EPA/600/A-96/116
Methods for Improving Emissions Estimates
H. L. Waters, V. E. McCormick, J. G. Cleland, and J. R. Youngberg
Research Triangle Institute
Research Triangle Park, NC 27709-2194
P. J. Chappell
U. S. Environmental Protection Agency
Air and Energy Engineering Research Laboratory
Research Triangle Park, NC 27711
ABSTRACT
The Environmental Protection Agency (EPA) is investigating ways to improve
methods for estimating volatile organic compound (VOC) emissions from area sources.
Using the automobile refinishing industry for a detailed area source case study, the
Research Triangle Institute (RTI) and the EPA's Air and Energy Engineering Research
Laboratory (AEERL) are developing an emission estimation method that utilizes both
advanced computational techniques and updated, comprehensive, emissions-related
information. This development includes a thorough characterization of the area source
industry, an analysis of current emission estimation methods, the development and
execution of a nationwide industry activity survey, and a compilation and analysis of the
survey results and other explanatory variables. Results are to be captured in a personal
computer-based VOC emissions estimation system called VOCEES. VOCEES is a dual-
use tool, the users of which can both prepare VOC emissions inventories and analyze the
impact of numerous factors on emissions. This methodology and VOCEES are readily
extendible to other area sources.
INTRODUCTION
Stationary sources of pollutant emissions are designated as either point sources or
area sources. Whereas point sources are inventoried on an individual basis, area source
emissions emanate from processes, activities, or businesses that are too small or too
numerous to be practically tracked as individual emission sources. The distinction
between point and area sources is an annual emissions cutoff, such as 10 tons (9090 kg)
of volatile organic compounds (VOCs) per year per source. The U.S. Environmental
Protection Agency's (EPA's) Air and Energy Engineering Research Laboratory (AEERL)
has initiated an effort to develop appropriate area source emission estimation
methodologies for solvent area source categories using available technologies and to
evaluate long-term informational needs. The Research Triangle Institute (RTI) and
AEERL are working together to establish an emission estimation methodology
development process that will result in accurate emission estimate methods for all area
sources. Results of this work will either validate existing area source emission estimation
methods or recommend alternative methods.
-------
The required components of an emission estimation methodology are: 1) the
calculation of the emission estimates (e.g., using emission and activity factors); 2) the
temporal and spatial allocation of the emissions; 3) the validation of the emission
estimates, and 4) the speciation of the emissions. The work presented in this paper
concentrates on the first three components. Tools are available from EPA that can be
used to develop a VOC inventory grouped into reactivity classes suitable for modeling
[1]. The criteria used in developing an emission estimation methodology include:
reasonable accuracy and cost; dynamic and robust behavior; use of readily available
information; and ease of use.
Why New Methods?
This work has been initiated for two reasons: 1) the current emission estimation
methods require further development to improve accuracy and ease of use; and 2)
innovative tools have emerged which may contribute significantly to improving
estimation methods. Among the issues and uncertainties are:
* Most solvents used by any one solvent area source category are also used by other
industries, preventing use of solvent manufacturer production figures alone for
estimating emissions from one solvent source category.
• There are a large number of very different solvent area sources categories.
Therefore, reasonably accurate emission estimates require area-source-category-
specific estimation methods.
Current methods endorse the use of data sources whose published data are from 2 to
5 years old and may misrepresent a significant segment of the area source industry.
For example, one-person establishments are a significant percentage of the estimated
65,000 automobile refinishing establishments in the U.S. Since most federal
information sources are derived from business payroll records, there is often no
record of these businesses in the endorsed information sources. Also, data at the
county level for certain Standard Industrial Classification (SIC) codes are often not
disclosed as policy.
Current methods do not consider the dynamic factors that impact area sources such as
local economics, changing technology, and regulatory influence. Nor do they
consider human behavior and consumption patterns. For example, it is difficult to
estimate automobile refinishing emissions from insurance payments because people
are more likely to "pocket" insurance payments for vehicle repair if they have more
pressing expenses.
With the emergence of new technologies, such as innovative computational
techniques, often comes the opportunity to improve on existing methods. Part of the
ongoing work being sponsored by AEERL is to investigate the appropriateness of
applying advanced, inference-based computational intelligence techniques, such as neural
networks, fuzzy logic, and genetic algorithms, to emissions estimation. These techniques
provide the ability to determine and utilize relationships between two domains (e.g.,
industry variables and emission levels) without developing traditional mathematical
2
-------
models. They also provide the ability to incorporate expertise expressed in inexact,
English-like terms, such as big, small, fast, or slow. Also, since all emissions have a
geographic component in that each emission has a source, it is helpful to capture and
represent emissions-related data using a geographic information system (GIS). A GIS
supports the organization, analysis, and visualization of data by their geographic
orientation. This is a powerful tool when determining the characteristics of a geographic
region, such as personal income levels, that may impact emission levels.
Relevant Work By Others
Most EPA-endorsed methods for estimating solvent emissions from area sources have
been derived from a methodology developed as part of the National Emissions Data
System (NEDS) [2] on sources of airborne pollutants. This methodology determines
national solvent emissions through material balance, allocating emissions to states and
counties using emission factors, which are based on source-specific emission
measurements as a function of source activity levels. The NEDS area source emission
estimation method has been used in several emission inventory efforts. These efforts
include the 1985 National Acid Precipitation Assessment Program (NAPAP) emissions
inventory prepared by EPA [3], the Regional Ozone Modeling for Northeast Transport
(ROMNET), and the Area and Mobile Source (AMS) Subsystem of the Aerometric
Information Retrieval System (AIRS) [4], which will replace NEDS.
