
concentrations above a health effect benchmark level would be determined, followed by a
derivation of relationships between these very short-term peaks and their corresponding 1-hr
concentrations. These relationships will then be used for extrapolation of hourly data to estimate
additional 5-minute peak concentrations.

3.2.1.1 1-Hour Average Ambient Monitoring Data Analysis

The following initial analyses will be performed by year at each monitor reporting 1-hr
average ambient SO2 concentrations:

•	Identify possible factors that could affect or influence measured concentrations at
each monitor, e.g., whether the monitor is in close proximity to a important emission
source of SO2 or could be used to represent background SO2 concentrations within
selected locations;

•	Estimate the annual average and daily average ambient SO2 concentration;

•	Calculate surrogate exposure metrics such as the frequency of concentrations above
current daily and annual standards and considering alternative averaging times (e.g.,
estimate the distribution of hourly concentrations in particular urban areas);

•	Identify potential factors influencing surrogate exposure estimates.

9


-------
A preliminary evaluation of the spatial and temporal variability in 1-hr average SO2
monitoring data is reported in the 1st draft ISA for several U.S. cities3 (US EPA, 2007b). A
strong west-to-east ambient concentration gradient exists across the U.S., with cities in
California reporting the lowest mean concentrations, and Pittsburgh and Steubenville in the
Eastern U.S. containing the highest. This concentration gradient is consistent with the national
pattern in S02 emissions. Correlations among multiple monitors within a city are generally low
and strength of the correlation decreased with decreasing mean concentrations. This lack of
correlation reflects the spatial heterogeneity in ambient S02 concentrations and indicates that
local sources may influence variability in exposure. Compositing the data to evaluate diurnal
variation indicate that, although concentrations below the 95th percentile for each hour are
relatively indistinguishable from one another within a day, a pattern may exist for the 1-hr peak
concentrations. Most of the highest measured ambient S02 concentrations occur either at mid-
day or during the middle of the night.

Urban area analyses will be performed in this review considering several averaging times
(e.g., 1-hr, 24-hr, annual average) using both the current air quality and air quality adjusted to
just meet the current and potential alternative 1-hr, 24-hr, and annual average standards.

3.2.1.2 5-Minute Ambient Monitoring Data Analysis

The following initial analyses will be performed by year at each monitor reporting 5-
minute ambient S02 concentrations:

•	Identify possible factors that could affect or influence measured concentrations at
each monitor (e.g., proximity to local source emissions, type of local sources);

•	Calculate peak-to-mean ratios (PMRs; 5-minute to 1-hr average concentrations);

•	Identify potential factors influencing PMRs;

•	Estimate surrogate exposure metrics such as the frequency of concentrations above
potential health effect benchmarks including

o the number of peak concentration exceedances per day or year;
o the probability of multiple 5-minute exceedances within an hour.

In the prior NAAQS review, two sources of data were identified that contained source-
relevant PMR data for use in estimating the probability of 5-minute peak concentrations from 1-
hr concentrations. The first study was conducted in Kincaid, Illinois and involved monitoring 5-
minute concentrations at 18 sites surrounding a coal-fired power plant (Thrall et al., 1982). The
second source of data was generated from two ambient monitors in Billings, Montana, one of
which was located 1.6 km from a coal-fired power utility (named 'Coburn Rd.') and the other
about 4.8 km from the power utility (named 'Scottish Rites') (Stoeckenius, 1990). Cumulative
density functions (CDFs) of the PMRs from each of these sources are summarized in Figure 3.
The CDFs from Thrall et al. (1982) were used in the each of the exposure analyses conducted in
the previous review and applied universally to both the utility and non-utility sources. The
Scottish Rites ratio data were used for estimating the lower bounds of exposure in the non-utility
analyses, since the data were noted as likely more representative of PMRs that would occur

3 Urban areas containing at least 4 ambient monitors include Philadelphia, Washington, Jacksonville, Tampa,
Pittsburgh, Steubenville, Chicago, Salt Lake City, Phoenix, San Francisco, Riverside, Los Angeles (US EPA,
2007b; Section 4.1.2) for ambient monitoring conducted during the years 2003-2005.

10


-------
where individuals reside in surrounding neighborhoods compared to the Kincaid data
(Stoeckenius et al., 1990). In review of the prior analyses, the Coburn road data do not appear to
have been used for obtaining any estimates of short-term peak concentrations.

10 :

0 -P	1	1	1	1	1	1	1	1	1	1	

1 2 3 4 5 6 7 8 9 10 11 12

5-min Peak/1-hour Mean Ratio (PMR)

Figure 3. Cumulative density functions (CDF) of 5-minute peak to 1-hr mean ratios for three locations.
Data obtained from Stoeckenius et. al (1990) Table 2-18.

Later analyses of PMRs indicated that the ratio is likely influenced by a few factors (SAI,
1995; Thompson, 2000). On average, the PMR is approximately two; however, much higher
PMR have been observed. There is greater variability in the ratio at lower 1-hr average
concentrations and has been described by an inverse relationship, i.e., there is decreasing
variability with increasing 1-hr average concentration. In addition, the location of the monitor
can be highly influential. The occurrence of short-term peak concentrations at ambient monitors
is likely to be influenced by their distance from local sources and source characteristics including
the magnitude of emissions, temporal operating patterns (e.g., seasonal, time-of-day), facility
maintenance, and other physical parameters (e.g., stack height, area terrain), as well as by local
meteorological conditions. As part of a sensitivity analysis conducted for copper-smelters, the
influence of PMRs were evaluated considering the distance from the source stratified by a
normalized 1-hr mean concentration (Sciences International, 1995)4. Distance was inversely
proportional to the PMR in all three of the 1-hr mean stratifications (i.e., <0.04 ppm, 0.04 to
<0.15ppm, and >0.15ppm), with the highest 1-hr category containing the lowest range of PMRs.

The current analysis will examine these issues again, and will benefit from the increased
5-minute monitoring effort that began as a result of negotiations with the American Lung
Association (ALA) following the 1998 Court of Appeals remand. Specifically, the current
review will have substantially more 5-minute S02 concentrations to analyze, and the benefit of
some monitors measuring all twelve 5-minute intervals in an hour, rather than just the 5-minute

4 In this analysis, normalized 1-hr concentrations were obtained by dividing by the maximum hourly concentration.

11


-------
maximum SO2 concentration in a given hour. Moreover, the current review will also analyze the
continuous monitoring data alone to determine the frequency distribution of 5-minute peaks
within a given hour and whether any temporal patterns in peak concentrations (e.g., within an
hour, particular hours of the day, etc.). Influential factors described above will be considered in
the development of PMR CDFs (e.g., 1-hr concentrations, particular source influence). The
approach here also extends the analysis beyond the characterization of the PMRs as done in the
previous review, by development of a statistical model to estimate the frequency of short-term
peak concentrations at monitors reporting only 1-hr average concentrations.

3.2.1.3 Statistical Model Development

The next step in this tier of the assessment is to develop a statistical model(s) accounting
for any important factors identified in the above analyses. The purpose of the model is to
estimate the frequency of short-term peak concentrations where only 1-hr average values were
reported, and for considering different averaging time scenarios. Thompson (2000) performed
an initial analysis of 1990-2000 air quality monitoring data to quantify the relationship between
1-hr mean and 5-minute peak concentrations. A single semi-empirical distribution of PMRs was
used based on assumptions regarding independence between the ratio and 1-hr average
concentration and grouping the 1-hr average concentrations into 6 bins (Figure 3).

1 2 3 4 5 6 7 8 9 10 11 12
5-min Peak/1-hour Mean Ratio (PMR)

Figure 4. Illustrative example of a semi-empirical cumulative density function (CDF) of peak-to-mean
ratios (PMRs). Figure modified using data from Thompson (2000) (Table 5) that were derived from 5-
minute ambient monitoring data, years 1990-2000.

The proposed analysis here builds upon that approach and includes the development of
PMR cumulative density fuctions (CDFs), including the addition of recent monitoring data and
considering any identified influential factors. Rather than assuming a single distribution is
representative of all PMRs as was done previously by Thompson (2000), this analysis would
include the development of multiple CDFs that account for influential attributes, where

12


-------
appropriate data exist. Currently, the level of the 1-hr mean concentration has been identified as
an important consideration in defining a PMR distribution. In addition, the type of source(s) and
operating conditions have also been identified as important in influencing temporal variability in
concentrations, thus possibly affecting the PMR. To address these issues, it is proposed that the
empirically-derived PMR CDFs will be stratified by 1-hr mean concentration ranges (e.g., 0.05
to 0.09 ppm 1-hr, 0.10 to 0.14 ppm 1-hr, etc.) and possibly stratified by a measure of variability
in the 1-hr mean concentrations (e.g., standard deviation (SD), coefficient of variation
(COV=mean/SD)). Staff will investigate if additional stratification of these distributions is
warranted, such as by season or possibly with distance from potential sources (i.e., power
utilites, refineries). For example, any 5-minute monitors that are identified in close proximity to
important facility types will be evaluated for possible influence by local sources by deriving
separate PMR CDFs and to be compared with those at CDFs from monitors at greater distances
from local sources. In considering these and any other identified influential factors, there will be
consistency in the assignment of PMRs developed from measurement data as applied to the 1-hr
average concentrations where 5-minute peak concentrations are not measured and in the
dispersion modeled 1-hr concentrations.

3.2.1.4 Statistical Model Application

The expected number of short-term peaks above a particular value can be estimated using
a derived ratio CDF as follows.

