A r nA United SlatB5
^•5# ^ Environmental Protection
I— I Agency
Bayesian space-time downscaling fusion model
(downscaler) -Derived Estimates of Air Quality
for 2009
-------
EPA-454/R-13-003
November 2013
Bayesian space-time downscaling fusion model (downscaler) -Derived
Estimates of Air Quality for 2009
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Air Quality Assessment Division
Research Triangle Park, NC
-------
Contributors
Ellen Baldridge (EPA/OAR)
Halil Cakir (EPA/OAR)
Alison Eyth (EPA/OAR)
Dave Holland (EPA/ORD)
David Mintz (EPA/OAR)
Sharon Phillips (EPA/OAR)
Adam Reff (EPA/OAR)
Acknowledgements
The following people served as reviewers of this document and provided valuable comments
Jan Cortelyou (EPA/OAR)
Dennis Doll (EPA/OAR)
Tyler Fox (EPA/OAR)
Neil Frank (EPA/OAR)
James Hemby (EPA/OAR)
Marc Houyoux (EPA/OAR),
Dr. Bryan Hubbell (EPA/OAR)
ii
-------
Contents
Contents 1
1.0 Introduction 2
2.0 Air Quality Data 5
2.1 Introduction to Air Quality Impacts in the United States 5
2.2 Ambient Air Quality Monitoring in the United States 7
2.3 Air Quality Indicators Developed for the EPHT Network 11
3.0 Emissions Data 13
3.1 Introduction to Emissions Data Development 13
3.2 2009 Emission Inventories and Approaches 13
3.3 Emissions Modeling Summary 34
4.0 CMAQ Air Quality Model Estimates 55
4.1 Introduction to the CMAQ Modeling Platform 55
4.2 CMAQ Model Version, Inputs and Configuration 57
4.3 CMAQ Model Performance Evaluation 62
5.0 Bayesian space-time downscaling fusion model (downscaler) -Derived Air Quality
Estimates 77
5.1 Introduction 77
5.2 Downscaler Model 77
5.3 Downscaler Output 78
5.4 Overview of Downscaler Model Results for 2009 79
5.5 Accuracy Assessment of Downscaler Model Results 105
5.6 Use of EPA Downscaler Model Predictions 116
Appendix A - Acronyms 117
1
-------
1.0 Intro due
This report describes estimates of daily ozone (maximum 8-hour average) and PM2.5 (24-hour
average) concentrations throughout the contiguous United States during the 2009 calendar
year generated by EPA's recently developed data fusion method termed the "downscaler
model" (DS). Air quality monitoring data from the National Air Monitoring Stations/State and
Local Air Monitoring Stations (NAMS/SLAMS) and numerical output from the Community
Multiscale Air Quality (CMAQ) model were both input to DS to predict concentrations at the
2010 US census tract centroids encompassed by the CMAQ modeling domain. Information on
EPA's air quality monitors, CMAQ model, and downscaler model is included to provide the
background and context for understanding the data output presented in this report. These
estimates are intended for use by statisticians and environmental scientists interested in the
daily spatial distribution of ozone and PM2.5.
DS essentially operates by calibrating CMAQ data to the observational data, and then uses the
resulting relationship to predict "observed" concentrations at new spatial points in the domain.
Although similar in principle to a linear regression, spatial modeling aspects have been
incorporated for improving the model fit, and a Bayesian1 approaching to fitting is used to
generate an uncertainty value associated with each concentration prediction. The uncertainties
that DS produces are a major distinguishing feature from earlier fusion methods previously
used by EPA such as the "Hierarchical Bayesian" (HB) model (McMillan et al, 2009). The
term "downscaler" refers to the fact that DS takes grid-averaged data (CMAQ) for input and
produces point-based estimates, thus "scaling down" the area of data representation. Although
this allows air pollution concentration estimates to be made at points where no observations
exist, caution is needed when interpreting any within-gridcell spatial gradients generated by
DS since they may not exist in the input datasets. The theory, development, and initial
evaluation of DS can be found in the earlier papers of Berrocal, Gelfand, and Holland (2009,
2010, and 2011).
The data contained in this report are an outgrowth of a collaborative research partnership
between EPA scientists from the Office of Research and Development's (ORD) National
Exposure Research Laboratory (NERL) and personnel from EPA's Office of Air and
Radiation's (OAR) Office of Air Quality Planning and Standards (OAQPS). NERL's Human
Exposure and Atmospheric Sciences Division (HEASD), Atmospheric Modeling Division
(AMD), and Environmental Sciences Division (ESD), in conjunction with OAQPS, work
together to provide air quality monitoring data and model estimates to the Centers for Disease
Control and Prevention (CDC) for use in their Environmental Public Health Tracking (EPHT)
Network.
1 Bayesian statistical modeling refers to methods that are based on Bayes' theorem, and model the world in terms
of probabilities based on previously acquired knowledge.
2
-------
CDC's EPHT Network supports linkage of air quality data with human health outcome data
for use by various public health agencies throughout the U.S. The EPHT Network Program is
a multidisciplinary collaboration that involves the ongoing collection, integration, analysis,
interpretation, and dissemination of data from: environmental hazard monitoring activities;
human exposure assessment information; and surveillance of noninfectious health conditions.
As part of the National EPHT Program efforts, the CDC led the initiative to build the National
EPHT Network (http:// www.cdc.gov/nceh/tracking/default.htm). The National EPHT
Program, with the EPHT Network as its cornerstone, is the CDC's response to requests calling
for improved understanding of how the environment affects human health. The EPHT
Network is designed to provide the means to identify, access, and organize hazard, exposure,
and health data from a variety of sources and to examine, analyze and interpret those data
based on their spatial and temporal characteristics.
Since 2002, EPA has collaborated with the CDC on the development of the EPHT Network.
On September 30, 2003, the Secretary of Health and Human Services (HHS) and the
Administrator of EPA signed a joint Memorandum of Understanding (MOU) with the
objective of advancing efforts to achieve mutual environmental public health goals2. HHS,
acting through the CDC and the Agency for Toxic Substances and Disease Registry
(ATSDR), and EPA agreed to expand their cooperative activities in support of the CDC
EPHT Network and EPA's Central Data Exchange Node on the Environmental Information
Exchange Network in the following areas:
• Collecting, analyzing and interpreting environmental and health data from both
agencies (HHS and EPA).
• Collaborating on emerging information technology practices related to building,
supporting, and operating the CDC EPHT Network and the Environmental
Information Exchange Network.
• Developing and validating additional environmental public health indicators.
• Sharing reliable environmental and public health data between their respective
networks in an efficient and effective manner.
• Consulting and informing each other about dissemination of results obtained through
work carried out under the MOU and the associated Interagency Agreement (IAG)
between EPA and CDC.
2 HHS and EPA agreed to extend the duration of the MOU, effective since 2002 and renewed in 2007, until June 29,
2017. The MOU is available at www.cdc.gov/nceh/tracking/partners/epa mou 2007.htm.
3
-------
The best available statistical fusion model, air quality data, and CMAQ numerical model
output were used to develop the 2009 estimates. Fusion results can vary with different inputs
and fusion modeling approaches. As new and improved statistical models become available,
EPA will provide updates.
Although these data have been processed on a computer system at the Environmental Protection
Agency, no warranty expressed or implied is made regarding the accuracy or utility of the data on
any other system or for general or scientific purposes, nor shall the act of distribution of the data
constitute any such warranty. It is also strongly recommended that careful attention be paid to the
contents of the metadata file associated with these data to evaluate data set limitations, restrictions
or intended use. The U.S. Environmental Protection Agency shall not be held liable for improper
or incorrect use of the data described and/or contained herein.
The four remaining sections and one appendix in the report are as follows.
• Section 2 describes the air quality data obtained from EPA's nationwide monitoring
network and the importance of the monitoring data in determining health potential
health risks.
• Section 3 details the emissions inventory data, how it is obtained and its role as a key
input into the CMAQ air quality computer model.
• Section 4 describes the CMAQ computer model and its role in providing estimates of
pollutant concentrations across the U.S. based on 12-km grid cells over the contiguous
U.S.
• Section 5 explains the downscaler model used to statistically combine air quality
monitoring data and air quality estimates from the CMAQ model to provide daily air
quality estimates for the 2010 US census tract centroid locations within the contiguous
U.S.
• The appendix provides a description of acronyms used in this report.
4
-------
lallty Data
To compare health outcomes with air quality measures, it is important to understand the origins
of those measures and the methods for obtaining them. This section provides a brief overview of
the origins and process of air quality regulation in this country. It provides a detailed discussion
of ozone (03) and particulate matter (PM). The EPHT program has focused on these two
pollutants, since numerous studies have found them to be most pervasive and harmful to public
health and the environment, and there are extensive monitoring and modeling data available.
2.1 Introduction to Air Quality Impacts in the United States
2.1.1 The Clean Air Act
In 1970, the Clean Air Act (CAA) was signed into law. Under this law, EPA sets limits on how
much of a pollutant can be in the air anywhere in the United States. This ensures that all
Americans have the same basic health and environmental protections. The CAA has been
amended several times to keep pace with new information. For more information on the CAA,
go to http://www.epa.gov/oar/caa/.
Under the CAA, the U.S. EPA has established standards or limits for six air pollutants, known as
the criteria air pollutants: carbon monoxide (CO), lead (Pb), nitrogen dioxide (NO2), sulfur
dioxide (S02), ozone (03), and particulate matter (PM). These standards, called the National
Ambient Air Quality Standards (NAAQS), are designed to protect public health and the
environment. The CAA established two types of air quality standards. Primary standards set
limits to protect public health, including the health of "sensitive" populations such as asthmatics,
children, and the elderly. Secondary standards set limits to protect public welfare, including
protection against decreased visibility, damage to animals, crops, vegetation, and buildings. The
law requires EPA to review periodically these standards. For more specific information on the
NAAQS, go to www.epa.gov/air/criteria.html. For general information on the criteria pollutants,
go to http://www.epa.gov/air/urbanair/6poll.html.
When these standards are not met, the area is designated as a nonattainment area. States must
develop state implementation plans (SIPs) that explain the regulations and controls it will use to
clean up the nonattainment areas. States with an EPA-approved SIP can request that the area be
designated from nonattainment to attainment by providing three consecutive years of data
showing NAAQS compliance. The state must also provide a maintenance plan to demonstrate
how it will continue to comply with the NAAQS and demonstrate compliance over a 10-year
period, and what corrective actions it will take should a NAAQS violation occur after
designation. EPA must review and approve the NAAQS compliance data and the maintenance
plan before designating the area; thus, a person may live in an area designated as non- attainment
even though no NAAQS violation has been observed for quite some time. For more information
on designations, go to http://www.epa.gov/ozonedesignations/ and
http://vvvvvv.epa.gov/pmdesignations.
5
-------
2.1.2 Ozone
Ozone is a colorless gas composed of three oxygen atoms. Ground level ozone is formed when
pollutants released from cars, power plants, and other sources react in the presence of heat and
sunlight. It is the prime ingredient of what is commonly called "smog." When inhaled, ozone can
cause acute respiratory problems, aggravate asthma, cause inflammation of lung tissue, and even
temporarily decrease the lung capacity of healthy adults. Repeated exposure may permanently scar
lung tissue. Toxicological, human exposure, and epidemiological studies were integrated by EPA
in "Air Quality Criteria for Ozone and Related Photochemical Oxidants." It is available at
http://www.epa.g0v/ttn/naaqs/standards/0z0ne/s o3 index.html. The current (as of October 2008)
NAAQS for ozone is a daily maximum 8-hour average of 0.075 parts per million [ppm] (for details,
see http://www.epa.gov/ozonedesignations/). The Clean Air Act requires EPA to review the
NAAQS at least every five years and revise them as appropriate in accordance with Section 108
and Section 109 of the Act.
2.1.3 Particulate Matter
PM air pollution is a complex mixture of small and large particles of varying origin that can
contain hundreds of different chemicals, including cancer-causing agents like polycyclic aromatic
hydrocarbons (PAH), as well as heavy metals such as arsenic and cadmium. PM air pollution
results from direct emissions of particles as well as particles formed through chemical
transformations of gaseous air pollutants. The characteristics, sources, and potential health effects
of particulate matter depend on its source, the season, and atmospheric conditions.
As practical convention, PM is divided by sizes into classes with differing health concerns and
potential sources32. Particles less than 10 micrometers in diameter (PMio) pose a health concern
because they can be inhaled into and accumulate in the respiratory system. Particles less than 2.5
micrometers in diameter (PM2.5) are referred to as "fine" particles. Because of their small size, fine
particles can lodge deeply into the lungs. Sources of fine particles include all types of combustion
(motor vehicles, power plants, wood burning, etc.) and some industrial processes. Particles with
diameters between 2.5 and 10 micrometers (PM10-2.5) are referred to as "coarse" or PMc. Sources
of PMc include crushing or grinding operations and dust from paved or unpaved roads. The
distribution of PMi0, PM2.5 and PMc varies from the Eastern U.S. to arid western areas.
Particle pollution - especially fine particles - contains microscopic solids and liquid droplets that
are so small that they can get deep into the lungs and cause serious health problems. Numerous
scientific studies have linked particle pollution exposure to a variety of problems, including
premature death in people with heart or lung disease, nonfatal heart attacks, irregular heartbeat,
aggravated asthma, decreased lung function, and increase respiratory symptoms, such as irritation
of airways, coughing or difficulty breathing. Additional information on the health effects of
particle pollution and other technical documents related to PM standards are available at
http://www.epa.gOv/ttn/naaqs/standards/pm/s pm index.html.
3 The measure used to classify PM into sizes is the aerodynamic diameter. The measurement instruments used for PM
are designed and operated to separate large particles from the smaller particles. For example, the PM2 5 instrument only
captures and thus measures particles with an aerodynamic diameter less than 2.5 micrometers. The EPA method to
measure PMc is designed around taking the mathematical difference between measurements for PMi0 and PM2 5
6
-------
The current NAAQS for PM2.5 includes both a 24-hour standard to protect against short-term effects, and
an annual standard to protect against long-term effects. The annual average PM2.5 concentration must not
exceed 12.0 micrograms per cubic meter (ug/m3), and the 24-hr average concentration must not exceed 35
ug/m3. More information is available at http://www.epa.gov/air/criteria.html and
http://www.epa.gov/oar/particlepollution/. The standards for PM2.5 values are shown in Table 2-1.
Table 2-1. PM2.5 Standards
Micrograms Per Cubic Meter:
Measurement - (ug/m3)
1997
2006
2012
Annual Average
15.0
15.0
12.0
24-Hour Average
65
35
35
2.2 Ambient Air Quality Monitoring in the United States
2.2.1 Monitoring Networks
The Clean Air Act requires every state to establish a network of air monitoring stations for criteria
pollutants, following specific guidelines for their location and operation. The monitoring stations in this
network have been called the State and Local Air Monitoring Stations (SLAMS). The SLAMS network
consists of approximately 4,000 monitoring sites whose distribution is largely determined by the needs of
State and local air pollution control agencies. All ambient monitoring networks selected for use in
SLAMS are tested periodically to assess the quality of the SLAMS data being produced. Measurement
accuracy and precision are estimated for both automated and manual methods. The individual results of
these tests for each method or analyzer are reported to EPA. Then, EPA calculates quarterly integrated
estimates of precision and accuracy for the SLAMS data.
The National Air Monitoring Station network (NAMS) is about a 1,000-site subset of the SLAMS
network, with emphasis on areas of maximum concentrations and high population density in urban and
multi-source areas. The NAMS monitoring sites are designed to obtain more timely and detailed
information about air quality in strategic locations and must meet more stringent monitor siting,
equipment type, and quality assurance criteria. NAMS monitors also must submit detailed quarterly and
annual monitoring results to EPA.
The SLAMS and NAMS networks experienced accelerated growth throughout the 1970s. The networks
were further expanded in 1999 following the 1997 revision of the CAA to include separate standards for
fine particles (PM2.5) based on their link to serious health problems ranging from increased symptoms,
hospital admissions, and emergency room visits, to premature death in people with heart or lung disease.
While most of the monitors in these networks are located in populated areas of the country, "background"
and rural monitors are an important part of these networks. For criteria pollutants other than ozone and
PM2.5, the number of monitors has declined. For more information on SLAMS and NAMS, as well as
EPA's other air monitoring networks go to www.epa.gov/ttn/amtic.
In 2009, approximately 43 percent of the US population was living within 10 kilometers of ozone and
PM2.5 monitoring sites. In terms of US Census Bureau tract locations, 31,341 out of 72,283 census tract
centroids were within 10 kilometers of ozone monitoring sites. Highly populated Eastern US and
California coasts are well covered by both ozone and PM2.5 monitoring network (Figure 2-1).
7
-------
Distance to the Nearest
• 41 -10.000 meters
• 10,001 -25.000 meters
25,001 - 50.000 meters
50,001 -75.000 meters
75,001 -100,000 meters
• 100,001 - 150,000 meters
• 150,001 - 333,252 meters
Distance to the Nearest PM2.5
• 41 - 10,000 meters
• 10,001 -25,000 meters
25,001 - 50.000 meters
50,001 - 75,000 meters
75,001 -100.000 meters
• 100,001 - 150,000 meters
• 150,001 -333,252 meters
Figure 2-1. Distances from US Census Tract centroids to the nearest monitoring site.
8
-------
In summary, state and local agencies and tribes implement a quality-assured monitoring network to
measure air quality across the United States. EPA provides guidance to ensure a thorough understanding
of the quality of the data produced by these networks. These monitoring data have been used to
characterize the status of the nation's air quality and the trends across the U.S. (see
www.epa.gov/airtrends).
2.2.2 Air Quality System Database
EPA's Air Quality System (AQS) database contains ambient air pollution data collected by EPA, state,
local, and tribal air pollution control agencies from thousands of monitoring stations. AQS also contains
meteorological data, descriptive information about each monitoring station (including its geographic
location and its operator), and data quality assurance and quality control information. State and local
agencies are required to submit their air quality monitoring data into AQS within 90 days following the
end of the quarter in which the data were collected. This ensures timely submission of these data for use
by state, local, and tribal agencies, EPA, and the public. EPA's Office of Air Quality Planning and
Standards and other AQS users rely upon the data in AQS to assess air quality, assist in compliance with
the NAAQS, evaluate SIPs, perform modeling for permit review analysis, and perform other air quality
management functions. For more details, including how users can retrieve data, go to
http://www.epa.gov/ttn/airs/airsaqs/index.htm.
2.2.3 Advantages and Limitations of the Air Quality Monitoring and Reporting System
Air quality data is required to assess public health outcomes that are affected by poor air quality. The
challenge is to get surrogates for air quality on time and spatial scales that are useful for Environmental
Public Health Tracking activities.
The advantage of using ambient data from EPA monitoring networks for comparing with health outcomes
is that these measurements of pollution concentrations are the best characterization of the concentration
of a given pollutant at a given time and location. Furthermore, the data are supported by a comprehensive
quality assurance program, ensuring data of known quality. One disadvantage of using the ambient data
is that it is usually out of spatial and temporal alignment with health outcomes. This spatial and temporal
'misalignment' between air quality monitoring data and health outcomes is influenced by the following
key factors: the living and/or working locations (microenvironments) where a person spends their time
not being co-located with an air quality monitor; time(s)/date(s) when a patient experiences a health
outcome/symptom (e.g., asthma attack) not coinciding with time(s)/date(s) when an air quality monitor
records ambient concentrations of a pollutant high enough to affect the symptom (e.g., asthma attack
either during or shortly after a high PM2.5 day). To compare/correlate ambient concentrations with acute
health effects, daily local air quality data is needed4. Spatial gaps exist in the air quality monitoring
network, especially in rural areas, since the air quality monitoring network is designed to focus on
measurement of pollutant concentrations in high population density areas. Temporal limits also exist.
Hourly ozone measurements are aggregated to daily values (the daily max 8-hour average is relevant to
the ozone standard). Ozone is typically monitored during the ozone season (the warmer months,
approximately April through October). However, year-long data is available in many areas and is
extremely useful to evaluate whether ozone is a factor in health outcomes during the non-ozone seasons.
PM2.5 is generally measured year-round. Most Federal Reference Method (FRM) PM2.5 monitors collect
4 EPA uses exposure models to evaluate the health risks and environmental effects associated with exposure. These models
are limited by the availability of air quality estimates, http://www.epa.gov/ttn/fera/index.html.
9
-------
data one day in every three days, due in part to the time and costs involved in collecting and analyzing the
samples. However, over the past several years, continuous monitors, which can automatically collect,
analyze, and report PM2.5 measurements on an hourly basis, have been introduced. These monitors are
available in most of the major metropolitan areas. Some of these continuous monitors have been
determined to be equivalent to the FRM monitors for regulatory purposes and are called FEM (Federal
Equivalent Methods).
2.2.4 Use of Air Quality Monitoring Data
Air quality monitoring data has been used to provide the information for the following situations:
(1) Assessing effectiveness of SIPs in addressing NAAQS nonattainment areas
(2) Characterizing local, state, and national air quality status and trends
(3) Associating health and environmental damage with air quality levels/concentrations
For the EPHT effort, EPA is providing air quality data to support efforts associated with (2), and (3)
above. Data supporting (3) is generated by EPA through the use of its air quality data and its downscaler
model.
Most studies that associate air quality with health outcomes use air monitoring as a surrogate for exposure
to the air pollutants being investigated. Many studies have used the monitoring networks operated by
state and federal agencies. Some studies perform special monitoring that can better represent exposure to
the air pollutants: community monitoring, near residences, in-house or work place monitoring, and
personal monitoring. For the EPHT program, special monitoring is generally not supported, though it
could be used on a case-by-case basis.
From proximity based exposure estimates to statistical interpolation, many approaches are developed for
estimating exposures to air pollutants using ambient monitoring data (Jerrett et al., 2005). Depending
upon the approach and the spatial and temporal distribution of ambient monitoring data, exposure
estimates to air pollutants may vary greatly in areas further apart from monitors (Bravo et al., 2012).
Factors like limited temporal coverage (i.e., PM2.5 monitors do not operate continuously such as recording
every third day or ozone monitors operate only certain part of the year) and limited spatial coverage (i. e.,
most monitors are located in urban areas and rural coverage is limited) hinder the ability of most of the
interpolation techniques that use monitoring data alone as the input. If we look at the example of
Voronoi Neighbor Averaging (VNA) (referred as the Nearest Neighbor Averaging in most literature),
rural estimates would be biased towards the urban estimates. To further explain this point, assume the
scenario of two cities with monitors and no monitors in the rural areas between, which is very plausible. ,
Since exposure estimates are guaranteed to be within the range of monitors in VNA, estimates for the
rural areas would be higher according to this scenario.
Air quality models may overcome some of the limitations that monitoring networks possess. Models such
as the Community Multi-Scale Air Quality (CMAQ) modeling systems can estimate concentrations in
reasonable temporal and spatial resolutions. However these sophisticated air quality models are prune to
systematic biases since they depend upon so many variables (i.e., metrological models and emission
models) and complex chemical and physical process simulations.
Combining monitoring data with air quality models (via fusion or regression) may provide the best results
10
-------
in terms of estimating ambient air concentrations in space and time. EPA's eVNA5 is an example of an
earlier approach for merging air quality monitor data with CMAQ model predictions. The downscaler
model attempts to address some of the shortcomings in these earlier attempts to statistically combine
monitor and model predicted data, see published paper referenced in section 1 for more information about
the downscaler model. As discussed in the next section, there are two methods used in EPHT to provide
estimates of ambient concentrations of air pollutants: air quality monitoring data and the downscaler
model estimate, which is a statistical 'combination' of air quality monitor data and photochemical air
quality model predictions (e.g., CMAQ).
2.3 Air Quality Indicators Developed for the EPHT Network
Air quality indicators have been developed for use in the Environmental Public Health Tracking Network
by CDC using the ozone and PM2.5 data from EPA. The approach used divides "indicators" into two
categories. First, basic air quality measures were developed to compare air quality levels over space and
time within a public health context (e.g., using the NAAQS as a benchmark). Next, indicators were
developed that mathematically link air quality data to public health tracking data (e.g., daily PM2.5 levels
and hospitalization data for acute myocardial infarction). Table 2-3 and Table 2-4 describe the issues
impacting calculation of basic air quality indicators.
Table 2-2. Public Health Surveillance Goals and Current Status
(io;il
Sialus
Air data sets and metadata required for air quality
indicators are available to EPHT state Grantees.
AQS data are available through state agencies and
EPA's Air Quality System (AQS). EPA and CDC
developed an interagency agreement, where EPA
provides air quality data along with statistically
combined AQS and Community Multiscale Air Quality
(CMAQ) Model data, associated metadata, and technical
reports that are delivered to CDC.
Estimate the linkage or association of PM2 5 and
ozone on health to:
Identify populations that may have higher risk of
adverse health effects due to PM2 5 and ozone,
Generate hypothesis for further research, and
Provide information to support prevention and
pollution control strategies.
Regular discussions have been held on health-air linked
indicators and CDC/HFI/EPA convened a workshop
January 2008. CDC has collaborated on a health impact
assessment (HIA) with Emory University, EPA, and
state grantees that can be used to facilitate greater
understanding of these linkages.
Produce and disseminate basic indicators and other
findings in electronic and print formats to provide
the public, environmental health professionals, and
policymakers, with current and easy-to-use
information about air pollution and the impact on
public health.
Templates and "how to" guides for PM2 5 and ozone
have been developed for routine indicators. Calculation
techniques and presentations for the indicators have been
developed.
5eVNA is described in the "Regulatory Impact Analysis for the Final Clean Air Interstate Rule", EPA-452/R-05-002, March
2005, http://www.epa.gov/cair/pdfs/finaltech08.pdf. Appendix F.
11
-------
Table 2-3. Basic Air Quality Indicators used in EPHT, derived from the EPA data delivered to
CDC
O/.onc (daily 8-hr period with maximum concentration—ppm—by Federal Reference Method (FRM))
• Number of days with maximum ozone concentration over the NAAQS (or other relevant benchmarks (by county
and MSA)
• Number of person-days with maximum 8-hr average ozone concentration over the NAAQS & other relevant
benchmarks (by county and MSA)
I'M idails 24-hi nik'uiak'd samples uu in -In I'kMi
• Average ambient concentrations of particulate matter (<2.5 microns in diameter) and compared to annual PM25
NAAQS (by state).
• % population exceeding annual PM2 5 NAAQS (by state).
• % of days with PM2 5 concentration over the daily NAAQS (or other relevant benchmarks (by county and MSA)
• Number of person-days with PM2 5 concentration over the daily NAAQS & other relevant benchmarks (by
county and MSA)
2.3.1 Rationale for the Air Quality Indicators
The CDC EPHT Network is initially focusing on ozone and PM2.5 These air quality indicators are based
mainly around the NAAQS health findings and program-based measures (measurement, data and analysis
methodologies). The indicators will allow comparisons across space and time for EPHT actions. They
are in the context of health-based benchmarks. By bringing population into the measures, they roughly
distinguish between potential exposures (at broad scale).
2.3.2 Air Quality Data Sources
The air quality data will be available in the US EPA Air Quality System (AQS) database based on the
state/federal air program's data collection and processing. The AQS database contains ambient air
pollution data collected by EPA, state, local, and tribal air pollution control agencies from thousands of
monitoring stations (SLAMS and NAMS).
2.3.3 Use of Air Quality Indicators for Public Health Practice
The basic indicators will be used to inform policymakers and the public regarding the degree of hazard
within a state and across states (national). For example, the number of days per year that ozone is above
the NAAQS can be used to communicate to sensitive populations (such as asthmatics) the number of days
that they may be exposed to unhealthy levels of ozone. This is the same level used in the Air Quality
Alerts that inform these sensitive populations when and how to reduce their exposure. These indicators,
however, are not a surrogate measure of exposure and therefore will not be linked with health data.
12
-------
3,0 Emissions Data
3.1 Introduction to Emissions Data Development
The U.S. Environmental Protection Agency (EPA) developed an air quality modeling platform based
primarily on the 2008 National Emissions Inventory (NEI), Version 2 to process year 2009 emission data
for this project. This section provides a summary of the emissions inventory and emissions modeling
techniques applied to Criteria Air Pollutants (CAPs) and the following select Hazardous Air Pollutants
(HAPs): chlorine (CI), hydrogen chloride (HC1), benzene, acetaldehyde, formaldehyde and methanol.
This section also describes the approach and data used to produce emissions inputs to the air quality
model. The air quality modeling, meteorological inputs and boundary conditions are described in a
separate section.
The Community Multi scale Air Quality (CMAQ) model (Ttftp://www.epa.gov/AMD/CMAQ/) is used to
model ozone (O?) and particulate matter (PM) for this project. CMAQ requires hourly and gridded
emissions of the following inventory pollutants: carbon monoxide (CO),nitrogen oxides (NOx), volatile
organic compounds (VOC), sulfur dioxide (SOj), ammonia (NH3), particulate matter less than or equal
to 10 microns (PM10), and individual component species for particulate matter less than or equal to 2.5
microns (PM2 5) In addition, the CMAQ CB05 with chlorine chemistry used here allows for explicit
treatment of the V OC HAPs benzene, acetaldehyde, formaldehyde and methanol (BAFM) and includes
anthropogenic HAP emissions of HC1 and CI.