There have also been detailed studies performed on individual area sources in order
to provide guidance on area source emission control technology. Methods for estimating
emissions are often included in these studies. A 1988 study of VOC emissions from
automobile refinishing area sources uses a rather complex calculation dependent upon the
final thickness of applied coatings, in thousandths of an inch, while categorizing the
entire industry into three types of establishments -small, medium, and volume shops [5],
This approach has been criticized as being inappropriate for an industry which has a
large variance in operating characteristics [6],
METHODOLOGY DEVELOPMENT PROCESS
The goal of this work is to establish an emissions estimation methodology
development process that can be applied to all solvent area source categories. The
proposed process is shown in Figure 1 and includes the following functions:
1) Select the area source category for which the emission estimation method is to be
developed.
2) Thoroughly characterize the chosen area source category so as to fully
understand the issues affecting emission estimation for that category.
3) Study all methods currently used for estimating the category's emissions, so as to
understand their strengths and weaknesses.
4) Develop an explanatory variables database ~ data which directly or indirectly
affect the area source category operations and assist in explaining its emission
levels and fluctuations.
5) Assemble a tool set containing statistical tools, computational intelligence tools,
3
-------
and geographic information system tools for analyzing and relating assembled
data.
6) Conduct the nationwide area source survey using a questionnaire reviewed and
endorsed by industry experts and by trade associations. Compile and analyze the
survey results.
7) Select readily available explanatory variables that most significantly impact
emission levels. Structure and complete the method which uses these variables to
estimate emissions from the area source category, and which satisfies the
specified criteria.
Area Source Selection
VOC-emitting area sources include gasoline distribution losses, stationary source
solvent evaporation, bioprocess emission sources, catastrophic or accidental releases,
solid waste incineration, and small stationary source fossil fuel use. Stationary solvent
evaporation sources include dry cleaning, surface cleaning, surface coating, graphic arts,
asphalt paving, pesticide application, commercial or consumer solvent use, and synthetic
organic chemical storage tanks. Automobile refinishing, a surface coating category, was
chosen for study because it is representative of VOC emission area source categories in
terms of relative environmental significance, national prevalence, and an accessible
information base representative of such sources.
Industry Characterization
A detailed characterization of the automobile refinishing industry was conducted in
order to fully consider the issues and variables associated with the industry's VOC
emissions. The number and employment of automobile refinishing establishments vary
across information sources. According to industry representatives, the impact of
technological and regulatory changes, along with economic factors, has provoked a
steady decrease in number of automobile refinishing establishments nationwide over the
past 20 years [5][7]. They attribute this reduction in establishments and paint use to: 1)
improved vehicle safety and more stringent enforcement of traffic laws; 2) more
corrosion-resistant finishes; 3) smaller cars with less surface area to paint per repair; and
4) more efficient spray guns which use less paint per job. Other information sources,
however, have reported a steady, yet gradual increase in number of establishments. The
difference may be explained in part by undocumented "backyard" establishments, which
may represent an additional 25% to 40% of the total number of establishments. These
"unseen shops" are potentially a large source of VOC emissions since they are less likely
to use emissions control technology or comply with laws and regulations, and more likely
to use high-VOC coatings.
With the passage of the Clean Air Act Amendments, and other EPA initiatives, new
regulations for the automobile refinishing industry are emerging. Regulations requiring
use of low-VOC coatings have forced the industry away from lacquer and enamel
coatings and toward urethanes. Additional emission control methods include:
• Use of enclosed equipment cleaning devices that support solvent reuse.
4
-------
• Increased paint (to surface) transfer efficiency through use of high-volume, low-
pressure (HVLP) spray guns.
• Addition to paint spray booths of emission controls which use the following
techniques: thermal incineration, catalytic incineration, and carbon adsorption.
All three methods may be too expensive to be considered as universal, reasonable,
control techniques.
• Better shop operations management, including: use of tight fitting containers;
reduction of spills; rigid control of inventory; tracking of worker use rates; mixture
of paint to need; provision of operator training; use of proper cleanup methods;
and use of in-house or leased solvent recycling.
Current Methods Review
Methods for estimating solvent emissions from area sources are based on use of
emission factors. Emission factors "are developed from only a limited sampling of the
emissions source population for any given category, and the values reported are an
average of those limited samples and may not be statistically representative of the
population" [4]. The basic approach in estimating emissions is derived from a simple
calculation that requires an estimate of an activity level, an emissions factor relating
emissions to activities, and, if the source has a pollution control device, a control factor:
Emissions = activity level x emission factor x control factor
Emission factors can be found in a number of EPA references. Table 1 contains the
emission and activity factors for automobile refinishing from AP-42 Compilation of Air
Pollutant Emission Factors [8] and its Supplement D [91, the EPA State Implementation
Plan (SIP) guidance document[10], and AIRS/AMS [I 1] [ 12], In analyzing these data,
note that:
• The AP-42 emission factors and national VOC emissions did not change for more
than 10 years. They may not have changed since the AP-42 was first published in
1972.
• The SIP guidance document's per capita emission factor is 21% greater than that
of the AP-42 document, while the per employee emission factor is 48% less than
that of the AP-42 document. Some difference may be explained by the 1987-1988
SIC code changes (i.e., using SIC 7532, Automotive Top and Body Repair and
Paint Shops, instead of SIC 7535, Automotive Paint Shops).
Examples of two problem areas are:
1) Difference in Emission Estimates. The two emission factors (per capita vs. per
employeej seldom produce the same emission estimate, as shown in the
example in Figure 2 for 26 California comities. These discrepancies arise from
the static nature of emission factors and the different characteristics of
individual geographic regions.
5
-------
2) Missing Data. County Business Patterns [13] exercises disclosure protection
for counties where revealing information may disclose details about the
operations of individual businesses. In 1990, SIC 7532 employment for 47 of
North Carolina's 100 counties was not disclosed. An emission estimation
method should utilize activity factors that are available for all areas required to
compile emission inventories.
Database Development
The success of any new method to estimate area source emissions is limited by the availability
and accessibility of information. The objective is to identify data that are updated at least annually
and represent county-level activity. Records should exist for at least the past 5 years and into the
foreseeable future. The data must be statistically defensible, representative, "universal"
(representative of national distributions and/or variations), and result in more accurate emissions
estimates. The data are then related to the primary area source variables of both a) solvent use
and b) emissions. These variables may be normalized — for example, on the basis of per capita,
per employee, or per operation.