If c =	short-term peak concentration (ppm)

m =	1-hr mean concentration (ppm)

r =	peak to mean ratio (PMR), or dm derived above

then p =	probability of an r given all possible r

If interested in a particular peak concentration such as a health effect benchmark level
that has the same averaging time as c, then this can be represented as

h = short-term health effect benchmark concentration (ppm)

It follows that since

h = c

then r = him

thus, P = probability of c > h given m
= 1 -p

The appropriate function(s) will be applied to the 1-hr mean ambient monitoring data
considering the potential health effect benchmark level of interest and as defined by the
stratification variable(s), limited to the criterion that the health effect benchmark is the same
averaging time as the PMR. Currently this would be 5-minutes in duration, however if
alternative short term peak concentrations and associated averaging times are identified as
important (e.g., 10- or 15-min), additional functions would be developed. Regardless, the

13


-------
resultant probabilities are used to determine whether a short-term peak concentration would
exceed a health effect benchmark level for each hour given the 1-hr mean concentration (i.e.,
either there is or is not an exceedance). Once the presence or absence of short-term peaks have
been generated for each 1-hr average ambient monitoring concentration, additional relationships
can be evaluated that account for longer-term average concentrations (e.g., daily and annual
averaged) and the associated frequency of short-term 5-minute peaks. Furthermore, simulated
air quality can be used to generate estimates of short-term peak concentrations, considering air
quality scenarios that deviate from the existing air quality concentrations.

An additional application of the statistical model and the estimates of short-term peak
concentrations would consider population densities within a given distance of the ambient
monitor. Thompson (2000) combined the population residing within 5 km of each ambient
monitor with the number of 5-minute S02 concentrations above 0.6 ppm generated from the
available 5-minute ambient monitoring data and those estimated from the 1-hr average ambient
monitoring data. Similarly, estimates of the number of short-term peaks using the recent
monitoring data could be associated with the population living within 5 km or any alternative
distance of interest (e.g., 2 km, 5 km, 10 km) of the ambient monitor, however the analysis here
would also account for the fraction of asthmatic individuals rather than considering the entire
population. This analysis would serve to place the frequency of peak concentrations in context
with possible contact by susceptible populations.

3.2.2	Generated Outcomes

Descriptive statistics (e.g., daily mean ambient concentrations, annual average ambient
concentrations, PMRs, their associated percentiles, etc.) will be summarized in tables and
figures, accounting for particular factors contributing to their variability (e.g., year, location),
where relevant data exist. Newly generated CDFs will then allow for the estimation of the
number of exceedances of short-term peak concentrations (e.g., 5-minute exceedances of 0.5
ppm) considering current air quality, upon just meeting the current SO2 standards (daily and
annual) and meeting other potential alternative standards that may be under consideration. All
results, including CDFs and exceedance estimates will be compared with that reported by SAI
(1996) and Thompson (2000), where appropriate comparisons can be made.

3.2.3	Variability and Uncertainty

One general assumption regarding the air quality characterization is that quality
assurance checks have been applied to air quality data. Reported concentrations contain only
valid measures, since values with quality limitations are either removed or flagged. Therefore,
the quality of the monitoring data used contributes minimally to uncertainty. Depending on the
data set used for analysis, the temporal variability in concentrations should be representative of
that observed for SO2. Depending on degree of completeness, the short-term monitoring data
should be representative of any longer duration averaging times. However, the limited number
of monitors may not account for some of the spatial and perhaps temporal variability in most
locations and therefore contribute to uncertainty. Other concerns could result from the exclusion
of any unidentified outdoor sources, the ability of ambient monitors to capture the effect of local
sources due to their siting location, and the effect of additional local sources on personal

14


-------
exposure estimates. Additionally, there is uncertainty in the application of the identified
potential health effect benchmark levels for potentially susceptible populations.

As mentioned in the overview, a tiered approach to assessing uncertainty will be
employed with the goal of progressing to a quantitative analysis if warranted and if data are
available to support such an analysis. The first step in the uncertainty analysis would be to
identify the components of the assessment that do or do not contribute to uncertainty and to
provide a rationale for why this is the case. This is described below for this particular tier of the
assessment, although the identified components are, in a broad sense, also relevant to subsequent
exposure analyses. The following includes a preliminary qualitative evaluation for the uncertain
components of the planned Tier I analysis, indicating the direction of influence (under- or over-
estimate) on exposure estimates.

•	Ambient SO? measurement: The draft ISA (Section 2.3) notes various positive and
negative sources of interference that could contribute to uncertainty in the
measurement of SO2. Many of the identified sources (e.g., polycyclic aromatic
hydrocarbons, stray light, collisional quenching) have limited impact to S02
measurement due to the presence of instrument controls that prevent the interference.
The actual impact on any individual monitor is uncertain, i.e., the presence of
negative and positive interferences has not been quantitated. Therefore, reported
ambient monitoring concentrations could be over- or under-estimated, but likely
minimally.

•	Ambient Monitor Siting: In general, the 5-minute monitors are located in areas
impacted by local sources and thus likely capture the anticipated occurrence of peak
concentrations well. Many of the 1-hr monitors are not located near large sources of
S02, thus using the 1-hr monitors for extrapolation may lead to the overestimation of
5-minute peak concentrations. Where such sources exist near 1-hr monitors, the level
of uncertainty may be limited by the characterization of factors influencing
monitoring concentrations.

•	Temporal Representativeness: Data are valid 5-minute and 1-hr average measures
and should be of the same temporal scale as identified health effect benchmarks.
While there may be missing values within a given year contributing to uncertainty
(data will not be interpolated in the Tier I analysis), temporal profiles will be assumed
complete and representative. Criteria typically used for establishing a valid year of 1-
hr average ambient monitoring data include 75% of valid days in a year, with at least
18 1-hr measurements for a valid day (thus at least 274 valid days, a minimum of
4,932 hours). The 5-minute monitoring has been performed less frequently than the
hourly monitoring, generally only a few years of data exist per 5-minute monitor.
Due to the limited data available and that they will be used primarily for development
of statistical relationships, all 5-minute data will be used as is provided the PMR
criteria defined below are met.

•	Spatial Representativeness: In general, there are a limited number of SO2 monitors in
a given area, particularly when considering the number of monitors that report 5-
minute peaks/values. There may be locations where 5-minute peak concentrations are
higher than those measured by a local monitor, e.g., locations in close proximity to a
particular source or locations impacted by other nearby sources not represented by the

15


-------
nearest monitor, or locations where 5-minute peak concentrations occur more
frequently above a health effect benchmark compared with the nearest monitor. This
may lead to an underestimate of both the frequency and magnitude of peak
concentrations.

•	Monitor to Exposure Representativeness: Human exposure is characterized by contact
of a pollutant with a person, and as such, this analysis contains the broad assumption
that the monitoring concentrations are serving as a surrogate for exposure. The ISA
reports that personal exposure data are of limited use since ambient concentrations are
typically below the detection limit of the personal samplers. There is not a method to
quantitatively assess the impact of the uncertain relationship between 5-minute
ambient monitoring data and personal exposures for the Tier I estimates, particularly
since personal exposures are time-averaged over hours or days, and never by 5-
minute averages. Therefore the relationship of short-term peak personal exposure
concentrations (i.e., attributed to ambient) to peak-ambient is largely unknown and
thus contributes to uncertainty. An evaluation in the ISA indicates the relationship
between longer-term averaged ambient monitoring concentrations and personal
exposures is reasonably strong, particularly when ambient concentrations are above
detection limits, however personal exposure concentrations are reportedly a small
fraction of ambient concentrations. This is because outdoor concentrations are
typically V2 of the ambient concentrations, and indoor concentrations about V2 of the
outdoor concentrations (USEPA, 2007a). Therefore, the use of monitoring data as a
surrogate for exposure would likely lead to an overestimate in the number of peak
concentrations that people might encounter.

•	Peak-to-Mean Ratios: The criterion used previously by Thompson (2000) centered
on data that contained a 5-minute peak to 1-hr mean ratio of at least 1 and less than
12. Values <1 would imply the 5-minute peak is less than the 1-hr average, a
physical impossibility, and values >12 are a mathematical impossibility. While data
can be screened for values outside of these bounds it raises an issue regarding the
certainty of values within the range of 1 to 12. Staff is investigating methods for
estimating confidence intervals around the number of surrogate peak exposures and
possibly performing a cross-validation of the PMRs using subsets of the data used to
construct the CDFs. In addition, use of the historical data in developing PMR CDFs
carries the assumption that the sources present at that time are similar as current
sources, adding uncertainty to results if this were not the case.

•	Single vs. Multiple Short-Term Peak Concentrations: The model is primarily
designed to estimate the frequency of a single exceedance of a particular health effect
benchmark. However, multiple short-term peak concentrations are possible in any
hour. Preliminary analysis of the 5-minute continuous monitoring data indicates that
multiple occurrences of concentrations above 0.6 ppm-5min within the same hour are
common. Using continuous monitoring data obtained from years 1990-2000,
multiple peak concentrations (i.e., 2 or more) at or above 0.6 ppm-5min within the
same hour occurred with a 70% frequency. Analysis of recent continuous monitoring
data (i.e., 1997-2006) indicate that the frequency of multiple peaks within the same
hour is still common but less frequent (i.e., about 35%). A single peak approach for
estimating surrogate exposures would likely lead to an underestimate in the number
of potential exposure events.

16


-------
•	Simulated Air Quality: The proportional simulation procedure assumes that potential
pollutant control strategies effect sources equally and that the observed impact at any
monitor would also be distributed equally, i.e., the likelihood for reduction in
emissions or concentrations is equal for both low and high levels. This would lead to
either under- or over-estimates in concentrations if the impact of control strategies is
not distributed evenly across the different parts of the air quality distribution. In
addition, use of historical data in some of the analyses here carries the assumption
that the sources present at that time are similar to current sources, adding uncertainty
to results if this is not the case.