The effort to create the 2009 emission inputs for this study included development of emission inventories
for a 2009 model evaluation case, and application of emissions modeling tools to convert the inventories
into the format and resolution needed by CMAQ. An evaluation case uses 2009-specific fire and
continuous emission monitoring (CEM) data for electric generating units (EGUs) whereas other types of
cases use averages for these sources. The primary emissions modeling tool used to create the CMAQ
model-ready emissions was the Sparse Matrix Operator Kernel Emissions (SMOKE) modeling system.
SMOKE version 3.1 was used to create emissions files for a 12-km national grid. Additional information
about SMOKE is available from http://www.smoke-model.org.
This chapter contains two additional sections. Section 3.2 describes the inventories input to SMOKE and
the ancillary files used along with the emission inventories. Section 3.3 describes the emissions modeling
performed to convert the inventories into the format and resolution needed by CMAQ.
3.2 2009 Emission Inventories and Approaches
This section describes the emissions inventories created for input to SMOKE. The 2008 NEI, which is the
primary basis for the input to SMOKE, includes five main categories of source sectors: a) nonpoint
(formerly called "stationary area") sources; b) point sources; c) nonroad mobile sources; d) onroad
mobile sources; and e) fires. The NEI data are largely compiled from data submitted by state, local and
tribal (S/L/T) agencies for CAPs. HAP emissions data are often augmented by EPA because they are a
voluntary component. The 2008 NEI was compiled using the Emissions Inventory System (EIS). EIS
13
-------
includes hundreds of automated QA checks to help improve data quality, and also supports release point
(stack) coordinates separately from facility coordinates. Improved EPA collaboration with S/L/T
agencies prevented duplication between point and nonpoint source categories such as industrial boilers.
Documentation for the 2008 NEI is available at http://www.epa.gOv/ttn/chief/net/2008inventory.html#
inventorvdoc.
2009-specific data submitted by S/L/T agencies was used for some large point sources. For EGU
emissions, 2009 continuous emissions monitoring (CEM) data was used where it was available. For fires,
EPA used the SMARTFIRE2 (SF2) system to develop 2009 emissions. SF2 was the first system to
assign all fires as either prescribed burning or wildfire categories and includes improved emission factor
estimates for prescribed burning. 2009-specific data for onroad, nonroad, and large commercial marine
sources was also developed. Some data obtained from regional planning organizations (RPOs) was
substituted for NEI data where the RPO data was more recently collected. California-provided mobile
source emissions were also used. For inventories outside of the United States, including Canada, Mexico
and offshore emissions, the latest available base year inventories were used.
The methods used to process emissions for this project are very similar to those documented for EPA's
Version 5, 2007 Emissions Modeling Platform. A technical support document (TSD) for this platform is
available at EPA's emissions modeling clearinghouse (EMCH):
http://www.epa.gOv/ttn/chief/emch/index.html#pmnaaqs. Electronic copies of inventories similar to those
used for this project are available in the same section of the EMCH.
The emissions modeling process, performed using SMOKE v3.1 apportions the emissions inventories
into the grid cells used by CMAQ and temporalizes the emissions into hourly values. In addition, the
pollutants in the inventories (e.g., NOx and VOC) are split into the chemical species needed by CM AQ.
For the purposes of preparing the CM AQ- ready emissions, the broader NEI emissions inventories are
split into emissions modeling "platform" sectors; and biogenic emissions are added along with emissions
from other sources other than the NEI, such as the Canadian, Mexican, and offshore inventories. The
significance of an emissions sector for the emissions modeling platform is that it is run through all of the
s programs, except the final merge, independently from the other sectors. The final merge program called
Mrggrid combines the sector- specific gridded, speciated and temporalized emissions to create the final
CMAQ-ready emissions inputs.
Table 3-1 presents the sectors in the emissions modeling platform used to develop 2009 emissions for this
project. The sector abbreviations are provided in italics; these abbreviations are used in the SMOKE
modeling scripts and inventory file names and throughout the remainder of this section. Annual 2009
emission summaries for the U.S. anthropogenic sectors are shown in Table 3-2 (i.e., biogenic emissions
are excluded). Table 3-3 provides a summary of emissions for the anthropogenic sectors containing
Canadian, Mexican and offshore sources. State total emissions for each sector are provided in Appendix
A, a workbook entitled "Appendix_A_2009_emissions_totals_by_sector.xlsx".
14
-------
Table 3-1. Platform Sectors Used in the Emissions Modeling Process
2009 Platform Sector
(Abbrev)
2009 NEI
Sector
Description and resolution of the data input to
SMOKE
IPM (ptipm)
Point
2009 NEI point source EGUs that can be mapped to
the Integrated Planning Model (IPM) model. NEI
values replaced with year 2009 hourly continuous
emission monitoring (CEM) NOx and SO2 emissions.
Other pollutants are scaled from 2008 NEI using heat
input.
Point non-IPM (ptnonipm)
Point
A mix of 2008 NEI point source emissions with some
2009 records where data was provided by states and
locals and 2006 WRAP oil and gas data; these are
emissions not matched to the ptipm sector, annual
resolution. Includes all aircraft emissions
Point source fire (ptfire)
Fires
Point source day-specific wildfires and prescribed fires
for 2009.
Agricultural (ag)
Nonpoint
2008 NEI nonpoint NH3 emissions from livestock and
fertilizer application; county and annual resolution
with some 2007 monthly resolution data provided by
the Midwest.
Area fugitive dust (a/dust)
Nonpoint
2008 NEI nonpoint PM10 and PM2.5 from fugitive dust
sources (e.g., building construction, road construction,
paved roads, unpaved roads, agricultural dust), county
and annual resolution. A land use-based transport
fraction and 2009-based precipitation zero-out is
applied.
Remaining nonpoint (nonpt)
Nonpoint
Primarily 2008 NEI nonpoint for sources not included
in other sectors mixed with 2006 WRAP oil and gas
data, county and annual resolution.
Nonroad (nonroad)
Nonroad
Year 2009 monthly nonroad emissions from the
National Mobile Inventory Model (NMIM) plus
California-provided data; county and annual
resolution.
C1 and C2 marine and
locomotive (clc2rail)
Nonroad
Year 2008 non-rail maintenance locomotives, and
category 1 and category 2 commercial marine vessel
(CMV) emissions sources; county and annual
resolution; year 2009 for California.
C3 commercial marine
(c3marine)
Nonroad
Non-NEI, year 2009 category 3 (C3) CMV emissions
projected from year 2002. Developed for the rule
called "Control of Emissions from New Marine
Compression-Ignition Engines at or Above 30 Liters
per Cylinder", usually described as the Emissions
Control Area- International Maritime Organization
(ECA-IMO) study:
http://www.epa.gov/otaq/oceanvessels.htm. (EPA-
420-F-10-041, August 2010). Annual resolution and
15
-------
treated as point sources.
Onroad (onroad)
Onroad
Onroad Refueling
(onroad rfl)
Biogenic (beis)
Onroad
Biogenic
Other point sources (othpt) N/A
Other nonpoint and nonroad
(othar)
N/A
Other onroad sources (othon) N/A
Year 2009 gridded hourly emissions from onroad
mobile gasoline and diesel vehicles from parking lots
and moving vehicles including exhaust, evaporative,
permeation, and brake and tire wear. Generated using
MOVES 2010b emission factors, 2009 VMT and
vehicle population data, and 2009 gridded met. data. In
California, adjusted to match CA-provided emissions.
Year 2009 gridded hourly emissions from onroad
mobile gasoline and diesel vehicles from parking lots
and moving vehicles for refueling only. Generated
using MOVES 2010b, emission actors, 2009 VMT and
vehicle population data, and 2009 gridded met. data.
Spatially allocated to gasoline station locations.
Hour- and grid cell-specific emissions for 2009
generated from the BEIS 3.14 model, including
emissions in Canada and Mexico.
Point sources not from the NEI, including Canada's
2006 inventory and a 2008 projection of Mexico's
Phase III 1999 inventory; annual resolution. Also
includes 2008 offshore oil point source emissions for
the U.S. from the 2008 NEI.
Nonpoint and nonroad sources not from the NEI,
including annual 2006 Canada sources at province
resolution and a 2008 projection of annual 1999
Mexico sources at municipio resolution.
Onroad sources not from the NEI, including annual
2006 Canada sources at province resolution and a
2008 projection of 1999 Mexico sources at municipio
resolution.
16
-------
Table 3-2. 2009 Continental United States Emissions by Sector (tons/yr in 48 states + D.C.)
Sector
CO nh3
NOx
PMio
pm25
so2
voc
afdust
5,823,635
816,524
ag
3,595,429
clc2rail
217,984 559
1,329,661
43,528
40,733
48,487
60,809
nonpt
4,336,565 155,317
1,230,624
767,225
676,243
402,633
6,456,455
nonroad
15,053,215 1,985
1,784,297
174,562
165,768
32,169
2,249,982
onroad w/rfl.
27,221,698 127,354
6,165,415
302,003
222,002
34,973
2,736,569
ptfire
12,378,697 203,630
192,773
1,280,587
1,085,244
100,324
2,927,182
ptipm
676,123 24,015
2,046,085
298,162
205,675
5,965,968
32,955
ptnonipm
2,573,239 70,131
1,905,593
512,816
360,946
1,329,436
1,065,623
c3 marine
14,757
160,083
14,515
13,326
121,120
5,725
Con.US Total
62,472,278 4,178,420
14,814,530 !
9,217,034 3,586,460
8,035,109
15,535,300
Table 3-3. 2009 Non-US Emissions by Sector within Modeling Domain (tons/yr for Canada,
Mexico, Offshore)
Country &
Sector
[tons/yr] [tons/yr] [tons/yr]
CO NH3 NOx
[tons/yr]
PMio
[tons/yr]
pm25
[tons/yr]
so2
[tons/yr]
VOC
Canada othar
3,747,303 537,912 718,757
1,421,686
393,642
97,709
1,267,472
Canada othon
4,513,915 21,810 537,704
15,004
10,634
5,430
277,874
Canada othpt
1,148,101 21,138 861,256
117,254
68,115
1,762,345
425,792
Canada Subtotal 9,409,320 580,860 2,117,717
1,553,944
472,390
1,865,484
1,971,138
Mexico othar
477,908 132,913 198,972
88,319
56,809
56,417
510,955
Mexico othon
659,536 2,971 93,839
7,935
7,348
5,738
96,218
Mexico othpt
101,309
0 344,896
122,654
90,304
740,238
78,465
Mexico Subtotal 1,238,753 135,884 637,708
218,908
154,460
802,393
685,639
Offshore othpt
82,133
0 74,277
780
769
1,021
60,756
Canada c3marine 13,394
0 160,983
13,434
12,311
99,644
5,690
Offshore c3marine 80,212
0 961,146
80,549
74,063
599,679
34,079
2009 TOTAL
10,823,812 716,744 3,951,831
1,867,615
713,994
3,368,221
2,757,303
3.2.1 Point Sources (ptipm andptnonipm)
Point sources are sources of emissions for which specific geographic coordinates (e.g., latitude/longitude)
are specified, as in the case of an individual facility. A facility may have multiple emission points, which
may be characterized as units such as boilers, reactors, spray booths, kilns, etc. A unit may have multiple
processes (e.g., a boiler that sometimes burns residual oil and sometimes burns natural gas). The point
sources used for this study include a limited set of emissions data for 2009 collected via the NEI process,
with 2008 NEI data for any sources that did not report in 2009. Note that only large sources are required
17
-------
to report annually as opposed to triennially. This section describes NEI point sources within the
contiguous United States. The offshore oil (othpt sector), fires (ptfire) and category 3 CMV emissions
(c3marine sector) are point source formatted inventories discussed later in this section. Full
documentation for the development of the 2008 NEI (EPA, 2012), is posted at:
http://www.epa.gOv/ttn/chief/net/2008inventorv.html#inventorvdoc.
After removing offshore oil platforms into the othpt sector, we created two platform sectors from the
remaining point sources for input into SMOKE: the EGU sector - also called the IPM sector (i.e., ptipm)
and the non-EGU sector - also called the non-IPM sector (i.e., ptnonipm). This split facilitates the use of
different SMOKE temporal processing and future-year projection techniques for each of these sectors.
The inventory pollutants processed through SMOKE for both the ptipm and ptnonipm sectors were: CO,
NOX, VOC, S02, NH3, PM10, and PM2.5 and the following HAPs: HC1 (pollutant code = 7647010),
and CI (code = 7782505). BAFM from these sectors was not utilized because VOC was speciated
without the use (i.e., integration) of VOC HAP pollutants from the inventory (integration is discussed in
detail in Section3.3.4).
In the 2009 model evaluation case used in this study, for ptipm sector sources with CEM data that could
be matched to the NEI, 2009 hourly S02 and NOx emissions were used alongside annual emissions of all
other pollutants. The hourly electric generating unit (EGU) emissions were obtained for SOj and NOx
emissions and heat input from EPA's Acid Rain Program. This data also contained heat input, which was
used to allocate the annual emissions for other pollutants (e.g., VOC, PM2 5, HQ) to hourly values. For
unmatched EGU units, annual emissions were temporalized to days using multi-year averages and to
hours using state-specific averages.
The Non-EGU Stationary Point Sources (ptnonipm) emissions were provided to SMOKE as annual
emissions. The emissions were developed as follows:
a. 2008 CAP and HAP data were provided by States, locals and tribes under the Consolidated
Emissions Reporting Rule
b. EPA corrected known issues and filled PM data gaps.
c. EPA added HAP data from the Toxic Release Inventory (TRI) where it was not provided by
states/locals.
d. EPA provided data for airports and rail yards.
e. Off-shore platform data was added from Mineral Management Services (MMS).
The changes made to the NEI point sources prior to modeling are as follows:
• The tribal data, which do not use state/county Federal Information Processing Standards (FIPS)
codes in the NEI, but rather use the tribal code, were assigned a state/county FIPS code of
88XXX, where XXX is the3-digit tribal code in the NEI. This change was made because SMOKE
requires the state/county FIPS code.
• Stack parameters for some point sources were defaulted when modeling in SMOKE. SMOKE uses
an ancillary file, called the PSTK file, which provides default stack parameters by SCC code to
either gap fill stack parameters if they are missing in the NEI or to correct stack parameters if they
are outside the ranges specified.
18
-------
• Replaced stack parameters with values from the 2008 NEI where 2008 values were determined to
be more realistic.
• Replaced facility emissions with 2008 NEI values where the 2009 NEI contained questionable
values.
3.2.1.11PM Sector (ptipm)
The ptipm sector contains emissions from EG Us in the 2009 NEI point inventory that could be matched to
the units found in the NEEDS database, version 4.10 (http://www.epa.gov/airmarkets/progsregs/epa-ipm/
index.html). IPM provides future year emission inventories for the universe of EGUs contained in the
NEEDS database. As described below, matching with NEEDS was done (1) to provide consistency
between the 2009 EGU sources and future year EGU emissions for sources which are forecasted by IPM,
and (2) to avoid double counting when projecting point source emissions.
The 2009 NEI point source inventory contains emissions estimates for both EGU and non-EGU sources.
When future years are modeled, IPM is used to predict the future year emissions for the EGU sources. The
remaining non-EGU point sources are projected by applying projection and control factors to the base
year emissions. It was therefore necessary to identify and separate into two sectors: (1) sources that are
projected via IPM (i.e., the "ptipm" sector) and (2) sources that are not (i.e., "the "ptnonipm" sector). The
two sectors are modeled separately in the base year as well as the future years.
A primary reason the ptipm sources were separated from the other point sources was due to the difference
in the temporal resolution of the data input to SMOKE. The ptipm sector uses the available hourly CEM
data via a method first implemented in the 2002 platform and still used for the 2009 platform. Hourly
CEM data for 2009 were obtained from the CAMD Data and Maps website3. For sources and pollutants
with CEM data, the actual year 2009 hourly CEM data were used. The SMOKE modeling system matches
the ORIS Facility and Boiler IDs in the NEI SMOKE-ready file to the same fields in the CEM data,
thereby allowing the hourly S02 and NOx CEM emissions to be read directly from the CEM data file. The
heat input from the hourly CEM data was used to allocate the NEI annual values to hourly values for all
other pollutants from CEM sources, because CEMs are not used to measure emissions of these
pollutants.
For this project, the point source inventory was reviewed to determine whether additional matches needed
to be made. Newly identified matches for CEM and NEEDS IDs were loaded into the Emissions
Inventory System (E1S) so they could then be written into the modeling files. Some matches were made
outside of E1S when IDs were not mapped one to one between the systems.
Emissions were scaled from 2008 levels to 2009 levels where possible based on CEM data, where
possible. For sources not matching the CEM data ("non-CEM" sources), daily emissions were computed
from the NEI annual emissions using a structured query language (SQL) program and state-average CEM
data. To allocate annual emissions to each month, state-specific, three-year averages of 2008-2010 CEM
data were created. These average annual- to-month factors were assigned to non-CEM sources by state.
To allocate the monthly emissions to each day, the 2009 CEM data were used to compute state-specific
month- to-day factors, which were then averaged across all units in each state. The resulting daily
emissions were input into SMOKE. The daily-to-hourly allocation was performed in SMOKE using
diurnal profiles. The development of these diurnal ptipm-specific profiles, considered ancillary data for
19
-------
SMOKE, is described in a later section.
3.2.1.2 Non-IPM Sector (ptnonipm)
The non-IPM (ptnonipm) sector contains all NEI point sources not included in the IPM (ptipm) sector
except for the offshore oil and day-specific fire emissions. For the most part, the ptnonipm sector reflects
the non-EGU component of the NEI point inventory; however, as previously discussed, it is likely that
some small low-emitting EG Us that are not reflected in the CEMs database are present in the ptnonipm
sector. The ptnonipm sector contains a small amount of fugitive dust PM emissions from vehicular traffic
on paved or unpaved roads at industrial facilities or coal handling at coal mines. In previous versions of
the platform, we would reduce these emissions prior to input to SMOKE. However, in this platform the
reduction is not made because of a new methodology used to reduce PM dust.
For some geographic areas, some of the sources in the ptnonipm sector belong to source categories that
are contained in other sectors. This occurs in the inventory when states, tribes or local programs report
certain inventory emissions as point sources because they have specific geographic coordinates for these
sources. They may use point source SCCs (8-digit) or they may use non- point, onroad or nonroad (10-
digit) SCCs. In the 2008 NEI, examples of these types of sources include: aircraft and ground support
emissions, livestock (i.e., cattle feedlots) in California, and rail yards.
Some adjustments were made to the point inventory prior to its use in modeling. These include:
• Removing sources with state county codes ending in '777'. These are used for 'portable' point
sources like asphalt plants.
• Removing sources with SCCs not typically used for modeling.
• Adjusting latitude-longitude coordinates for sources identified to be substantially outside the
county in which they reside.
• Removed all offshore oil records as reflected by FIPS=85000 because these sources are processed
in the othpt sector.
• Added 2008 ethanol facilities provided by EPA's OTAQ that were not already included in the
2008 NEI.
• Corrected stack parameters for some units with missing or invalid parameter assignments.
• Added South Dakota emissions because they did not submit to the 2008 NEI.
• Added MeadWestVaco facility in Covington, VA because it was missing in the 2008 NEI.
• Added oil and gas emissions that were not otherwise included in the NEI from the Western
Regional Air Partnership (WRAP) RPO created year 2006 "Phase III" oil and gas inventory
project.
• Removed onroad refueling emissions that some states included in the point sector because these
are modeled nationwide using MOVES2010b.
3.2.2 Nonpoint Sources (afdust, ag, nonpt)
The nonpoint emissions sources used in this study are primarily from the 2008 NEI. Documentation for
the 2008 NEI is available at http://www.epa. uov/ttn/chief/net/2008inventorv.html#inventorvdoc. Prior to
modeling, the nonpoint portion of the 2008 NEI was divided into the following sectors for which the data
20
-------
is processed in consistent ways: area fugitive dust (afdust), agricultural ammonia (ag), and the other
nonpoint sources (nonpt). This section describes stationary nonpoint sources only. Class 1 & Class 2
(clc2) and Class 3 (c3) coniniercial marine vessels and locomotives are also in the 2008 NEI nonpoint
data category, but these sources are included in the mobile source portion of this documentation.
Nonpoint tribal-submitted emissions were removed to prevent possible double counting with county-level
emissions. Because the tribal nonpoint emissions are small, these omissions should not impact results at
the 12-ktn scale used for modeling. This omission also eliminated the need to develop costly spatial
surrogate data to allocate tribal data to grid cells during the SMOKE processing. Some specific types of
nonpoint sources were not included in the modeling due to one of the following reasons: 1) the sources
are only reported by a few states or agencies, 2) the sources are 'atypical" and small, and/or 3) there are
other data available that appears to be more accurate. Additional details on nonpoint source processing
can be found in the Version 5, 2007 Emissions Modeling Platform documentation discussed earlier.
In the rest of this section, each of the platform sectors into which the 2008 nonpoint NEI was divided is
described, along with any changes made to these data.
3.2.2.1 Area Fugitive Dust Sector (afdust)
The area-source fugitive dust (afdust) sector contains PM emission estimates for 2008 NEI nonpoint
SCCs identified by EPA staff as fugitive dust sources. Categories included in this sector are paved roads,
unpaved roads and airstrips, construction (residential, industrial, road and total), agriculture production
and all of the mining 10-digit SCCs beginning with the digits "2325." It does not include fugitive dust
from grain elevators because these are elevated point sources.
This sector is separated from other nonpoint sectors to allow for the application of "transport fraction,"
and meteorology/precipitation ("MET") reductions. These adjustments are applied via sector-specific
scripts and make use of land use-based gridded transport fractions. The land use data used to reduce the
NEI emissions explains the amount of emissions that are subject to transport. This methodology is
discussed in (Pouliot, et. al., 2010),
http://www.epa.gov/ttn/chief/conference/eil9/session9/pouliot pres.pdf. and in Fugitive Dust Modeling
for the 2008 Emissions Modeling Platform (Adelman, 2012). The precipitation adjustment is then
applied to remove all emissions for days on which measureable rain occurs or there is snow on the
ground. Both the transport fraction and MET adjustments are based on the gridded meteorological data;
therefore, different emissions could result from different grid resolutions. Application of the transport
fraction and MET adjustments reduces the overestimation of fugitive dust impacts in the grid modeling as
compared to ambient samples.
3.2.2.2 Agricultural Ammonia Sector (ag)
The agricultural NH3 "ag" sector is comprised of livestock and agricultural fertilizer application emissions
from the nonpoint sector of the 2008 NEI. The livestock and fertilizer emissions were extracted based on
SCC. The "ag" sector includes all of the NH3 emissions from fertilizer contained in the NEI. However,
the "ag" sector does not include all of the livestock ammonia emissions, as there are also some NH3
emissions from feedlot livestock in the point source inventory. To prevent double-counting, emissions
were not included in the nonpoint ag inventory for counties in which they were in the point source
inventory. A significant error in the 2008 NEI was corrected in the modeling platform ag sector. A
fertilizer application source "N-P-K (multi-grade nutrient fertilizers)" (SCC=2801700010) in Luna
county New Mexico (FIPS=35025), was 6,953 tons of NH3 in the 2008 NEI. This source was corrected
21
-------
by a factor of 1,000 to be 6.953 tons in the modeling platform.
Monthly NH3 emissions provided by the Lake Michigan Air Directors Consortium were used to replace
NEI ag sector emissions in that region due to the improved temporal resolution. 2008 NEI (annual) ag
sector emissions were used in all other states. A new temporal allocation methodology for animal NH3
was implemented for this modeling platform that allocates monthly emissions down to the hourly level by
taking into account temperature and wind speed. This method is discussed in more detail in the emission
modeling portion of this chapter.
3.2.2.3 Other Nonpoint Sources (nonpt)
Stationary nonpoint sources that were not subdivided into the afdust, ag or nonpt sectors were assigned to
the "nonpt" sector. In preparing the nonpt sector, catastrophic releases were excluded since these
emissions were dominated by tire burning, which is an episodic, location-specific emissions category. Tire
burning accounts for significant emissions of particulate matter in some parts of the country. Because such
sources are reported by a very small number of states, and are inventoried as county/annual totals without
the information needed to temporally and spatially allocate the emissions to the time and location where
the event occurred, catastrophic releases were excluded. All fire emissions, including agricultural,
wildfire, and prescribed burning, were removed and substituted with SMARTFIRE emissions (see the
"ptfire" sector). Locomotives and CMV mobile sources from the 2008 NEI nonpoint inventory are
described in the mobile sources section.
The nonpt sector includes emission estimates for Portable Fuel Containers (PFCs), also known as "gas
cans." The PFC inventory consists of five distinct sources of PFC emissions, further distinguished by
residential or commercial use. The five sources are: (1) displacement of the vapor within the can; (2)
spillage of gasoline while filling the can; (3) spillage of gasoline during transport; (4) emissions due to
evaporation (i.e., diurnal emissions); and (5) emissions due to permeation. Note that spillage and vapor
displacement associated with using PFCs to refuel nonroad equipment are included in the n on road
inventory.
Some adjustments to the 2008 NEI nonpoint data were made using data from regional planning
organizations (RPOs) as follows:
• Replaced 2008 NEI oil and gas emissions (SCCs beginning with "23100") with year 2006 Phase
III oil and gas emissions for several basins in the WRAP RPO states. These WRAP Phase III
emissions contain point and nonpoint formatted data are discussed in greater detail at:
http://www.wrapair2.org/PhaseIII.aspx. These changes were made only in counties for which
there was WRAP data.
• Replaced 2008 NEI nonpoint agriculture burning emissions with year 2008 SMARTFIRE day-
specific county-based emissions aggregated to monthly totals.
• Replaced open burning "land clearing" (SCC=2610000500) emissions in Florida and Georgia
with SESARM-provided daily point data, but aggregated to county and monthly resolution.
• Replaced open burning data (SCCs beginning with 261000x) in MARAMA states with RPO-
proved data.
• Removed industrial coal combustion emissions (SCC=2102002000) in Tennessee.
• Replaced, removed and modified much of the residential wood combustion (RWC) emissions in
22
-------
the MARAMA, MWRPO and SESARM states with RPO data and non-RPO corrections,
modified the outdoor hydronic heater (OHH) emissions in all states and indoor furnaces in
MWRPO states.
• Removed EPA-estimated commercial cooking (SCCs 2302002100 and 2302002200) duplicate
PM emissions in California.
• Removed duplicate "Industrial Processes; Food and Kindred Products; Total" source
(SCC=23020000000) in Maricopa county Arizona (FIPS=04013).
The oil and gas changes were already discussed in the ptnonipm section. Other significant changes are
discussed below.
Ag burning
2008 NEI agricultural burning estimates were replaced with more specific data from the Fire
Characteristic Classification System (FCCS) module fuel loadings map in the BlueSky Framework
(http://blueskyframework.org/modules/fuel-loading/fccs). Year 2008-specific fire locations from
SMARTFIRE version 1 (Sullivan, et al., 2008) were read into the FCCS module and intersected with the
FCCS fuel-loading dataset. The module assigned an FCCS code to each fire record that reflects the
ecosystem geography and potential natural vegetation based on remote sensing data. Prescribed or
unclassified fires having an FCCS code equal to zero (0) were assumed to be agricultural fires. Arc GIS
was used to categorize the fires as occurring on rangeland, cropland or other land use via USGS 2006
National Land Cover Database (NLCD). Activity data were analyzed to restrict to cropland fires and
assign state and crop-specific emission factors. Emissions were then appropriately weighted based on
known statistics about each state's crop mix.
These SMARTFIRE-based ag burning emissions were provided in at 1km point source and day-specific
resolution. State-county FIPS codes were assigned using GIS. The emissions were aggregated to county
and monthly resolution and converted to SMOKE nonpoint FF10 format. This SMARTFIRE-based ag
burning dataset includes emissions for all but these 7 of the lower 48 states: CT, DC, MA, ME, NH, RI
and VT. These 7 states did not contain any cropland burning estimates for year 2008 based on this
SMARTFIRE approach.
Open burning RPO data
All 2008 NEI open burning emissions (CAPs only) were replaced in the MARAMA states with the 2007
MARAMA open burning inventory. These MARAMA open burning emissions include estimates for
household waste (SCC=2610030000), land clearing (2610000500) and yard waste leaf and brush
(2610000100 and 2610000400 respectively).
The 2008 NEI land clearing emissions in Georgia and Florida were replaced with SESARM-based year-
2007 data. The SESARM land clearing emissions are based on daily point emissions from the
CONSUME v3.0 model (SESARM, 2012a). These daily point-format emissions were aggregated to
county and monthly resolution as a separate FF10 nonpoint monthly inventory.
TN coal combustion
Tennessee nonpoint industrial coal combustion (SCC=2102002000) emissions are significantly
overestimated in the 2008 NEI because of incorrect reconciliation with the point source inventory.
Nonpoint industrial coal combustion emissions were estimated by subtracting point source emissions
rather than activity. By not accounting for controlled sources, the remaining activity for nonpoint coal
23
-------
combustion is significantly overestimated. EPA NEI experts determined that it would be more
appropriate to completely remove the nonpoint component of this sector than to leave the values as they
were. The reality for TN industrial coal combustion nonpoint sector emissions is likely much closer to
zero than the value in the 2008 NEI because these emissions are accounted for in the point source
inventory.