The database assembled for this study has both geographic or spatial components (e.g.,
nation, state, county, or city) and temporal components (e.g., year or month). The current data
set includes:
• 60 variables over 12 years for the United States
• 25 variables over 12 years for 51 states (including DC)
• 1 1 variables over 12 years for 3126 counties
• 10 variables for 1993 for 64,524 automobile refinishing establishments
Analytical Tool Set Selection
An analytical tool set has been assembled for the purpose of capturing the relationships
between explanatory variables and emission levels. This set includes both Computational
Intelligence tools and traditional statistical tools. Computational Intelligence is a term adopted
by the Institute for Electrical and Electronics Engineers (IEEE) for innovative computational
techniques that include artificial neural networks, fuzzy logic, and genetic algorithms.
Artificial Neural Networks. An artificial neural network (ANN) is an analysis tool that is
modeled after the massively parallel structure of the brain. It simulates a highly interconnected,
parallel computational structure with many relatively simple, individual processing elements or
neurons. Feed-forward ANN paradigms are capable of learning or extracting a relationship
between two domains. ANNs are best applied where there are no known rules or mathematical
models that accurately relate the independent and dependent variables of interest, but there is an
abundance of data representing the relationship.
Fuzzv Logic Expert System. An expert system is a knowledge-based system that captures
human expertise in a specific knowledge domain. A fuzzy logic expert system is being developed
to augment estimation of VOC emissions from the automobile refinishing area sources. Fuzzy
logic is an approximate reasoning technique used in processing inexact information. While a
typical expert system may be thought of as defining "true or false" conditions, fuzzy systems allow
for varying degrees of truth, or "shades of gray," more like human reasoning.
For example, if climate is a factor in an area's emission levels, then it could be classified as
dry, moderate, or rainy. The type of area might also be loosely classified into three fuzzy sets:
rural, suburban, and urban. A typical fuzzy rule, based on expert opinion, may be expressed as,
6
-------
"IF the climate is dry AND the area is rural THEN emissions are low " Another rule may state
"IF the climate is wet AND the area is urban then emissions are high." These rules describe the
increased likelihood of accidents and auto refinishing in a congested area with poor weather and
vice versa. The fuzzy system uses the degree of membership of an input in a given set to
determine to what degree the output belongs in any set (e.g., low, medium, high). This type of
reasoning can augment the emissions prediction based on optimally correlated data.
Genetic Algorithms. Genetic algorithms (GAs) are a class of machine learning search
algorithms based on the mechanics of natural selection and natural genetics. GAs combine
Darwin's "survival of the fittest" with structured, yet randomized information exchange to form a
search algorithm with some innovative flair of human search [141. Created by John Holland at
the University of Michigan [15], GAs are robust, general-purpose problem solvers especially
suited for optimization and classification. GAs can develop better solutions from thousands of
choices more effectively than other techniques for a host of problems.
Use of GAs in the emissions problem will be confined to methodology development.
Examples of using GAs would be in the design and training of neural networks or development of
optimum fuzzy logic membership functions. To properly optimize a system, the GA must know
what constitutes good performance. In estimating emissions, this would be a knowledge of the
VOC emission level for a corresponding set of explanatory variables resulting from the nationwide
industry survey.
Statistics. Simple statistical techniques have been employed as the preliminary step for
analyzing variable relationships and selecting a priority explanatory variable set. Multivariate
regression analyses (linear correlations) have been applied and correlation indicators derived.
There is a high correlation between the automobile refinishing industry and several of the
explanatory variables included in the database. The variables have been analyzed by examining
their distribution across different regions of the country, their change over time, and their ratio
one to another.
Geographic Information System. Emissions are characterized by their levels and by their
distribution in space and time. A geographic information system (GIS) is being used to assign
emissions-related explanatory variable values, such as number of employees and annual sales, to
the actual location of an individual business. These values can then be aggregated to combine
values within a ZIP Code, county, non-attainment area, metropolitan statistical area (MSA), state,
EPA Region, or nation. For State Implementation Plan (SIP) emission inventory purposes, for
example, area source emissions would be aggregated to county and to non-attainment area
boundaries.
Nationwide Industry Survey
One of the most important steps in the emission estimation method development process is
the nationwide survey of the chosen area sources. The Automobile Refinishing Solvent Use
Survey (ARSUS) is the first survey of its kind. It is also essential for validation of the new
emission estimation method. Important features of ARSUS are:
• The survey is designed based on a detailed knowledge of the industry built on numerous
contacts with shop operators, paint manufacturers, and association representatives.
• Results will be statistically defensible, based on random probability sets, with the results
represented by statistically correct accuracy estimations and confidence levels.
7
-------
The map shown in Figure 3 represents the scope of coverage that the 5900 samples will
provide, ARSUS includes local-area intensive surveys of six high-population areas. The survey
data are divided into two independent sets with probability proportional to population, each
containing 30 areas or Primary Sample Units (PSUs). One set is assigned for estimating model
parameters (developing the method), while the second set is for developing a comparison variable
(evaluating and validating the method). The estimated accuracies of solvent use data from the
surveys are shown in Table 2.
Information services and computerized "Yellow Pages" are used to retrieve information
abstracts on firms with a range of SIC codes for automobile refinishing in the counties selected.
This file is stratified by SIC and number of employees, with a probability sample of organizations
in each of the sample PSUS. A detailed file for each sample organization is then retrieved with
the names and addresses for the mail survey and auxiliary data for use in the final statistical
analyses.
The survey combines mail and telephone contacts to maximize response rates, minimize
respondent burden, and complete the data collection efficiently. Survey results are entered using
bar-coded identification labels and event codes which indicate the pending or final status. Results
are to be tabulated in data files for analysis and inclusion in the developed method.