•	Health Effect Benchmark Representativeness: Potential health effect benchmarks
will be based on the assessment of the science as documented in the 1st and 2nd drafts
of the ISA. Since potential health effect benchmarks are derived from controlled
human exposure studies, the uncertainty about the exposure and resultant response is
primarily limited to the extrapolation from the study subjects to the modeled
population. For example, uncertainties in the exposure characterization and/or in the
susceptibility of specific populations could contribute to the overall uncertainty. As
discussed in section 4, uncertainties associated with identification of potential health
effect benchmarks will be discussed qualitatively based on information provided in
the ISA. In addition, alternative potential health effect benchmark levels will be
included in the analysis to illustrate the impact of alternative benchmark levels on the
risk characterization.

3.3 TIER II: EXPOSURE ASSESSMENT

The Tier II exposure assessment is intended to build upon exposure analyses conducted
for the previous S02 NAAQS review (Burton et al., 1987; Stoeckenius et al., 1990; Rosenbaum
et al., 1992) and an industry sponsored supplemental exposure analysis (Sciences International,
1995). The objectives of a Tier II exposure assessment would be (1) to improve the
spatial/temporal resolution of ambient concentration fields surrounding important local sources
of S02 considering current emissions, (2) to account for human attributes that influence short-
term (e.g., 5-minute) personal exposure, and (3) to account for physical factors that may
contribute to lessened or greater personal exposures.

As was done in the previous SO2 NAAQS review, a combined dispersion modeling and
exposure modeling approach would be used to simulate personal exposures of individuals
residing in close proximity to important utility and non-utility SO2 emission sources. The result
of this analysis would be the generation of person-based exposure profiles for a given population
under direct impact from these local sources of SO2, centered on the number of 5-minutes peak
exposure events in an entire year. General steps are as follows:

1.	Estimate 1-hr SO2 concentrations at receptors with varying distance from selected
facilities using the most recent emissions estimates, local meteorology, and facility
parameters as input to a dispersion model;

2.	Estimate short-term peak (5-minute) concentrations from 1-hr concentrations at
modeled receptors using PMRs, accounting for any important influential factors,
where possible;

17


-------
3. Estimate individual exposure concentrations by using the dispersion model receptor
concentrations as input to an exposure model accounting for time-location-activity
patterns of simulated persons.

While the general approach is similar to that performed in the prior review, there are
planned improvements to the approach, models used, and data available for this review. Details
of improvements are discussed in the following subsections (3.3.1 through 3.3.3).

3.3.1 Dispersion Modeling Approach

The first component in the Tier II analysis involves estimating local ambient
concentrations, given that there may be locations which people reside and visit that are not well
represented by ambient monitoring alone. When considering important local sources of SO2, it is
anticipated that short-term peak concentrations will be higher within close proximity (generally
within 20 km) of power generating utilities and non-utility emission sources. Due to the large
number of power facilities and other emission sources across the U.S., some simplifying
assumptions may be applied for simulating concentrations associated with each facility, albeit
indirectly using facility prototypes and by considering influential facility attributes and other
local features. This may involve grouping facilities into a number of defining bins based on
ranges of dispersion characteristics, meteorological/climatic conditions, specific source
characteristics, and possibly land use patterns/topography.

For example, previous analyses for power utilities employed 24 bins (Burton et al., 1987;
Rosenbaum et al., 1992) defined by dispersion characteristics (ambient concentration-to-source
emission rate ratios, or X/Q ratios), atmospheric stability classes (a measure of local atmospheric
turbulence and wind speed), and load categories representing general patterns in facility
operation (base-load vs. daily cycling or peaking units). Prototype stacks were defined to
represent each bin, hourly concentrations were estimated at receptors at a distance from the
prototype, followed with an estimation of the frequency of 5-minute peak concentration
exceedances of 0.5 ppm.5 The binning approach and use of prototypes to represent the
individual sources at that time was justified by the limited computational resources, availability
of emissions estimates, and the capabilities of the dispersion models used (CRSTER, ISCST)6.

The approach for estimating short-term concentrations in this NAAQS review would
instead use AERMOD, the EPA-approved, steady-state, Gaussian plume model (US EPA, 2004).
Relevant model input data and other modeling information include, but are not limited to the
following:

5	Following estimation of peak concentration probabilities, Burton et al. (1987) first extrapolated from the prototype
probabilities to the individual stacks based on defining characteristics, then followed with exposure modeling. In a
slightly different approach, the occurrence of 5-minute peak exposures were estimated for each prototype using an
exposure model in the analysis by Rosenbaum et al. (1992). Then exposures associated with a prototype were then
interpolated to the other stacks/facilities in each bin using a functional relationship between the fuel sulfur level and
the number of 5-minute peak exposures exceedances. The revised approach resulted in about 13% more exposure
events. See Rosenbaum et al. (1992) for a discussion of the differences in the approaches and results.

6	CRSTER is EPA's Single Source Dispersion Model. ISCST is the EPA's Industrial Source Complex Short Term
model.

18


-------
•	The most recent emissions data from the latest National Emissions Inventory (NEI)7
will be used. Sources identified as contributing greatest to SO2 emissions will be
evaluated (e.g., those contributing to >1,000 tons per year) and may include

o power-generating facilities (e.g., >25 Megawatt (MW) capacity)
o coal and oil-fired industrial boilers
o petroleum refineries
o pulp and paper mills
o copper smelters
o sulfuric acid plants
o aluminum production facilities.

•	The most recent meteorological data contained in the National Climatic Data Center
(NCDC) as reported by the National Weather Service (NWS) would be used to
perform recent year air quality simulations (e.g., 2004-2006).

•	Receptor grids may range from 100 to 500 meter polygons based on estimated spatial
variability in SO2 concentrations and then equally spaced outwards from the
facility/stack to a maximum distance of 25 km;

•	1-hr background S02 ambient concentrations obtained from local ambient monitoring
data and characterized as not directly impacted by any local sources.

A binning approach along with the use of stack/facility prototypes, if determined
appropriate, will be employed in this review to best represent the large number of individual
emission sources. Simulating each emission source individually, while technically possible, may
not be practical given the input data requirements (e.g., site topography, specific release points
for certain non-utility sources). Parameters representing site topography will be investigated as
an additional binning attribute in this review, rather than assuming flat terrain for all sources as
modeled in the prior review. Output from this analysis would include hourly SO2 concentrations
at defined receptor locations in close proximity of selected utility and non-utility sources (i.e.,
prototypes) over the duration of the simulation.

3.3.2 Approach for Estimating 5-Minute Peak Concentrations

A less intensive post-processing of the dispersion model estimated hourly concentrations
may be required compared with that performed for the previous review, due to the current
dispersion modeling capabilities (e.g., capable of incorporating fuel sulfur distributions,
background concentrations, and varying load levels) and the form of the output data (time series
of hourly concentrations for the simulation period). The estimation of 5-minute peak
concentrations at receptor points located at a distance from the source is dependent on the
development of appropriate PMR CDFs, and is described above in Section 3.2.1.2 in detail.

7 The NEI is EPA's comprehensive national emission inventory and contains emission measurements and estimates
for the criteria air pollutants and others. The NEI includes air pollution emissions from point sources (e.g., electric
utilities, petroleum refineries), mobile sources (e.g., cars, trucks, nonroad engines such as construction equipment,
etc.), and nonpoint sources (e.g., residential fuel use). The NEI is developed using the latest data and best
estimation methods including data from Continuous Emissions Monitors (CEMs), data collected from all 50 States,
as well as many local and tribal air agencies, and data using EPA's latest models such as the MOBILE and
NONROAD models. See http://www.epa.gov/ttn/chief/net/index.html for more.

19


-------
Briefly, the 5-minute peak concentrations will be estimated probabilistically considering
empirically-derived PMR CDFs obtained from either the historic source-specific data
(Stoeckenius et al., 1990) and/or that developed from recent 5-minute ambient monitoring
conducted in close proximity to particular S02 emission sources, where identified. Compared to
what was done in the prior review, the analysis of recent 5-minute monitoring data proposed in
this review could result in the development of refined PMR distributions, possibly accounting for
influential factors (e.g., considering the relationship between the 1-hr mean concentration level
and respective PMRs). Regardless, the peak concentrations associated with all 1-hr average
receptor concentrations would be estimated by random sampling from the most relevant and
available PMR CDF. Thus for every 1-hr concentration estimated at each receptor, an associated
5-minute peak SO2 concentration would be generated as the primary output of this analysis.

The approach is designed to generate 5-minute peak concentrations to use in estimating
potential exposures of concern (EOC) within an hour. In general, it is not an objective to
estimate each of the other eleven 5-minute concentrations within the hour with a high degree of
certainty. While the occurrence of multiple peak concentrations is possible, the potential health
effect benchmark levels are related to single peak exposures. Currently the exposure model uses
1-hr ambient concentrations, however if all twelve 5-minute values are determined necessary as
an input, the additional concentrations within an hour at each receptor could be approximated
using the following:

X =	eq (1)

n-1

where,

X = 5-minute concentration in each of non-peak concentration periods in the
hour at a receptor (ppm or ppb)

C = 1-hr mean concentration estimated at a receptor (ppm or ppb)
P = estimated peak concentration at a receptor (ppm or ppb)
n = number of time periods within the hour (or 12)

In addition to the level of the peak concentration, the actual time of when the contact
occurs with a person is also of importance. The ISA indicates that adverse health effects
associated with short-term peak exposures occurs with moderate to heavy exertion. Human
activities are variable over time, a wide range of activities are possible even within a single hour
of the day. The type of activity an individual performs, such as sleeping or jogging, will
influence their breathing rate. Therefore, a general strategy is needed to set the 5-minute peak
concentrations within the hour. If there are no observed short-term temporal relationships
following the evaluation of continuous PMR data, then clock times for peak values would be
estimated either randomly (i.e., any one of the 12 possible time periods within the hour) or at an
assigned time (e.g., at 35 minutes past the hour).

Multiple peak exposures may be of interest, given that there is a need to match the peak
concentration with elevated activity levels. The frequency multiple peak concentrations could be
estimated using the appropriate probabilities generated from the PMR analysis, applied here as

20


-------
an additional step in populating 5-minute peak concentrations within the hour, and modification
of equation (1).