Residential Wood Combustion
There were many modifications to the RWC emissions data. First, all RWC outdoor wood burning
devices such as "fire pits and chimeas" (SCC=2104008700) were removed because they were only
reported in a couple of states, RPO inventories did not include them for most states and emissions were
generally insignificant. A market research report (Frost and Sullivan, 2010) developed in support of the
potential RWC New Source Performance Standard (NSPS) indicated slower sales of outdoor hydronic
heaters compared to what was assumed for growth estimates in the 2008 NEI. Therefore, outdoor
hydronic heater appliance counts and emissions estimates (SCC=2104008610) were recomputed for all
states, resulting in a 51% reduction to outdoor hydronic heater emissions for all states.
In addition, all emissions in the SESARM states (i.e., AL, FL, GA, KY, MS, NC, SC, TN, VA, WV),
including Virginia, were replaced with the SESARM year-2007 inventory (SESARM, 2012b). Urban area
RWC were lower than the NEI estimates partially because of the assumptions about greater penetration of
natural gas fireplaces, less access to inexpensive wood supplies and a lower proportion of housing units
with wood burning appliances as primary heating units than rural areas. Overall, the SESARM RWC
estimates are considerably lower than the 2008 NEI estimates for several states, particularly for
"uncertified" and "general" wood stoves and insert categories: FL, KY, NC, TN, VA and WV. However,
emissions in Mississippi are only slightly reduced and emissions in AL, GA and SC are very similar to
those in the 2008NEIv2.
The Midwest RPO (LADCO) states (i.e., IL, IL, MI, OH, WI, MN) year-2007 RWC inventory was
similar to the 2008 NEI for most source types. However, the pellet stoves (SCC=2104008400), indoor
furnaces (2104008510), and outdoor hydronic heater (OHH, SCC=2104008610) estimates were updated
to reallocate the indoor furnaces and OHHs to non-MSA counties (LADCO, 2012) for several urban
areas. Some double counting of appliances was also fixed in Wisconsin and Michigan. Overall, the
MWRPO states totals are very similar to the 2008 NEI; however, emissions are spatially redistributed
from urban to rural areas. Therefore, for the MWRPO states, the 2008 NEI emissions were used for all
RWC sources except the three aforementioned SCCs that use the 2007 MWRPO data.
Emissions from indoor wood fired furnaces (SCC=2104008510) in several MWRPO states based were
also recomputed based on newer, improved survey data from Minnesota. The 2008 NEI for these sources
started with an assumption of year 2002 Minnesota wood burning survey data of 38 indoor furnaces per
100 woodstoves for Illinois, Indiana, Michigan, Ohio, and Wisconsin. More recent year 2007 MN survey
data resulted in the much lower ratio of 7.3 indoor furnaces per 100 wood stove units. Thus, for the other
five MWRPO states previously listed, the indoor furnace emissions are normalized by setting the indoor
furnace count ratio to wood stoves to match the 7.6% reported value in Minnesota. The resulting
adjustment factors reduce the indoor furnace emissions in these states by 67% (Wisconsin) to as much as
83% in Ohio.
The MARAMA states (i.e., CT, DE, DC, ME, MD, MA, NH, NJ, NY, PA, RI, VT) year 2007 RWC
24
-------
inventory was either unchanged from the 2008 NEI, or was missing for most states. The exceptions were
New York and Pennsylvania which includes significantly revised RWC estimates compared to the 2008
NEI. For New York, the MARAMA estimates were not split out into the refined set of 10 RWC
appliance types/SCCs in the NEI. New York only reported "general" fireplaces (SCC=2104008100) and
"EPA certified, non-catalytic" woodstoves (SCC=2104008320). However, similar to the SESARM and
MWRPO improvements, the MARAMA NY RWC estimates were spatially reallocated from urban to
more rural areas and were also lower state-wide than the NEI. For Pennsylvania, MARAMA RWC
estimates were not much different state-wide on the aggregate, but were refined by SCC and spatially
compared to the 2008 NEI. Therefore, the MARAMA 2007 RWC data is used for New York and
Pennsylvania and the 2008 NEI emissions are used for all RWC sources in the rest of the MARAMA
states.
The uniform temporalization from month to day was modified to be day-of-year specific as discussed in
more detail in the emissions modeling section. In short, the SMOKE program (GenTPRO) is used to
distribute annual RWC emissions to the coldest days of the year, using maximum temperature thresholds
by-state and/or by-county. On days where the low temperature does not drop below this threshold, RWC
emissions are zero. Conversely, the program temporally allocates the most relative emissions to the
coldest days. This meteorological-based temporal allocation can have a substantial impact on the amount
of RWC emissions in an area on any given day.
3.2.4 Day-Specific Point Source Fires (ptfire)
Wildfire and prescribed burning emissions are contained in the ptfire sector. The ptfire sector has
emissions provided at geographic coordinates (point locations) and has daily estimates of the emissions
from each fires value. The ptfire sector for the 2009 Platform excludes agricultural burning and other open
burning sources, which are included in the nonpt sector. The agricultural burning and other open burning
sources are in the nonpt sector because these categories were not factored into the development of the
ptfire sector. Additionally, their year-to-year impacts are not as variable as wildfires and non-agricultural
prescribed/managed burns.
The ptfire sector includes a satellite derived latitude/longitude of the fire's origin and other parameters
associated with the emissions such as acres-burned and fuel load, which allow estimation of plume rise.
Note that agricultural burning is not included in the ptfire sector but is included in the nonpt sector. The
point source day-specific emission estimates for 2009 fires rely on the Satellite Mapping Automated
Reanalysis Tool for Fire Incident Reconciliation Version 2 (SMARTFIRE2) system (Raffuse, et al.,
2012). Activity data was used from the Monitoring Trends in Burn Severity (MTBS) project. Incident
Command Summary Reports (1CS-209), and the National Oceanic and Atmospheric Administration's
(NOAA's) Hazard Mapping System (HMS).
The method involves the reconciliation of 1CS-209 reports (Incident Status Summary Reports) with
satellite-based fire detections to determine spatial and temporal information about the fires. The ICS-209
reports for each large wildfire are created daily to enable fire incident commanders to track the status and
resources assigned to each large fire (100 acre timber fire or 300 acre rangeland fire). The SM A RTF I RE
system of reconciliation with ICS-209 reports is described in an Air and Waste Management Association
report (Raffuse, et al., 2007). Once the fire reconciliation process is completed, the emissions are
calculated using the U.S. Forest Service's CONSUMEv3.0 fuel consumption model and the FCCS fuel-
loading database in the BlueSky Framework (Ottmar, et. al., 2007). The detection of fires with this
25
-------
method is satellite-based. Additional sources of information used in the fire classification process
included MODIS satellite and fuel moistures derived from fire weather observational data.
The ICS-209 reports for each large wildfire are created daily to enable fire incident commanders to track
the status and resources assigned to each large fire (100 acre timber fire or 300 acre rangeland fire). Note
that the distinction between wildfire and prescribed burn is not as precise as with ground-based methods.
The fire size was based on the number of satellite pixels and a nominal fire size of 100 acres/pixel was
assumed for a significant number of fire detections when the first detections were not matched to ICS 209
reports, so the fire size information is not as precise as ground-based methods.
The activity data and other information were used within the BlueSky Framework to model vegetation
distribution, fuel consumption, and emission rates, respectively. Latitude and longitude locations were
incorporated as a post processing step. The method to classify fires as WF, WFU, RX (FCCS > 0), and
unclassified (FCCS > 0) involves the reconciliation of ICS-209 reports (Incident Status Summary
Reports) with satellite-based fire detections to determine spatial and temporal information about the fires.
Because the HMS satellite product from NOAA is based on daily detections, the emission inventory
represents a time-integrated emission estimate. For example, a large smoldering fire will show up on
satellite for many days and would count as acres burned on a daily basis; whereas a ground-based method
would count the area burned only once even it burns over many days.
The SMOKE-ready "ORL" inventory files created from the raw daily fires contain both CAPs and HAPs.
The BAFM HAP emissions from the inventory were obtained using VOC speciation profiles (i.e., a "no-
integrate noHAP" use case). The BEIS3.14 model creates gridded, hourly, model-species emissions from
vegetation and soils. It estimates CO, VOC, and NO\ emissions for the U.S., Mexico, and Canada. The
BEIS3.14 model is described further in
http://vvvvvv.cmascenter.oru/conference/2008/slides/pouliot tale two cmasOS.ppt. Additional references
for this method are provided in (McKenzie, et al., 2007), (Ottmar, et al., 2003), (Ottmar, et al., 2006), and
(Anderson et al., 2004).
3.2.5 Biogenic Sources (beis)
For CMAQ, biogenic emissions were computed with the BEIS3.14 model within SMOKE using 2009
meteorological data. The BEIS3.14 model creates gridded, hourly, model-species emissions from vegetation
and soils. It estimates CO, VOC (most notably isoprene, terpine, and sesquiterpene), and NO emissions for the
U.S., Mexico, and Canada. The BEIS3.14 model is described further in:
http://www.cmascenter.org/conference/2008/slides/pouliot tale two cmasOS.ppt.
The inputs to BEIS include:
• Temperature data at 2 meters from the CM AQ meteorological input files,
• Land-use data from the Biogenic Emissions Landuse Database, version 3 (BELD3) that provides
data on the 230 vegetation classes at 1 -km resolution over most of North America.
3.2.6 Mobile Sources (onroad, onroadrfl, nonroad, clc2rail, c3marine)
The 2009 onroad emissions are broken out into two sectors: "onroad" and "onroad rfl". Aircraft
26
-------
emissions are in the nonEGU point inventory. The locomotive and commercial marine emissions are
divided into two sectors: "clc2rail" and "c3 marine", and the "nonroad" sector contains the remaining
nonroad emissions. Note that the 2008 NEI includes state-submitted emissions data for nonroad, but the
modeling performed for this platform does not incorporate state-submitted emissions for the onroad or
nonroad sectors, except for California. All tribal data from the mobile sectors have been dropped because
we do not have spatial surrogate data, and the emissions are small.
The onroad and onroadrfl sectors are processed separately to allow for different spatial allocation to be
applied to onroad refueling via a gas station surrogate, versus onroad vehicles that are spatially allocated
based on roads and population. Except for California, all onroad and onroad refueling emissions are
generated using the SMOKE-MOVES emissions modeling framework that leverages MOVES201 Ob-
generated outputs (http://www.epa.gov/otaq/models/moves/index.htm) and hourly meteorology.
Emissions for onroad (including refueling), nonroad and clc2rail sources in California were provided by
the California Air Resources Board (CARB).
The nonroad sector is based on NMIM except for California which uses data provided by the California
Air Resources Board (CARB). NMIM (EPA, 2005) creates the nonroad emissions on a month-specific
basis that accounts for temperature, fuel types, and other variables that vary by month. The 2009 NMIM
nonroad emissions were generated using updated activity (fuels, vehicle population, etc) data, but are
otherwise similar in methodology to those generated for the 2005 NEI. All nonroad emissions are
compiled at the county/SCC- level. Detailed inventory documentation for the 2008 NEI nonroad sectors is
available at http://www.epa.gOv/ttn/chief/net/2008inventorv.html#inventorvdoc. Neither NMIM nor
MOVES generates tribal data.
The locomotive and commercial marine vessel (CMV) emissions are divided into two nonroad sectors:
"clc2rail" and "c3marine". The c 1 c2rail sector includes all railway and most rail yard emissions as well
as the gasoline and diesel-fueled Class 1 and Class 2 CMV emissions. The c3marine sector emissions
contain the larger residual fueled ocean-going vessel Class 3 CMV emissions and are treated as point
emissions with an elevated release component; all other nonroad emissions are treated as county-specific
low-level emissions (i.e., are in model layer 1). The 2008 NEI c3marine emissions were replaced with a
set of approximately 4-km resolution point source format emissions. These data are used for all states,
including California, as well as offshore and international emissions within our air quality modeling
domain, and are modeled separately as point sources in the "c3marine" sector.
3.2.7 Onroad non-refueling (onroad)
For the Version 5 modeling platform, EPA estimated emissions for every county in the continental U.S.
except for California using similar methods as for the 2008 NEI Versions 2 and 3. The modeling
framework took into account the strong temperature sensitivity of the onroad emissions. Specifically,
county-specific inputs and tools were used that integrated the MOVES model with the SMOKE emission
inventory model to take advantage of the gridded hourly temperature information available from
meteorology modeling used for air quality modeling. This integrated "SMOKE-MOVES" tool was
developed by EPA in 2010 and is in use by states and regional planning organizations for regional air
quality modeling. SMOKE-MOVES requires emission rate "lookup" tables generated by MOVES that
differentiate emissions by process (running, start, vapor venting, etc.), vehicle type, road type,
temperature, speed, hour of day, etc.
27
-------
To generate the MOVES emission rates that could be applied across the U.S., EPA used an automated
process to run MOVES to produce emission factors by temperature and speed for 146 "representative
counties," to which every other county could be mapped as detailed below. Using the MOVES emission
rates, SMOKE selected appropriate emissions rates for each county, hourly temperature, SCC, and speed
bin and multiplied the emission rate by activity (i.e., VMT (vehicle miles travelled) or vehicle
population) to produce emissions. These calculations were done for every county, grid cell, and hour in
the continental United States. SMOKE-MOVES can be used with different versions of the MOVES
model. For the Version 5 modeling platform, EPA used the latest publically released version:
MOVES2010b (http://www.epa.gov/otaq/models/moves/index.htm). The MOVES default database used
was named movesdb20120410.
Using SMOKE-MOVES for creating emissions for modeling requires numerous steps, as described in the
sections below:
• Determine which counties will be used to represent other counties in the MOVES runs.
• Determine which months will be used to represent other month's fuel characteristics.
• Create MOVES inputs needed only for MOVES runs. MOVES requires county-specific
information on vehicle populations, age distributions, and inspection-maintenance programs for
each of the representative counties.
• Create inputs needed both by MOVES and by SMOKE, including a list of year-specific
temperatures and activity data.
• Run MOVES to create emission factor tables using year-specific fuel information.
• Run SMOKE to apply the emission factors to activities to calculate emissions.
• Aggregate the results at the county-SCC level for summaries and quality assurance.
Some data used in the SMOKE-MOVES process is year-specific. When MOVES was run to generate the
emission factors, gasoline and diesel properties for representing counties were based on 2009 fuel
information (i.e., RegionalFuels_2009_20120323). The temperature and humidity inputs were also based
on 2009 values. The VMT used by SMOKE-MOVES was generated by taking 2009 VMT by state and
freeway/non-freeway from FHWA VM-2 tables and allocating to county and month and roadtype using
the 2008 NEI VMT. The VMT was allocated to vehicle type using FHWA's VM-4 table and to MOVES
sourcetype using ratios from MOVES. Vehicle populations were then generated by applying
VMT/vehicle default ratios from MOVES to the VMT. The same speed data used for the 2008 NEI were
also used for this study.
The California emissions were post-processed to incorporate both CARB supplied inventories and the
shape of the meteorologically-based SMOKE-MOVES results by scaling the SMOKE-MOVES
generated totals to match CARB-provided totals. Because CARB provide 2007 and 2011 emissions data,
the data for 2009 were linearly interpolated between 2007 and 2011 levels. For more details on this
process, see the Version 5 platform documentation.
3.2.8 Onroad Refueling (onroad rfl)
Onroad refueling was modeled very similarly to the other onroad emissions. MOVES2010b was used
28
-------
produce emission factors (EFs) for refueling. These EFs are at the resolution of the onroad SCC and
were run separately from the other onroad mobile sources to allow for different spatial allocation. To
facilitate this, the EFs were separated into refueling and non-refueling tables. SMOKE-MOVES was
then run using these EF tables as inputs and the results spatially allocated based on a gas stations spatial
surrogate. For California, the SMOKE-MOVES generated emissions were used for onroad refueling
without any adjustments because there were no CARB-supplied refueling emissions.
3.2.9 Nonroad Mobile Sources — NMIM-Based (nonroad)
The nonroad sector includes monthly exhaust, evaporative and refueling emissions from nonroad engines
(not including commercial marine, aircraft, and locomotives) that are derived from NMIM for all states
except California. NMIM 20090504 was run using 2009 meteorological and fuel data to create county-
SCC emissions by month for the 2009 nonroad mobile CAP and HAP sources. This version of NMIM ran
the NR08a version of NONROAD. The nonroad county database was labeled 20101201 2009. The run
incorporated Bond rule revisions to some of the base case inputs and the Bond Rule controls did not take
effect until future years. NMIM provides nonroad emissions for VOC by three emission modes: exhaust,
evaporative and refueling. Unlike the onroad sector, refueling emissions from nonroad sources are not
separated into a different sector.
EPA default inputs were replaced by state inputs where such data were provided via the 2008 NEI process.
The 2008 NEI documentation describes this and other details of the NMIM nonroad emissions
development. CAPs and only the necessary HAPs for the nonroad sector (i.e., BAFM, butadiene, and
naphthalene) were included. For this study, NMIM was run separately for each county. To aid with the
processing by SMOKE, the mode was appended to the pollutant name and the California NMIM data was
replaced with state-supplied data.
For California, year 2009 nonroad emissions values were interpolated between the 2007 and 201 1
emissions provided by CARB. The CARB-supplied nonroad annual inventory to monthly emissions
values by using the aforementioned EPA NMIM monthly inventories to compute monthly ratios by
pollutant and SCC. Some adjustments to the CARB inventory were needed to convert the provided total
organic gas (TOG) to the VOC that was needed by SMOKE.
3.2.10 Nonroad Mobile Sources: Commercial Marine CI, C2, and Locomotive (clc2rail)
The clc2rail sector contains CAP and HAP emissions from locomotive and commercial marine sources,
except for the category 3/residual-fuel (C3) commercial marine vessels (CMV) found in the c3marine
sector. The "clc2" portion of this sector name refers to the Class I/I I CMV emissions, not the railway
emissions. Railway maintenance emissions are included in the nonroad sector because these are included
in the nonroad NMIM monthly inventories. The C3 CMV emissions are in the c3marine sector. Except
for California, the emissions in the clc2rail sector are year 2008 and are composed of the following
SCCs: 2280002100 (CMV diesel, ports), 2280002200 (CMV diesel, underway), 2285002006
(locomotives diesel line haul Class I), 2285002007 (locomotives diesel line haul Class I I/I 11),
2285002008 (locomotives diesel line haul passenger trains), 2285002009 (locomotives diesel line haul
commuter lines), and 2285002010 (locomotives diesel, yard).
The 2008 NEI Version 2 was the starting point for this sector, but several adjustments were made. First,
29
-------
the 2008 NEI point inventory contains rail yard emissions for several states and counties. The NEI point
and nonpoint inventories were reviewed for counties with significant rail yard emissions in both
inventories. It was assumed that the point inventory contained more accurate information when both
inventories contained rail yard emissions. Therefore, nonpoint rail yards were removed from the clc2rail
sector for certain counties in California, Maryland, Oregon and Arizona. For more information, see the
Version 5 2007 platform documentation.
Analysis of the total rail emissions in the 2008 NEI showed what appeared to be missing rail line
emissions in Texas. It was determined that line haul emissions from Texas were essentially zero in the
2008 NEI. Therefore, all line haul emissions from the 2008 NEI were removed and information from an
EPA default dataset of Texas line haul emissions was added. These EPA line haul emissions are
restricted to the Class I and Class I I/I 11 operations and add approximately 52,000 tons of NOX to Texas
that would otherwise be missing.
For several Texas counties, the C1/C2 CMV emissions in the 2008 NEI included EPA gap filled values
where shape IDs were not populated on submittal. The intended Texas submittal was often much smaller
than the EPA-estimated default value for several counties. An example of this is Harris county
(FIPS=48201) where the Texas submittal was approximately 1,200 tons of NOX for port and underway
emissions but not all shape IDs were included. The NEI methodology used EPA emissions where Texas
did not provide estimates and the resulting double count and overestimate of this top-down method
resulted in over 49,000 tons of NOX in the 2008 NEI in Harris County, Texas. Therefore, the modeling
platform used the original Texas submittal, did not append any EPA emissions, and summed up port and
underway for the modeling files to the county level. Similar corrections to these may have been included
in Version 3 of the 2008 NEI. Other states were impacted by a similar error in the 2008 NEI Version 2,
but for many of these states alternative data were used as discussed below.
For California, the California Air Resources Board (CARB) provided year 2007 and 201 1 emissions for
all mobile sources, including C1/C2 CMV and rail. These emissions are documented in a staff report
available at: http://www.arb.ca.gov/regact/2010/offroadlsi 10/offroadisor.pdf. The modeling platform uses
2009 emissions interpolated between the 2007 and 201 1 emissions. The C1/C2 CMV emissions were
obtained from the CARB nonroad mobile dataset and include the regulations to reduce emissions from
diesel engines on commercial harbor craft operated within California waters and 24 nautical miles of the
California baseline. These emissions were developed using Version 1 of the CEPAM that supports
various California off-road regulations. The locomotive emissions were obtained from the CARB trains
dataset "ARMJ_RF#2002_ANNU AL_TRAINS.txt". Documentation of the CARB offroad mobile
methodology, including clc2rail sector data, is provided here:
http://www.arb.ca.gOv/msei/categories.htm#offroad motor vehicles. The CARB inventory TOG
emissions were converted to VOC by dividing the inventory TOG by the available VOC-to-TOG
speciation factor.
Year-2007 inventories provided by MARAMA, SESARM and the MWRPO were used for the clc2rail
sector emissions in their respective states. Emissions data from MARAMA rather than SESARM was
used for Virginia because the SESARM data included some rather large emissions for Commuter Lines
(SCC=2285002009) that were not reflected in the 2008 NEI nor the M AR AM A dataset. The MWRPO
year-2007 clc2rail data were obtained from a subset of their version 7 emissions modeling file
"nrinv.mwrpo_alm.baseCv7.annual.orl.txt", where MWRPO NEI Inventory Format (NIF)-formatted data
30
-------
were converted to SMOKE ORL format. The MAR AM A dataset was obtained from a subset of their
version 3.3 January 27, 2012 vintage file "ARINV_2007_MAR_Jan2012.txt". The SESARM dataset
was obtained from a subset of the file "nrinv.aim.semap.base07.v093010.orl.txt" developed for the
Southeastern Modeling, Analysis, and Planning (SEMAP) project. All RPO datasets were edited to
remove non-c 1 c2rail sources.
3.2.11 Nonroad mobile sources: C3 commercial marine (c3marine)
The c3marine sector emissions data were developed based on a 4-km resolution ASCII raster format
dataset used since the Emissions Control Area-International Marine Organization (ECA-IMO) project
began in 2005, then known as the Sulfur Emissions Control Area (SECA). These emissions consist of
large marine diesel engines (at or above 30 liters/cylinder) that until very recently, were allowed to meet
relatively modest emission requirements, often burning residual fuel. The emissions in this sector are
comprised of primarily foreign-flagged ocean-going vessels, referred to as Category 3 (C3) CMV ships.
The c3marine inventory includes these ships in several intra-port modes (cruising, hoteling, reduced
speed zone, maneuvering, and idling) and underway mode and includes near-port auxiliary engines. An
overview of the C3 ECA Proposal to the International Maritime Organization (EPA-420-F-10-041,
August 2010) project and future-year goals for reduction of NOX, S02, and PM C3 emissions can be
found at: http://www.epa.gov/oms/regs/nonroad/marine/ci/420r09019.pdf. The resulting ECA-IMO
coordinated strategy, including emission standards under the Clean Air Act for new marine diesel engines
with per-cylinder displacement at or above 30 liters, and the establishment of Emission Control Areas is
at: http://www.epa.gov/oms/oceanvessels.htm.
The ECA-IMO emissions data were converted to SMOKE point-source ORL input format as described in
http://vvvvvv.epa.gov/ttn/chief/conference/ei 17/session6/mason.pdf. thereby allowing for the emissions to
be allocated to modeling layers above the surface layer. As described in the paper, the ASCII raster
dataset was converted to latitude-longitude, mapped to state/county FIPS codes that extended up to 200
nautical miles (nm) from the coast, assigned stack parameters, and monthly ASCII raster dataset
emissions were used to create monthly temporal profiles. Counties were assigned as extending up to
200nm from the coast because this was the distance to the edge of the U.S. Exclusive Economic Zone
(EEZ), a distance that defines the outer limits of ECA-IMO controls for these vessels. All non-US
emissions (i.e., in waters considered outside of the 200nm EEZ, and hence out of the U.S. territory) are
assigned a dummy state/county FIPS code=98001. The SMOKE-ready data were cropped from the
original EC A-IMO data to cover only the 36-ktn CMAQ domain, which is the largest domain used for this
effort, and larger than the 12km domain used in this project.
The base year ECA inventory is 2002 and consists of these CAPs: PM10, PM2.5, CO, C02, NH3, NOX,
SOX (assumed to be S02), and Hydrocarbons (assumed to be VOC). The EPA developed regional
growth (activity-based) factors that we applied to create the 2007v5 inventory from the 2002 data. These
growth factors are provided in Table 3-4. The East Coast and Gulf Coast regions were divided along a
line roughly through Key Largo (longitude 80° 26' West).
31
-------
Table 3-4. Growth factors to project the 2002 ECA inventory to 2009
Region
EEZ FIPS
NOx
PMio
pm25
VOC
CO
so2
East Coast (EC)
85004
1.284
1.374
1.376
1.374
1.374
1.374
Gulf Coast (GC)
85003
1.137
1.217
1.214
1.216
1.217
1.217
North Pacific (NP)
85001
1.193
1.268
1.250
1.268
1.268
1.268
South Pacific (SP)
85002
1.334
1.429
1.427
1.417
1.415
1.434
Great Lakes (GL)
n/a
1.108
1.137
1.137
1.138
1.137
1.137
Outside ECA
98001
1.252
1.338
1.338
1.338
1.338
1.338
A modification to the original ECA-IMO c3marine dataset include updating the state of Delaware county
total emissions to reflect comments received during the Cross-State Air Pollution Rule (CSAPR)
emissions modeling platform development: http://www.epa.gOv/ttn/chief/emch/index.html#final. The
original ECA-IMO inventory also did not delineate between ports and underway (or other C3 modes such
as hoteling, maneuvering, reduced-speed zone, and idling) emissions; however, we used a U.S. ports
spatial surrogate dataset to assign the ECA-IMO emissions to ports and underway SCCs - 2280003100
and 2280003200, respectively. This has no effect on temporal allocation or speciation because all C3
emissions, unclassified/total, port and underway, share the same temporal and speciation profiles.
Canadian near-shore emissions were assigned to province-level FIPS codes and paired those to region
classifications for British Columbia (North Pacific), Ontario (Great Lakes) and Nova Scotia (East Coast).
The assignment of U.S. FIPS was also restricted to state-federal water boundaries data from the Mineral
Management Service (MMS) that extended only (approximately) 3 to 10 miles off shore. Emissions
outside the 3 to 10 mile MMS boundary but within the approximately 200 nm EEZ boundary in Figure 2
8 were projected to year 2009 using the same regional adjustment factors as the U.S. emissions; however,
the FIPS codes were assigned as "EEZ" FIPS. Note that state boundaries in the Great Lakes are an
exception, extending through the middle of each lake such that all emissions in the Great Lakes are
assigned to a U.S. county or Ontario. The classification of emissions to U.S. and Canadian FIPS codes is
primarily needed only for inventory summaries and is irrelevant for air quality modeling except
potentially for source apportionment of states contributions to transport.
Factors were applied to compute HAP emissions (based on emissions ratios) to VOC to obtain HAP
emissions values. Table 3-5 below shows these factors. Because HAPs were computed directly from the
CAP inventory and the calculations are therefore consistent, the entire c3marine sector utilizes CAP-HAP
VOC integration to use the VOC HAP species directly, rather than VOC speciation profiles.
Table 3-5. HAP emission ratios for generation of HAP emissions from criteria emissions for C3
commercial marine vessels
Pollutant
Apply to
Pollutant
Code
Factor
Acetaldehyde
VOC
75070
0.0002286
Benzene
VOC
71432
9.80E-06
Formaldehyde
VOC
50000
0.0015672
32
-------
3.2.12 Emissions from Canada, Mexico and Offshore Drilling Platforms (othpt, othar, othon)
The emissions from Canada, Mexico, and offshore drilling platforms are included as part of three
emissions modeling sectors: othpt, othar, and othon. The "oth" refers to the fact that these emissions are
usually "other" than those in the U.S. state-county geographic FIPS code, and the third and fourth
characters provide the SMOKE source types: "pt" for point, "ar" for "area and nonroad mobile", and
"on" for onroad mobile. All "oth" emissions are CAP-only inventories.
For Canada, year-2006 Canadian emissions were used but several modifications were applied to the
inventories:
1. Wildfires or prescribed burning were not included because Canada does not include these
inventory data in their modeling.
2. In-flight aircraft emissions were not included because we do not include these for the U.S. and we
do not have a finalized approach to include in our modeling.
3. A 75% reduction ("transport fraction") was applied to PM for the road dust, agricultural, and
construction emissions in the Canadian "afdust" inventory. This approach is more simplistic than
the county-specific approach used for the U.S., but a comparable approach was not available for
Canada.