Method Implementation
The final step of the emission estimation method development consists of bringing together
all explanatory variable and survey data, analyzing them, and processing them with the assembled
tool set. More than one computational intelligence tool is needed to meet the specified method
development criteria. The tool set's statistical components are being used to identify the best data
for use in development. Also, fiizzy set clustering is being used during preprocessing to group the
geographic regions (e.g., counties) with similar characteristics for additional analysis. The
preprocessed survey results and explanatory variable data can be presented to a neural network as
illustrated in Figure 4. Initially, half of the available data will be used to train the neural network
while the other half will be used to evaluate that which the network has learned. Genetic
algorithms will be used during this phase of development to identify the optimum neural network
architecture, learning algorithm, and other critical parameters.
Finally, the entire database will be used to train the network, to continue to identify and
minimize the input variable set which best estimates emission levels. Analysis of the trained
network will identify the input variables which are most important in estimating emission levels,
reducing the number of input variables required. A sensitivity analysis will also be performed to
determine the error introduced by this minimized artificial neural network.
A personal-computer-based VOC emission estimation system (VOCEES) will automate the
developed method. The components of the system are: 1) the essential explanatory variable
database; 2) basic algorithms and possibly an artificial neural network-based computational
component; 3) the supplemental fiizzy logic expert system; and 4) the GIS-based user interface.
PRELIMINARY RESULTS AND CONCLUSIONS
Practically all of the emission estimation method case study is complete. The hardware and
software implementation of the method is called the Volatile Organic Compound Emission
Estimation System or VOCEES. The system is a dual-use tool, enabling the user both to prepare
standard VOC emission inventories for geographic regions of various sizes, and also to analyze
how emission levels are influenced by various factors. VOCEES currently combines the emission-
factor-based VOC estimation with the predicted impact of current and anticipated VOC-limiting
8
-------
regulations. The system supports tabular, bar graph, and map formatted analysis and reporting
facilities. An example is presented in Figure 6.
Of the seven functions specified by the proposed development process shown in Figure 1,
work performed to date has completed the first five functions and the preparation and review of
the survey questionnaire as described in the sixth function. It is anticipated that all seven steps
will be completed for the automobile refinishing case study and the process will be extended to
other area source categories. No technical barriers to this process have been identified, and the
early results demonstrate a more efficient and cost-effective emissions estimation methodology.
DISCLAIMER
This paper has been reviewed in accordance with the U.S. Environmental Protection
Agency's peer and administrative review policies and approved for presentation and publication.
9
-------
REFERENCES
1. VOCIPM Speciation Data System Documentation and User's Guide, Version 1.32a; EPA-
450/291-002; U. S Environmental Protection Agency: Research Triangle Park, November
1990.
2. Myers, J.P., Benesh, F.; Methodologies for Countywide Estimation of Coal, Gas, and
Organic Solvent Consumption; EPA-450/3-75-086 (NTIS PB259909); U. S. Environmental
Protection Agency, Office of Air Quality Planning and Standards: Research Triangle Park,
December 1975.
3. Demmy, J., Tax W., Warn, T.; Area Source Documentation for the 1985 National Acid
Precipitation Assessment Program Inventory; EPA-600/8-88-106 (NTIS'PB89-151427);
U.S. Environmental Protection Agency, Air and Energy Engineering Research Laboratory:
Research Triangle Park, December 1988.
4. Seinfeld, J., Atkinson, R., Berglund, R., et al.; Rethinking the Ozone Problem in Urban and
Regional Air Pollution; National Academy Press: Washington, 1991.
5. Athey, C., Hester, C., McLaughlin, M., et al.; Reduction of Volatile Organic Compound
Emissions from Automobile Refinishing; EPA-450/3-88-009 (NTIS PB89-148282); U.S.
Environmental Protection Agency, Control Technology Center: Research Triangle Park,
October 1988.
6. National Air Pollution Control Techniques Advisory Committee, Meeting Minutes, Volume
2; U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards:
Research Triangle Park, November 19-21, 1991; p 1008.
7. M. Hunke, Carstar Automotive, Inc., Shawnee Mission, KS, personal communication, 1992.
8. Compilation of Air Pollution Emission Factors, Volume /.- Stationary Point and Area
Sources, AP-42, Fourth Edition (GPO 055-000-00251-7); U.S. Environmental Protection
Agency, Office of Air Quality Planning and Standards: Research Triangle Park, September
1985; pp. 2, 4.2.1-1.
9. Joyner, W.; Supplement D to Compilation of Air Pollution Emission Factors, Volume
I.Stationary Point and Area Sources, AP-42, Fourth Edition (GPO 055-00-00391-2); U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards: Research
Triangle Park, September 199 1; pp 2, 4.2. 1-1.
10. Procedures for the Preparation of Emission Inventories for Carbon Monoxide and
Precursors Of Ozone, Volume /.- General Guidance For Stationary Sources; EPA-450/4-
91-016 (NTIS PB92112168); U.S. Environmental Protection Agency: Research Triangle
Park, May 1991; pp. 4-24, 4-47.
11. Kimbrough, E.; Documentation of AIRS AMS National Methodologies; EP A-600-R-92-
001 (NTIS PB92-132869); U.S. Environmental Protection Agency, Air and Energy
10
-------
Engineering Research Laboratory: Research Triangle Park, January 1992.
12. Aerometric Information Retrieval System (AIRS): "Short List" of AMS SCCs and Emission
Factors; U.S. Environmental Protection Agency, Office of Air Quality Planning and
Standards: Research Triangle Park, July 1992.
13. County Business Patterns; U.S. Department of Commerce, Bureau of the Census:
Washington; Annual Publication.
14. Goldberg, D.E.; Genetic Algorithms in Search Optimization and Machine Learning;
AddisonWesley Publishing Company: Reading, 1989.
15. Holland, J.H.; Adaptation in Natural and Artificial Systems; University of Michigan Press:
Ann Arbor, 1973.
11
-------
Table 1. Recommended emission factors for the automobile refinishing area source.