'•C-ZP,

X=-

n-m

eq (2)

where,

X = 5-minute concentration in each of the n-m non-peak concentration periods
in the hour at a receptor (ppm or ppb)

C	=	1-hr mean concentration estimated at a receptor (ppm or ppb)

Pi	=	estimated peak concentration at a receptor (ppm or ppb)

n	=	number of time periods within the hour (or 12)

m	=	number of peak concentrations

The outcome of this analysis would be 5-minute ambient SO2 concentrations for each
grid receptor in the facility/stack(s) across the simulation period. This may be in the form of all
twelve 5-minute values that would occur within an hour (including the peak concentration(s)) or
the 5-minute peak concentration(s) within an hour alone, dependent on the utilization of such
data in the exposure model.

3.3.3 Exposure Modeling Approach

The exposure modeling approach would use EPA's Air Pollutants Exposure (APEX)
model (US EPA, 2006a; 2006b).8 APEX is a Monte Carlo simulation model used to simulate a
large number of randomly sampled individuals within an area reflecting population
demographics, thus generating area-wide estimates of population exposure. The PC-based
probabilistic model was recently used to estimate population exposures in 12 urban areas for the
O3 NAAQS review (US EPA, 2007d). The modeling approach and exposure results have been
peer-reviewed by the CASAC O3 panel as part of that NAAQS review.

APEX simulates exposures that occur in indoor, outdoor, and in-vehicle
microenvironments (MEs). The model stochastically generates simulated individuals using
Census-derived probability distributions from the 2000 Census, typically at the tract level. A
national commuting database based on 2000 Census data provides home-to-work commuting
flows between tracts. Any number of simulated individuals can be modeled, and collectively
they represent a random sample of the study area population in the modeled area.

APEX draws human time-location-activity data from EPA's Consolidated Human
Activity Database (CHAD; McCurdy et al., 2000) and generates longitudinal activity sequences
to represent the movement of simulated individuals through time and space, accounting for the
effects of particular day-types (e.g., weekday versus weekend) and temperature on daily

8 APEX is also referred to as the Total Risk Integrated Methodology/Exposure (TRIM.Expo) model (see
http://www.epa.gov/ttn/fera/trim_gen.html for general details on TRIM).

21


-------
activities. APEX calculates the concentration in the ME associated with each event9 in an
individual's activity pattern and sums the event-specific exposures by hour to obtain a
continuous time series of hourly exposures spanning the time period of interest.

The concentrations in each ME are estimated using either a mass-balance or a factors
approach, and the user specifies the probability distributions of the parameters used for the
concentration calculations (e.g., indoor-outdoor air exchange rates). These distributions can also
depend on the values of other variables in the model. For example, the distribution of air
exchange rates in a home, office, or car depends on the type of heating and air conditioning
present, which are also stochastic inputs to the model. The user can choose to retain the value of
a stochastic parameter constant for the entire simulation (e.g., house volume would remain the
same throughout the exposure period), or can specify that a new value shall be sampled hourly,
daily, or seasonally from specified distributions. APEX also allows the user to specify diurnal,
weekly, or seasonal patterns for certain ME parameters.

The calculation of ME concentrations in APEX is dependent not only on the parameter
distributions for the mass balance and factors approaches, but also on the ambient (outdoor) SO2
concentrations and temperatures. Surface temperatures would be obtained from the NCDC/NWS
and spatially interpolated for each study area as input to APEX. For the application to SO2, MEs
such as the following would be modeled, depending on available data:

•	Indoors - residence

•	Indoors - bars and restaurants

•	Indoors - schools

•	Indoors - day care centers (commercial)

•	Indoors - other (e.g., offices, shopping)

•	Outdoors - residence

•	Outdoors - other (e.g., playgrounds, parks)

•	In vehicles - cars, trucks, others

One particular consideration in this tier of the exposure assessment involves addressing
the population fraction living within the tracts containing identified sources and therefore, the
assignment of ambient concentrations to individuals in these locations. The receptor grid
concentration estimates will be at a finer spatial resolution than that commonly employed in the
exposure model, even if the exposure model used the Census scale of block group or block. The
proportion of the population in a grid may be defined based on dividing the Census scale (e.g.,
block) equally by the number of grids falling within that Census scale. The number of grids
within the Census scale would be variable since the spatial dimension of the selected census
region is also variable. Therefore, individuals residing in a given Census scale will have an
equal probability of exposure to any of the grid concentrations estimated for that Census scale.

In addition, the time associated with the input data (i.e., 5-minute) is at a finer scale than
used in the APEX (currently it is 1-hr). In the latest review completed in 1994, the Human

9 Exposure events are when an individual is in a single microenvironment, exposed to a constant concentration for a
definite time duration, and while at a constant activity level. The duration of the event is a maximum of 1-hr due to
the structure of the time-location activity diary, but could be as short as 1-minute.

22


-------
Exposure Model (HEM) was used to estimate the probability of short-term peak exposures. For
any given 5-minute period within an hour where a short-term peak concentration of concern was
estimated, it was assumed the individual was at moderate or greater exertion for the entire hour.
This approach would tend to overestimate the number of short-term peak exposures since it is
unlikely that all individuals are at heightened exertion levels for the entire 1-hr period. Staff
proposes to address this as part of this review by estimating the peak concentration in a specific
5-minute time period and correlating that with the time-location-activity pattern of the individual
within the hour. The result will be more reasonable estimates of the number of EOC compared
to those estimated previously, since the occurrence of peak concentrations will be paired with
variable breathing rates (as would be expected) within an hour, rather than assuming elevated
breathing rates for the entire hour.

Another important consideration in the design of the exposure assessment involves the
linking of particular activity levels with the potential EOC. The ISA indicated the requisite for
adverse health effects in certain individuals is exposure while performing moderate to heavy
exertion activities. Conceptually, the exertion level of an individual engaged in a particular
activity can be estimated either by energy expended or associated ventilation rates. In the
previous review, the clinical studies indicated that short-term exposures were associated with
adverse health effects at breathing rates of 35 L/min or greater. Rather than using specific
breathing rates, the exposure model(s) assumed probabilities of persons/cohorts being outdoors
performing high-level activities.10 APEX can estimate an individual's activity-specific
metabolic equivalents of work (or METS) which is used in generating associated ventilation
rates (e.g., expiratory ventilation (VE) or oxygen consumption (V02)).

Staff is investigating an improved approach to classifying when an individual is
performing elevated activities. One method would consider using the exceedance of a set METS
value to identify when the simulated individual is at the target activity level. This METS level
would be specific for the individual and likely fall between commonly used ranges for moderate
(METS from 3.0 to 6.0) and vigorous (>6.0 METS) physical activity (e.g., Ainsworth et al.,
2000). A second approach would consider whether an individual's activity-specific rate of VO2
exceeded a percentage of their normalized V02max (mL-02/min-kg). Regardless of the
approach, the selected target level would correspond to the ventilation or activity level where
adverse health effects were observed in the clinical studies. Simulated persons would be
assigned variable target levels (one per individual), accounting for any influential factors such as
age and gender.

Finally, an extrapolation of exposure estimates to the facilities not modeled would need
to be performed. As was done a previous assessment, a relationship would be developed from
the prototype emissions parameters (e.g., fuel sulfur level) and estimates of the number of
exceedances generated from the exposure model (see Rosenbaum et. al, 1992). Then facility-
specific values would be used to generate exposure estimates at each facility where exposure
modeling was not conducted.

10 The only reference with any detailed classification was Burton et al. (1987). Appendix A notes three activity
levels were assigned to a cohort (using NEM), classified as either low (200 cal/hr), med (200-500 cal/hr) and high
(>500 cal/hr) for each hour. It is unclear as to how this was proportionally applied to the probability of individuals
exercising outdoors at a particular clock hour.

23


-------
3.3.4	Populations Modeled

A detailed consideration of the population residing in each modeled area would be
included, where exposure modeling is performed. The assessment would consider susceptible
and vulnerable populations as identified in the ISA, defined from either an exposure or
susceptibility perspective. The population subgroups identified by the ISA (US EPA, 2007b)
that we plan to include in an exposure assessment include:

•	Children (birth to age 18 - subdivided to preschool/school age)

•	Asthmatic children (birth to age 18 - subdivided to preschool/school age)

•	Asthmatic adults (>19 years)

•	Elderly (> 65 years)

The proportion of the population of individuals characterized as being asthmatic will be
estimated by statistics on asthma prevalence rates recently used in the NAAQS review for O3
(US EPA, 2007d). Where sufficient data are available, region-specific data would be applied.

3.3.5	Generated Outcomes

Exposure estimates would use the most recent SO2 emissions data. The exposure
assessment would take into account several important factors including the magnitude and
duration of exposures, frequency of repeated peak exposures, and breathing rate of individuals at
the time of exposure. Estimates of exposure would include (1) temporally and spatially resolved
hourly and 5-minute peak ambient concentrations for areas surrounding local stationary sources,
(2) counts of people exposed one or more times to a given short-term peak S02 concentration at
a particular exertion level, and (3) counts of person-occurrences of particular exposures at a
given exertion level.

3.3.6	Variability and Uncertainty

The principle objective of a refined exposure assessment would be to estimate exposures
by representing the variability in a given population's characteristics that influence its exposure,
while minimizing the uncertainties. Variability can be described in terms of the empirical
quantities that are important in estimating exposure and are inherently variable across time and
space, or when considering a group of individuals (Cullen and Frey, 1999). For example, body
mass is a measurable quantity that differs for individuals within a population (depending on a
number of factors) and can be represented by a frequency distribution(s). Uncertainty tends to
reflect the degree of confidence in the use of or the representativeness of models or model
components, for the purposes of which they were designed. For example, uncertainties arise in
body mass distributions due to random or systematic measurement error, or perhaps uncertainty
is introduced by the application of a body mass distribution obtained using one population of
individuals to extrapolate to another distinct population of individuals. In this example using a
distribution of measured body mass, uncertainty can be present as apparent variability or exist as
unaccounted variability. It is within this general context that variability and uncertainty would
be addressed in this tier of the assessment.