4. Speciated V OC emissions from the A DOM chemical mechanism were not included because we
use speciated emissions from the CB5 chemical mechanism that Canada also provided.
5. Residual fuel CMV (C3) SCCs (22800030X0) were removed because these emissions are
included in the c3marine sector, which covers not only emissions close to Canada but also
emissions far at sea. Canada was involved in the inventory development of the c3marine sector
emissions.
6. Wind erosion (SCC=2730100000) and cigarette smoke (SCC=2810060000) emissions were
removed from the nonpoint (nonpt) inventory; these emissions are also absent from our U.S.
inventory.
7. Quebec PM2.5 emissions (2,000 tons/yr) were removed for one SCC (2305070000) for Industrial
Processes, Mineral Processes, Gypsum, and Plaster Products due to corrupt fields after conversion
to SMOKE input format. This error should be corrected in a future inventory.
8. Excessively high CO emissions were removed from Babine Forest Products Ltd (British
Columbia SMOKE plantid="5 188") in the point inventory.
9. The county part of the state/county FIPS code field in the SMOKE inputs were modified in the
point inventory from "000" to "001" to enable matching to existing temporal profiles.
For Mexico, year 2008 emissions were used that are projections of their 1999 inventory originally
developed by Eastern Research Group Inc., (ERG, 2006) as part of a partnership between Mexico's
Secretariat of the Environment and Natural Resources (Secretaria de Medio Ambiente y Recursos
Naturales-SEMARNAT) and National Institute of Ecology (Instituto National de Ecologia-INE), the
U.S. EPA, the Western Governors' Association (WGA), and the North American Commission for
Environmental Cooperation (CEC). This inventory includes emissions from all states in Mexico. A
background on the development of year-2008 Mexico emissions from the 1999 inventory is available at:
http://www.wrapair.org/forums/ef/inventories/MNEI/index.html.
33
-------
The offshore emissions include point source offshore oil and gas drilling platforms. We used emissions
from the 2008 NEI point source inventory. The offshore sources were provided by the Mineral
Management Services (MMS).
3.2.13 SMOKE-ready non-anthropogenic chlorine inventory
The ocean chlorine gas emission estimates are based on the build-up of molecular chlorine (C12)
concentrations in oceanic air masses (Bullock and Brehme, 2002). Data at 36 km and 12 km resolution
were available and were not modified other than the name "CHLORINE" was changed to "CL2" because
that is the name required by the CMAQ model. The same data was used as in the CAP and HAP 2002-
based Platform was used. See ftp://ftp.epa.gov/EmisInventory/2002v3CAPHAP/ documentation for
additional details.
3.3 Emissions Modeling Summary
CMAQ requires emissions data to be input as hourly rates of specific gas and particle species for the
horizontal and vertical grid cells contained within the modeled region (i.e., modeling domain). To
provide emissions in the form and format required by the model, it is necessary to "pre-process" the
"raw" emissions (i.e., emissions input to SMOKE) for the sectors described above. In brief, the process
of emissions modeling transforms the emissions inventories from their original temporal resolution,
pollutant resolution, and spatial resolution into the hourly, speciated, gridded resolution required by the
air quality model. The pre-processing steps involving temporal allocation, spatial allocation, pollutant
speciation, and vertical allocation of point sources are referred to as emissions modeling.
The temporal resolution of the emissions inventories input to SMOKE for the modeling platform varies
across sectors, and may be hourly, monthly, or annual total emissions. The spatial resolution, which also
can be different for different sectors, may be at the level of individual point sources, county totals,
province totals for Canada, or municipio totals for Mexico. This section provides some basic information
about the tools and data files used for emissions modeling as part of the Version 5 platform. The
emissions inventories were discussed in detail earlier. Therefore, we have limited the descriptions of data
in this section to the ancillary data SMOKE uses to perform the emissions modeling steps.
3.3.1 The SMOKE Modeling System
For this study, emission inventories were processed into CMAQ-ready inputs using SMOKE version 3.1.
SMOKE executables and source code are available from the Community Multiscale Analysis System
(CMAS) Center at http://www.cinascenter.org. Additional information about SMOKE is available from
http ://www. smoke-model. org. For sectors that have plume rise, the in-line emissions capability of CMAQ
was used, and therefore source-based emissions files were created rather than the much larger three-
dimensional files. For quality assurance purposes, emissions totals by specie for the entire model domain
are output as reports that are then compared to inventory level reports generated by SMOKE to ensure
mass is not lost or gained during this conversion process.
3.3.2 Key Emissions Modeling Settings
When preparing emissions for the air quality model, emissions for each sector are processed separately
34
-------
through SMOKE, and then the final merge program (Mrggrid) is run to combine the model-ready, sector-
specific emissions across sectors. The SMOKE settings in the run scripts and the data in the SMOKE
ancillary files control the approaches used for the individual SMOKE programs for each sector. Table 3-
6 summarizes the major processing steps of each platform sector. The "Spatial" column shows the spatial
approach: "point" indicates that SMOKE maps the source from a point location (i.e., latitude and
longitude) to a grid cell; "surrogates" indicates that some or all of the sources use spatial surrogates to
allocate county emissions to grid cells; and "area-to-point" indicates that some of the sources use the
SMOKE area-to-point feature to grid the emissions. The "Speciation" column indicates that all sectors
use the SMOKE speciation step, though biogenics speciation is done within BEIS3 and not as a separate
SMOKE step. The "Inventory resolution" column shows the inventory temporal resolution from which
SMOKE needs to calculate hourly emissions. Note that for some sectors (e.g., onroad, beis), there is no
input inventory. Instead activity data and emission factors are used in combination with meteorological
data to compute hourly emissions.
Finally, the "plume rise" column indicates the sectors for which the "in-line" approach is used. These
sectors are the only ones which will have emissions in aloft layers, based on plume rise. The term "in-
line" means that the plume rise calculations are done inside of the air quality model instead of being
computed by SMOKE. The air quality model computes the plume rise using the stack data and the
hourly air quality model inputs found in the SMOKE output files for each model-ready emissions sector.
The height of the plume rise determines the model layer into which the emissions are placed. The
c3marine and ptfire sectors are the only sectors with only "in-line" emissions, meaning that all of the
emissions are placed in aloft layers and thus there are no emissions for those sectors in the two-
dimensional, layer-1 files created by SMOKE. In addition to the other settings, no grouping of stacks was
performed using the PELVCONFIG file because grouping done for "in-line" processing will not give
identical results as "offline" (i.e., processing whereby SMOKE creates 3-dimensional files). The only
way to get the same results between in-line and offline is to choose to have no grouping.
Table 3-6. Key emissions modeling steps by sector
Platform sector
Spatial
Speciation
Inventory
resolution
Plume rise
Ptipm
Point
Yes
daily & hourly
in-line
Ptnonipm
Point
Yes
annual
in-line
Ptfire
Point
Yes
Daily
in-line
Othpt
Point
Yes
annual
in-line
c3 marine
Point
Yes
annual
in-line
Ag
Surrogates
Yes
annual & monthly
Afdust
Surrogates
Yes
annual
Beis
pre-gridded landuse
in BEIS
computed hourly
clc2rail
Surrogates
Yes
annual
Nonpt
surrogates & area-to-
point
Yes
annual & monthly for
ag burning and
SESARM open
Nonroad
surrogates & area-to-
nnint
Yes
monthly
point
35
-------
Onroad
Surrogates
Yes
computed hourly
onroadrfl
Surrogates
Yes
computed hourly
Otliar
Surrogates
Yes
annual
Othon Surrogates Yes annual
3.3.3 Spatial Configuration
For this study, SMOKE and CMAQ were run for a 12-km modeling domain shown in Figure 3-1
(12US1). The grid used a Lambert-Conformal projection, with Alpha = 33, Beta = 45 and Gamma = -97,
with a center of X = -97 and Y = 40. Later sections provide details on the spatial surrogates and area-to-
point data used to accomplish spatial allocation with SMOKE.
12km CONUS nationwic
x,y: -2556000.-1728000
col: 459 row: 299
Figure 3-1. CMAQ Modeling Domain
3.3.4 Chemical Speciation Configuration
The emissions modeling step for chemical speciation creates "model species" needed by the air quality
model for a specific chemical mechanism. These model species are either individual chemical compounds
or groups of species, called "model species." The chemical mechanism used for this study is the Carbon
Bond 05 (CB05) mechanism (Yarwood, 2005) with secondary organic aerosol (SOA) and FfONO
36
-------
enhancements as described in http://www.cmascenter.org/help/model docs/cinaq/4.7/
RELEASE N0TES.txt. The mapping of inventory pollutants to model species is shown in Table 3-7.
From the perspective of emissions preparation, the CB05 with SOA mechanism is the same as was used
in the 2005 platform. It should be noted that the BENZENE model species is not part of CB05 in that the
concentrations of BENZENE do not provide any feedback into the chemical reactions (i.e., it is not
"inside" the chemical mechanism). Rather, benzene is used as a reactive tracer and as such is impacted
by the CB05 chemistry. BENZENE, along with several reactive CB05 species (such as TOL and XYL)
plays a role in SOA formation in CMAQ 4.7.
Table 3-7. Model Species Produced by SMOKE for CB05
Inventory Pollutant
Model Species
Model Species Description
CO
CO
Carbon monoxide
NOx
NO
Nitrogen oxide
no2
Nitrogen dioxide
S02
S02
Sulfur dioxide
SULF
Sulfuric acid vapor
nh3
nh3
Ammonia
VOC
ALD2
Acetaldehyde
ALDX
Propionaldehyde and higher aldehydes
ETH
Ethene
ETHA
Ethane
ETOH
Ethanol
FORM
Formaldehyde
IOLE
Internal olefin carbon bond (R-C=C-R)
ISOP
Isoprene
MEOH
Methanol
OLE
Terminal olefin carbon bond (R-C=C)
PAR
Paraffin carbon bond
TOL
Toluene and other monoalkyl aromatics
XYL
Xylene and other polyalkyl aromatics
Various additional VOC
species from the biogenics
model which do not map to
TERP
Terpenes
the above model species
PMio
PMC
Coarse PM >2.5 microns and <10 microns
pm2,
PEC
Particulate elemental carbon <2.5 microns
pno3
Particulate nitrate <2.5 microns
POC
Particulate organic carbon (carbon only) <2.5
microns
PS04
Particulate sulfate < 2.5 microns
PMFINE
Other particulate matter <2.5 microns
The approach for speciating PM2.5 emissions supports both CMAQ 4.7.1 with five species (i.e., AE5)
and CMAQ 5.0 that includes speciation of PM2.5 into 17 PM model species (i.e., AE6). The TOG and
37
-------
PM2.5 speciation factors that are the basis of the chemical speciation approach were developed from the
SPECIATE4.3 database (http ://www.epa.gov/ttn/chief/software/speciate) and is the EPA's repository of
TOG and PM speciation profiles of air pollution sources. A few of the profiles used in the v5 platform
will be published in later versions of the SPECIATE database. The SPECIATE database development
and maintenance is a collaboration involving the EPA's ORD, OTAQ, and the Office of Air Quality
Planning and Standards (OAQPS), and Environment Canada (EPA, 2006a). The SPECIATE database
contains speciation profiles for TOG, speciated into individual chemical compounds, VOC-to-TOG
conversion factors associated with the TOG profiles, and speciation profiles for PM2 5. The database also
contains the PM2.5, speciated into both individual chemical compounds (e.g., zinc, potassium, manganese,
lead) and into the "simplified" PM2 5 components used in the air quality model. These simplified
components for AE5 are:
• PSO4 : primary particulate sulfate
• PN03: primary particulate nitrate
• PEC: primary particulate elemental carbon
• POC: primary particulate organic carbon
• PMFINE: other primary particulate, less than 2.5 micrograms in diameter
NOX can be speciated into NO, N02, and/or HONO. For the non-mobile sources, a single profile is used
"NHONO" to split NOX into NO and N02 with 10% N02 and 90% NO. For the mobile sources except
for onroad (including nonroad, clc2rail, c3marine, othon sectors) and for specific SCCs in othar and
ptnonipm, the profile "HONO" splits NOX into NO, N02, and HONO with 90% NO, 9.2% N02 and
0.8% HONO. The onroad sector does not use the "HONO" profile to speciate NOX. Instead,
MOVES2010b produces speciated NO, N02, and HONO by source, including emission factors for these
species in the emission factor tables used by SMOKE-MOVES. Within MOVES, the HONO fraction is a
constant 0.008 of NOX. The NO fraction varies by heavy duty versus light duty, fuel type, and model
year. The N02 fraction = 1 - NO - HONO. For more details on the NOX fractions within MOVES, see
http://vvvvvv.epa.uov/otaq/models/moves/documents/420rl2022.pdf. The SMOKE-MOVES system is
configured to model these species directly without further speciation.
The approach for speciating VOC emissions from non-biogenic sources has the following characteristics:
1) for some sources, HAP emissions are used in the speciation process to allow integration of VOC and
HAP emissions in the NEI; and, 2) for some mobile sources, "combination" profiles are specified by
county and month and emission mode (e.g., exhaust, evaporative). SMOKE computes the resultant profile
on-the-fly given the fraction of each specific profile specified for the particular county, month and
emission mode. The SMOKE feature called the GSPROCOMBO file supports this approach.
The VOC speciation approach for the 2009 Platform includes HAP emissions from the NEI in the
speciation process for some sectors. That is instead of speciating VOC to generate all of the species
needed by the model, emissions of the 4 HAPs, benzene, acetaldehyde, formaldehyde and methanol
(BAFM) from the NEI were integrated with the NEI VOC. The integration process combines the BAFM
HAPs with the VOC in a way that does not double-count emissions and uses the BAFM directly in the
speciation process. Generally, the HAP emissions from the NEI are believed to be more representative of
emissions of these compounds than their generation via VOC speciation.
38
-------
The BAFM HAPs were chosen for this special treatment because, with the exception of BENZENE, they
are the only explicit VOC HAPs in the base version of CMAQ 4.7 model. By "explicit VOC HAPs," we
mean model species that participate in the modeled chemistry using the CB05 chemical mechanism. The
use of these HAP emission estimates along with VOC is called "HAP-CAP integration". BENZENE was
chosen because it was added as a model species in the base version of CMAQ 4.7, and there was a desire
to keep its emissions consistent between multi- pollutant and base versions of CMAQ.
For specific sources, especially within the on road and onroadrfl sectors, we included ethanol in our
integration. To differentiate when a source was integrating BAFM versus EBAFM (ethanol in addition to
BAFM), the speciation profiles which do not include ethanol are referred to as an "E-profile", for
example E10 headspace gasoline evaporative speciation profile 8763 where ethanol is speciated from
VOC, versus 8763 E where ethanol is obtained directly from the inventory. The specific profiles used in
2009 are the same as used for the 2007 platform (see 2007 speciation in Table 3-6 in the 2007v5 TSD).
The only differences between 2009 and 2007 are the GSPRO COMBOs, which represent a different
mixture of E0 and E10 by county between the two modeling years.
The integration of HAP VOC with VOC is a feature available in SMOKE for all inventory formats other
than PTDAY (the format used for the ptfire sector). SMOKE allows the user to specify the particular
HAPs to integrate and the particular sources to integrate. The HAPs to integrate are specified in the
INVTABLE file, and the sources to integrate are based on the NHAPEXCLUDE file (which lists the
sources that are excluded from integration). For the "integrate" sources, SMOKE subtracts the "integrate
" HAPs from the VOC (at the source level) to compute emissions for the new pollutant
"NONHAPVOC." The user provides NONHAPVOC-to-NONHAPTOG factors and NONHAPTOG
speciation profiles. SMOKE computes NONHAPTOG and then applies the speciation profiles to allocate
the NONHAPTOG to the other CMAQ VOC species not including the integrated HAPs.
CAP-HAP integration was considered for all sectors and "integration criteria" were developed for some of
those. Table 3-8 summarizes the integration approach for each platform sector. For the clc2rail sector, the
integration criteria were (1) that the source had to have at least one of the 4 HAPs and (2) that the sum of
BAFM could not exceed the VOC emissions. For the nonpt sector, the following integration criteria were
used to determine the sources to integrate:
1. Any source for which the sum of B, A, F, or M is greater than the VOC was not integrated, since
this clearly identifies sources for which there is an inconsistency between VOC and VOC HAPs.
2. For some source categories (those that comprised 80% of the V OC emissions), sources were
selected for integration in the category per specific criteria. For most of these source categories,
sources may be integrated if they had the minimum combination of B, A, F, and M. For some
source categories, all sources were designated as "no-integrate".
3. For source categories that do not comprise the top 80% of VOC emissions, as long as the source
has emissions of one of the B, F, A or M pollutants, then it can be integrated.
39
-------
Table 3-8. Integration status of benzene, acetaldehyde, formaldehyde and methanol (BAFM) for
each platform sector
Platform
Sector
Approach for Integrating NEI emissions of Benzene (B), Acetaldehyde (A), Formaldehyde (F) and
Methanol (M)
Ptipm
No integration because emissions of BAFM are relatively small for this sector
Ptnonipm
No integration because emissions of BAFM are relatively small for this sector and it is not
expected that criteria for integration would be met by a significant number of sources
Ptfire
No integration.
Ag
N/A—sector contains no VOC
Afdust
N/A—sector contains no VOC
Biog
N/A—sector contains no inventory pollutant "VOC"; but rather specific VOC species
Clc2rail
Partial integration
C3 marine
Full integration
Nonpt
Partial integration
Nonroad
Partial integration—did not integrate California emissions, CNG orLPG sources (SCCs
beginning with 2268 or 2267) because NMIM computed only VOC and not any HAPs for these
SCCs.
Onroad
Full integration
Othar
No integration—not the NEI
Othon
No integration—not the NEI
Othpt
No integration—not the NEI
The SMOKE feature to compute speciation profiles from mixtures of other profiles in user-specified
proportions was used in this project. The combinations are specified in the GSPROCOMBO ancillary
file by pollutant (including pollutant mode, e.g., EXH VOC). state and county (i.e., state/county FIPS
code) and time period (i.e., month). This feature was used for onroad and nonroad mobile and gasoline-
related related stationary sources. Since the ethanol content varies spatially (e.g., by state or sources use
fuels with varying ethanol content, and therefore the speciation profiles require different combinations of
gasoline and E10 profiles by county), temporally (e.g., by month) and by modeling year (i.e., future years
have more thanol) the combo feature allows combinations to be specified at various levels for different
years.
The INVTABLE and NHAPEXCLUDE SMOKE input files have a critical function in the VOC
speciation process for emissions modeling cases utilizing HAP-CAP integration, as is done for the 2009
Platform. Two different types of INVTABLE files were developed to use with different sectors of the
platform. For sectors in which we chose no integration across the entire sector a "no HAP use"
INVTABLE was developed in which the "KEEP" flag is set to "N" for BAFM pollutants. Thus, any
BAFM pollutants in the inventory input into SMOKE are dropped. This both avoids double-counting of
these species and assumes that the VOC speciation is the best available approach for these species for the
sectors using the approach. The second INVTABLE is used for sectors in which one or more sources are
integrated and causes SMOKE to keep the BAFM pollutants and indicates that they are to be integrated
with VOC (by setting the "VOC or TOG component" field to "V" for all four HAP pollutants. This
integrate INVTABLE is further differentiated into sectors that integrate BAFM versus those that integrate
EBAFM (e.g., the onroad and onroad rfl sectors).
40
-------
Unlike other sectors, the onroad sector has pre-speciated PM. This speciated PM comes from the
MOVES model and is processed through the SMOKE-MOVES system. Unfortunately, the
MOVES2010b speciated PM does not map 1-to-l to either the AE5 or AE6 species. Table 3-9 shows the
relationship between MOVES2010b exhaust PM2.5 related species and CMAQ AE5 PM species.
Table 3-9. MOVES exhaust PM species versus AE5 species
MOVES2010b Pollutant Name
Variable
name for
Equations
Relation to AE5 model species
Primary Exhaust PM2.5 - Total
PM25 TOTAL
Primary PM2.5 - Organic Carbon
PM250M
Sum of POC, PN03 and
PMFINE
Primary PM2.5 - Elemental
Carbon
PM25EC
PEC
Primary PM2.5 - Sulfate
Particulate
PM25S04
PS04
MOVES species are related as follows:
PM25TOTAL = PM25EC + PM250M + PS04
The five CMAQ AE5 species also sum to total PM2.5:
PM2.5 = P0C+PEC+PN03+PS04+PMFINE
The basic problem is to differentiate MOVES species "PM250M" into the component AE5 species
(POC, PN03 and PMFINE). The Moves2smkEF post-processor script takes the MOVES2010b species
(EF tables) and calculates the appropriate AE5 PM2.5 species and converts them into a format that is
appropriate for SMOKE (see http://www.smoke-model.org/version3. l/html/ch05s02s04.html for details
on the Moves2smkEF script).
For brake wear and tire wear PM, total PM2.5 (not speciated) comes directly from MOVES2010b. These
PM modes are speciated by SMOKE. PMFINE from onroad exhaust is further speciated by SMOKE into
the component AE6 species.
Speciation profiles for use with BEIS are not included in SPECIATE. The 2009 Platform uses BE1S3.14
and includes a species (SESQ) that was not in B El S3.13 (the version used for the 2002 Platform). This
species was mapped to the CMAQ species SESQT. The profile code associated with BE1S3.14 profiles for
use with CB05 was "B10C5."
3.3.4 Temporal Processing Configuration
Temporal allocation or temporalization is the process of distributing aggregated emissions to a finer
temporal resolution, such as converting annual emissions to hourly emissions. While the total emissions
are important, the timing of the occurrence of emissions is also essential for accurately simulating ozone,
PM, and other pollutant concentrations in the atmosphere. Typically, emissions inventories are annual or
monthly in nature. Temporalization takes these annual emissions and distributes them to the month, the
monthly emissions to the day, and the daily emissions to the hour. This process is typically done by
41
-------
applying temporal profiles—monthly, day of the week, and diurnal—to the inventories.
The monthly, weekly, and diurnal temporal profiles and associated cross references used to create the
hourly emissions inputs for the air quality model were similar to those used for the 2005v4.3 platform.
Some new methodologies are introduced in this platform and updated profiles are discussed. Temporal
factors are typically applied to the inventory by some combination of country, state, county, SCC, and
pollutant.
Table 3-10 summarizes the temporal aspect of the emissions processing configuration. It compares the key
approaches used for temporal processing across the sectors. The temporal aspects of SMOKE processing
are controlled through (a) the scripts T TYPE (Temporal type) and M TYPE (Mergetype) settings and (b)
ancillary data files. In the table, "Daily temporal approach" refers to the temporal approach for getting
daily emissions from the inventory using the Temporal program. The "Merge processing approach" refers
to the days used to represent other days in the month for the merge step. If not "all", then the SMOKE
merge step runs only for representative days, which could include holidays as indicated by the right-most
column. In addition to the resolution, temporal processing includes a ramp-up period for several days
prior to January 1, 2009, intended to mitigate the effects of initial condition concentrations. The ramp up
period for the national 12km grid was 10 days. For most sectors, the emissions from late December of
2008 were used to provide emissions for the end of December, 2009.
The Flat File 2010 format (FF10) is a new inventory format for SMOKE that provides a more
consolidated format for monthly, daily, and hourly emissions inventories. Previously, 12 separate
inventory files would be required to process monthly inventory data. With the FF 10 format, a single
inventory file can contain emissions for all 12 months and the annual emissions in a single record. This
helps simplify the management of numerous inventories. Similarly, individual records contain data for
all days in a month and all hours in a day in the daily and hourly FF 10 inventories, respectively.
SMOKE 3.1 prevents the application of temporal profiles on top of the "native" resolution of the
inventory. For example, a monthly inventory should not have annual to month temporalization applied;
rather, it should only have month to day and diurnal temporalization. This becomes particularly
important when specific sectors have a mix of annual, monthly, daily, and/or hourly inventories (e.g. the
nonpt sector). The flags that control temporalization for a mixed set of inventories are discussed in the
SMOKE documentation.
42
-------
Table 3-10. Temporal Settings Used for the Platform Sectors in SMOKE
Daily
Process
Monthly
temporal
Merge
Holidays as
Inventory
profiles
approach
processing
separate
Platform sector
resolution
used?
1,2
approach 1,3
days?
Ptipm
daily & hourly
All
all
yes
Ptnonipm
annual
yes
Mwdss
all
yes
Ptfire
Daily
All
all
yes
Ag
annual & monthly
yes
all
all
yes
Afdust
annual
yes
Week
all
yes
Beis
hourly
n/a
all
yes
c3 marine
annual
yes
Aveday
aveday
clc2rail
annual
yes
Mwdss
mwdss
Nonpt
annual & monthly
yes
All
all
yes
Nonroad
monthly
Mwdss
mwdss
yes
Onroad
annual & monthly3
all
all
yes
onroadrfl
annual & monthly3
All
all
yes
Othar
annual
yes
Week
week
Othon
annual
yes
Week
week
Othpt
annual
yes
Mwdss
mwdss
1 Definitions for processing resolution:
all = hourly emissions computed for every day of the year
week = hourly emissions computed for all days in one "representative" week, representing all weeks for each month, which means
emissions have day-of-week variation, but not week-to-week variation within the month
mwdss = hourly emissions for one representative Monday, representative weekday, representative Saturday and representative Sunday for
each month, which means emissions have variation between Mondays, other weekdays, Saturdays and Sundays within the month, but not
week-to-week variation within the month. Also Tuesdays, Wednesdays and Thursdays are treated the same.
aveday = hourly emissions computed for one representative day of each month, which means emissions for all days of each month are the
same.
2 Daily temporal approach refers to the temporal approach for getting daily emissions from the inventory using the Temporal program. The
values given are the values of the L_TYPE setting.
3 Merge processing approach refers to the days used to represent other days in the month for the merge step. If not "all", then the SMOKE
merge step just rim for representative days, which could include holidays as indicated by the rightmost column. The values given are the
values of the M_TYPE setting.
" For onroad and onroad_rfl, the annual and monthly refers to activity data (VMT and VPOP). Emissions are computed on an hourly basis.
For the EGU emissions in the ptipm sector, hourly CEM NOx and SO2 data were used directly for
sources that match CEMs. For other pollutants, hourly CEM heat input data were used to allocate the
NEI annual values. For sources not matching CEM data ("non-CEM" sources), daily emissions were
computed from the NEI annual emissions using a structured query language (SQL) program and state-
average CEM data. To allocate annual emissions to each month, state-specific three-year averages of
2008-2010 CEM data were created. These average annual-to-month factors were assigned to non-CEM
sources within each state. To allocate the monthly emissions to each day, the 2009 CEM data to compute
state-specific month-to-day factors, averaged across all units in each state. These daily emissions wee
calculated outside of SMOKE and the resulting daily inventory is used as an input into SMOKE.
The daily-to-hourly allocation was performed in SMOKE using diurnal profiles. The state-specific and
pollutant-specific diurnal profiles for use in allocating the day-specific emissions for non-CEM sources in
43
-------
the ptipm sector were updated. The 2009 CEM data was used to create state-specific, day-to-hour
factors, averaged over the whole year and all units in each state. Diurnal factors were calculated using
CEM SO2 and NO\ emissions and heat input. SO2 and NOx-specific factors were computed from the
CEM data for these pollutants. All other pollutants used factors created from the hourly heat input data.
The resulting profiles were assigned by state and pollutant.
Two updated diurnal temporal profiles were incorporated into the 2009 modeling platform. For all
agricultural burning, we used a diurnal temporal profile (McCarty et al., 2009) that puts more of the
emissions during the actual work day and suppresses the emissions during the middle of the night was
used. Note that all states used a uniform day of week profile for all agricultural burning emissions,
except for the following states that for which state-specific day of week profiles were used: Arkansas,
Kansas, Louisiana, Minnesota, Missouri, Nebraska, Oklahoma, and Texas. For residential wood
combustion, a profile was used that placed more of the emissions in the morning and the evening when
people are typically using these sources. This profile is based on an average of 2004 MANE-VU survey
based temporal profiles (see http://www.marama.org/publications folder/ResWoodCombustion/
Final report.pdf). When this profile was compared to a concentration-based analysis of aethalometer
measurements in Rochester, NY (Wang et al. 2011) for various seasons and day of the week it was found
that the updated RWC profile generally tracked the concentration based temporal patterns.
The temporal profile assignments for the Canadian 2006 inventory were provided by Environment
Canada along with the inventory. They provided profile assignments that rely on the existing set of
temporal profiles in the 2002 Platform. For point sources, they provided profile assignments by
PLANTID.
3.3.5 Meteorological-based Temporal Profiles
A significant improvement over previous platforms is the introduction of meteorologically-based
temporalization. We recognize that there are many factors that impact the timing of when emissions
occur. The benefits of utilizing meteorology as method of temporalizing are: (1) a consistent
meteorological dataset as is used by the AQ model (e.g. WRF) is available; (2) the meteorological model
data is highly resolved in terms of spatial resolution; and (3) the meteorological variables vary at hourly
resolution which can translate to hour-specific temporalization.