National VOC VQC emission factors Nnmber of U.S.
emissions
-------
u
o
>
•8
o
a,
.S
u
u
9000000
8000000
7000000
6000000
5000000
4000000
3000000
2000000
1000000
0
-1000000
i in
8
Difference in SIP Guidance estimate*
s §§
m
s
M
W
P
W W
26 California Counties
i
«
1
w
Figure 2. The difference in automobile refmishing VOC emission estimates, using the two SIP
Guidance emission factors, per employee and per capita.
Figure 3. Preliminary targets of nationwide automobile refmishing survey: six highly populated
areas and 60 primary sample units (PSUs).
13
-------
VI
p
o
EXPLANATORY
VARIABLES
Resident
Population
Licensed
Drivers
Registered Motor
Vehicles
Civilian Labor
Force, Total
Motor Vehicle
Deaths I |
_o.
o __
0
Use of Control
Equipment
Network learns inter-variable relationships
without mathematical models
UNKNOWN
£vog
I Emissions
L
Survey data will be used to train the network
Figure 4. A use of neural networks in the estimation of VOC emissions.
14
-------
VOCEES SIP Generation
VOC Emissions
(tons) ~ 14
¦ 96
¦ 226
Raleigh-Durham-Chapel Hill, NC,Nonattainment Area
o
u
n
t
y
Durtiam
Gfanvflle
Wake
100 150
1993 VOC Emissions (tons)
250
Figure 5. Example emission reporting format using VOCEES for a three-county
nonattainment area in North Carolina.
15
-------
-------
Reproduced by NTIS
National Technical Information Service
U.S. Department of Commerce
Springfield, VA 22161
u
u O
° t
a «
Si
{J U-4
TS
U *,p
C/5 U
C3
o "O
«a a>
j_,
z *
# st
a a
»h y
S ^
O O)
fl
c/3
0> ^
o .
*3 "d
C/5 S
0
•*H
z 2J
t*-< 0>
•»H .—I
-------
-------
-------
-------
EPA/600/A-96/116
Methods for Improving Emissions Estimates
H. L. Waters, V. E. McCormick, J. G. Cleland, and J. R. Youngberg
Research Triangle Institute
Research Triangle Park, NC 27709-2194
P. J. Chapped
U. S, Environmental Protection Agency
Air and Energy Engineering Research Laboratory
Research Triangle Park, NC 27711
ABSTRACT
The Environmental Protection Agency (EPA) is investigating ways to improve
methods for estimating volatile organic compound (VOC) emissions from area sources.
Using the automobile refmishing industry for a detailed area source case study, the
Research Triangle Institute (RTI) and the EPA's Air and Energy Engineering Research
Laboratory (AEERL) are developing an emission estimation method that utilizes both
advanced computational techniques and updated, comprehensive, emissions-related
information. This development includes a thorough characterization of the area source
industry, an analysis of current emission estimation methods, the development and
execution of a nationwide industry activity survey, and a compilation and analysis of the
survey results and other explanatory variables. Results are to be captured in a personal
computer-based VOC emissions estimation system called VOCEES. VOCEES is a dual-
use tool, the users of which can both prepare VOC emissions inventories and analyze the
impact of numerous factors on emissions. This methodology and VOCEES are readily
extendible to other area sources.
INTRODUCTION
Stationary sources of pollutant emissions are designated as either point sources or
area sources. Whereas point sources are inventoried on an individual basis, area source
emissions emanate from processes, activities, or businesses that are too small or too
numerous to be practically tracked as individual emission sources. The distinction
between point and area sources is an annual emissions cutoff, such as 10 tons (9090 kg)
of volatile organic compounds (VOCs) per year per source. The U.S. Environmental
Protection Agency's (EPA's) Air and Energy Engineering Research Laboratory (AEERL)
has initiated an effort to develop appropriate area source emission estimation
methodologies for solvent area source categories using available technologies and to
evaluate long-term informational needs. The Research Triangle Institute (RTI) and
AEERL are working together to establish an emission estimation methodology
development process that will result in accurate emission estimate methods for all area
sources. Results of this work will either validate existing area source emission estimation
methods or recommend alternative methods.
-------
The required components of an emission estimation methodology are: 1) the
calculation of the emission estimates (e.g., using emission and activity factors); 2) the
temporal and spatial allocation of the emissions; 3) the validation of the emission
estimates, and 4) the speciation of the emissions. The work presented in this paper
concentrates on the first three components. Tools are available from EPA that can be
used to develop a VOC inventory grouped into reactivity classes suitable for modeling
[1], The criteria used in developing an emission estimation methodology include:
reasonable accuracy and cost; dynamic and robust behavior; use of readily available
information; and ease of use.
Why New Methods?
This work has been initiated for two reasons: 1) the current emission estimation
methods require further development to improve accuracy and ease of use; and 2)
innovative tools have emerged which may contribute significantly to improving
estimation methods. Among the issues and uncertainties are:
• Most solvents used by any one solvent area source category are also used by other
industries, preventing use of solvent manufacturer production figures alone for
estimating emissions from one solvent source category.
• There are a large number of very different solvent area sources categories.
Therefore, reasonably accurate emission estimates require area-source-category-
specific estimation methods.
Current methods endorse the use of data sources whose published data are from 2 to
5 years old and may misrepresent a significant segment of the area source industry.
For example, one-person establishments are a significant percentage of the estimated
65,000 automobile refinishing establishments in the U.S. Since most federal
information sources are derived from business payroll records, there is often no
record of these businesses in the endorsed information sources. Also, data at the
county level for certain Standard Industrial Classification (SIC) codes are often not
disclosed as policy.
Current methods do not consider the dynamic factors that impact area sources such as
local economics, changing technology, and regulatory influence. Nor do they
consider human behavior and consumption patterns. For example, it is difficult to
estimate automobile refinishing emissions from insurance payments because people
are more likely to "pocket" insurance payments for vehicle repair if they have more
pressing expenses.