Uncertainty would be assessed quantitatively through individual parameter analyses and
possibly a unified uncertainty analysis as described previously in the recent 03 NAAQS review

24


-------
(US EPA, 2007d; Langstaff, 2007). Briefly, there are two primary sources of uncertainty that
would be addressed in this type of a quantitative analysis. The first is uncertainty associated
with the modeling inputs (e.g., emissions estimates, peak-to-mean ratios, time-location-activity
diaries, microenvironmental factor distributions). The second is uncertainty associated with
model formulation (e.g., algorithms included in the model).

For the dispersion modeling, binning and use of prototypes in estimating receptor
concentrations carries assumptions regarding similarities in parameters such as meteorology,
dispersion characteristics, and load levels to the extrapolated stacks/facilities. This combined
with limited data on 5-minute peak to 1-hr average concentrations could result in over or under
estimation in concentrations. These and other potential sources of uncertainty could be evaluated
to determine the magnitude of impact to receptor concentration estimations through individual
sensitivity analyses. In addition, the overall approach may be evaluated by comparing dispersion
model concentrations estimates with available 5-minute ambient monitoring data that are located
within defined dispersion model grids. Recent 5-minute monitoring conducted in two counties
of Missouri (Iron and Jefferson) could provide a benchmark for comparison based on the type of
source and monitor proximities.

For APEX, a 2-dimensional Monte Carlo Latin hypercube sampling approach could be
used as a combined variability and uncertainty analysis for APEX. A Monte Carlo approach
entails performing a large number of model runs with inputs randomly sampled from specified
distributions that reflect the variability and uncertainty of the model inputs. The 2-dimensional
Monte Carlo method allows for the separate characterization of variability and uncertainty in the
model results (Morgan and Henri on, 1990). If this approach were taken, developing appropriate
distributions representing both variability and uncertainty in model inputs (e.g., dispersion model
input data, air exchange rates, S02 decay rates, physiological parameters) would be a key part of
the effort.

In the case of model formulation, the preferred approach would be to compare model
predictions with measured values; however, according to the draft ISA, these data are limited to
a few locations in the U.S., none of which have averaging times of 5 minutes. In the absence of
measurements that can be used to estimate model uncertainty, the analysis must rely on informed
judgment. The approach would be to partition the model formulation uncertainty into that of the
components, or sub-models, of APEX (e.g., microenvironemental concentrations, ventilation
estimates). For each of the sub-models, we would discuss the simplifying assumptions and the
uncertainties associated with those assumptions. Where possible, we would evaluate these sub-
models by comparing their predictions with measured data. Where this is not possible, we would
formulate an informed judgment regarding a range of plausible uncertainties for the sub-models.

3.4 CRITERIA FOR DETERMINING APPROACH

Criteria have been developed to determine the tier level of the assessment to be
performed. The criteria are designed to determine the value added to the overall assessment as
measured by assumptions retained in each tier and, either the reduction of uncertainty or the
improved characterization of uncertainty in the exposure estimates. The factors identified below
will be considered in the progression from one tier to the next.

25


-------
Outcome of the ambient air quality characterization, including the estimated number

of peak concentrations using current ambient concentrations and those assuming any

potential alternative standards that may be under consideration;

Availability of information and data defining the potential impact of important local

sources on nearby residents (e.g., time spent outdoors while at elevated exertion);

Existence of the data required to perform the analyses in each subsequent tier of the

assessment;

Representation of identified susceptible populations in the current review.

26


-------
4. RISK ASSESSMENT SCOPE AND METHODS

4.1 OVERVIEW

A three-tiered approach to characterizing health risks will be utilized. Tier I of the
assessment will review the health effects evidence in the 1st and 2nd draft ISA and identify the
health endpoints that are judged to be causal or likely causal with respect to ambient SO2 levels
at specific averaging times. A key part of the Tier I assessment will be judgments as to which of
the identified health endpoints are likely candidates for progression to a Tier II or III risk
characterization.

The Tier II risk assessment would build upon the information gathered in the Tier I
assessment. For health endpoints based on findings from controlled human exposure studies
identified during the Tier I assessment, a Tier II analysis will first determine whether potential
health effect benchmarks can be developed based on the evidence and evaluation presented in the
draft ISA. If potential benchmarks can be developed, exceedances of these health benchmarks
will be evaluated based on air quality (serving as a surrogate for exposure), or estimates from the
exposure assessment described in the previous section.

For health endpoints based on findings from epidemiological studies identified during the
Tier I assessment, a Tier II analysis will involve a more extensive evaluation of the ambient air
quality levels for SO2, and co-pollutants where possible, to see if there are any trends or patterns
with respect to the effect estimates reported. This evaluation will include examination of
whether there are any trends or patterns in the reported concentration-response relationships with
respect to the use of different averaging times and air quality metrics as well as the use of single
versus multi-pollutant models.

A Tier III risk assessment, if conducted, would involve an estimation of the number of
people expected to experience specific health effects and total number of occurrences of these
effects. This type of assessment would only be conducted for those health effects identified as
having sufficient basis to provide quantitative estimates based on the previous tiers of the
assessment. More specifically, a Tier III assessment would estimate the number of people
estimated to have one or more occurrences in a year and the total number of annual occurrences
of health effects associated with recent ambient SO2 levels and with SO2 levels that just meet the
current and alternative standards. For health effects based on evidence from controlled human
exposure studies, a Tier III risk assessment requires combining estimated exposure-response
relationships with exposure estimates for the relevant averaging time(s) and population(s)
associated with recent air quality and air quality simulated to just meet the current and potential
alternative standards to generate population risk estimates for one or more health endpoints. For
health effects based on evidence from epidemiological studies, a Tier III assessment requires
combining estimated concentration-response (C-R) relationships with either recent ambient air
quality data or simulated ambient air quality data representing just meeting the current or
potential alternative standards to generate population risk estimates for one of more health
endpoints.

27


-------
Decisions on whether or not to conduct a Tier III risk assessment will take into account
the following considerations: (1) whether the weight of the evidence supports conducting a
quantitative assessment for specific health endpoints, (2) whether the data needed to conduct
such quantitative assessments are available, (3) the anticipated utility of results to inform
decisions on the adequacy of the current SO2 NAAQS and to provide insights related to potential
alternative standards, and (4) whether or not there is adequate time and resources to complete
such assessments under the current schedule.

Ultimately, we believe this tiered approach to assessing risk will accomplish the
following goals: (1) to provide an overall characterization of the health effects associated with
ambient S02 exposures including a summary discussion of all significant health effects,
including those which are of public health concern but which are not judged appropriate for
inclusion in quantitative assessment, (2) to estimate the number of occurrences of short-term air
quality events (i.e., on the order of 5 to 10 minutes) at or above potential health effect
benchmarks associated with various air quality levels, including recent levels and air quality
levels meeting potential alternative SO2 standards, (3) to estimate the number of people exposed
at or above potential health effect benchmarks, for effects based on controlled human exposure
studies, associated with recent air quality levels and air quality levels just meeting the current
and potential alternative S02 standards, (4) to provide insights about whether or not there are
patterns of exposure in terms of differences in levels, averaging times, and/or air quality metrics
for the health effects based on epidemiological studies, (5) to provide distributions of population
health risk estimates for health endpoints based on community epidemiologic studies reporting
associations between respiratory effects at ambient S02 levels for recent air quality and air
quality levels just meeting potential alternative 1-hr, 24-hr, and annual SO2 standards if a Tier III
assessment is judged appropriate and conducted, and (6) to identify and discuss key assumptions,
degree of variability, and nature and extent of uncertainties in the estimates and to characterize
quantitatively, where feasible, the uncertainties and variability in such estimates.

Conceptually, if there were sufficient scientific data available, the objective of the health
risk assessment would be to develop population-based health risks for various health effect
endpoints in at-risk11 population groups associated with recent air quality levels and just meeting
the current and potential alternative SO2 NAAQS. In addition, the health risk assessment would
include a quantitative characterization of the uncertainties in those risk estimates and key
assumptions underlying such estimates. We recognize that the current state-of-knowledge about
S02-related health effects, as reflected in the evaluation contained in the first draft ISA, likely
precludes the development of quantitative health risk estimates for most health endpoints
discussed in the ISA. Our initial judgments about health effect categories and appropriate
approaches to conduct the assessments are presented below and are based on the current draft
ISA, recognizing that the 1st draft risk assessment will be informed by CASAC and public review
of the current draft of the ISA, in addition to the information and evaluation contained in the 2nd
draft ISA and associated Annexes.

11 At risk is used here to include both susceptible populations (i.e., those who are likely to be inherently more
sensitive) and vulnerable populations (i.e., those who are at greater risk due to increased exposure).

28


-------
4.2 TIER I: HEALTH EFFECTS EVALUATION

The Tier I assessment will essentially be a qualitative analysis of the health effects
information presented in the 1st and 2nd drafts of the ISA and associated Annexes. The first step
will be to identify key studies, both controlled human exposure and epidemiologic studies, that
are presented in the ISA and its Annexes as providing evidence for adverse health effects
associated with specific averaging times. This includes an analysis of studies that examined S02
exposure alone, as well as those that looked for SO2 associated health effects after adjustment for
co-pollutants (e.g., 03 or PM). Once potential health effect endpoints are identified, the Tier I
characterization will involve making judgments as to which of these endpoints are likely
candidates for progression to be addressed more quantitatively in Tier II and/or Tier III
assessments.