The SMOKE program GenTPRO provides a method for developing meteorologically-based
temporalization. Currently, the program can utilize three types of temporal algorithms: RWC,
agricultural livestock ammonia, and a generic meteorology based algorithm. For the 2007 platform, we
used the RWC and ag NH3 GenTPRO generated profiles. GenTPRO reads in gridded meteorology data
(MCIP) and spatial surrogates and uses the specified algorithm to produce a new temporal profile that can
be input into SMOKE. The meteorological variables and the resolution of the generated temporal profile
(hourly, daily, etc.) depend on the algorithm and the run parameters. For more details on the
development of these algorithms and running GenTPRO, see the GenTPRO documentation
http://www.smoke-model.Org/version3.l/GenTPRQ TechnicalSummary Aug2012 Final.pdf and the
SMOKE manual section http://www.smoke-model.org/version3.1 /html/ch05s03s07.html.
For the RWC algorithm, GenTPRO uses the daily minimum temperature to determine the temporal
allocation of emissions to days. GenTPRO was run to create an annual-to-day temporal profile for the
RWC sources within the nonpt sector. These generated profiles distribute annual RWC emissions to the
44
-------
coldest days of the year. On days where the minimum temperature does not drop below a user-defined
threshold, RWC emissions are zero. Conversely, the program temporally allocates the largest percentage
of emissions to the coldest days. Similar to other temporal allocation profiles, the total annual emissions
do not change, just the distribution of the emissions within the year. Initially, the RWC algorithm used a
the default temperature threshold of 50 °F. For most of the country, this produced a reasonable
distribution of emissions, but for a few Southern counties all of the emissions were compressed into a few
days creating excessively high daily emissions. GenTPRO was then modified to accept an optional input
that defines a county/state specific alternative temperature threshold. In addition, an alternative RWC
algorithm was created to avoid negative RWC emissions when the daily minimum temperature was
greater than 53.3 °F. For the v5 platform, the alternative RWC algorithm was used for the whole country,
with the default 50 F threshold for the majority of the states, and a 60 °F threshold for the following
states: Alabama, Arizona, California, Florida, Georgia, Louisiana, Mississippi, South Carolina, and
Texas.
For the agricultural livestock NH3 algorithm, GenTPRO algorithm is based on the Russel and Cass
(1986) equation. This algorithm uses county-average hourly temperature and wind speed to calculate the
temporal profile. GenTPRO was run to create month-to-hour temporal profiles for these sources.
Because these profiles distribute to the hour based on monthly emissions, the emissions will either come
from a monthly inventory or from an annual inventory that has been temporalized already to the month.
For the on road and onroadrfl sectors, meteorology is not used in the development of the temporal
profiles; rather, but meteorology impacts the calculation of the hourly emissions through the program
Movesmrg. The result is that the emissions will vary at the hourly level by grid cell. More specifically,
the on-network (RPD) and the off-network (RPV) exhaust, evaporative, and evaporative permeation
modes use the gridded meteorology (MCIP) directly. Movesmrg determines the temperature for each
hour and grid cell and uses it to select the appropriate EF for that SCC/pollutant/mode. For the off-
network rate per profile (RPP) emissions, Movesmrg uses the Met4moves output for SMOKE (daily
minimum and maximum temperatures by county) to determine the appropriate EF for that hour and
SCC/pollutant. The result is that the emissions will vary hourly by county. The combination of these
three processes (RPD, RPV, and RPP) is the total onroad emissions, while the combination of the two
processes (RPD, RPV) for the refueling mode only is the total onroadrfl emissions. Both sectors will
show a strong meteorological influence on their temporal patterns.
3.3.6 Vertical Allocation of Emissions
Table 3-6 specifies the sectors for which plume rise is calculated. If there is no plume rise for a sector, the
emissions are placed into layer 1 of the air quality model. Vertical plume rise was performed in-line
within CMAQ for all of the SMOKE point-source sectors (i.e., ptipm, ptnonipm, ptfire, othpt, and
c3marine). The in-line plume rise computed within CM AQ is nearly identical to the plume rise that would
be calculated within SMOKE using the Lay point program. See http ://www. smoke-
model .org/version2.7/html/ch06s07.html for full documentation of Laypoint. The selection of point
sources for plume rise is pre-determined in SMOKE using the Elevpoint program (http://www.smoke-
model, org/ version2.7/html/ch06s03.htmlY The calculation is done in conjunction with the CM AQ model
time steps with interpolated meteorological data and is therefore more temporally resolved than when it is
done in SMOKE. Also, the calculation of the location of the point source is slightly different than the one
used in SMOKE and this can result in slightly different placement of point sources near grid cell
boundaries.
45
-------
For point sources, the stack parameters are used as inputs to the Briggs algorithm, but point fires do not
have stack parameters. However, the ptfire inventory does contain data on the acres burned (acres per day)
and fuel consumption (tons fuel per acre) for each day. CMAQ uses these additional parameters to
estimate the plume rise of emissions into layers above the surface model layer. Specifically, these data are
used to calculate heat flux, which is then used to estimate plume rise. In addition to the acres burned and
fuel consumption, heat content of the fuel is needed to compute heat flux. The heat content was assumed
to be 8000 Btu/lb of fuel for all fires because specific data on the fuels were unavailable in the inventory.
The plume rise algorithm applied to the fires is a modification of the Briggs algorithm with a stack height
of zero.
CMAQ uses the Briggs algorithm to determine the plume top and bottom, and then computes the plumes"
distributions into the vertical layers that the plumes intersect. The pressure difference across each layer
divided by the pressure difference across the entire plume is used as a weighting factor to assign the
emissions to layers. This approach gives plume fractions by layer and source.
3.3.7 Emissions Modeling A ncillary Files
The methods used to perform spatial allocation for the 2007 platform are summarized in this section. For
the 2007 platform, spatial factors are typically applied by country and SCC. As described earlier, spatial
allocation was performed for a national 12-km domain. To accomplish this, SMOKE used national 12-
km spatial surrogates and a SMOKE area-to-point data file. For the U.S., the spatial surrogates used
2010-based data (e.g., population) wherever possible. For Mexico, the same spatial surrogates were used
in the 2005 platform. For Canada we used a set of Canadian surrogates provided by Environment
Canada, also unchanged from the 2005v4.3 platform. The U.S., Mexican, and Canadian 12-km
surrogates cover the entire CON US domain 12US1 shown in Figure 3-1. The remainder of this
subsection provides further details on the origin of the data used for the spatial surrogates and the area-to-
point data.
The SMOKE ancillary data files, particularly the cross-reference files, provide the specific inventory
resolution at which spatial, speciation, and temporal factors are applied. For the 2009 Platform, spatial
factors were generally applied by country/SCC, speciation factors by pollutant/SCC- or (for combination
profiles) state/ county FIPS code and month, and temporal factors by some combination of country, state,
county, SCC, and pollutant.
3.3.7.1 Surrogates for U.S. Emissions
More than sixty spatial surrogates were used to spatially allocate U.S. county-level emissions to the
CMAQ 12-km grid cells. The Surrogate Tool was used to generate all of the surrogates. The shapefiles
input to the Surrogate Tool are provided and documented at
tittp://www.epa.gov/ttn/clr :h/spatial/spatial surrogate.htm 1. The tool and updated documentation for
it is available at http://www.ie.unc.edu/cempd/projects/mims/spatial/ and
http://www.cmascenter.org/help/documentation.cfm?MODEL=spatial allocator&VERSION=3.6&temp
id=99999. The detailed steps in developing the county boundaries for the surrogates are documented at
ftp://ftp.epa.gov/Emisitivetitorv/ emiss shp2006/us/metadata for 2002
county boundary shapefiles_rev.pdf.
Table 3-11 lists the codes and descriptions of the surrogates. The surrogates in bold have been updated
46
-------
with 2010-based data, including 2010 census data at the block group level, 2010 American Community
Survey Data for heating fuels, 2010 TIGER/Line data for railroads and roads, and 2010 National
Transportation Atlas Data for ports and navigable waterways. For this project "Version 3" of the 2010-
baed spatial surrogates was used. Not all of the available surrogates are used to spatially allocate sources
in the 2007 platform; that is, some surrogates shown in Table 3-11 were not assigned to any SCCs. An
area-to-point approach overrides the use of surrogates for some airport-related sources.
Alternative surrogates for ports (801) and shipping lanes (802) were developed from the 2008 NEI
shapefiles: Ports_032310_wrf and ShippingLanes l 11309FINAL_wrf. These surrogates were used for
cl and c2 commercial marine emissions instead of the standard 800 and 810 surrogates, respectively.
For the on road sector, the on-network (RPD) emissions were spatially allocated to roadways, which the
off-network (RPP and RPV) emissions were allocated to parking areas. For the onroadrfl sector, the
emissions were spatially allocated to gas station locations.
For the oil and gas sources in the nonpt sector, the WRAP Phase III sources have detailed basin-specific
spatial surrogates shown in Table 3-12. The remaining oil and gas sources used the 2005-based surrogate
"Oil & Gas Wells, IHS Energy, Inc. and USGS" (680) developed for oil and gas SCCs. The surrogates in
Table 3-12 were applied for the counties listed in Table 3-13.
3.3.7.3 Allocation Methodfor Airport-Related Sources in the U.S.
There are numerous airport-related emission sources in the 2005 NEI, such as aircraft, airport ground
support equipment, and jet refueling. In the 2002 platform most of these emissions were contained in
sectors with county-level resolution — aim (aircraft), nonroad (airport ground support) and nonpt (jet
refueling), but in the 2005 and 2008 platforms aircraft emissions are included as point sources as part of
the ptnonipm sector.
For the 2009 platform, the SMOKE "area-to-point" approach was used for airport ground support
equipment (nonroad sector), and jet refueling (nonpt sector). The approach is described in detail in the
2002 Platform documentation: http://www.epa.gov/scram001 /reports/ Emissions%20TSD%20Vo11 02-
28-08.pdf.
Nearly the same ARTOPNT file was used to implement the area-to-point approach as was done for the
CAP and HAP-2002-based Platform. This was slightly updated from the CAP-only 2002 Platform by
further allocating the Detroit-area airports into multiple sets of geographic coordinates to support finer
scale modeling. The updated file was retained for the 2009 Platform.
3.3.7.4 Surrogates for Canada and Mexico Emission Inventories
The Mexican emissions and single surrogate (population) were the same as those used in the 2002 and
2005 Platforms. For Canada, surrogates provided by Environment Canada with the 2006 emissions were
used to spatially allocate the 2006 Canadian emissions for the 2005 and 2009 Platforms.
The Canadian surrogate data described in Table 3-14 came from Environment Canada. They provided
both the surrogates and cross references; the surrogates were outputs from the Surrogate Tool (previously
referenced). Per Environment Canada, the surrogates are based on 2001 Canadian census data. The cross-
references that Canada originally provided were updated as follows: all assignments to surrogate '978'
(manufacturing industries) were changed to '906' (manufacturing services), and all assignments to '985'
47
-------
(construction and mining) and '984' (construction industries) were changed to '907' (construction
services) because the surrogate fractions in 984, 978 and 985 did not sum to 1. Codes for surrogates other
than population that did not begin with the digit "9" were also changed.
Table 3-11. U.S. Surrogates Available for the 2009 Platform
ICode
Surrogate Description
Code
Surrogate Description 1
N/A
Area-to-point approach (see 3.3.1.2)
520
Commercial plus Industrial plus
100
Population
525
Golf Courses + Institutional +Industrial +
Commercial
110
Housing
527
Single Family Residential
120
Urban Population
530
Residential - High Density
130
Rural Population
535
Residential + Commercial + Industrial +
Institutional
137
Housing Change
540
Retail Trade
140
Housing Change and Population
545
Personal Repair
150
Residential Heating - Natural Gas
550
Retail Trade plus Personal Repair
160
Residential Heating - Wood
555
Professional/Technical plus General
Government
165
0.5 Residential Heating - Wood plus 0.5 Low
Intensity Residential
560
Hospital
170
Residential Heating - Distillate Oil
565
Medical Office/Clinic
180
Residential Heating - Coal
570
Heavy and High Tech Industrial
190
Residential Heating - LP Gas
575
Light and High Tech Industrial
200
Urban Primary Road Miles
580
Food, Drug, Chemical Industrial
210
Rural Primary Road Miles
585
Metals and Minerals Industrial
220
Urban Secondary Road Miles
590
Heavy Industrial
230
Rural Secondary Road Miles
595
Light Industrial
240
Total Road Miles
596
Industrial plus Institutional plus Hospitals
250
Urban Primary plus Rural Primary
600
Gas Stations
255
0.75 Total Roadway Miles plus 0.25 Population
650
Refineries and Tank Farms
260
Total Railroad Miles
675
Refineries and Tank Farms and Gas Stations
270
Class 1 Railroad Miles
680
Oil and Gas
280
Class 2 and 3 Railroad Miles
700
Airport Areas
300
Low Intensity Residential
710
Airport Points
310
Total Agriculture
720
Military Airports
312
Orchards/Vineyards
800
Marine Ports
320
Forest Land
801
NEI Ports
330
Strip Mines/Quarries
802
NEI Shipping Lanes
340
Land
807
Navigable Waterway Miles
350
Water
810
Navigable Waterway Activity
400
Rural Land Area
850
Golf Courses
500
Commercial Land
860
Mines
48
-------
505 Industrial Land
510 Commercial plus Industrial
515 Commercial plus Institutional Land
870 Wastewater Treatment Facilities
880 Drvcleancrs
890 Commercial Timber
Table 3-12. Spatial Surrogates for WRAP Oil and Gas Data
Country
Code
Surrogate Description
USA
699
Gas production at CBM wells
USA
698
Well count - gas wells
USA
697
Oil production at gas wells
USA
696
Gas production at gas wells
USA
695
Well count - oil wells
USA
694
Oil production at Oil wells
USA
693
Well count - all wells
USA
692
Spud count
USA
691
Well count - CBM wells
USA
690
Oil production at all wells
USA
689
Gas production at all wells
Table 3-13. Counties included in the WRAP Dataset
FIPS
State
County
8001
Colorado
Adams
8005
Colorado
Arapahoe
8007
Colorado
Archuleta
8013
Colorado
Boulder
8014
Colorado
Broomfield
8029
Colorado
Delta
8031
Colorado
Denver
8039
Colorado
Elbert
8043
Colorado
Fremont
8045
Colorado
Garfield
8051
Colorado
Gunnison
8063
Colorado
Kit Carson
8067
Colorado
La Plata
8069
Colorado
Larimer
8073
Colorado
Lincoln
8075
Colorado
Logan
8077
Colorado
Mesa
8081
Colorado
Moffat
8087
Colorado
Morgan
8095
Colorado
Phillips
8103
Colorado
Rio Blanco
FIPS
State
County
30075
Montana
Powder River
35031
New Mexico
Mc Kinley
35039
New Mexico
Rio Arriba
35043
New Mexico
Sandoval
35045
New Mexico
San Juan
49007
Utah
Carbon
49009
Utah
Daggett
49013
Utah
Duchesne
49015
Utah
Emery
49019
Utah
Grand
49043
Utah
Summit
49047
Utah
Uintah
56001
Wyoming
Albany
56005
Wyoming
Campbell
56007
Wyoming
Carbon
56009
Wyoming
Converse
56011
Wyoming
Crook
56013
Wyoming
Fremont
56019
Wyoming
Johnson
56023
Wyoming
Lincoln
56025
Wyoming
Natrona
49
-------
8107
Colorado
Routt
8115
Colorado
Sedgwick
8121
Colorado
Washington
8123
Colorado
Weld
8125
Colorado
Yuma
30003
Montana
Big Horn
56027
Wyoming
Niobrara
56033
Wyoming
Sheridan
56035
Wyoming
Sublette
56037
Wyoming
Sweetwater
56041
Wyoming
Uinta
56045
Wyoming
Weston
Table 3-14. Canadian Spatial Surrogates for Canadian Emissions
Code
Description
Code
Description
9100
Population
9493
Warehousing and storage
9101
Total dwelling
9494
Total Transport and warehouse
9102
Urban dwelling
9511
Publishing and information services
9103
Rural dwelling
9512
Motion picture and sound recording
industries
9104
Total Employment
9513
Broadcasting and
tel ecommuni cati ons
9106
ALLINDUST
9514
Data processing services
9111
Farms
9516
Total Info and culture
9113
Forestry and logging
9521
Monetary authorities - central bank
9114
Fishing hunting and trapping
9522
Credit intermediation activities
9115
Agriculture and forestry activities
9523
Securities commodity contracts and
other financial investment activities
9116
Total Resources
9524
Insurance carriers and related
activities
9211
Oil and Gas Extraction
9526
Funds and other financial vehicles
9212
Mining except oil and gas
9528
Total Banks
9213
Mining and Oil and Gas Extract
activities
9531
Real estate
9219
Mining-unspecified
9532
Rental and leasing services
9221
Total Mining
9533
Lessors of non-financial intangible
assets (except copyrighted works)
9222
Utilities
9534
Total Real estate
9231
Construction except land subdivision
and land development
9541
Professional scientific and technical
services
9232
Land subdivision and land
development
9551
Management of companies and
enterprises
9233
Total Land Development
9561
Administrative and support services
9308
Food manufacturing
9562
Waste management and remediation
services
9309
Beverage and tobacco product
manufacturing
9611
Education Services
9313
Textile mills
9621
Ambulatory health care services
50
-------
Code
Description
Code
Description
9314
Textile product mills
9622
Hospitals
9315
Clothing manufacturing
9623
Nursing and residential care facilities
9316
Leather and allied product
manufacturing
9624
Social assistance
9321
Wood product manufacturing
9625
Total Service
9322
Paper manufacturing
9711
Performing arts spectator sports and
related industries
9323
Printing and related support activities
9712
Heritage institutions
9324
Petroleum and coal products
manufacturing
9713
Amusement gambling and recreation
industries
9325
Chemical manufacturing
9721
Accommodation services
9326
Plastics and rubber products
manufacturing
9722
Food services and drinking places
9327
Non-metallic mineral product
manufacturing
9723
Total Tourism
9331
Primary Metal Manufacturing
9811
Repair and maintenance
9332
Fabricated metal product
manufacturing
9812
Personal and laundry services
9333
Machinery manufacturing
9813
Religious grant-making civic and
professional and similar
organizations
9334
Computer and Electronic
manufacturing
9814
Private households
9335
Electrical equipment appliance and
component manufacturing
9815
Total other services
9336
Transportation equipment
manufacturing
9911
Federal government public
administration
9337
Furniture and related product
manufacturing
9912
Provincial and territorial public
administration (9121 to 9129)
9338
Miscellaneous manufacturing
9913
Local municipal and regional public
administration (9131 to 9139)
9339
Total Manufacturing
9914
Aboriginal public administration
9411
Farm product wholesaler-distributors
9919
International and other extra-
territorial public administration
9412
Petroleum product wholesaler-
distributors
9920
Total Government
9413
Food beverage and tobacco
wholesaler-distributors
9921
Commercial Fuel Combustion
9414
Personal and household goods
wholesaler-distributors
9922
TOTAL DISTRIBUTION AND
RETAIL
9415
Motor vehicle and parts wholesaler-
distributors
9923
TOTAL INSTITUTIONAL AND
GOVERNEMNT
51
-------
Code
Description
Code
Description
9416
Building material and supplies
wholesaler-distributors
9924
Primary Industry
9417
Machinery equipment and supplies
wholesaler-distributors
9925
Manufacturing and Assembly
9418
Miscellaneous wholesaler-distributors
9926
Distribution and Retail (no
petroleum)
9419
Wholesale agents and brokers
9927
Commercial Services
9420
Total Wholesale
9928
Commercial Meat cooking
9441
Motor vehicle and parts dealers
9929
HIGHJET
9442
Furniture and home furnishings stores
9930
LOWMEDJET
9443
Electronics and appliance stores
9931
OTHERJET
9444
Building material and garden
equipment and supplies dealers
9932
CANRAIL
9445
Food and beverage stores
9933
Forest fires
9446
Health and personal care stores
9941
PAVED ROADS
9447
Gasoline stations
9942
UNPAVED ROADS
9448
clothing and clothing accessories
stores
9943
HIGHWAY
9451
Sporting goods hobby book and music
stores
9944
ROAD
9452
General Merchandise stores
9945
Commercial Marine Vessels
9453
Miscellaneous store retailers
9946
Construction and mining
9454
Non-store retailers
9947
Agriculture Construction and mining
9455
Total Retail
9950
Intersection of Forest and Housing
9481
Air transportation
9960
TOTBEEF
9482
Rail transportation
9970
TOTPOUL
9483
Water Transportation
9980
TOTSWIN
9484
Truck transportation
9990
TOTFERT
9485
Transit and ground passenger
transportation
9993
Trail
9486
Pipeline transportation
9994
ALLROADS
9487
Scenic and sightseeing transportation
9995
3 0UNP A VED_7 Otrail
9488
Support activities for transportation
9996
Urban area
9491
Postal service
9997
CHBOISQC
9492
Couriers and messengers
9991
Traffic
52
-------
REFERENCES
Adelman, Z. 2012. Memorandum: Fugitive Dust Modeling for the 2008 Emissions Modeling Platform. UNC
Institute for the Environment, Chapel Hill, NC. September, 28, 2012.
Anderson, G.K.; Sandberg, D.V; Norheim, R.A., 2004. Fire Emission Production Simulator (FEPS) User's
Guide. Available at http://www.fs.fed.us/pnw/fera/feps/FEPS users guide.pdf
Bullock Jr., R, and K. A. Brehme (2002) "Atmospheric mercury simulation using the CMAQ model:
formulation description and analysis of wet deposition results." Atmospheric Environment 36, pp 2135—
2146.
ERG, 2006. Mexico National Emissions Inventory, 1999: Final, prepared by Eastern Research Group for
Secratariat of the Environment and Natural Resources and the National Institute of Ecology, Mexico,
October 11, 2006. Available at:
http://www.epa.gov/ttn/chief/net/mexico/1999 mexico nei final report.pdf
Environ Corp. 2008. Emission Profiles for EPA SPECIATE Database, Part 2: EPAct Fuels (Evaporative
Emissions). Prepared for U. S. EPA, Office of Transportation and Air Quality, September 30, 2008.
EPA, 2005. EPA 's National Inventory Model (NMIM), A Consolidated Emissions Modeling System for
MOBILE6 andNONROAD, U.S. Environmental Protection Agency, Office of Transportation and Air
Quality, Assessment and Standards Division. Ann Arbor, MI 48105, EPA420-R-05-024, December 2005.
Available at http://www.epa.gov/otaq/models/nmim/420r05024.pdf.
EPA 2006a. SPECIATE 4.0, Speciation Database Development Document, Final Report, U.S. Environmental
Protection Agency, Office of Research and Development, National Risk Management Research Laboratory,
Research Triangle Park, NC 27711, EPA600-R-06-161, February 2006. Available at
http://vvvvvv.epa.gov/ttn/chief/softvvare/speciate/speciate4/documentation/speciatedoc 1206.pdf.
EPA, 2012a. 2008 National Emissions Inventory, version 2 Technical Support Document. Office of Air Quality
Planning and Standards, Air Quality Assessment Division, Research Triangle Park, NC. Available at:
http://www.epa.gOv/ttn/chief/nety2008inventorv.html#inventorvdoc
Frost & Sullivan, 2010. "Project: Market Research and Report on North American Residential Wood Heaters,
Fireplaces, and Hearth Heating Products Market (P.O. # PO1-IMP403-F&S). Final Report April 26,
2010". Prepared by Frost & Sullivan, Mountain View, CA 94041.
Joint Fire Science Program, 2009. Consume 3.0—a software tool for computing fuel consumption. Fire Science
Brief. 66, June 2009. Consume 3.0 is available at:
http://www.fs.fed.us/pnw/fera/research/smoke/consume/index.shtml
Kochera, A., 1997. "Residential Use of Fireplaces," Housing Economics, March 1997, 10-11. Also see:
http://www.epa.gov/ttnchie 1/conference/ei 10/area/houck.pdf.
LADCO, 2012. "Regional Air Quality Analyses for Ozone, PM2.5, and Regional Haze: Base C Emissions
Inventory (September 12, 2011)". Lake Michigan Air Directors Consortium, Rosemont, IL 60018.
Available at: http://www.ladco.org/tech/emis/basecv8/Base C Emissions Documentation Sept 12.pdf
McCarty, J.L., Korontzi, S., Jutice, C.O., and T. Loboda. 2009. The spatial and temporal distribution of crop
residue burning in the contiguous United States. Science of the Total Environment, 407 (21): 5701-5712.
McKenzie, D.; Raymond, C.L.; Kellogg, L.-K.B.; Norheim, R.A; Andreu, A.G.; Bayard, A.C.; Kopper, K.E.;
Elman. E. 2007. Mapping fuels at multiple scales: landscape application of the Fuel Characteristic
Classification System. Canadian Journal of Forest Research. 37:2421-2437. Oak Ridge National
Laboratory, 2009. Analysis of Fuel Ethanol Transportation Activity and Potential Distribution
Constraints. U.S. Department of Energy, March 2009. Docket No. EPA-HQ-OAR-2010-0133.
Ottmar, R.D.; Sandberg, D.V.; Bluhm, A. 2003. Biomass consumption and carbon pools. Poster. In: Galley,
K.E.M., Klinger, R.C.; Sugihara, N.G. (eds.) Proceedings of Fire Ecology, Prevention, and Management.
53
-------
Misc. Pub. 13, Tallahassee, FL: Tall Timbers Research Station.
Ottmar, R.D.; Prichard, S.J.; Vihnanek, R.E.; Sandberg, D.V. 2006. Modification and validation of fuel
consumption models for shrub and forested lands in the Southwest, Pacific Northwest, Rockes, Midwest,
Southeast, and Alaska. Final report, JFSP Project 98-1-9-06.
Ottmar, R.D.; Sandberg, D.V.; Riccardi, C.L.; Prichard, S.J. 2007. An Overview of the Fuel Characteristic
Classification System - Quantifying, Classifying, and Creating Fuelbeds for Resource Planning.
Canadian Journal of Forest Research. 37(12): 2383-2393. FCCS is available at:
http://www.fs.fed.us/pnw/fera/fccs/index.shtml
Pouliot, G., H. Simon, P. Bhave, D. Tong, D. Mobley, T. Pace, and T. Pierce . (2010) "Assessing the
Anthropogenic Fugitive Dust Emission Inventory and Temporal Allocation Using an Updated Speciation
of Particulate Matter." International Emission Inventory Conference, San Antonio, TX. Available at
http://www.epa.gov/ttn/chief/conference/eil9/session9/pouliot.pdf
Raffuse, S., N. Larkin, P. Lahm, Y. Du, 2012. Development of Version 2 of the Wildland Fire Portion of the
[2011] National Emissions Inventory. International Emission Inventory Conference, Tampa, FL.
Available at: http://www.epa.gov/ttn/chief/conference/ei20/session2/sraffuse.pdf
Raffuse, S., D. Sullivan, L. Chinkin, S. Larkin, R. Solomon, A. Soja, 2007. Integration of Satellite-Detected and
Incident Command Reported Wildfire Information into BlueSky, June 27, 2007. Available at:
http:// getbl uesk v. org/ sm artfi re/docs. cfm
Russell, A.G. and G.R. Cass, 1986. Verification of a Mathematical Model for Aerosol Nitrate and Nitric Acid
Formation and Its Use for Control Measure Evaluation, Atmospheric Environment, 20: 2011-2025.
SESARM, 2012a. "Development of the 2007 Base Year and Typical Year Fire Emission Inventory for the
Southeastern States", Air Resources Managers, Inc., Fire Methodology, AMEC Environment and
Infrastructure, Inc. AMEC Project No.: 6066090326, April, 2012
SESARM, 2012b. "Area and Nonroad 2007 Base Year Inventories. Revised Final Report", Contract No. S-2009-06-01,
Prepared by Transystems Corporation, January 2012. Available at:
http://www. google. com/url?sa=t&rct=i&q=&esrc=s&source=web&cd=3&cad=ria&ved=OCDAOFiAC&url=ftp%
3 A%2F%2Fwsip-70-164-45 -
196.dc.dc.cox.net%2Fpublic%2FSESARM%2FRevised%2520Final%2FSESARM%2520Base%2520Year%2520
Revised%2520Final%2520Report Jan2012.docx&ei=xU-AUPulF4WA0AHC5YHYCg&usg=AFQiCNFhigx3Ei-
hbfYmMUP4zGI HBiqZA&sig2=hWWN0m3WYPSQ28QSzn5BIA.
Skamarock, W., J. Klemp, J. Dudhia, D. Gill, D. Barker, M. Duda, X. Huang, W. Wang, J. Powers, 2008. A
Description of the Advanced Research WRF Version 3. NCAR Technical Note. National Center for
Atmospheric Research, Mesoscale and Microscale Meteorology Division, Boulder, CO. June 2008.