With the emergence of new technologies, such as innovative computational
techniques, often comes the opportunity to improve on existing methods. Part of the
ongoing work being sponsored by AEERL is to investigate the appropriateness of
applying advanced, inference-based computational intelligence techniques, such as neural
networks, fuzzy logic, and genetic algorithms, to emissions estimation. These techniques
provide the ability to determine and utilize relationships between two domains (e.g.,
industry variables and emission levels) without developing traditional mathematical
2
-------
models. They also provide the ability to incorporate expertise expressed in inexact,
English-like terms, such as big, small, fast, or slow. Also, since all emissions have a
geographic component in that each emission has a source, it is helpful to capture and
represent emissions-related data using a geographic information system (GIS). A GIS
supports the organization, analysis, and visualization of data by their geographic
orientation. This is a powerful tool when determining the characteristics of a geographic
region, such as personal income levels, that may impact emission levels.
Relevant Work By Others
Most EPA-endorsed methods for estimating solvent emissions from area sources have
been derived from a methodology developed as part of the National Emissions Data
System (NEDS) [2] on sources of airborne pollutants. This methodology determines
national solvent emissions through material balance, allocating emissions to states and
counties using emission factors, which are based on source-specific emission
measurements as a function of source activity levels. The NEDS area source emission
estimation method has been used in several emission inventory efforts. These efforts
include the 1985 National Acid Precipitation Assessment Program (NAPAP) emissions
inventory prepared by EPA [3], the Regional Ozone Modeling for Northeast Transport
(ROMNET), and the Area and Mobile Source (AMS) Subsystem of the Aerometric
Information Retrieval System (AIRS) [4], which will replace NEDS.
There have also been detailed studies performed on individual area sources in order
to provide guidance on area source emission control technology. Methods for estimating
emissions are often included in these studies. A 1988 study of VOC emissions from
automobile refinishing area sources uses a rather complex calculation dependent upon the
final thickness of applied coatings, in thousandths of an inch, while categorizing the
entire industry into three types of establishments -small, medium, and volume shops [5],
This approach has been criticized as being inappropriate for an industry which has a
large variance in operating characteristics [6],
METHODOLOGY DEVELOPMENT PROCESS
The goal of this work is to establish an emissions estimation methodology
development process that can be applied to all solvent area source categories. The
proposed process is shown in Figure 1 and includes the following functions:
1) Select the area source category for which the emission estimation method is to be
developed.
2) Thoroughly characterize the chosen area source category so as to fully
understand the issues affecting emission estimation for that category.
3) Study all methods currently used for estimating the category's emissions, so as to
understand their strengths and weaknesses.
4) Develop an explanatory variables database - data which directly or indirectly
affect the area source category operations and assist in explaining its emission
levels and fluctuations.
5) Assemble a tool set containing statistical tools, computational intelligence tools,
3
-------
and geographic information system tools for analyzing and relating assembled
data.
6) Conduct the nationwide area source survey using a questionnaire reviewed and
endorsed by industry experts and by trade associations. Compile and analyze the
survey results.
7) Select readily available explanatory variables that most significantly impact
emission levels. Structure and complete the method which uses these variables to
estimate emissions from the area source category, and which satisfies the
specified criteria.
Area Source Selection
VOC-emitting area sources include gasoline distribution losses, stationary source
solvent evaporation, bioprocess emission sources, catastrophic or accidental releases,
solid waste incineration, and small stationary source fossil fuel use. Stationary solvent
evaporation sources include dry cleaning, surface cleaning, surface coating, graphic arts,
asphalt paving, pesticide application, commercial or consumer solvent use, and synthetic
organic chemical storage tanks. Automobile refinishing, a surface coating category, was
chosen for study because it is representative of VOC emission area source categories in
terms of relative environmental significance, national prevalence, and an accessible
information base representative of such sources.
Industry Characterization
A detailed characterization of the automobile refinishing industry was conducted in
order to fully consider the issues and variables associated with the industry's VOC
emissions. The number and employment of automobile refinishing establishments vary
across information sources. According to industry representatives, the impact of
technological and regulatory changes, along with economic factors, has provoked a
steady decrease in number of automobile refinishing establishments nationwide over the
past 20 years [5][7], They attribute this reduction in establishments and paint use to: 1)
improved vehicle safety and more stringent enforcement of traffic laws; 2) more
corrosion-resistant finishes; 3) smaller cars with less surface area to paint per repair; and
4) more efficient spray guns which use less paint per job. Other information sources,
however, have reported a steady, yet gradual increase in number of establishments. The
difference may be explained in part by undocumented "backyard" establishments, which
may represent an additional 25% to 40% of the total number of establishments. These
"unseen shops" are potentially a large source of VOC emissions since they are less likely
to use emissions control technology or comply with laws and regulations, and more likely
to use high-VOC coatings.
With the passage of the Clean Air Act Amendments, and other EPA initiatives, new
regulations for the automobile refinishing industry are emerging. Regulations requiring
use of low-VOC coatings have forced the industry away from lacquer and enamel
coatings and toward urethanes. Additional emission control methods include:
• Use of enclosed equipment cleaning devices that support solvent reuse.
4
-------
* Increased paint (to surface) transfer efficiency through use of high-volume, low-
pressure (HVLP) spray guns.
* Addition to paint spray booths of emission controls which use the following
techniques: thermal incineration, catalytic incineration, and carbon adsorption.
All three methods may be too expensive to be considered as universal, reasonable,
control techniques.
* Better shop operations management, including: use of tight fitting containers;
reduction of spills; rigid control of inventory; tracking of worker use rates; mixture
of paint to need; provision of operator training; use of proper cleanup methods;
and use of in-house or leased solvent recycling.