Staff has reviewed the information and evaluation of the health effects evidence
presented in the 1st draft of the ISA and we have some initial observations and judgments about
potential health effect categories and endpoints that should be considered for Tier II and/or Tier
III assessments. The 1st draft ISA identifies several health endpoints from controlled human
exposure and epidemiologic studies that are associated with ambient levels of SO2. The 1st draft
ISA notes adverse health effects associated with exposure to short-term peaks (5-10 minutes) of
SO2 based on findings from controlled human exposure studies, most of which were conducted
in the 1980's and were evaluated in previous criteria documents. Specifically, the health
endpoints identified include changes in respiratory function indicative of bronchoconstriction as
measured by specific airway resistance (SRaw) and decreased Forced Expiratory Volume in one
second (FEVi) observed in asthmatic subjects. The draft ISA also identifies asthmatics (children
and adults) as the population group most at-risk from respiratory-related effects associated with
these short-term peak SO2 exposures.

The 1st draft ISA also highlights recent epidemiologic studies that provide evidence for
an association between 24-hr and 3-hr average SO2 concentrations and increased respiratory
symptoms in children, particularly those with asthma or chronic respiratory symptoms.
Additionally, the ISA notes that the SO2 effect was generally found to be robust after adjusting
for particulate matter (PM) and other co-pollutants. Similarly, the ISA presents a large number
of epidemiologic studies that provide evidence of positive, but not always statistically
significant, associations between ambient 24-hr and 1-hr SO2 concentrations and ED visits and
hospitalizations for all respiratory causes and asthma, particularly among children and older
adults. These findings were generally robust when additional co-pollutants are included in the
model. Moreover, the 1st draft ISA notes biological plausibility for increased ED visits and
hospitalizations is found in both the and human clinical studies (mentioned above) and the
epidemiologic studies that observed increased respiratory symptoms and decreased lung
function, as well as the animal toxicological studies that observed SCVinduced altered lung host
defense.

In contrast, the 1st draft ISA concludes that the overall epidemiologic evidence is
inconclusive regarding the effect of short-term exposures (typically 24-hr) to SO2 on the
cardiovascular and nervous systems. The 1st draft ISA also concludes that the epidemiologic
evidence is suggestive of associations between short-term exposures to SO2 and non-accidental
and cardiopulmonary-related mortality, but notes the limited experimental evidence to support

29


-------
judgments about biological plausibility while raising concerns about confounding from other
pollutants, including PM and N02. The 1st draft ISA also concludes that the evidence is
inconclusive regarding the associations between long-term exposure to SO2 and morbidity and
mortality.

Based on the evaluation of the health effects evidence in the 1st draft ISA, the following
health effect endpoints are initially judged to be the most appropriate candidates to focus on in
Tier II and/or Tier III assessments:

•	Changes in respiratory function indicative of bronchoconstriction as measured by SRaw
and decreased FEVi observed in asthmatic subjects exposed to 5 to 10 minute S02
concentrations;

•	Respiratory symptoms in children, particularly those with asthma or chronic respiratory
symptoms associated with 3- and 24-hr average ambient S02 concentrations;

•	Respiratory-related emergency department visits, especially for asthmatic children
associated with 1- and 24-hr average ambient SO2 concentrations;

•	Respiratory-related hospital admissions, especially for asthmatics associated with 1- and
24-hr average ambient SO2 concentrations.

Given the conclusion in the 1st draft ISA that the evidence is inconclusive regarding the
associations between long-term exposure to SO2 and morbidity and mortality, we do not
anticipate focusing on these long-term exposure-related health effects in the Tier II and/or Tier
III assessments.

4.3 TIER II ASSESSMENT

As noted above, the type of health effects evidence providing the basis for concerns about
respiratory-related outcomes associated with very short-term (i.e., 5 to 10 minute exposures) are
fundamentally different from the type of health effects evidence that serves as the basis for
concerns about respiratory-related effects associated with exposures on the order of 1 to 24-
hours. More specifically, the evidence finding respiratory effects in asthmatics for very short-
term exposures involves controlled chamber studies of exercising subjects generally exposed to
SO2 alone. For this type of evidence, one can more clearly attribute observed effects as being
causally related to exposure to SO2 and these studies also provide direct relationships between
exposure to SO2 and effects. In contrast, for health effects based on findings from epidemiologic
studies, it is much more difficult to attribute the effects as being due to exposure to SO2 and to
sort out the relative contribution of S02 relative to the many other pollutants and possible
contributors in an ambient real world setting. In addition, the epidemiologic studies do not
provide a direct relationship between exposure and response, but rather provide relationships
between ambient concentrations, as measured at fixed-site monitors, and response, which we
refer to in this plan as concentration-response (C-R) relationships. Consequently, as described
below, the approach to Tier II assessments is fundamentally different depending on whether the
health effects identified as candidates for further quantitative treatment are based on evidence
from controlled human exposure or epidemiologic studies.

In addition to consideration of the type of health effect relationship provided, there is a
difference in the geographic unit of concern that also affects the approach taken in this risk

30


-------
assessment. If very short-term SO2 exposures are of concern only at levels exceeding 0.5 ppm
for 5 minute exposures, current air quality data from the fixed-site monitoring network suggest
that these levels are most likely to occur around point sources and will not be observed in the
routine urban area-oriented S02 monitoring network. Thus, concerns about very short-term peak
exposures require a source-oriented focus, in contrast to concerns about 1- to 24-hr averaging
time exposures to S02 which require a more urban area-oriented focus.

4.3.1 Approaches

Two different approaches to the Tier II assessment, depending on whether the evidence
for a given health effect is based on controlled human exposure studies or epidemiologic studies,
are described in this section. As noted above, the two different approaches also reflect a
different focus in terms of geographic scope.

For respiratory health effects observed in controlled human exposure studies of
excerising asthmatics exposed for very short durations (i.e., 5 to 10 minutes) to SO2, the
approach is similar to calculating a hazard quotient, which is the ratio of the air quality
concentration or exposure concentration (either population-weighted or individual exposure
depending on the Tier exposure assessment output) to the potential health effect benchmark
concentration. Counts would be obtained for the number of times the various potential health
effect benchmarks are exceeded. The estimation of short-term peak concentrations exceedances
at all source- and population-oriented hourly ambient monitors generated from the air quality
characterization provides a broad context for potential populations at risk of short-term peak
exposures including both urban and non-urban locations (Section 3.2). A refined source-oriented
exposure analysis, designed to address the occurrence of short-term peak SO2 levels in the
vicinity of major point sources, is also on a national scale, however would not be considered an
urban-oriented analysis.

Based on our initial evaluation of the controlled human exposure studies in the draft ISA,
we have identified bronchoconstriction in exercising asthmatics as a candidate health effect
category for Tier II assessment. Moreover, we have tentatively identified potential health effect
benchmarks in the range of 0.5 to 0.6 ppm (5-minute averaging time), and have identified
asthmatics (children and adults) as the population group most at-risk from these potential SO2
benchmark levels.

As noted in the section on Tier I above, several respiratory health effect categories have
been identified as being the most strongly supported effects associated with ambient 1-hr and 24-
hr SO2 concentrations. These health effect categories include: respiratory symptoms,
respiratory-related emergency department visits (particularly for asthmatics), and respiratory-
related hospital admissions. Since epidemiologic studies generally consider the entire
distribution of pollutant levels and report effect estimates in terms of a given response per unit
change in pollution, one cannot develop potential health effect benchmarks directly from this
type of study. Thus, a Tier II analysis for these health endpoints will involve a more extensive
evaluation of the ambient air quality levels for SO2 and co-pollutants, where possible, to see if
there are any trends or patterns in the reported concentration-response relationships. Similar to
the approach taken in the recent O3 and PM NAAQS reviews, EPA will gather additional
information to characterize the SO2 ambient air quality that existed at the time the various key

31


-------
U.S. and Canadian studies were conducted. The goal is to see if there are any patterns with
respect to the levels and associated averaging times and the reported effect estimates across the
studies reporting a given health effect. We will also examine the studies addressing these
specific health effects to see if there are any differences with respect to geographic location
and/or season. In addition, another goal is to evaluate whether there are any patterns in the
reported C-R relationships with respect to the inclusion of various co-pollutants in the effect
estimate models.

For the respiratory health effects associated with 1- to 24-hr averaging times that are
based on the epidemiologic studies conducted in urban areas, the geographic focus of the
assessment would use urban area S02 concentrations. As part of the air quality characterization,
1-hr ambient monitoring concentrations will be evaluated focusing on urban areas and
considering several averaging times (e.g., 1-hr, 24-hr, and annual average concentrations).

4.3.2	Generated Outcomes

For the respiratory-related effects based on evidence from controlled human exposure
studies for very short-term SO2 exposures, outcomes would be the number of occurrences that air
quality exceeds a potential health effects benchmark, as well as the number of times a population
or an individual experiences an exposure of concern (EOC) in a given year, considering recent
air quality levels and air quality levels just meeting the current and potential alternative SO2
standards that may be considered. Frequencies would be given for each population subgroup
analyzed and the particular locations of interest.

The outcome of the Tier II assessment for the respiratory-related effects based on
evidence from epidemiologic studies providing effect estimates for 1- to 24-hr averaging time
S02 levels, will be more qualitative. We anticipate the presentation of this part of the
assessment will include tables and/or graphs illustrating the existence or absence of any patterns
found between concentrations at which various studies were conducted and health effect
estimates.

4.3.3	Variability and Uncertainty

Variability in the context of the Tier II risk assessment can be described in terms of the
empirical quantities and relationships that are important in estimating health risks and are
inherently variable across time and space, or when considering a group of individuals (Cullen
and Frey, 1999). For the initial Tier II screening level assessment that estimates the number of
exceedances of alternative potential health effect benchmarks across all areas selected for the
assessment, results for the individual locations incorporate and illustrate the variability due to
differences in air quality patterns and distributions. A second phase of the Tier II level risk
assessment would use the results of the Tier II exposure assessment to generate estimates of the
number of people exposed to levels at or above the various potential health effect benchmark
levels. Results for all areas included in this assessment would incorporate and reflect the
variability in air quality and the variability in key inputs that may influence the estimation of
population exposures including, but not limited to, the spatial pattern of the population, time-
location-activity patterns, and proximity to local sources.