Available at: http://www.mmm.ucar.edu/wrf/users/docs/arw v3.pdf
Sullivan D.C., Raffuse S.M., Pryden D.A., Craig K.J., Reid S.B., Wheeler N.J.M., Chinkin L.R., Larkin N.K.,
Solomon R., and Strand T. (2008) Development and applications of systems for modeling emissions and
smoke from fires: the BlueSky smoke modeling framework and SMARTFIRE: 17th International
Emissions Inventory Conference, Portland, OR, June 2-5. Available at:
http://vvvvvv.epa.gov/ttn/chief/conferences.html
Wang, Y., P. Hopke, O. V. Rattigan, X. Xia, D. C. Chalupa, M. J. Utell. (2011) "Characterization of Residential
Wood Combustion Particles Using the Two-Wavelength Aethalometer", Environ. Sci. Technol., 45 (17),
pp 7387-7393
Yarwood, G., S. Rao, M. Yocke, and G. Whitten, 2005: Updates to the Carbon Bond Chemical Mechanism:
CB05. Final Report to the US EPA, RT-0400675. Available at
http://www.camx.com/publ/pdfs/CB05 Final Report 120805.pdf.
54
-------
\*f - '/icy Model Estimates
4.1 Introduction to the CMAQ Modeling Platform
The Clean Air Act (CAA) provides a mandate to assess and manage air pollution levels to protect human
health and the environment. EPA has established National Ambient Air Quality Standards (NAAQS),
requiring the development of effective emissions control strategies for such pollutants as ozone and
particulate matter. Air quality models are used to develop these emission control strategies to achieve the
objectives of the CAA.
Historically, air quality models have addressed individual pollutant issues separately. However, many of
the same precursor chemicals are involved in both ozone and aerosol (particulate matter) chemistry;
therefore, the chemical transformation pathways are dependent. Thus, modeled abatement strategies of
pollutant precursors, such as volatile organic compounds (VOC) and NOx to reduce ozone levels, may
exacerbate other air pollutants such as particulate matter.
To meet the need to address the complex relationships between pollutants, EPA developed the Community
Multi scale Air Quality (CMAQ) modeling system. The primary goals for CMAQ are to:
• Improve the environmental management community's ability to evaluate the impact of air quality
management practices for multiple pollutants at multiple scales.
• Improve the scientist's ability to better probe, understand, and simulate chemical and physical
interactions in the atmosphere.
The CMAQ modeling system brings together key physical and chemical functions associated with the
dispersion and transformations of air pollution at various scales. It was designed to approach air quality as
a whole by including state-of-the-science capabilities for modeling multiple air quality issues, including
tropospheric ozone, fine particles, toxics, acid deposition, and visibility degradation. CMAQ relies on
emission estimates from various sources, including the U.S. EPA Office of Air Quality Planning and
Standards" current emission inventories, observed emission from major utility stacks, and model estimates
of natural emissions from biogenic and agricultural sources. CMAQ also relies on meteorological
predictions that include assimilation of meteorological observations as constraints. Emissions and
meteorology data are fed into CMAQ and run through various algorithms that simulate the physical and
chemical processes in the atmosphere to provide estimated concentrations of the pollutants. Traditionally,
the model has been used to predict air quality across a regional or national domain and then to simulate
the effects of various changes in emission levels for policymaking purposes. For health studies, the model
can also be used to provide supplemental information about air quality in areas where no monitors exist.
CMAQ was also designed to have multi-scale capabilities so that separate models were not needed for
urban and regional scale air quality modeling. The grid spatial resolutions in past annual CMAQ runs
have been 36 km x 36 km per grid for the "parent" domain, and nested within that domain are 12 km x 12
km grid resolution domains. The parent domain typically covered the continental United States, and the
55
-------
nested 12 km x 12 km domain covered the Eastern or Western United States. The CMAQ simulation
performed for this 2009 assessment used a single domain that covers the entire continental U.S. (CONUS)
and large portions of Canada and Mexico using 12 km by 12 km horizontal grid spacing. For urban
applications, CMAQ has also been applied with a 4-km x 4-km grid resolution for urban core areas;
however, the uncertainties in emissions and meteorology information can actually increase at this high of
a resolution. Currently, 12 km x 12 km resolution is recommended for most applications as the highest
resolution. With the temporal flexibility of the model, simulations can be performed to evaluate longer
term (annual to multi-year) pollutant climatologies as well as short-term (weeks to months) transport from
localized sources. By making CMAQ a modeling system that addresses multiple pollutants and different
temporal and spatial scales, CMAQ has a "one atmosphere" perspective that combines the efforts of the
scientific community. Improvements will be made to the CMAQ modeling system as the scientific
community further develops the state-of-the-science. For more information on CMAQ, go to
http ://www. epa. gov/asm dnerl/CM AQ or http ://www. cmascenter. ore.
4.1.1 Advantages and Limitations of the CMA Q Air Quality Model
An advantage of using the CMAQ model output for comparing with health outcomes is that it has the
potential to provide complete spatial and temporal coverage. Additionally, meteorological predictions,
which are also needed when comparing health outcomes, are available for every grid cell along with the
air quality predictions.
A disadvantage of using CMAQ is that, as a deterministic model, it has none of the statistical qualities of
interpolation techniques that fit the observed data to one degree or another. Furthermore, the emissions
and meteorological data used in CMAQ each have large uncertainties, in particular for unusual emission
or meteorological events. There are also uncertainties associated with the chemical transformation and
fate process algorithms used in air quality models. Thus, emissions and meteorological data plus modeling
uncertainties cause CMAQ to predict best on longer time scale bases (e.g., synoptic, monthly, and annual
scales) and be most error prone at high time and space resolutions compared to direct measures.
One practical disadvantage of using CMAQ output is that the regularly spaced grid cells do not line up
directly with counties or ZIP codes which are the geographical units over which health outcomes are
likely to be aggregated. But it is possible to overlay grid cells with county or ZIP code boundaries and
devise means of assigning an exposure level that nonetheless provides more complete coverage than that
available from ambient data alone. Another practical disadvantage is that CMAQ requires significant data
and computing resources to obtain results for daily environmental health surveillance.
This section describes the air quality modeling platform used for the 2009 CMAQ simulation. A modeling
platform is a structured system of connected modeling-related tools and data that provide a consistent and
transparent basis for assessing the air quality response to changes in emissions and/or meteorology. A
platform typically consists of a specific air quality model, emissions estimates, a set of meteorological
inputs, and estimates of "boundary conditions" representing pollutant transport from source areas outside
the region modeled. We used the CMAQ6 model as part of the 2009 Platform to provide a national scale
6Byun, D.W., and K. L. Schere, 2006: Review of the Governing Equations, Computational Algorithms, and Other
Components of the Models-3 Community Multiscale Air Quality (CMAQ) Modeling System. Applied Mechanics Reviews,
Volume 59, Number 2 (March 2006), pp. 51-77.
56
-------
air quality modeling analysis. The CMAQ model simulates the multiple physical and chemical processes
involved in the formation, transport, and destruction of ozone and fine particulate matter (PM2 5)
This section provides a description of each of the main components of the 2009 CMAQ simulation along
with the results of a model performance evaluation in which the 2009 model predictions are compared to
corresponding measured concentrations.
4.2 CMAQ Model Version, Inputs and Configuration
4.2.1 Model Version
CMAQ is a non-proprietary computer model that simulates the formation and fate of photochemical
oxidants, including PM2.5 and ozone, for given input sets of meteorological conditions and emissions. The
CMAQ model version 4.7 was most recently peer-reviewed in February of 2009 for the U.S. EPA7. As
mentioned previously, CMAQ includes numerous science modules that simulate the emission,
production, decay, deposition and transport of organic and inorganic gas-phase and pollutants in the
atmosphere. This analysis employed a version of CMAQ based on the latest publicly released version of
CMAQ (i.e., version 4.7.18) at the time of the 2009 air quality modeling. CMAQ version 4.7.1 reflects
updates to version 4.7 to improve the underlying science which include aqueous chemistry mass
conservation improvements and improved vertical convective mixing. The model enhancements in
version 4.7.1 also include:
1. Aqueous chemistry
• Mass conservation improvements
Imposed one second minimum timestep for remainder of the cloud lifetime after 100
'iterations" in the solver
Force mass balance for the last timestep in the cloud by limiting oxidized amount to mass
available
• Implemented steady state assumption for OH
• Only allow sulfur oxidation to control the aqueous chemistry solver timestep (previously,
reactions of OH, GLY, MGLY, and Hg for multipollutant model also controlled the timestep)
2. Advection
• Added additional divergence-based constraint on advection timestep
• Vertical advection in the Yamo module is now represented with the PPM scheme to limit
numerical diffusion
7 Allen. D., Burns, D.. Chock. D.. Kumar, N., Lamb. B.. Moran, M. (February 2009 Draft Version). Report on the Peer
Review of the Atmospheric Modeling and Analysis Division, NERL/ORD/EPA. U.S. EPA, Research Triangle Park, NC.
CMAQ version 4.7 was released on December, 2008. It is available from the Community Modeling and Analysis System
(CMAS) as well as previous peer-review reports at: http://www.cmascenter.org.
8 CMAQ version 4.7.1 model code is available from the Community Modeling and Analysis System (CMAS) at:
http://www.cmascenter.org.
57
-------
3. Model time step determination
• Fixed a potential advection time step error
The sum of the advection steps for a given layer time step might not equal the output time
step duration in some extreme cases
- Ensured that the advection steps sum up to the synchronization step
4. Horizontal diffusion
• Fixed a potential error
Concentration data may not be correctly initialized if multiple sub-cycle time steps are
required
- Fix to initialize concentrations with values calculated in the previous sub-time step
5. Emissions
• Bug fix in EMIS DEFN.F to include point source layer 1 NH3 emissions
• Bug fix to calculate soil NO "pulse" emissions in BEIS
• Remove excessive logging of cases where ambient air temperature exceeds 315.0 Kelvin. When
this occurs, the values are just slightly over 315
• Bug fix for parallel decomposition errors in plume rise emissions
6. Photolysis
• JPROC/phottable and photsat options
- Expanded lookup tables to facilitate applications across the globe and vertical extent to
20km
- Updated temperature adjustments for absorption cross sections and quantum yields
- Revised algorithm that processes TOMS datasets for OMI data format
• In-line option
Asymmetry factor calculation updated using values from Mie theory integrated over log
normal particle distribution; added special treatment for large particles in asymmetry factor
algorithm to avoid numerical instabilities
4.2.2 Model Domain and Grid Resolution
The CMAQ modeling analyses were performed for a domain covering the continental United States, as
shown in Figure 4-1. This single domain covers the entire continental U.S. (CONUS) and large portions
of Canada and Mexico using 12 km by 12 km horizontal grid spacing. The model extends vertically from
the surface to 50 millibars (approximately 19 km) using a sigma-pressure coordinate system. Air quality
conditions at the outer boundary of the 12 km domain were taken from a global model. Table 4-1 provides
some basic geographic information regarding the 12 km CMAQ domain.
58
-------
Table 4-1. Geographic Information for 12 km Modeling Domain
1 National 12 km CMAQ Modeling Configuration 1
Map Projection
Map Projection
Grid Resolution
Grid Resolution
Coordinate
Coordinate Center
Center
True Latitudes
True Latitudes
Dimensions
Dimensions
Vertical Extent
Vertical Extent
^'% ( I
It?\
12km CONUS nationwid^dwman^^ 5 sk if IV \ •
x,y: -2556000,-1728000 J V V y J \ „ *
col: 459 row; 299 W M^r S 7 7 " -
O \\ / y\> (
r* J.. ink
Figure 4-1. Map of the CMAQ Modeling Domain. The blue box denotes the 12 km national
modeling domain. (Same as Figure 3-1.)
4.2.3 Modeling Period/ Ozone Episodes
The 12 km CMAQ modeling domain was modeled for the entire year of 2009. The 2009 annual
simulation was performed in two half-year segments (i.e., January through June, and July through
December) for each emissions scenario. With this approach to segmenting an annual simulation we were
able to reduce the overall throughput time for an annual simulation. The annual simulation included a
"ramp-up" period, comprised of 10 days before the beginning of each half-year segment, to mitigate the
effects of initial concentrations. All 365 model days were used in the annual average levels of PM2.5- For
59
-------
the 8-hour ozone, we used modeling results from the period between May 1 and September 30. This 153-
day period generally conforms to the ozone season across most parts of the U.S. and contains the majority
of days that observed high ozone concentrations.
4.2.4 Model Inputs: Emissions, Meteorology and Boundary Conditions
2009 Emissions: The emissions inventories used in the 2009 air quality modeling are described in Section
3, above.
Meteorological Input Data: The gridded meteorological data for the entire year of 2009 at the 12 km
continental United States scale domain was derived from version 3.2 of the Weather Research and
Forecasting Model (WRF), Advanced Research WRF (ARW) core.9 Previous CMAQ annual simulations
have typically utilized meteorology provided by the 5th Generation Mesoscale Model (MM5).10 The WRF
Model is a next-generation mesoscale numerical weather prediction system developed for both operational
forecasting and atmospheric research applications (http://wrf-model.orgy The 2009 WRF simulation
included the physics options of the Pleitn-Xiu land surface model (LSM), Asymmetric Convective Model
version 2 planetary boundary layer (PBL) scheme, Morrison double moment microphysics, Kain- Frit sell
cumulus parameterization scheme and the RRTMG long-wave and shortwave radiation (LWR/SWR)
scheme.11
The WRF meteorological outputs were processed to create model-ready inputs for CMAQ using the
Meteorology- Chemistry Interface Processor (MCIP) package12, version 3.6, to derive the specific inputs
to CMAQ: horizontal wind components (i.e., speed and direction), temperature, moisture, and its related
speciated components was conducted for vertical diffusion rates, and rainfall rates for each grid cell in
each vertical layer. The WRF simulation used the same CMAQ map projection, a Lambert Conformal
projection centered at (-97, 40) with true latitudes at 33 and 45 degrees north. The 12 km WRF domain
consisted of 459 by 299 grid cells. The WRF simulation utilized 34 vertical layers with a surface layer of
approximately 38 meters. Table 4-2 shows the vertical layer structure used in WRF and the layer
collapsing approach to generate the CMAQ meteorological inputs. CMAQ resolved the vertical
atmosphere with 24 layers, preserving greater resolution in the PBL.
In terms of the 2009 WRF meteorological model performance evaluation, an approach which included a
combination of qualitative and quantitative analyses was used to assess the adequacy of the WRF
simulated fields. The qualitative aspects involved comparisons of the model-estimated synoptic patterns
against observed patterns from historical weather chart archives. Additionally, the evaluations compared
9 Skamarock, W.C., Klemp, J.B., Dudhia, J., Gill, D.O., Barker, D.M., Duda, M.G., Huang, X., Wang, W., Powers, J.G., 2008.
A Description of the Advanced Research WRF Version 3.
' "Grell, G. A., Dudhia, A. J., and StaulTer. D. R.. 1994. A description of the Fifth-Generation PennState/NCAR Mesoscale
Model (MM5). NCAR Technical Note NCAR/TN-398+STR. Available at http://www.niniiii. ucar.edii/mniS/doe 1.html.
11 Gilliam. R.C.. Pleim. I.E., 2010. Performance Assessment of New Land Surface and Planetary Boundary Layer Physics in
the WRF- ARW. Journal of Applied Meteorology and Climatology 49, 760-774.
12Otte T.L., Pleim. I.E., 2010. The Meteorology-Chemistry Interface Processor (MCIP) for the CMAQ modeling system:
updates through v3.4.1. Geoscientific Model Development 3, 243-256.
60
-------
spatial patterns of monthly average rainfall and monthly maximum planetary boundary layer (PBL)
heights. The statistical portion of the evaluation examined the model bias and error for temperature, water
vapor mixing ratio, solar radiation, and wind fields. These statistical values were calculated on a monthly
basis.
Table 4-2Vertical layer structure for 2009 WRF and CMAQ simulations (heights are layer top)
Height
(m)
Pressure
(mb)
WRF
Depth
(m)
CMAQ
Depth
(m)
17,145
50
34
2,655
24
4,552
14,490
95
33
1,896
12,593
140
32
1,499
23
2,749
11,094
185
31
1,250
9,844
230
30
1,078
22
2,029
8,766
275
29
951
7,815
320
28
853
21
1,627
6,962
365
27
775
6,188
410
26
711
20
1,368
5,477
455
25
657
4,820
500
24
612
19
1,185
4,208
454
23
573
3,635
590
22
539
18
539
3,095
635
21
509
17
509
2,586
680
20
388
16
388
2,198
716
19
281
15
281
1,917
743
18
273
14
273
1,644
770
17
178
13
178
1,466
788
16
174
12
174
1,292
806
15
171
11
171
1,121
824
14
168
10
168
952
842
13
165
9
165
787
860
12
82
8
163
705
869
11
81
624
878
10
80
7
160
544
887
9
80
465
896
8
79
6
157
386
905
7
78
307
914
6
78
5
78
230
923
5
77
4
77
153
932
4
38
3
76
114
937
3
38
76
941
2
38
2
38
38
946
1
38
1
38
61
-------
Initial and Boundary Conditions: The lateral boundary and initial species concentrations are provided by
a three- dimensional global atmospheric chemistry model, the GEOS-CHEM13 model (standard version 8-
03-02 with 8-02-03 chemistry). The global GEOS-CHEM model simulates atmospheric chemical and
physical processes driven by assimilated meteorological observations from the NASA's Goddard Earth
Observing System (GEOS). This model was run for 2009 with a grid resolution of 2.0 degrees x 2.5
degrees (latitude-longitude) and 46 vertical layers up to 0.01 hPa. The predictions were processed using
the GEOS-2-CMAQ tool and used to provide one-way dynamic boundary conditions at one-hour
intervals.14 Ozone was evaluated from these GEOS-Chem runs by comparing to satellite vertical profiles
and ground-based measurements and found acceptable model performance. More information is available
about the GEOS-CHEM model and other applications using this tool at: http://www-
as. harvard. edit/chem i strv/trop/geos.
4.3 CMAQ Model Performance Evaluation
An operational model performance evaluation for ozone and PM2 5 and its related speciated components
was conducted for the 2009 simulation using state/local monitoring sites data in order to estimate the
ability of the CMAQ modeling system to replicate the 2009 base year concentrations for the 12 km
continental U.S. domain.
There are various statistical metrics available and used by the science community for model performance
evaluation. For a robust evaluation, the principal evaluation statistics used to evaluate CMAQ
performance were two bias metrics, normalized mean bias and fractional bias; and two error metrics,
normalized mean error and fractional error. Normalized mean bias (NMB) is used as a normalization to
facilitate a range of concentration magnitudes. This statistic averages the difference (model - observed)
over the sum of observed values. NMB is a useful model performance indicator because it avoids
overinflating the observed range of values, especially at low concentrations. Normalized mean bias is
defined as:
pp-o)
NMB = - *100, where P = predicted concentrations and O = observed
1(0)
1
Normalized mean error (NME) is also similar to NMB, where the performance statistic is used as a
normalization of the mean error. NME calculates the absolute value of the difference (model - observed)
over the sum of observed values. Normalized mean error is defined as:
13 Yantosca, B., 2004. GEOS-CHEMv7-01-02 User's Guide, Atmospheric Chemistry Modeling Group, Harvard University,
Cambridge, MA, October 15, 2004.
14Akhtar, F., Henderson, B., Appel, W., Napelenok, S., Hutzell, B., Pye, H., Foley, K., 2012. Multiyear Boundary Conditions
for CMAQ 5.0 from GEOS-Chem with Secondary Organic Aerosol Extensions, 11th Annual Community Modeling and
Analysis System conference, Chapel Hill, NC, October 2012.
62
-------
YJ\p-o\
NME = - *100
n
1(0)
1
Fractional bias is defined as:
/ \
1
FB = -
n
\ i \ ^ / j
FB is a useful model performance indicator because it has the advantage of equally weighting positive and
negative bias estimates. The single largest disadvantage in this estimate of model performance is that the
estimated concentration (i.e., prediction, P) is found in both the numerator and denominator.
Fractional error (FE) is similar to fractional bias except the absolute value of the difference is used so that
the error is always positive. Fractional error is defined as:
/ \
1
FE = —
n
V 1 \ ^ / J
In addition to the performance statistics, regional maps which show the normalized mean bias and error
were prepared for the ozone season. May through September, at individual monitoring sites as well as on
an annual basis for PM2 ? and its component species.
Evaluation for 8-hour Daily Maximum Ozone: The operational model performance evaluation for eight-
hour daily maximum ozone was conducted using the statistics defined above. Ozone measurements for
2009 in the continental U.S. were included in the evaluation and were taken from the 2009 State/local
monitoring site data in the Air Quality System (AQS) Aerometric Information Retrieval System (AIRS).
The performance statistics were calculated using predicted and observed data that were paired in time and
space on an 8-hour basis. Statistics were generated for the following geographic groupings in the 12-km
continental U.S. domain15: five large sub regions: Midwest, Northeast, Southeast, Central and Western
U.S.
The 8-hour ozone model performance bias and error statistics for each sub region and each season are
provided in Table 4-4. Seasons were defined as: winter (December-January- February), spring (March-
April-May), summer (June, July, August), and fall (September-October-November). Spatial plots of the
normalized mean bias and error for individual monitors are shown in Figures 4-2 through 4-3. The
statistics shown in these two figures were calculated over the ozone season using data pairs on days with
observed 8-hour ozone of greater than or equal to 60 ppb.
15 The sub regions are defined by States where: Midwest is IL. IN, MI, OH, and WI; Northeast is CT, DE, MA, MD. ME, NH,
NJ, NY, PA. Rl. and VT; Southeast is AL, FL. GA. KY. MS. NC. SC. TN. VA, and WV; Central is AR. IA, KS. LA. MN. MO.
NE. OK, and TX; West is AK, CA, OR. WA. AZ. NM, CO. UT. WY. SD. ND. MT, ID. and NV.
H(p-o)
1
(P+O)
= 100
£\p-o\
1
(P+O)
= 100
63
-------
In general, the model performance statistics indicate that the 8-hour daily maximum ozone concentrations
predicted by the 2009 CMAQ simulation closely reflect the corresponding 8-hour observed ozone
concentrations in space and time in each subregion of the 12 km modeling domain. As indicated by the
statistics in Table 4-4, bias and error for 8-hour daily maximum ozone are relatively low in each
subregion, not only in the summer when concentrations are highest, but also during other times of the
year. Specifically, 8-hour ozone in the summer is slightly over predicted with the greatest over prediction
in the Southeast (NMB is 23.1 percent). Ozone performance in Spring shows better performance with
slight over predictions in most of the subregions except in the West (slight under prediction of 0.6). In the
winter, when concentrations are generally low, the model slightly over predicts 8-hour ozone with the
exception of the Northeast (NMB is -11.2). In the fall, when concentrations are also relatively low, ozone
is also slightly over predicted; with NMBs less than 24 percent in each subregion.
Model bias at individual sites during the ozone season is similar to that seen on a subregional basis for the
summer. The information in Figure 4-2 indicates that the bias for days with observed 8-hour daily
maximum ozone greater than 60 ppb is within ± 20 percent at the vast majority of monitoring sites across
the U.S. domain. The exceptions are sites in and/or near Chicago, IL, Baton Rouge, LA, Tampa and
Orlando, FL, northern (St. Lawrence/ Franklin counties) NY, Greenville, WV, Brunswick, GA; as well as
a few areas along the southern California coast. At these sites observed concentrations greater than 60 ppb
are generally predicted in the range of ±20 to 40 percent. Looking at the map of bias, Figure 4-2 indicates
that the low bias at these sites is not evident at other sites in these same areas. This suggests that the under
prediction at these sites is likely due to very local features (e.g., meteorology and/or emissions) and not
indicative of a systematic problem in the modeling platform. Model error, as seen from Figure 4-3, is 30
percent or less at most of the sites across the U.S. modeling domain. Somewhat greater error is evident at
sites in several areas most notably along portions of the Northeast Corridor and in portions of Florida,
Louisiana, Texas, Mississippi, Alabama, South Carolina and along the California coastline.
Table 4-3. Summary of CMAQ 2009 8-Hour Daily Maximum Ozone Model Performance Statistics
by Subregion, by Season
Subregion
Season
No. of
Obs
NMB
(%)
NME
(%)
FB (%)
FE (%)
Northeast
Winter
5,472
-11.2
19.2
-11.9
21.7
Spring
11,995
0.8
12.0
1.7
12.7
Summer
15,215
14.6
19.4
14.7
19.1
Fall
11,070
18.1
24.3
18.3
24.1
Midwest
Winter
2,708
0.5
23.9
-4.3
23.0
Spring
11,616
2.4
13.0
3.4
13.7
Summer
15,914
13.2
18.3
13.4
18.1
Fall
9,350
15.8
20.9
16.4
22.0
Central States
Winter
11,083
4.4
16.5
5.7
18.2
Spring
14,851
5.0
14.7
6.7
15.6
Summer
16,464
21.1
26.2
21.0
25.3
Fall
14,495
11.0
20.2
12.8
21.5
64
-------
Southeast Winter 6,536 5.5 14.9 11.6 6.3
Spring
17.194
9.8
16.4
11.1
17.0
Summer
19,395
23.1
26.0
22.8
25.2
Fall
15,308
24.4
28.7
23.9
27.6
West Winter
22,813
14.4
22.7
16.4
24.5
Spring
26,499
-0.6
12.4
-0.3
13.0
Summer
29,460
6.8
16.5
7.4
16.6
Fall
26,324
8.2
17.1
9.7
17.9
03_8hrmax NMB (%) tor run 2009at2_v5_Q9d_12USl tor May-Sep lor 12US1 [Q3_8hrmax_ot»=60ppbl
CiRCLE=AOS_Deply;
Figure 4-2. Normalized Mean Bias (%) of 8-hour daily maximum ozone greater than 60 ppb over
the period May- September 2009 at monitoring sites in the continental U.S. modeling domain
65
-------
03_8hrmax NME {%) for run 2009ef2_v5_09d_12US1 for 20090501 to 20090930
units « %
coverage limit ¦ 75%
CIRCLE=AQS_Daily;
Figure 4-3. Normalized Mean Error (%) of 8-hour daily maximum ozone greater than 60 ppb over
the period May-September 2009 at monitoring sites in the continental U.S. modeling domain
Evaluation for Annual PM > 5: The PM evaluation focuses on PM2.5 total mass and its components
including sulfate (SO4), nitrate (NO3), total nitrate (TNO3 = NO3 + HNO3), ammonium (NH4), elemental
carbon (EC), and organic carbon (OC).
The PM2.5 bias and error performance statistics were calculated on an annual basis for each subregion
(Table 4-5). PM2.5 measurements for 2009 were obtained from the following networks for model
evaluation: Chemical Speciation Network (CSN, 24 hour average), Interagency Monitoring of PROtected
Visual Environments (IMPROVE, 24 hour average, and Clean Air Status and Trends Network
(CASTNet, weekly average). For PM2.5 species that are measure by more than one network, we
calculated separate sets of statistics for each network by subregion. For brevity, Table 4-5 provides
annual model performance statistics for PM2.5 and its component species for the five sub-regions in the 12
km continental U.S. domain defined above (Northeast, Midwest, Southeast, Central, and West). In
addition to the tabular summaries of bias and error statistics, annual spatial maps which show the
normalized mean bias and error by site for each PM2.5 species are provided in Figures 4-4 through 4-17.
As indicated by the statistics in Table 4-5, annual CMAQ PM2.5 for 2009 shows under predictions at rural
IMPROVE monitoring sites and urban CSN monitoring sites in each subregion except in the Northeast
and Midwest at CSN sites which shows a slight over prediction in NMB of 0 to 3 percent. Although not
shown here, the mean observed concentrations of PM2 5 are more than twice as high at the CSN sites
(~10(.ig m"3) as the IMPROVE sites (~5 jig m"3), thus illustrating the statistical differences between the
urban CSN and rural IMPROVE networks.
Annual average sulfate is consistently under predicted at CSN, IMPROVE, and CASTNet monitoring
sites across the modeling domain, with NMB values ranging from -14 percent to -41 percent. Overall,
sulfate bias performance is slightly better at rural IMPROVE sites than at urban CSN and/or suburban
66
-------
CASTNet sites. Sulfate performance shows moderate error, ranging from 28 to 45 percent. Figures 4-6
and 4-7, suggest spatial patterns vary by region. The model bias for most of the Southeast, Central and
Southwest states are within -20 to -40 percent. The model bias appears to be much less (±20 percent)
in the Northeast, and Northwest states. A few sites in the West and in the Central U.S. have biases
much greater than 20 percent. Model error also shows a spatial trend by region, where much of the
Eatern states are 20 to 40 percent, the Western and Central U.S. states are 30 to 60 percent.
Annual average nitrate is over predicted at the urban and rural monitoring sites in most of the
subregions in thel2 km modeling domain (NMB in the range of 19% to 47%), except in the West
where nitrate is under predicted (NMB in the range of -20% to -32%). The bias statistics indicate that
the model performance for nitrate is generally best at the urban CSN monitoring sites. Model
performance of total nitrate at sub-urban CASTNet monitoring sites shows an over prediction across
all subregions. Model error for nitrate is somewhat greater for each subregion as compared to sulfate.
Model bias at individual sites indicates mainly over prediction of greater than 20 percent at most
monitoring sites in the Eastern half of the U.S. as well and in the extreme Northwest, as indicated in
Figure 4-8. The exception to this is in the Florida and the Southwest of the modeling domain where
there appears to be a greater number of sites with under prediction of nitrate of 20 to 80 percent.