Current Methods Review
Methods for estimating solvent emissions from area sources are based on use of
emission factors. Emission factors "are developed from only a limited sampling of the
emissions source population for any given category, and the values reported are an
average of those limited samples and may not be statistically representative of the
population" [4], The basic approach in estimating emissions is derived from a simple
calculation that requires an estimate of an activity level, an emissions factor relating
emissions to activities, and, if the source has a pollution control device, a control factor:
Emissions = activity level x emission factor x control factor
Emission factors can be found in a number of EPA references. Table 1 contains the
emission and activity factors for automobile refinishing from AP-42 Compilation of Air
Pollutant Emission Factors [8] and its Supplement D [91, the EPA State Implementation
Plan (SIP) guidance documentf 10], and AIRS/AMS [1 1] [ 12]. In analyzing these data,
note that:
• The AP-42 emission factors and national VOC emissions did not change for more
than 10 years. They may not have changed since the AP-42 was first published in
1972.
• The SIP guidance document's per capita emission factor is 21% greater than that
of the AP-42 document, while the per employee emission factor is 48% less than
that of the AP-42 document. Some difference may be explained by the 1987-1988
SIC code changes (i.e., using SIC 7532, Automotive Top and Body Repair and
Paint Shops, instead of SIC 7535, Automotive Paint Shops).
Examples of two problem areas are:
1) Difference in Emission Estimates. The two emission factors (per capita vs. per
employee) seldom produce the same emission estimate, as shown in the
example in Figure 2 for 26 California counties. These discrepancies arise from
the static nature of emission factors and the different characteristics of
individual geographic regions.
5
-------
2) Missing Data. County Business Patterns [13 ] exercises disclosure protection
for counties where revealing information may disclose details about the
operations of individual businesses. In 1990, SIC 7532 employment for 47 of
North Carolina's 100 counties was not disclosed. An emission estimation
method should utilize activity factors that are available for all areas required to
compile emission inventories.
Database Development
The success of any new method to estimate area source emissions is limited by the availability
and accessibility of information. The objective is to identify data that are updated at least annually
and represent county-level activity. Records should exist for at least the past 5 years and into the
foreseeable future. The data must be statistically defensible, representative, "universal"
(representative of national distributions and/or variations), and result in more accurate emissions
estimates. The data are then related to the primary area source variables of both a) solvent use
and b) emissions. These variables may be normalized — for example, on the basis of per capita,
per employee, or per operation.
The database assembled for this study has both geographic or spatial components (e.g.,
nation, state, county, or city) and temporal components (e.g., year or month). The current data
set includes:
• 60 variables over 12 years for the United States
• 25 variables over 12 years for 51 states (including DC)
• 1 1 variables over 12 years for 3126 counties
• 10 variables for 1993 for 64,524 automobile refinishing establishments
Analytical Tool Set Selection
An analytical tool set has been assembled for the purpose of capturing the relationships
between explanatory variables and emission levels. This set includes both Computational
Intelligence tools and traditional statistical tools. Computational Intelligence is a term adopted
by the Institute for Electrical and Electronics Engineers (IEEE) for innovative computational
techniques that include artificial neural networks, fuzzy logic, and genetic algorithms.
Artificial Neural Networks. An artificial neural network (ANN) is an analysis tool that is
modeled after the massively parallel structure of the brain. It simulates a highly interconnected,
parallel computational structure with many relatively simple, individual processing elements or
neurons. Feed-forward ANN paradigms are capable of learning or extracting a relationship
between two domains. ANNs are best applied where there are no known rules or mathematical
models that accurately relate the independent and dependent variables of interest, but there is an
abundance of data representing the relationship.
Fii77v T .ogic Expert System. An expert system is a knowledge-based system that captures
human expertise in a specific knowledge domain. A fuzzy logic expert system is being developed
to augment estimation of VOC emissions from the automobile refinishing area sources. Fuzzy
logic is an approximate reasoning technique used in processing inexact information. While a
typical expert system may be thought of as defining "true or false" conditions, fuzzy systems allow
for varying degrees of truth, or "shades of gray," more like human reasoning.
For example, if climate is a factor in an area's emission levels, then it could be classified as
dry, moderate, or rainy. The type of area might also be loosely classified into three fuzzy sets:
rural, suburban, and urban. A typical fuzzy rule, based on expert opinion, may be expressed as,
6
-------
"IF the climate is dry AND the area is rural THEN emissions are low." Another rule may state
"IF the climate is wet AND the area is urban then emissions are high." These rules describe the
increased likelihood of accidents and auto refinishing in a congested area with poor weather and
vice versa. The fuzzy system uses the degree of membership of an input in a given set to
determine to what degree the output belongs in any set (e.g., low, medium, high). This type of
reasoning can augment the emissions prediction based on optimally correlated data.
Genetic Algorithms. Genetic algorithms (GAs) are a class of machine learning search
algorithms based on the mechanics of natural selection and natural genetics. GAs combine
Darwin's "survival of the fittest" with structured, yet randomized information exchange to form a
search algorithm with some innovative flair of human search [141. Created by John Holland at
the University of Michigan [15], GAs are robust, general-purpose problem solvers especially
suited for optimization and classification. GAs can develop better solutions from thousands of
choices more effectively than other techniques for a host of problems.
Use of GAs in the emissions problem will be confined to methodology development.
Examples of using GAs would be in the design and training of neural networks or development of
optimum fuzzy logic membership functions. To properly optimize a system, the GA must know
what constitutes good performance. In estimating emissions, this would be a knowledge of the
VOC emission level for a corresponding set of explanatory variables resulting from the nationwide
industiy survey.
Statistics. Simple statistical techniques have been employed as the preliminary step for
analyzing variable relationships and selecting a priority explanatory variable set. Multivariate
regression analyses (linear correlations) have been applied and correlation indicators derived.
There is a high correlation between the automobile refinishing industry and several of the
explanatory variables included in the database. The variables have been analyzed by examining
their distribution across different regions of the country, their change over time, and their ratio
one to another.
Geographic Information System. Emissions are characterized by their levels and by their
distribution in space and time. A geographic information system (GIS) is being used to assign
emissions-related explanatory variable values, such as number of employees and annual sales, to
the actual location of an individual business. These values can then be aggregated to combine
values within a ZIP Code, county, non-attainment area, metropolitan statistical area (MSA), state,
EPA Region, or nation. For State Implementation Plan (SEP) emission inventory purposes, for
example, area source emissions would be aggregated to county and to non-attainment area
boundaries.