32


-------
Consistent with the approach described above, a tiered approach to assessing uncertainty
will be employed with the goal of progressing to a quantitative analysis if warranted and if data
are available to support such an analysis. The first step in the uncertainty analysis would be to
identify the components of the assessment that contribute the most to uncertainty. Sections 3.2.3
and 3.3.6 provide a preliminary qualitative evaluation for the uncertain components of the
planned air quality analysis and exposure assessment, generally indicating the direction of
influence (under- or over-estimate) on air quality concentration and exposure estimates that
would be used in a Tier II health risk assessment.

In addition to uncertainties related to the air quality analysis and/or exposure assessment
components of a Tier II risk assessment, there is uncertainty related to the potential health effect
benchmark levels used in the assessment. The use of any specific potential health effect
benchmark assumes that the level is appropriate for application to all susceptible individuals
equally, between and within each population subgroup. Recognizing that there is both
considerable variability in responsiveness and uncertainty associated with the use of any single
potential health effect benchmark, a range of potential health effect benchmarks will be included
in the Tier II assessments. This will allow the decision maker to gain insight into the impact that
uncertainty about the level at which adverse health effects are likely to occur has on the Tier I
estimates. From a directional perspective, we have confidence that higher potential health effect
benchmarks are associated with susceptible individuals being adversely affected and that a larger
fraction of the population is likely to experience adverse health effects. Conversely, we have
less confidence that adverse health effects will occur at lower benchmark levels and a smaller
fraction of the population is likely to experience adverse health effects.

4.4 TIER III ASSESMENT

As noted above, based on review of the scientific evidence from controlled human
exposure studies regarding respiratory health effects associated with very short averaging times
(5-10 minutes), we believe that there is insufficient information to develop credible exposure-
response relationships for SCVrelated respiratory health effects for use in a quantitative risk
assessment. Thus, the discussion below focuses on a possible Tier III assessment which would
focus on respiratory health effects associated with 1- and 24-hr ambient SO2 concentrations in
urban areas based on evidence from epidemiologic studies.

As discussed above, based on evidence from epidemiologic studies, health responses
most strongly related to SO2 include respiratory symptoms in asthmatic children, asthma
emergency department visits, and respiratory related hospital admissions. A risk assessment
based on epidemiologic studies typically requires baseline incidence rates for the specific health
endpoints to be analyzed and population data for the specific risk assessment locations.

4.4.1 Approach

As noted earlier in this plan, previous reviews of the SO2 primary NAAQS completed in
1982, 1986, and 1994 did not include quantitative health risk assessments. Thus, the planned
risk assessment described in this Scope and Methods Plan builds upon the methodology and
lessons learned from the risk assessment work conducted for the recent PM and current O3
NAAQS reviews (Abt Associates, 2005; Abt Associates, 2007). Many of the same

33


-------
methodological issues are present for each of these criteria air pollutants where epidemiologic
studies provided the basis for the C-R relationships used in the quantitative risk assessment. The
plans discussed below are based on the information and evaluation contained in the 1st draft ISA
and some aspects of these plans may change based on CAS AC and public comments on the 1st
draft ISA and changes that will be incorporated in the 2nd draft ISA. The discussion below
represents current staff thinking with respect to health effect endpoints that are candidates for
inclusion in a possibleTier III risk assessment and those health endpoints for which there is
insufficient evidence to warrant inclusion in a Tier III quantitative risk assessment.

4.4.1.1	Selection of Health Effect Endpoints

In selecting potential health endpoints to include in a Tier III risk assessment, staff plans
to focus on health endpoints identified during the Tier I and II risk assessment that have well-
defined health consequences (i.e., where there is consensus about the degree of response that
represents an adverse health effect). In addition, we are focusing on health endpoint categories
identified in the ISA where the weight of evidence supports the inference of a likely causal
relationship. As discussed below, once we identify candidate health endpoints based on these
criteria, there are additional factors that must be considered in deciding whether to proceed with
a quantitative Tier III risk assessment. These include: (1) the likely utility of such information in
the decision, (2) the availability of sufficient concentration-response data that is relevant to
locations in the U.S., and (3) the availability of baseline incidence data for the health effects.

Based on the evaluation of the health effects evidence in the 1st draft ISA, the following
health effect endpoints are judged to be the most appropriate candidates for developing
quantitative risk estimates:

•	Respiratory symptoms (e.g., cough, wheeze), particularly in children and asthmatics

•	Respiratory-related hospital admissions, especially for asthmatics

•	Respiratory-related emergency department visits, especially for asthmatics

Generally, for a Tier III quantitative risk assessment based on C-R relationships derived
from epidemiological studies, it is preferable to use C-R relationships based on studies that were
conducted in the same location chosen for the risk assessment. Using C-R relationships from
studies conducted in locations different than the risk assessment locations introduces additional
uncertainty into the risk assessment due to potential differences in population, S02 and co-
pollutant air quality patterns, exposure patterns, and other factors that may have influenced the
relationship between exposure to the pollutant of interest and the health effect outcome.
Following review of the 1st draft ISA and considering any comments and recommendations by
CASAC and the public, we plan to evaluate whether the existing epidemiological studies provide
C-R relationships that are judged suitable for applying in selected U.S. urban locations.

4.4.1.2	Selection of Concentration-Response Functions

If a Tier III risk assessment is judged to be both feasible and of sufficient utility, then
appropriate C-R relationships will have to be selected for inclusion in the assessment. Studies
often report more than one estimated C-R function for the same location and health endpoint.
Sometimes models include different sets of co-pollutants and/or different time lags. For some

34


-------
health endpoints, there are studies that estimated multi-city SO2 C-R functions, while other
studies estimated single-city functions.

As noted above, all else being equal, staff judges that a C-R function estimated in the
assessment location is preferable to a function estimated in some other location, to avoid any
uncertainties that may exist due to differences associated with geographic location. There are
several advantages, however, to using estimates from multi-city studies versus from studies
carried out in single cities. Multi-city studies are applicable to a variety of settings, since they
estimate a central tendency across multiple locations. Multi-city studies also tend to have more
statistical power and provide effect estimates with relatively greater precision than single-city
studies due to larger sample sizes, reducing the uncertainty around the estimated health
coefficient. Because single-city and multi-city studies have different advantages, staff plans to
include both types of functions, where they are available.

Most SO2 epidemiological studies include C-R functions in which SO2 was the only
pollutant entered in the model as well as other C-R functions in which S02 and one or more co-
pollutants (e.g., PM, NO2, CO, O3) were entered into the health effects model (i.e., multi-
pollutant models). To the extent that any of the co-pollutants present in the ambient air may
have contributed to the health effects attributed to SO2 in single pollutant models, risks attributed
to S02 might be overestimated where C-R functions are based on single pollutant models.
However, if co-pollutants are highly correlated with SO2, their inclusion in an SO2 model can
lead to misleading conclusions in identifying a specific causal pollutant. When collinearity
exists, inclusion of multiple pollutants in models often produces unstable and statistically
insignificant effect estimates for both S02 and the co-pollutants. Given that single and multi-
pollutant models each have both potential advantages and disadvantages, with neither type
clearly preferable over the other in all cases, if a Tier III risk assessment is developed, staff plans
to report risk estimates based on both types of models where both are available.

4.4.1.3 Baseline Health Effects Incidence Considerations

The most common epidemiological-based health risk model expresses the reductions in
health risk (Ay) associated with a given reduction in S02 concentrations (Ax) as a percentage of
the baseline incidence (y). Thus, information on the baseline incidence of health effects (i.e., the
incidence under as is air quality conditions) in each location is needed. Where at all possible,
staff plans to use county-specific incidences or incidence rates (in combination with county-
specific population data). Staff is investigating whether recent baseline incidence data is
available for respiratory-related emergency department visits and respiratory-related hospital
admissions for potential assessment locations.

For respiratory symptoms, there may be no information on baseline incidence other than
that reported in the original epidemiological study. We recognize that lack of recent location-
specific incidence data will increase the uncertainty surrounding any risk estimates that may be
generated in a Tier III risk assessment for this health endpoint.

4.4.2 Generated Outcomes

If a Tier III risk assessment were to be developed, both central tendency and 95%
confidence interval estimates would be provided and such estimates would be expressed using

35


-------
several risk metrics. These risk metrics would include the estimated incidence (i.e., number of
cases), percent of total incidence, and incidence per 100,000 relevant population for each health
endpoint and location included in the assessment. Results would also be presented for just
meeting any potential alternative standards identified for consideration.

4.4.3 Variability and Uncertainty

There are several uncertainties that affect the inputs to any Tier III S02 risk assessment
based on C-R functions derived from epidemiological studies. These include uncertainties in the
procedures used to simulate just meeting the current and potential alternative S02 standards,
baseline incidence rates, and appropriate model form for the C-R relationships used in a risk
assessment. There also is city-to-city variability in C-R relationships due to variability in air
quality and exposure patterns and population differences. Presentation of separate risk results for
selected example urban areas would incorporate and reflect variability in several key inputs to
the health risk assessment (e.g., variability in air quality patterns and baseline incidence data).

Consistent with the approach used in the recent O3 and PM NAAQS risk assessments, the
uncertainty resulting from the statistical uncertainty associated with the estimate of the SO2
health coefficient in the C-R function can be characterized by confidence intervals around the
corresponding point estimates of risk. However, these confidence intervals only address
sampling error and do not address broader uncertainties concerning the overall shape or form of
the C-R relationships. As noted above, if a Tier III assessment is conducted, staff plans to
include results using both single- and multi-city models, and single- and multi-pollutant models
and C-R functions based on different epidemiological studies. Presentation of a range of results
would provide decision makers with some perspective on the impact of alternative models and
the degree of uncertainty associated with any risk estimates.