Model error for annual nitrate, as shown in Figure 4-9, is least at sites in portions of the Midwest and
extending eastward to the Northeast corridor. Nitrate concentrations are typically higher in these areas
than in other portions of the modeling domain.
Annual average ammonium model performance as indicated in Table 4-5 has a tendency for the model
to under predict across the CSN and CASTNet sites (ranging from -1 to -25 percent). Ammonium is
slightly over predicted in the Midwest at CASTNet sites (NMB = ~ 3 percent). There is not a large
variation from subregion to subregion or at urban versus rural sites in the error statistics for ammonium.
The spatial variation of ammonium across the majority of individual monitoring sites shows bias within
±20 percent.
Annual average elemental carbon is over predicted in all subregions at urban and rural sites with the
exception of the near negligible bias in the Central U.S. at IMPROVE sites. Similar to ammonium error
there is not a large variation from subregion to subregion or at urban versus rural sites.
Annual average organic carbon is under predicted at both urban and rural monitoring sites in all
subregions of the U.S. (NMB ranging from -4 to 45 percent). Similar to ammonium and elemental carbon,
error model performance does not show a large variation from subregion to subregion or at urban versus
rural sites (48 to 67 percent).
Table 4-4. Summary of CMAQ 2009 Annual PM Species Model Performance Statistic
Pollutant
Monitor
Network
Subregion
No. of
Obs
NMB
(%)
NME
(%)
FB (%)
FE (%)
CSN
Northeast
2,754
3.0
37.7
0.4
36.6
Midwest
2,087
0.2
30.3
-4.6
33.2
pm25
Southeast
2,345
-23.7
40.0
-32.8
46.1
Central
1,891
-14.8
41.7
-20.2
46.8
West
2,986
-14.1
49.5
-15.2
51.0
67
-------
n II . . Monitor
Pollutant .. , .
Network
Subregion
No. of
Obs
NMB
(%)
NME
(%)
FB (%)
FE (%)
IMPROVE
Northeast
2,317
-1.3
43.2
-9.0
43.3
Midwest
577
-8.3
33.4
-15.4
39.4
Southeast
1,950
-25.2
44.3
-35.9
54.1
Central
2,500
-17.9
41.3
-24.2
49.3
West
10,295
-27.9
56.3
-34.7
62.9
CSN
Northeast
3,131
-20.5
34.3
-15.4
35.7
Midwest
2,238
-24.5
33.3
-24.7
36.4
Southeast
2,837
-28.4
36.8
-32.7
42.9
Central
2,295
-30.8
42.2
-31.4
46.9
West
3,196
-22.9
42.8
-15.6
44.7
IMPROVE
Northeast
2,307
-15.3
32.0
-6.3
33.7
Midwest
571
-25.8
35.4
-19.1
40.5
Sulfate
Southeast
1,951
-27.1
37.0
-27.8
43.2
Central
2,446
-29.1
37.9
-24.5
41.6
West
10,030
-14.4
44.9
-0.4
49.1
CASTNet
Northeast
769
-26.5
28.1
-26.7
29.9
Midwest
614
-30.5
31.2
-35.8
36.5
Southeast
1,096
-35.1
36.2
-42.5
44.6
Central
381
-41.3
42.0
-46.4
48.3
West
1,043
-29.8
39.2
-24.8
44.5
CSN
Northeast
3,143
27.9
65.5
-11.7
71.1
Midwest
2,325
23.4
56
-8.7
67.8
Southeast
2,851
25.2
93.3
56
106
Central
1,641
19.3
57.3
-24.5
80.6
West
3,164
-32.4
61.8
-69.6
94.1
Nitrate
IMPROVE
Northeast
2,308
46.8
91.0
-13.5
91.2
Midwest
570
33.9
69.8
-19.7
90.7
Southeast
1,951
24.7
108.0
-58.6
118.0
Central
2,445
33.9
74.1
-16.0
93.5
West
10,016
-20.9
88.7
-76.2
122.0
CASTNet
Northeast
769
53.4
59.1
41.5
50.7
Total Nitrate
(NO3 +
HNO3)
Midwest
614
33.9
43.6
33.8
40.7
Southeast
1,096
35.6
54.3
27.0
49.9
Central
381
16.1
36.9
13.6
36.8
68
-------
n II . . Monitor
Pollutant .. , .
Network
Subregion
No. of
Obs
NMB
(%)
NME
(%)
FB (%)
FE (%)
West
1,043
4.8
40.3
15.4
43.0
CSN
Northeast
3,131
-8.4
37.3
3.5
39.4
Midwest
2,238
-3.6
31.3
3.1
34.5
Southeast
2,837
-8.1
38.9
-7.7
41.3
Central
2,295
-7.1
41.1
-6.5
46.0
West
3,196
-25.9
58.1
-8.7
58.2
Ammonium
CASTNet
Northeast
769
-1.6
28.5
-1.3
30.4
Midwest
614
3.0
25.8
3.4
25.9
Southeast
1,096
-15.9
30.6
-18.4
35.1
Central
381
-2.0
36.9
-5.8
40.4
West
1,043
-12.8
45.3
-12.4
50.2
CSN
Northeast
2,978
36.1
61.5
25.1
49.2
Midwest
2,212
52.2
71.6
33.6
53.8
Southeast
2,823
18.8
54.8
13.2
49.3
Central
2,284
48.8
79.7
33.8
60.7
West
3,117
15.9
68.8
11.7
62.0
Elemental
Carbon
til Ly v_/ 11
IMPROVE
Northeast
2,320
27.7
61.6
6.0
51.1
Midwest
577
4.1
38.9
-0.5
44.8
Southeast
1,949
3.8
46.4
-6.5
47.4
Central
2,501
-0.3
40.9
-1.3
43.3
West
10,403
15.1
77.7
3.7
60.9
CSN
Northeast
2,900
-4.9
59.3
-6.3
60.5
Midwest
2,155
-14.9
53.6
-15.6
58.3
Southeast
2,779
-39.3
55.6
-48.7
68.7
Central
2,239
-31.5
56.9
-35.2
66.7
West
3,058
-21.5
57.1
-16.4
60.2
Organic
Carbon
til Ly v_/ 11
IMPROVE
Northeast
2,314
-3.8
63.0
-24.8
63.8
Midwest
574
-31.5
48.3
-41.9
64.3
Southeast
1,951
-34.1
55.5
-61.1
74.6
Central
2,499
-45.2
55.9
-60.7
72.4
West
10,238
-34.8
67.3
-37.5
72.0
69
-------
PM_TOT NMB (%) for run 2009ef2_v5_09d_12US1
for Annual for 12US1
units = %
coverage limit ¦ 75%
>100
80
60
40
20
0
-20
-40
-60
-80
<-100
CIRCLE=CSN; TRIANGLE=IMPROVE;
Figure 4-4. Normalized Mean Bias (%) of annual PM2.5 mass at monitoring sites in the
continental U.S. modeling domain
PM TOT NME (%) for run 2009ef2_v5_09d
12US1 for Annual for12US1
units = %
coverage limit - 75%
> 100
90
80
70
60
50
40
30
20
10
0
CIRCLE=CSN; TRIANGLE=IMPROVE;
Figure 4-5. Normalized Mean Error (%) of annual PM2.smass at monitoring sites in the
continental U.S. modeling domain
70
-------
S04 NMB
for run 2009ef2 v5 09d 12US1 for Annual for 12US1
units = %
coverage limit - 75%
> 100
80
60
40
20
0
-20
-40
-60
-80
<-100
CIRCLE=CSN; TRIANGLE=IMPROVE; SQUARE=CASTNET;
Figure 4-6. Normalized Mean Bias (%) of annual Sulfate at monitoring sites in the continental
U.S. modeling domain
units » %
coverage limit. 75%
S04 NME (%) for run 2009ef2_y5_09d_12US1 for Annual for 12US1
CIRCLE=CSN; TRIAI\IGLE=IMPROVE; SQUARE=CASTNET;
Figure 4-7. Normalized Mean Error (%) of annual Sulfate at monitoring sites in the
continental U.S. modeling domain
71
-------
(%) for run 2009ef2_v5__09d
units = %
coverage limit ¦ 75%
>100
80
60
40
20
0
-20
-40
-60
-80
<-100
N03 NMB
12US1 for Annual for 12US1
CIRCLE=IMPROVE; TRIANGLE=CASTNET;
Figure 4-8. Normalized Mean Bias (%) of annual Nitrate at monitoring sites in the continental U.S.
modeling domain.
units » %
coverage limit - 75%
> 100
CIRCLE=IMPROVE; TRIANGLE=CASTNET;
Figure 4-9. Normalized Mean Error (%) of annual Nitrate at monitoring sites in the continental
U.S. modeling domain
N03 NME (%) for run 2009ef2_v5_09d_12US1 for Annual for 12US1
72
-------
(%) for run 2009ef2_v5_09d
™i J—-'
TN03 NMB
12US1 for Annual for 12US1
CIRCLE=CASTNET;
Figure 4-10. Normalized Mean Bias (%) of annual Total Nitrate at monitoring sites in the
continental U.S. modeling domain
units ¦ %
coverage limit ¦ 75%
>100
80
60
40
20
0
-20
-40
-60
-80
<-100
(%) for run 2009ef2_v5_09d_12US1
units • %
coverage limit» 75%
> 100
CIRCLE=CASTNET;
Figure 4-11. Normalized Mean Error (%) of annual Total Nitrate at monitoring sites in the
continental U.S. modeling domain
TN03NME
for Annual for 12US1
73
-------
(%) lor run 2009ef2_v5 _09d
NH4NMB
units = %
coverage limit = 75%
>100
80
60
40
20
0
-20
-40
-60
-80
<-100
12US1 for Annual for 12US1
for run 2009ef2 v5 09d
CIRCLE=CSN; TRIANGLE=CASTNET;
Figure 4-12. Normalized Mean Error (%) of annual Total Nitrate at monitoring sites in the
continental U.S. modeling domain
NH4NME
12US1 for Annual for 12US1
units - %
coverage limit = 75%
¦100
CIRCLE=CSN; TRIANGLE=CASTNET;
Figure 4-13. Normalized Mean Error (%) of annual Ammonium at monitoring sites in the
continental U.S. modeling domain
74
-------
units = %
coverage limit - 75%
>100
80
60
40
20
0
-20
^0
-60
-80
<-100
EC NMB (%) for run 2009ef2_v5 09d 12US1 for Annual for 12US1
CIRCLE=CSN; TRIANGLE=IMPROVE;
Figure 4-14. Normalized Mean Bias (%) of annual Elemental Carbon at monitoring sites in the
continental U.S. modeling domain
(%) for run 2009ef2_
CIRCLE=CSN; TRIANGLE=IMPROVE;
Figure 4-15. Normalized Mean Error (%) of annual Elemental Carbon at monitoring sites in the
continental U.S. modeling domain
EC NME
09d_12US1 for Annual for 12US1
75
-------
units = %
coverage limit ¦ 75%
>100
80
60
40
20
0
-20
-40
-60
-80
<-100
OC NMB (%) for run 2009ef2_v5_09d_12US1 for Annual for 12US1
CIRCLE=CSN; TRIANGLE=IMPROVE;
Figure 4-16. Normalized Mean Bias (%) of annual Organic Carbon at monitoring sites in the
continental U.S. modeling domain
OC NME (%) for run 2008ab_08c_12US1 for 20080101 to 20081231
units = %
coverage limit ¦ 75%
CIRCLE=IMPROVE; TRIANGLE=CSN;
Figure 4-17. Normalized Mean Error (%) of annual Organic Carbon at monitoring sites in the
continental U.S. modeling domain
76
-------
5,0 Bayesiansj< 1 *, 1 J,«wnscaIIegfti.^«*
-------
point are the "posterior" distributions, and the means and standard distributions of these are used to
predict concentrations and associated uncertainties at new spatial points.
2) The model is "heirarchical" in structure, meaning that the top level parameters in Equation 1 (ie
~Po(s,t), Pi(t), ~x(s,t)) are actually defined in terms of further parameters and sub-parameters in the DS
code. For example, the overall slope and intercept is defined to be the sum of a global (one value for the
entire spatial domain) and local (values specific to each spatial point) component. This gives more
flexibility in fitting a model to the data to optimize the fit (i.e. minimize s(s,t)).
Further information about the development and inner workings of the current version of DS can be found
in Berrocal, Gelfand and Holland (2011) and references therein. The DS outputs that accompany this
report are described below, along with some additional analyses that include assessing the accuracy of the
DS predictions. Results are then summarized, and caveats are provided for interpreting them in the
context of air quality management activities.
5.3 Downscaler Output
In this application, DS was used to predict daily concentration and associated uncertainty values at the
2010 US census tract centroids across the continental U.S. using 2009 measurement and CMAQ data as
inputs. For ozone, the concentration unit is the daily maximum 8-hour average in ppb and for PM2.5 the
concentration unit is the 24-hour average in |J,g/m3. DS output is in the form of a comma-delimited table.
Example output of the 2009 ozone DS run is shown in Table 5-1. Each row is specific to date and census
tract. The columns of the output files are:
• Date - represented by the data given in this row, in MM/DD/YYYY format.
• Census TractFIPS code (http://quickfacts.census.gov/qfd/meta/long fips.html
• Latitude: The y-coordinate value transformed to latitude (degrees).
• Longitude: The x-coordinate value transformed to longitude (degrees).
• Prediction: Daily maximum estimated 8-hour ozone concentration in ppb or 24 hour average
PM2 5 in ug/m3.
• Uncertainty. The posterior standard deviation (error) of the estimated ozone or PM2.5
concentration.
78
-------
Table 5-1. Downscaler Model Prediction: Example Data File (Ozone)
Date
2010 US Census
Latitude
Longitude
Daily Maximum 8-
Standard Error
Tract FIPS
Hour Concentration
of
Code
(PPb)
Concentration
Jan-01-2002
1001020100
32.47718
-86.49001
26.122939
7.303598
Jan-01-2002
1001020200
32.47425
-86.47339
25.900202
7.373431
Jan-01-2002
1001020300
32.47544
-86.4602
26.130488
7.248804
Jan-01-2002
1001020400
32.47204
-86.4437
26.085497
7.312579
Jan-01-2002
1001020500
32.45892
-86.42271
25.942581
7.297929
Jan-01-2002
1001020600
32.44253
-86.47877
25.854266
7.347192
Jan-01-2002
1001020700
32.42723
-86.44118
25.534966
7.311914
Jan-01-2002
1001020801
32.41336
-86.5261
25.707846
7.50416
Jan-01-2002
1001020802
32.53474
-86.51259
26.253984
7.41871
Jan-01-2002
1001020900
32.64296
-86.52377
26.212948
7.552828
Jan-01-2002
1001021000
32.60895
-86.75607
25.718179
7.573458
Jan-01-2002
1001021100
32.45595
-86.73223
26.065351
7.432891
5.4 Downscaler Model Results for the 2009 Application
Monitoring data for 2009 from the AQS database described in Chapter 2 and output from the 2009 12 km
resolution CMAQ run described in Chapter 4 were input to DS to produce daily spatial predictions at the
2010 continental U.S. census tract centroids. The following summary information was extracted and
calculated for the DS ozone and PM2.5 inputs and outputs:
• Days with the highest and greatest spatial extent of high pollution
• Locations with the most days above the NAAQS
• Comparison between daily AQS observations and the nearest DS census tract prediction for
selected sites
5.4.1 Summary of 8-hour Ozone Results
As a summary of the overall year, Figure 5.1 shows the 4th max daily maximum 8-hour average ozone for
AQS observations, CMAQ model predictions and downscaler model results. Based on downscaler model
estimates for 2009, approximately 26 percent of the US Census tracts (17,280 out 66,186) have at least
one day with an ozone value above 75 ppb.
79
-------
AQS
CMAQ
2009
4'th Max, Daily max
8-hour avg
ozone (ppb)
(-lnf,55]
(55,60]
(60,65]
(65,70]
(70,75]
(75,80]
¦ (80,85]
¦ (85,90]
- (90, Inf]
Figure 5-1. Observed, modeled and predicted annual 4"' max (daily max 8-hour ozone
concentrations)
80
-------
Figure 5-2 shows the location of the 4,952 census tracts that were predicted by the downscaler model to
have annual 4th max daily max 8 hour average ozone concentrations above 75 ppb. Approximately
24,638,377 million people (1,871,717 are 65 years in age or older) live in these 4,952 census tracts. Most
of the high ambient ozone concentrations are in California and followed by Texas.
Census tract centroids where "4 high daily maximum
8-Hour average" values exceed 75 ppb
Figure 5-2. Census tract locations where annual 4ih max daily maximum 8-hour average ozone
concentrations estimates are above 75 ppb in 2009
Table 5-2 ranks the days in 2009 based on the combined spatial extent and intensity of the ozone
estimates above 75 ppb (only top 25 days are shown). There are 132 days on which at least one census
tract was predicted to have an ozone concentration above 75 ppb. This approach ranks the days of the
year based on two criteria: (1) spati al extent in terms of the number of census tracts where ozone
concentrations are above 75 ppb (spatial extent criterion), and (2) average ozone concentrations at those
locations (intensity criterion). Sunday, August 30, 2009 is the most intense day. It covers 2,696 census
tracts with average ozone concentrations of 91 ppb. Figure 5-2 shows the location of these census tracts
in yellow and red. August 3GUl is ranked second in terms of spatial extent. The combined spatial extent
and average ozone concentration scores make this day the highest ranked ozone day in 2009. Figure 5-3
shows that the high ozone concentrations are concentrated in a relatively small geographical area.
However, it is spatially coincident with a highly urbanized/populated area (Los Angeles, CA) thus
explaining the high number of census tracts. It is important to distinguish here that the spatial extent
criterion used above is based on the number of census tracts not the geographical extent.
81
-------
Table 5-2. Rank order of days in 2009 based on combined spatial extent and intensity of ozone
estimates (only top 25 days are shown)
Day
Spatial Extent in
Spatial Extent
Average
Intensity
Overall
Terms of Census
Ranking
Ozone (ppb)
Ranking
Rank
Tract Count
(truncated)
Sunday, August 30,2009
2696
2
91
1
1
Saturday, August 29,2009
2543
4
86
5
2
Saturday, July 18,2009
2177
10
89
2
3
Sunday, September 27,2009
2418
6
85
7
4
Sunday, June 28,2009
1700
15
85
9
5
Wednesday, July 01,2009
2622
3
83
23
6
Wednesday, August 19,2009
1686
16
84
12
7
Monday, August 31,2009
1497
20
85
10
8
Tuesday, August 18,2009
2078
11
84
19
9
Thursday, September 03,2009
1656
18
84
13
10
Friday, August 28,2009
1421
21
84
14
11
Saturday, June 27,2009
2483
5
82
31
12
Friday, September 18,2009
1509
19
83
22
13
Saturday, May 30,2009
2036
12
82
29
14
Thursday, July 02,2009
2380
7
82
35
15
Sunday, July 19,2009
1675
17
82
26
16
Saturday, September 26,2009
1976
13
82
32
17
Wednesday, August 12,2009
821
40
85
6
18
Tuesday, August 11,2009
1156
27
84
20
19
Sunday, May 17,2009
1754
14
81
37
20
Thursday, August 20,2009
810
42
85
11
21
Tuesday, July 07,2009
1075
30
83
24
22
Monday, August 17,2009
3024
1
80
53
23
Saturday, August 01,2009
744
44
84
17
24
Saturday, July 25,2009
714
45
84
18
25
Bottom part of the Figure 5-3 shows the uncertainties (posterior standard errors) associated with the
predictions made in August 30, 2009. Posterior standard errors are lower in the North Eastern quadrant
of the US aided by good monitor coverage (Figure 2-1) and low observed ozone concentrations. Also
elevations in the magnitude of errors can be seen over areas where ozone concentrations are high and the
prediction locations are not far from the monitoring sites such as Los Angeles, C A. General trend with
the standard errors of prediction is that the magnitude increases with high ozone concentrations and
predictions locations further away from monitoring locations. This can be best seen in the scatter plot of
August 30 predictions and associated posterior standard error with each prediction is color coded based
on its distance to the nearest ozone monitor (Figure 5-4). Similar patterns observed over the predictions
made in others days and will be discussed throughout the document.
82
-------
August 30, 2009 - Ozone (ppb)
30 August 2009 - UNCERTAINTY
Posterior Standard Deviation (Error)
Figure 5-3. August 30, 2009, ozone concentrations for the 2010 US Census Tract locations predicted
by downscaler model (Top) and posterior standard deviation of the predictions (Bottom)
83
-------
August 30, 2009
Distance to the Nearest Ozone Monitor
* 41 -10,000 meters
* 10,001 -25,000 meters
25,001 - 50,000 meters
50,001 - 75,000 meters
75,001 -1 00,000 meters
* 100,001 -150,000 meters
* 150,001 -333,252 meters
i 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 r
20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105
Predicted Ozone Concentrations (ppb)
Figure 5-4. Scatter plot of predicted ozone concentrations in August 30, 2009 and associated
posterior standard deviations. Each prediction is color coded based on its distance to the nearest
ozone monitor.
Monday, August 17, 2009 is the highest ozone day in terms of spatial extent covering 3,024 census tracts
over the two largest metropolitan areas in the U.S., New York, NY and Los Angeles, CA where
approximately 13,190,042 people live (Figure 5-5). Ozone concentrations are averaging 79.91 ppb with a
maximum concentration of 96 ppb. New York City, NY and surrounding areas are the only East Coast
areas predicted by the downscaler model to have concentrations above 75 ppb. Regarding to associated
uncertainty with August 17 predictions (Figure 5-6), standard error of the predictions are elevated by high
ozone concentrations observed over New York area. Contrary to August 30, errors in the South East are
less than the North East of the United States on August 17. Scatter plot of the predicted ozone
concentrations and posterior standard deviations (Figure 5-6) shows similar pattern observed in August
30th (Figure 5-4).
84
-------
»KiY!
Sck- **
• •••*.» /¦*1
•>. 4v:l
August 17, 2009 - Ozone (ppb)
17 August 2009 - UNCERTAINTY
Posterior Standard Deviation (Error)
Figure 5-5. August 17, 2009 ozone concentrations for the 2010 US Census Tract locations predicted
by downscaler model (Top) and standard deviations of the predictions (Bottom).
85
-------
August 17, 2009
Distance to the Nearest Ozone Monitor
• 41 -10,000 meters
• 10,001 -25,000 meters
25,001 - 50,000 meters
50,001 - 75,000 meters
75,001 -1 00,000 meters
• 100,001 -150,000 meters
• 150,001 -333,252 meters
20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
Predicted Ozone Concentrations (ppb)
Figure 5-6. Scatter plot of predicted ozone concentrations and associated standard deviations in
August 17, 2009. Each prediction is color coded based on its distance to the nearest ozone monitor.
Shown in Figure 5-7 July 181h is another highly ranked ozone day, 2nd in intensity and 10th in spatial
extent covering Dallas-Fort Worth, TX and Los Angeles, CA). On this day, 2177 census tracts are
estimated to have an average ozone concentration of 89 ppb.
During Friday, May 15, 2009 and Tuesday, March 10, 2009, only one census tract is predicted to be
above 75 ppb, which is considered to be the lowest amongst the high ozone days. In general, August is
the month with the highest ozone followed by July and September, respectively. In August, average
ambient ozone concentrations are estimated to be 83 ppb on days with ozone above 75 ppb.
86
-------
. •••• xy <'i^t * ••Jrirj^rli'
&*' * *
July 18, 2009 - Ozone (ppb)
:• **?&* • V:' *.: i *"ri *
J*.' s' •* . 4 •%!%• . Ju,!^
\L*#c '•:?•
• •~. • .wll ••rdk*
* * •*!. 4J; • •.
e.36i•••*
f • > * 5 ®T* •*>•!# A -
18 July 2009 - UNCERTAINITY
Posterior Standard Deviation (Error)
• • • • • • • i
O ^ Q O <5 ^ ^ n5* nS
«>• V - V1 v'
Figure 5-7. July 18, 2009 ozone concentrations for the 2010 US Census Tract locations predicted by
downscaler model (Top) and standard deviations of the predictions (Bottom).
87
-------
July 18, 2009
Distance to the Nearest Ozone Monitor
• 41 -10,000 meters
• 10,001 -25,000 meters
25,001 - 50,000 meters
50,001 -75,000 meters
75,001 -1 00,000 meters
• 100,001 -150,000 meters
• 150,001 -333,252 meters
—i 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 r-
15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120
Predicted Ozone Concentrations (ppb)
Figure 5-8. Scatter plot of predicted ozone concentrations and associated standard errors in July
18, 2009. Each prediction is color coded based on its distance to the nearest ozone monitor.
Table 5-3 ranks each census tract based on the number of days that the daily maximum 8 hour
concentrations are above 75 ppb. Associated Figure 5-9 displays the number of ozone days above 75 ppb
and their location. Based on the downscaler model estimates, census tracts "06071008602" and
"06071008706" have the highest number of days (72) with ozone above 75 ppb (72 days each). For both
census tracts the average ozone concentrations for those days is 85 ppb. On those 72 days the maximum
ozone concentrations were 116 and 115 ppb for the two tracts, respectively. The top 18 tracts are in San
Bernardino County in California (FIPS code "071" identified by 3rd through 5th character in Census Tract
ID) followed by tracts that are located in Riverside County, CA (FIPS code "065").
88
-------
Table 5-3. Census tract rankings based on ozone estimates (out of 17,280 census tracts that
predicted to have at least one day with ozone concentration above 75 ppb, only top 30 are shown)
Census Tract ID
Number of Days
Above 75 ppb
Ranking based on
Ozone Days
Average Ozone Above
75 ppb (truncated)
Maximum
Ozone
06071008602
72
1
85
116.1
06071008706
72
1
85
115.2
06071007904
70
3
86
118.8
06071008705
70
3
85
115.3
06071008401
69
5
86
119.0
06071008402
69
5
86
118.6
06071008404
69
5
85
118.1
06071008601
69
5
86
118.0
06071008703
69
5
85
111.6
06071008704
69
5
85
113.7
06071008710
69
5
85
113.6
06071007901
68
12
86
119.1
06071007903
68
12
86
119.2
06071008403
68
12
86
117.9
06071008500
68
12
85
115.8
06071008800
68
12
86
113.4
06071008708
67
17
86
113.0
06071011101
67
17
86
117.3
06065043809
66
19
85
108.0
06065043811
66
19
85
110.1
06065044104
66
19
85
106.2
06071007604
66
19
86
119.2
06071008002
66
19
86
118.2
06071008200
66
19
85
117.1
06071008709
66
19
86
112.9
06071011002
66
19
86
117.2
06065043802
65
27
85
111.7
06065043823
65
27
85
112.4
06065044200
65
27
85
103.0
89
-------
\* w/—T'J
W
^ J ^>X j
/ i V y-
Ozone days above 75ppb
:--'r 1 V -s-7 1 CT —^
1-3 days
4-15 days
16-30 days
• 31-45 days
• 46-72 days
Figure 5-9. Number of ozone days above 75 ppb for US census tracts predicted by DS.
The downscaler model estimates can track the AQS observations and CMAQ predictions, and the
downscaler model estimates can differ from either the AQS observations or the CMAQ predictions. To
see how the daily downscaler model estimates compare to the AQS observations, we selected the
monitors that are within 100 meters of a census tract centroid. Census tract and AQS site pairs are shown
in Table 5.4. The associated Figure 5-10 shows the time series data for the listed sites. As shown the
downscaler model estimates generally follow the AQS and CMAQ data. Keep in mind that the
downscaler concentrations are point estimates trying to replicate the point measurement conditions of the
AQS monitoring site. CMAQ concentrations represent the average conditions within 12 by 12 km grid
cells. To further elaborate this condition, Figure 5-11 shows the relationships among the AQS ozone
monitoring site locations, CMAQ grid cells and the US census tract centroids in the Los Angeles area.
The CMAQ cells are on a continuous grid. EPA ozone monitor siting criteria requires States to place
monitors in and around the urban areas with high populations. The US Census Bureau uses population
size to define the census tract boundaries. Census tract population sizes vary between 1,200 and 8,000.
The optimum size is 4,000 people. Therefore, in the Los Angeles area, it is not a surprise to see how
dense the census tract locations are in the urban areas, where the monitoring sites and high population
areas are located. Also, not surprising is how less dense the census tracts are in the rural areas where
there are few, if any ozone monitors.
90
-------
Table 5-4. List of AQS sites that are within 100 meters of a census tract centroid
040134004
04013811200
Maricopa
Arizona
120712002
12071010801
Lee
Florida
320032002
32003004000
Clark
Nevada
420950025
42095017800
Northampton
Pennsylvania
421010004
42101019000
Philadelphia
Pennsylvania
421250200
42125754400
Washington
Pennsylvania
170310064
17031836200
Cook
Illinois
91
-------
Northampton County. PA (AQS ID 420950025; Census Tract
ID 42095017800)
Clark County, IMV (AQS ID 320032002; Census Tract ID
32003004000)
a*0 sV rt* vV *1* Jy Jy (v
^ ^ & <3? ^ sT ^ ^ ^ ^
Philadelphia County, PA (AQS ID 421010004; Census Tract
ID 42101019000)
AQS CMAQ Downscaler AQS CMAQ -Downscaler
Figure 5-10. Daily 8-hour maximum ozone concentrations measured by AQS monitor and
estimated by CMAQ model and Downscaler fusion for selected sites in Table 5-4 (Census centroids
that are within 100 meters of and AQS site).
— so
-Q
a, 70
a
T 60
c
.2 50
2 40
§ 30
£ 20
Maricopa County, AZ (AQS ID 040134004; Census Tract ID
04013811200)
—. 80
-G
5- 70
Washington County, PA (AQS ID 421250200; Census Tract
ID 42125754400)
, Cook County, IL (AQS ID 170310064; Census Tract ID 17031836200)
xv .y Jr
Lee County, FL(AQS ID 120712002; Census Tract ID 12071010801)
92
-------
+ AQS Slles Census Tract Centroids
r~] CMAQ 12 km Grids Auflus'17' 2009 "03 pred,c,ed °ZOm 1 ppb» #
County Boundaries gSgggSSSSS8
Urban Areas »ES5SPE5$58
Figure 5-11. Downscaler model predictions over Los Angeles, CA and surrounding areas
5.4.2 Summary of PM2 5 Results
As a summary of the overall year, Figure 5-12 and Figure 5.13 show the annual means and the 98lh
percentile 24-hour average PM2.5 concentrations for AQS observations, CMAQ runs and downscaler
predictions. Based on downscaler model estimates for 2009, the 98th percentile of PM2.5 values are above
35 j.ig/m3 for the 1,137 census tracts (Figure 5-14) averaging 43.8 ug/m \ 18,056 Census tracts have at
least one day with a PM2.5 concentration above 35 ,ug/m3 7,526 census tracts have PM2.5 annual average
concentrations above 12 ug/irr (mean 13.2 jig/m3).
93
-------
AQS
CMAQ
2009
Annual mean,
24-hour avg
PM2.5 (ug/m3)
(0,3]
(3,5]
(5,8]
(8,10]
(10,12]
(12,15]
(15,18]
¦ (18,lnf]
DS
Figure 5-12. Observed, modeled and predicted annual mean PM2.5 concentrations)
94
-------
AQS
Figure 5.13 Observed, modeled and predicted 98th percentile 24-hour average PM2.5
concentrations.
CMAQ
2009
98'th percentile,
24-hour avg
PM2.5 (ug/m3)
(0,10]
(10,15]
(15,20]
(20,25]
(25,30]
(30,35]
(35,40]
¦ (40,45]
¦ (45,50]
" (50,lnf]
95
-------
*
*
US Census Tract Locations
u12.5 1
\
v.-
US Census Tract Locations
• PM2 5 Annual Average above 12 |jg/m3
Figure 5-14. Census tract locations (centroid) where the 98th percentile of 24-hour average and
annual average PM2.5 concentrations are above 35 fig/nr (top) and 12 jig/m3 (bottom) respectively.
96
-------
Table 5-5 ranks the days in 2009 based on the combined spatial extent and intensity of the 24-Hour
average PM2.5 concentration estimates above 35 |ig/m3 (only top 25 days are shown). There are 102 days
on which at least one census tract was predicted to have a 24-hour average PM2.5 concentration above 35
|ig/m3. Similar to ozone in section 5.4.1, this approach ranks the days of the year based on two criteria:
(1) spatial extent in terms of the number of census tracts where 24-hour average PM2.5 concentration
estimates are above 35 |ig/m3 (spatial extent criterion), and (2) average PM2.5 concentrations at those
locations {intensity criterion). In overall, Thursday, January 01, 2009 is the highest ranked PM2.5 day
based on the two criterions mentioned above (Figure 5-15). Tuesday, October 13, 2009, is the most
intense day with average concentrations of 57 |ig/m3 covering 107 census tracts (Figure 5-17). Thursday,
January 22, 2009, is the highest PM2.5 day in terms of spatial extent covering 4,589 census tracts (Figure
5-19).
Table 5-5. Rank order of days in 2009 based on combined spatial extent and intensity of 24-Hour
average PM2.5 concentration estimates (only top 25 days are shown)
Day
Spatial Extent in
Terms of Census
Tract Count
Spatial
Extent
Ranking
Average PM25
(jig/m3)
(truncated)
Intensity
Ranking
Overall
Rank
Thursday, January 01,2009
4202
2
54
6
1
Thursday, December 10,2009
1706
9
47
13
2
Friday, December 11,2009
1391
12
48
11
3
Tuesday, January 20,2009
836
21
55
5
4
Wednesday, January 21,2009
746
24
55
3
5
Friday, January 02,2009
3330
3
43
24
5
Thursday, January 22,2009
4589
1
43
26
5
Thursday, December 03,2009
652
29
52
7
8
Friday, December 04,2009
2534
5
43
31
8
Saturday, December 05,2009
1142
14
44
23
10
Friday, January 16,2009
1163
13
43
27
11
Monday, January 19,2009
1987
7
42
33
11
Tuesday, January 13,2009
725
25
46
17
13
Wednesday, December 02,2009
454
37
50
8
14
Sunday, January 18,2009
688
28
46
18
15
Friday, December 25,2009
1977
8
42
38
15
Friday, December 18,2009
2572
4
41
43
17
Tuesday, December 29,2009
505
35
47
14
18
Saturday, January 17,2009
806
22
43
29
19
Sunday, August 30,2009
1137
15
42
37
20
Thursday, January 15,2009
439
39
46
15
21
Sunday, December 20,2009
1410
11
41
46
22
Saturday, January 31,2009
452
38
44
22
23
Tuesday, October 13,2009
107
60
57
1
24
Wednesday, January 14,2009
419
41
44
21
25
97
-------
January 1, 2009 - PM2.5
January 1, 2009 - UNCERTAINLY
Posterior Standard Deviation (Error)
Figure 5-15. January 1, 2009 24-Hour PM2.s concentrations for the 2010 US Census Tract locations
predicted by downscaler model (Top) and posterior standard error of the predictions (Bottom)
98
-------
January 1, 2009 - PM2.5
11 ¦
-1 1 ! 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 T1
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
Predicted PM2.5 Concentrations (pg/m3)
10.5
10
9.5
9
8.5
Distance to the Nearest PM2 5 Monitor
4) 41 -10,000 meters
10,001 -25,000 meters
25,001 - 50,000 meters
50,001 - 75,000 meters
75,001 -100,000 meters
100,001 -1 50,000 meters
• 150,001 -333,252 meters
Figure 5-16. Scatter plot of predicted PM2.5 concentrations and associated posterior standard
errors in January 1, 2009. Each prediction is color coded based on its distance to the nearest ozone
monitor.
99
-------
. • •*,.'* C VjX-T•••>•!
•"••V
October 13, 2009 - UNCERTAINITY
Posterior Standard Deviation (Error)
Figure 5-17. October 13, 2009 24-Hour PM2.5 concentrations for the 2010 US Census Tract
locations predicted by downscaler model (Top) and posterior standard error of the predictions
(Bottom). Kern County, CA is highlighted with red rectangle.
100
-------
October 13, 2009 -PM2.5
.. 5i,V ;•
Kern, CA
Distance to the Nearest PIV12 5 Monitor
£ 41 -1 0,000 meters
10,001 - 25,000 meters
25,001 - 50,000 meters
50,001 - 75,000 meters
75,001 -100,000 meters
| 100,001 -1 50,000 meters
• 150,001 -333,252 meters
1 -*5 1 1 1 1 1 1 1 i 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 r-
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66
Predicted PM2.5 Concentrations (pg/m3)
1 5.5-
Figure 18. Scatter plot of predicted PM2.5 concentrations and associated posterior standard errors
in October 13, 2009. Each prediction is color coded based on its distance to the nearest ozone
monitor.
101
-------
January 22, 2009 - PM2.5
Salt Lake City, UT
January 22, 2009 - UNCERTAINTY
Posterior Standard Deviation (Error)
Figure 5-19. January 22, 2009 24-Hour PM2.5 concentrations for the 2010 US Census Tract
locations predicted by downscaler model (Top) and posterior standard error of the predictions
(Bottom). Salt lake City, UT is highlighted with red rectangle.
102
-------
January 22. 2009 - PM2.5
Distance to the Nearest PM,
0 41 -10,000 meters
0 10,001 - 25,000 meters
25,001 - 50,000 meters
50,001 - 75,000 meters
75,001 - 100,000 meters
100,001 -150,000 meters
• 150,001 -333,252 meters
Monitor
—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—f
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 48 48 50 52 54 56 58 60 62 64 66 68 70 72 74
Predicted PM2.5 Concentrations (|jgfrn3)
Figure 5-20. Scatter plot of predicted PM2.s concentrations and associated posterior standard
errors in January 22, 2009. Each prediction is color coded based on its distance to the nearest ozone
monitor.
103
-------
Table 5-6 ranks each census tract based on the number of days and average concentrations that the 24-
Hour average PM2.5 concentration estimates above 35 |ig/m3.
Associated Figure 5-21 displays the number of PM2.5 days above 35 |ig/m3 and their location. Based on
the downscaler model estimates, census tracts "06029002814" has the highest number of days (37) with
an average concentration of 50.8 |ig/m3. The top 30 tracts are in Kern County in California (FIPS code
"029" identified by 3rd through 5th character in Census Tract ID).
Table 5-6 . Census tract rankings based on PM2.5 estimates (out of 18,056 census tracts that
predicted to have at least one high concentration day of PM2i5 , only top 30 are shown)
Census Tract ID
Number of Days
Above 35 fig/m3
Average PM2.5
(jig/m3)
Maximum PM2.5
(jig/m3)
Rank
06029002814
37
50.8
75.3
1
06029002815
37
50.5
74.5
2
06029002816
37
50.4
74.5
3
06029002812
36
51.6
75.8
4
06029003112
37
50.1
73.7
5
06029002813
36
51.4
75.5
6
06029001802
36
51.4
75.6
7
06029001801
36
51.3
75.7
8
06029002804
36
51.3
75.4
9
06029001901
36
51.2
75.1
10
06029003113
37
49.8
73.4
11
06029002807
36
51.1
75.1
12
06029002700
36
51.0
74.9
13
06029002806
36
51.0
74.6
14
06029002818
36
51.0
74.5
15
06029003812
36
50.9
74.4
16
06029002808
36
50.9
74.3
17
06029002900
36
50.8
74.2
18
06029001902
36
50.8
74.6
19
06029003114
37
49.4
72.6
20
06029002817
36
50.7
74.3
21
06029003811
36
50.7
73.8
22
06029000507
36
50.6
74.3
23
06029002819
36
50.5
73.9
24
06029002811
36
50.5
73.7
25
06029000506
36
50.4
74.0
26
06029003808
36
50.4
73.5
27
06029001700
36
50.4
74.3
28
06029002821
36
50.4
73.5
29
104
-------
\
V / I " ~1
PM2.5 days above 35 pg/m3
1 - 7 days
8-10 days
• 11-20 days
• 21-30 days
• 31-37 days
Figure 5-21 Number of days 24-hour average PM2i5 concentrations are above 35 jiig/ni ' for the US
census tracts predicted by DS
5.5 Accuracy Assessment of Downscaler Model Results
This section describes the predictive performance of DS in the 2009 application. The general approach
for this involves running DS with a subset (in this case 10%) of monitors removed from the moni toring
data set and predicting to the spatial points of the removed monitors. This approach is sometimes called
"cross-validation" (CV). Errors and biases can then be calculated for each prediction point by comparing
the resulting prediction to the actual monitoring data. For this application, the default CV method in the
DS software was followed, which involves leaving out a random sample of 10% of the monitors in each
day of data. The sites left out on one day are not necessarily the same set of sites on another day.
Monitor- and day-specific errors and biases were then aggregated into the metrics below to provide an
overview of the model's accuracy.
First, day-specific Root Mean Square Error (RMSEd), Mean Absolute Error (PMAEd), and biasd (or Mean
Bias Error) were calculated to evaluate the predictive capability of the Downscaler model. Daily RMSE
is defined as
I72
i2
RMSE„ =
n
7 = 1
105
-------
where Pj and (), are the DS prediction and observed concentrations respectively in location j. d is the
specific day of the year. Even though both the RMSE and the MAE measure the average magnitude of
the errors, it is useful to report both to diagnose the variation in the errors. Daily MAE is defined as
MAE„ =
n
Oj,d |
y=i
While the MAE gives equal weight to all, the RMSE emphasizes large errors and is most useful when
large errors are undesirable. Given the fact that the RMSE will always be larger or equal to the MAE, the
difference between the two highlights the magnitude of the variance in the individual errors in such the
greater the difference, the greater the variance. Daily bias is defined as
bias,
— n 1 Pj.d ~ Oj.d) — Pd~
0r
y=i
where Pd and 0d are the model-predicted and observed daily mean concentrations respectively.
Secondly, location-specific Root Mean Square Error {RMSE,), Mean Absolute Error {MAE,), and bias,
were calculated as:
RMSEj =
n
'Zlp.
d,j ®d,j |
d=1
V2
where Pd and Od are the DS prediction and observed concentrations respectively in day d. j is the specific
observation (monitor) location.
MAEj =
n
"Zk
J ®d,j |
d=1
and
biasj = n 1 ^(Pdj ~ 0dJ) = Pj - Oj
d=1
where Pj and Oj are the model-predicted and observed location-specific mean concentrations
correspondingly.
Thirdly, to further analyze how the downscaler performed over different locations, the Getis-Ord Gi*
statistic (Getis and Ord, 1992)16 for each RMSE, and MAE, are calculated, which returned a z-score for
each monitor location. For statistically significant positive z-scores, the larger the z-score, the more
intense the clustering of high values indicating relatively poor model performance. For statistically
significant negative z-scores, the smaller the z-score is, the more intense the clustering of low values
which indicates better model performance. Getis-Ord local statistics is calculated as17:
16 Getis, A. and J.K. Ord. 1992. "The Analysis of Spatial Association by Use of Distance Statistics" in Geographical Analysis
24(3).
17The ArcGIS 10.1 Resources: How Hot Spot Analysis works:
http://resources.arcgis.eom/en/help/main/10.l/index.html#/How Hot Spot Analysis Getis Ord Gi works/QQ5pQQQQQQll
000000/
106
-------
G- =
Tj=iWj,jXj -X^iWij
nzy=1w5 - (Z?=1wu)2
n — 1
where Xy is either MA/i, or RMSEj for monitor /, Wjj is the spatial weight between monitor i and /, n is
equal to total number of AQS monitors. X and S2 are sample mean and variance:
X = V->xl/r
and
n
5 =
Z/=i*/2
n
W2
5.5.1 Assessment of 8-hour Ozone Run
Daily RMSEs, MAEs and Bias values are depicted in Figure 5-22. Daily Bias values are ranging from -
1.5 to 2 ppb. Ranges for the daily RMSEj and the MAE^ are 2.7 to 7.5 ppb and 2.1 to 5.4 ppb,
respectively. On January 25th, the variance in the individual errors, the difference between the RMSE and
the MAE were minimal. On April 5th, however, the variance was the highest. These results are
somewhat aligned with the test results reported in the Berrocal et. al. (2009) paper which documents
overall performance of the downscaler model. This provides some confidence that in general the 2009
application of the downscaler model is performing reasonably well.
107
-------
Daily Bias Daily RMSE Daily MAE
Figure 5-22. Daily validation results
In Figure 5-23 the location specific RMSE,, and MAE, values are presented over 1207 monitoring
locations. Both the RMSE, and the MAE, show similar patterns over the US domain with a slightly better
fit for the Eastern US than the Western part of the country.
108
-------
%
MAE
• I 0 * 1 9 pbb
O 2.0-2.4 pbb
O 2 6 2 9 pbb
O 3.0-3.3 pbb
O 3,6 • 4.2 pbb
O 4.3 • 5.3 pbb
O 5.3 *0,6 pbb
O 6.7 - 92 pbb
O 9 3-13 4 pbb
• 15.3 - 26.7 pbb
- srJfW**' *>
^ 0®00."j& n . *
ep®
J o* £
^ ..
o
'|0^o ^
•!° °0 „ • -
^£0Oc% 0 o QO • ° ® o°
& tSL«*» o&g^®0*%
o
%
(Jto
•' ':• !
%
0 3.1-3.7
0 3.6 • 4.4
~ 4.5 - 5.2
O 5.3 ¦ 6.2
O 6.3.7,7
~ 7.6-10.6
¦ 10T-I6I
CP O
s !
Oyffl " n
o*bo ^ <*
. % o *f I o
°°! ° ° °
& «a
o o oo & ° o
o «
Figure 5-23. The average magnitude of the errors in 2009 predictions based on the spatio-temporal
Downscaler model: the Mean Absolute Error (Top) and the Root mean Square Error (Bottom).
109
-------
The Getis-Ord Gi* statistic (pronounced G-i-star) for each RMSE, and MAE,- are calculated, which
returned a z-score for each monitor location (Figure 5-24). Clearly, the downscaler model performs
better over the Eastern US than the Western part of the country. There was intense clustering of high
RMSE, and MAE,- values in the West (statistically significant positive z-scores) indicating relatively poor
model performance. On the other hand over the Eastern US, there were statistically significant negative z-
scores indication of the clustering of low RMSE,- and MAE, values which is most likely due to the higher
density of the ozone monitoring network in the East.
Lastly, overall DQO metrics across locations and time are calculated. For 2009, the RMSE, MAE and
bias values are 4.7, 3.3, and 0.01 ppb, respectively. The difference between overall RMSE and MAE isn't
big enough to indicate the presence of very large errors however there is some variation in the magnitude
of the errors.
110
-------
i*
iSr/-
fi.. /
- m
r
i; *
Model Validation
Gi* Z-Score for the RMSE
• <-2 58SW Dev
• -258 • -1 96 SM Dev
• -1 96--166SM Dev
-1.65- 1 65 Sd Dev
• 1.66-1.96 SW Dev
• 1 96 - 258 SW, Dev
• >256 Sid Dev
I*
ft
9
Model Validation
Gi* Z-Score for the MAE
• < -2 58 Sid Dev
• -258- 196SM Oev
• -1 96 • -1.66 SM Dev
-1 86-1 65SM Dev
• 1.66 -1 96 SM Dev
• 1.96 - 258 Sid Dev
• > 258 Sid Dev
•
•••4 v/y J \
* m >1, » A '
. A . <
•), f • • s %
°' Tw
#
4?
&
Figure 5-24. The Gi* statistic returned z-scores for each monitor locations over the United States.
Both the RMSE/ (Top) and the MAE; (Bottom) based Gi* z-scores show similar patterns with very
slight differences.
Ill
-------
5.5.2 Assessment of PM2 5
Daily RMSEs, MAEs and Bias values are depicted Figure 5-25. Daily Bias values are ranging from -3.8
to 2 |ig/m3. Ranges for the daily RMSE and the MAE are 0.5 to 13.3 |ig/m3 and 0.4 to 5.2 |ig/m3,
respectively. On September 27th, the variance in the individual errors, the difference between the RMSE
and the MAE were minimal. On January 17th, the variance was the highest; it was not enough to be great
concern however.
14
12
10
8
6
4
2
0
2
-4 H 1 1 1 1 1 1 1 1 1 1
f J* ,4?
PM2.5 Daily Bias ^—PM2.5 Daily RMSE PM2.5 Daily MAE
Figure 5-25 Daily validation results for PM2.s
In Figure 5-26 the location specific RMSE, and MAE values are presented over 929 monitoring locations.
Both the RMSE and the MAE show similar patterns over the US domain with a slightly better fit for the
Eastern US than the Western part of the country.
112
-------
<3°
* 9
CO
S§ *0
(0
c*o
o° °
£ o®
• ^3
° 030 CP
<^QP
®s
o
3
0 o
o
o
o
O •
%
c9
o
o
o
o
D^O
*• o Q
O
MAE (PM2.5)
•
0.15-0.72
O
0.73 - 0.99
O
1.00-1.25
O
1.26-1.56
O
1.57-1.94
O
1.95-2.39
O
2.40 - 3.01
0
3.02 - 3.94
0
3.95 - 5.43
0
5.44-10.40
OO
*
2?
o
•c
— •
Q° 0
°o °°® °®<
o OO®
8>
o
o
o
99
<2>
*3
O
a »
03
044 o.o° a ° g°
° ° « o °
o oo ° e^°. <9°
^3000
I 8
go cP<£>0 Q °
o
#0 G{
bo
o •
o
o o
W>
.CP
7
§
1 •
@
o
0
1
cS
c*o
Sg
cP
o° 0
o o
O O
c9.
o
o %,
0 0)0 CP
o% ^
9* ®#
cP o
§00
•rf"\0.
cP
.0)
@°Q0
RMSE (PM2.5)
• 0.2-1.0
O 1.0-1.3
O 1.3-1.7
O 1.7-2.2
O 2.2-2.8
O 2.8 - 3.6
O 3.6-4.8
O 4.9-6.6
O 6.9-9.7
• 11.2-13.8
v_/
0 0
0
p 0
<9 0
0
• 0 °
"0
0
f
O
°n °
°o °°« °*<
O 00®
v0>- f
&
O
O
o
o
6
op o
Figure 5-26 The average magnitude of the errors for PM2.5 in 2009 predictions based on the spatio-
temporal downscaler model: the Mean Absolute Error (Top) and the Root mean Square Error
(Bottom).
113
-------
Similar to ozone assessment, the Getis-Ord Gi* statistic for each RMSE and MAE are calculated for
PM2.5, which returned a z-score for each monitor location (Figure 5-27). Statistically significant
clustering of high values are observed in the West coast indicating relatively poor model performance.
Significant clustering of low RMSE and MAE values are observed on the East coast indicating better
model performance. Similar to ozone application, the downscaler application of PM2.5 performs better
over the Eastern US than the Western part of the country.
Lastly, overall DQO metrics across locations and time are calculated for PM2.5. For 2009, the RMSE,
MAE and bias values are 2.8, 1.7, and 0.03 |ig/m3 respectively. The difference between overall RMSE
and MAE isn't big enough to indicate the presence of very large errors however there is some variation in
the magnitude of the errors.
114
-------
K .•
-
• I* °
o. ° •
/<£> --V #
• •
•N •
• V / °1
~v
Sk f o- I
*•«. *•
Model Validation (PM 2.5) ^
Gi* Z-Score for the MAE
• <-2.58 Std. Dev.
• -2.58--1.96 Std. Dev.
• -1.96--1.65 Std. Dev.
-1.65-1.65 Std. Dev
• 1.65 - 1.96 Std. Dev.
• 1.96-2.58 Std. Dev.
• > 2.58 Std. Dev.
•7
K ••
•S»V
• V
It*
•• •
i • .
-
• •
• C
*1
*•«. i i f *
• 1 ~ 1 f3 «
~ • • • «*f JU V
* i J * r. Q
Model Validation (PM 2.5) J W j: c ) #,
y
1
• <-2.58 Std. Dev. ) °
• -2.58--1.96 Std. Dev. • 9
» -1.96--1.65 Std. Dev.
Gi* Z-Score for the RMSE
-1.65-1.65 Std. Dev
9 1.65-1.96 Std. Dev. \ J •
• 1.96-2.58 Std. Dev.
• >2.58 Std. Dev. * ¦-*
•» \
Figure 5-27 The Gi* statistic returned z-scores for each PM2.5 monitoring locations over the United
States. Both the RMSE (Top) and the MAE (Bottom) based Gi* z-scores show similar patterns with
very slight differences.
115
-------
5.6 Summary and Conclusions
The results presented in this report are from an application of the DS fusion model for characterizing
national air quality for Ozone and PM2.5. DS provided spatial predictions of daily ozone and PM2.5 at
2010 U.S. census tract centroids by utilizing monitoring data and CMAQ output for 2009. Large-scale
spatial and temporal patterns of concentration predictions are generally consistent with those seen in
ambient monitoring data. Both ozone and PM2.5 were predicted with greater accuracy in the eastern
versus the western U.S., presumably due to the greater monitoring density in the east. Another way of
summarizing results is shown in Figure 5-28, which plots the DS predictions and ambient measurements
paired together by the census tracts containing them. Data is plotted for all days and split out (in each
row) by the NOAA climate regions. The outliers seen to the right of the 1:1 lines in Figure 5-28 were
found to arise in census tracts where the AQS value was substantially higher than the surrounding CMAQ
values. Sampling more points from the DS prediction surface and averaging across the census tracts may
better characterize the census tract area averages, which can be explored in future analyses with the DS
model.
A major distinguishing feature of the DS output is the standard errors accompanying each concentration
prediction. These standard errors give information that complements the cross-validation (CV)-
determined errors and biases. Whereas CV provides measures of accuracy, the DS-produced
uncertainties give a measure of prediction precision. Figures 5-4, 5-6, 5-8, 5-16, 5-18, and 5-20 illustrate
their utility: the errors demonstrate a clear increase in magnitude as distance from the nearest monitor
increases. This numerically demonstrates the intuitively expected decrease in confidence of the
relationship between observed and CMAQ data that DS models as monitor network density decreases,
e.g. as in the western U.S. A total uncertainty could theoretically be constructed by combining the
precision and bias, which could be a potentially useful tool in future network assessment and other
sampling designing activities.
An additional caution that warrants mentioning is related to the capability of DS to provide predictions at
multiple spatial points within a single CMAQ gridcell. Care needs to be taken not to over-interpret any
within-gridcell gradients that might be produced by a user. Fine-scale emission sources in CMAQ are
diluted into the gridcell averages, but a given source within a gridcell might or might not affect every
spatial point contained therein equally. Therefore DS-generated fine-scale gradients are not expected to
represent actual fine-scale atmospheric concentration gradients, unless possibly multiple monitors are
present in the gridcell.
116
-------
03 PM25
80-
60-
40-
20-
0 -
75-
50-
25
0
75-
50
25-
0-
80-
60-
40-
20-
108 =
"g 75
£ 50-
C/5 yn _
Q
o -
75-
50
25-
0 -
75-
50
25-
12§
100
75
50
25
0
Figure 5-28 Downscaler predictions in each census tract versus the AQS Monitoring value in the
same census tract. Each row pools all annual data for the specified NOAA Climate Region.
count
i 5000
4000
3000
2000
B 1000
20-
¦ Central Northwest
NOAA
Climate H EastNorthCentral South
Region
NorthEast SouthEast
|
50 100 150 0 50 100 150 200
AQS Concentration
117
-------
Appendix A - Acronyms
Acronyms
ARW
Advanced Research WRF core model
BEIS
Biogenic Emissions Inventory System
BlueSky
Emissions modeling framework
CAIR
Clean Air Interstate Rule
CAMD
EPA's Clean Air Markets Division
CAP
Criteria Air Pollutant
CAR
Conditional Auto Regressive spatial covariance structure (model)
CARB
California Air Resources Board
CEM
Continuous Emissions Monitoring
CHIEF
Clearinghouse for Inventories and Emissions Factors
CMAQ
Community Multiscale Air Quality model
CMV
Commercial marine vessel
CO
Carbon monoxide
CSN
Chemical Speciation Network
DQO
Data Quality Objectives
EGU
Electric Generating Units
Emission Inventory
Listing of elements contributing to atmospheric release of pollutant
substances
EPA
Environmental Protection Agency
EMFAC
Emission Factor (California's onroad mobile model)
FAA
Federal Aviation Administration
FDDA
Four Dimensional Data Assimilation
FIPS
Federal Information Processing Standards
HAP
Hazardous Air Pollutant
HMS
Hazard Mapping System
ICS-209
Incident Status Summary form
IPM
Integrated Planning Model
ITN
Itinerant
LSM
Land Surface Model
MOBILE
OTAQ's model for estimation of onroad mobile emissions factors
118
-------
MODIS Moderate Resolution Imaging Spectroradiometer
MOVES Motor Vehicle Emission Simulator
NEEDS National Electric Energy Database System
NEI National Emission Inventory
NERL National Exposure Research Laboratory
NESHAP National Emission Standards for Hazardous Air Pollutants
NH Ammonia
NMIM National Mobile Inventory Model
NONROAD OTAQ's model for estimation of nonroad mobile emissions
NO Nitrogen oxides
OAQPS EPA's Office of Air Quality Planning and Standards
OAR EPA's Office of Air and Radiation
ORD EPA's Office of Research and Development
ORIS Office of Regulatory Information Systems (code) - is a 4 or 5 digit
number assigned by the Department of Energy's (DOE) Energy
Information Agency (EIA) to facilities that generate electricity
ORL One Record per Line
OTAQ EPA's Office of Transportation and Air Quality
PAH Polycyclic Aromatic Hydrocarbon
PFC Portable Fuel Container
PM2.5 Particulate matter less than or equal to 2.5 microns
PM10 Particulate matter less than or equal to 10 microns
PMc Particulate matter greater than 2.5 microns and less than 10 microns
Prescribed Fire Intentionally set fire to clear vegetation
RIA Regulatory Impact Analysis
RPO Regional Planning Organization
RRTM Rapid Radiative Transfer Model
SCC Source Classification Code
SMARTFIRE Satellite Mapping Automatic Reanalysis Tool for Fire Incident
Reconciliation
SMOKE Sparse Matrix Operator Kernel Emissions
TCEQ Texas Commission on Environmental Quality
TSD Technical support document
VOC Volatile organic compounds
VMT Vehicle miles traveled
Wildfire Uncontrolled forest fire
WRAP Western Regional Air Partnership
WRF Weather Research and Forecasting Model
119
------- |