4.4 CRITERIA FOR DETERMINING APPROACH

The factors identified below will be considered in deciding whether to conduct a Tier III
quantitative risk assessment.

•	Outcome of the Tier I and Tier II risk assessments with respect to the magnitude of the
estimated number of concentrations and/or exposures exceeding several potential health
effect benchmark levels associated with current ambient concentrations and with SO2
levels just meeting the current and any potential alternative standards that may be
considered and kind and extent of the uncertainties associated with these estimates;

•	Availability of information and data required to conduct a Tier III risk assessment,
including baseline incidence data and concentration-response relationships that are
judged suitable for applying in several example U.S. urban areas;

•	The utility or value-added to the decision process of a Tier III risk assessment, beyond
that provided by the Tier I and II assessments. For example, is a Tier III risk assessment
likely to reduce or better characterize uncertainties in the characterization of S02-related
health risks;

•	The feasibility of conducting a credible Tier III risk assessment within the consent decree
schedule and available resources.

36


-------
4.5 BROADER HEALTH RISK CHARACTERIZATION

The exposure/health risk assessment document will include both summary air quality
information for the U.S. and summary information and discussion of the various health effects
identified in the 2nd draft ISA to help provide a broad context for the quantitative exposure and
risk estimates that are provided in the Tier II and/or Tier III exposure and risk assessments.
Thus, air quality statistics for all areas with S02 monitoring data will be presented to provide a
broad perspective of potential populations at risk, considering several potential health effect
benchmark levels for short-term peak concentrations. National scale information on the size of
various at-risk populations will also be presented for where short-term peak
concentrations/exposures may be of greatest concern, namely those individuals residing in close
proximity to fossil-fueled emission sources.

37


-------
5. SCHEDULE AND MILESTONES

Table 6 lists the key milestones for the risk/exposure assessment that will be conducted as
part of the current SO2 NAAQS review. Consultation with the CASAC NOx/SOx Panel is
planned for December 5-6, 2007 to obtain input on the 1st draft ISA and this draft Scope and
Methods Plan. Staff will then proceed to develop exposure and health risk estimates associated
with recent SO2 ambient concentrations and levels representing just meeting the current SO2
standard. These estimates and the methodology used will be presented in the first draft SO2
risk/exposure assessment and technical support documents. The draft report will be released for
CASAC and public review in May 2008. EPA will receive comments on these draft documents
from the CASAC NOx/SOx Panel and general public at a meeting in July 2008. A revised
assessment, including assessment for just meeting potential alternative standards will be released
in October 2008 for review by CASAC and public at a meeting to be held in December 2008.
Staff will consider these review comments and prepare a final risk/exposure assessment by
January 2009.

Table 2. Key Milestones for the Exposure and Health Risk Assessment for the SP2 NAAQS Review.

Milestone

Date

Release 1st draft SOx ISA

September 2007

Release 1st draft S02 Risk/Exposure Scope and Methods Plan

November 2007

CASAC/public review and meeting on 1st draft SOx ISA

December 5-6, 2007

CASAC consultation on 1st draft S02 Risk/Exposure Scope and Methods Plan

December 5-6, 2007

Release 2nd draft SOx ISA

April 2008

Release 1st draft of the S02 Risk/Exposure Assessment

May 2008

CASAC/public review and meeting on 2nd draft SOx ISA and 1st draft of the
Risk/Exposure Assessment

July 2008

Final SOx ISA

September 2008

Release 2nd draft of the S02 Risk/Exposure Assessment

October 2008

CASAC/public review and meeting on 2nd draft of the S02 Risk/Exposure
Assessment

December 2008

Final S02 Risk/Exposure Assessment

January 2009

38


-------
6. REFERENCES

Abt Associates Inc. (2005). Particulate Matter Health Risk Assessment for Selected Urban Areas. Prepared for
Office of Air Quality Planning and Standards, U.S. Environmental Protection Agency, Research Triangle Park,
NC. June 2005. Available at: http://www.epa.gOv/ttn/naaas/standards/pm/s pm pr td.html.

Abt Associates Inc. (2007). Ozone Health Risk Assessment for Selected Urban Areas. Prepared for Office of Air
Quality Planning and Standards, U.S. Environmental Protection Agency, Research Triangle Park, NC. July
2007. Available at: http://www.epa.gov/ttn/naaqs/standards/o3/s_o3_cr_td.html.

Ainsworth BE, Bassett Jr DR, Strath SJ, Swartz AM, O'Brien WL, et al. (2000). Comparison of three methods for
measuring the time spent in physical activity. Med Sci Sports Exer. 32(Suppl):S498-S516.

Burton CS, Stockenius TE, Stocking TS, Carr EL, Austin BS, Roberson RL (1987). Assessment of Exposures of
Exercising Asthmatics to Short-term S02 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.

Cullen AC and Frey HC (1999). Probabilistic Techniques in Exposure Assessment. A handbook for dealing with
variability and uncertainty in models and inputs. New York, NY. Plenum Press.

Graham SE and McCurdy T (2004). Developing meaningful cohorts for human exposure models. J Expos Anal
Environ Epidemiol. 14(1)23-43.

Langstaff JE (2007). OAQPS Staff Memorandum to Ozone NAAQS Review Docket (OAR-2005-0172). Subject:
Analysis of Uncertainty in Ozone Population Exposure Modeling. [January 31, 2007], Available at:
http://www.epa.gOv/ttn/naaas/standards/ozone/s o3 cr td.html.

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.

Morgan, GM and Henrion M (1990). Uncertainty: A guide to dealing with uncertainty on quantitative risk and
policy analysis. New York, NY: Cambridge University Press.

SAS (2007). SAS/STAT Software (Version 9.1.3). SAS Institute, Cary, NC.

Rosenbaum AS, Hudischewskyj AB, Roberson RL, Burton CS (1992). Estimates of Future Exposures of Exercising
Asthmatics to Short-term Elevated S02 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 S02 Concentration Data. Systems Applications
International, Research Triangle Park, NC. Final Report. May 8, 1996.

Science 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.

Thompson R (2000). Preliminary Analysis of 5-Minute Maximum Ambient S02 Concentrations. OAQPS.

12/21/00. Office of Air Quality Planning and Standards, U.S. Environmental Protection Agency, Research
Triangle Park, NC. Available at: http://www.epa.gov/ttn/amtic/so2data.html.

Thrall AD, Langstaff J, Liu MK, Burton CS (1982). On the Variability in Peak Sulfur Dioxide Concentrations
Contained in Longer-Term Averages: An Empirical Study of an Isolated Power Plant. Systems Appplications,
Inc., San Rafael, CA.

US EPA (1982a). Review of the National Ambient Air Quality Standards for Sulfur Oxides: Assessment of
Scientific and Technical Information. EPA-450/5-82-007.

US EPA (1982b). Air Quality Criteria for Particulate Matter and Sulfur Oxides (1982) Final Report. EPA/600/8-
82/029.

US EPA (1986a). 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. EPA-450/05-86-013.

US EPA (1986b). An Analysis of Short-term Sulfur Dioxide Population Exposures in the Vicinity of Utility Power
Plants. U.S. Environmental Protection Agency, Research Triangle Park, NC. September 12, 1986.

US EPA (1994a). Review of the National Ambient Air Quality Standards for Sulfur Oxides: Assessment of
Scientific and Technical Information. Supplement to the 1986 OAQPS Staff Paper Addendum. EPA-452/R-
94-013. September 1994.

39


-------
US EPA (1994b). Supplement to the Second Addendum (1986) to Air Quality Criteria for Particulate Matter and
Sulfur Oxides (1982): Assessment of New Findings on Sulfur Dioxide and Acute Exposure Health Effects in
Asthmatic Individuals. EPA/600/FP-93/002.

US EPA (2004). AERMOD: Description of Model Formulation. Office of Air Quality Planning and Standards.
EPA-454/R-03-004. Available at: http://www.epa.gov/scram001/7thconf/aermod/aermod mfd.pdf.

US EPA (2006a). Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model Documentation
(TRIM.Expo / APEX, Version 4) Volume I: User's Guide. Office of Air Quality Planning and Standards, U.S.
Environmental Protection Agency, Research Triangle Park, NC. June 2006. Available at:
http://www.epa.gov/ttn/fera/human apex.html.

US EPA (2006b). Total Risk Integrated Methodology (TRIM) - Air Pollutants Exposure Model Documentation
(TRIM.Expo / APEX, Version 4) Volume II: Technical Support Document. Office of Air Quality Planning and
Standards, U.S. Environmental Protection Agency, Research Triangle Park, NC. June 2006. Available at:
http://www.epa.gov/ttn/fera/human apex.html.

US EPA (2007a). Plan for Review of the Primary National Ambient Air Quality Standard for Sulfur Dioxide.
Available at: http://www.epa.gOv/ttn/naaas/standards/so2/s so2 cr pd.html.

US EPA (2007b). Integrated Science Assessment for Oxides of Sulfur - Health Criteria (First External Review
Draft) and Annexes. National Center for Environmental Assessment. September 2007. Available at:
http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=182057.

US EPA (2007c). Air Trends. Sulfur Dioxide, http://www.epa.gov/air/airtrends/sulfur.html.

US EPA (2007d). Review of the National Ambient Air Quality Standards for ozone: assessment of scientific and
technical information. OAQPS Staff paper (June 2007). Research Triangle Park, NC: Office of Air Quality
Planning and Standards. EPA-452/R-07-003. Available at:
http://epa.gOv/ttn/naaas/standards/ozone/s o3 cr sp.html.

40


-------