Nationwide Industry Survey
One of the most important steps in the emission estimation method development process is
the nationwide survey of the chosen area sources. The Automobile Refinishing Solvent Use
Survey (ARSUS) is the first survey of its kind. It is also essential for validation of the new
emission estimation method. Important features of ARSUS are:
• The survey is designed based on a detailed knowledge of the industry built on numerous
contacts with shop operators, paint manufacturers, and association representatives.
• Results will be statistically defensible, based on random probability sets, with the results
represented by statistically correct accuracy estimations and confidence levels.
7
-------
The map shown in Figure 3 represents the scope of coverage that the 5900 samples will
provide. ARSUS includes local-area intensive surveys of six high-population areas. The survey
data are divided into two independent sets with probability proportional to population, each
containing 30 areas or Primary Sample Units (PSUs). One set is assigned for estimating model
parameters (developing the method), while the second set is for developing a comparison variable
(evaluating and validating the method). The estimated accuracies of solvent use data from the
surveys are shown in Table 2.
Information services and computerized "Yellow Pages" are used to retrieve information
abstracts on firms with a range of SIC codes for automobile refinishing in the counties selected.
This file is stratified by SIC and number of employees, with a probability sample of organizations
in each of the sample PSUS. A detailed file for each sample organization is then retrieved with
the names arid addresses for the mail survey and auxiliary data for use in the final statistical
analyses.
The survey combines mail and telephone contacts to maximize response rates, minimize
respondent burden, and complete the data collection efficiently. Survey results are entered using
bar-coded identification labels and event codes which indicate the pending or final status. Results
are to be tabulated in data files for analysis and inclusion in the developed method.
Method Implementation
The final step of the emission estimation method development consists of bringing together
all explanatory variable and survey data, analyzing them, and processing them with the assembled
tool set. More than one computational intelligence tool is needed to meet the specified method
development criteria. The tool set's statistical components are being used to identify the best data
for use in development. Also, fuzzy set clustering is being used during preprocessing to group the
geographic regions (e.g., counties) with similar characteristics for additional analysis. The
preprocessed survey results and explanatory variable data can be presented to a neural network as
illustrated in Figure 4. Initially, half of the available data will be used to train the neural network
while the other half will be used to evaluate that which the network has learned. Genetic
algorithms will be used during this phase of development to identify the optimum neural network
architecture, learning algorithm, and other critical parameters.
Finally, the entire database will be used to train the network, to continue to identify and
minimize the input variable set which best estimates emission levels. Analysis of the trained
network will identify the input variables which are most important in estimating emission levels,
reducing the number of input variables required. A sensitivity analysis will also be performed to
determine the error introduced by this minimized artificial neural network.
A personal-computer-based VOC emission estimation system (VOCEES) will automate the
developed method. The components of the system are: 1) the essential explanatory variable
database; 2) basic algorithms and possibly an artificial neural network-based computational
component; 3) the supplemental fuzzy logic expert system; and 4) the GIS-based user interface.
PRELIMINARY RESULTS AND CONCLUSIONS
Practically all of the emission estimation method case study is complete. The hardware and
software implementation of the method is called the Volatile Organic Compound Emission
Estimation System or VOCEES. The system is a dual-use tool, enabling the user both to prepare
standard VOC emission inventories for geographic regions of various sizes, and also to analyze
how emission levels are influenced by various factors. VOCEES currently combines the emission-
factor-based VOC estimation with the predicted impact of current and anticipated VOC-limiting
8
-------
regulations. The system supports tabular, bar graph, and map formatted analysis and reporting
facilities. An example is presented in Figure 6.
Of the seven functions specified by the proposed development process shown in Figure 1,
work performed to date has completed the first five functions and the preparation and review of
the survey questionnaire as described in the sixth function. It is anticipated that all seven steps
will be completed for the automobile refinishing case study and the process will be extended to
other area source categories. No technical barriers to this process have been identified, and the
early results demonstrate a more efficient and cost-effective emissions estimation methodology.
DISCLAIMER
This paper has been reviewed in accordance with the U.S. Environmental Protection
Agency's peer and administrative review policies and approved for presentation and publication.
9
-------
REFERENCES
1. V0C.1PM Speciation Data System Documentation and User's Guide, Version 1.32a; EPA-
450/291-002; U. S. Environmental Protection Agency: Research Triangle Park, November
1990.
2. Myers, J.P., Benesh, F.; Methodologies for Countywide Estimation of Coal, Gas, and
Organic Solvent Consumption; EPA-450/3-75-086 (NTIS PB259909); U. S. Environmental
Protection Agency, Office of Air Quality Planning and Standards: Research Triangle Park,
December 1975.
3. Demmy, J., Tax W., Warn, T.; Area Source Documentation for the 1985 National Acid
Precipitation Assessment Program Inventory; EPA-600/8-88-106 (NTIS'PB89-151427);
U.S. Environmental Protection Agency, Air and Energy Engineering Research Laboratory:
Research Triangle Park, December 1988.
4. Seinfeld, J., Atkinson, R., Berglund, R., et al.; Rethinking the Ozone Problem in Urban and
Regional Air Pollution; National Academy Press: Washington, 1991.
5. Athey, C., Hester, C., McLaughlin, M., et al.; Reduction of Volatile Organic Compound
Emissions from Automobile Refinishing; EPA-450/3-88-009 (NTIS PB89-148282); U.S.
Environmental Protection Agency, Control Technology Center: Research Triangle Park,
October 1988.
6. National Air Pollution Control Techniques Advisory Committee, Meeting Minutes, Volume
2; U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards:
Research Triangle Park, November 19-21, 1991; p 1008.
7. M. Hunke, Carstar Automotive, Inc., Shawnee Mission, KS, personal communication, 1992.
8. Compilation of Air Pollution Emission Factors, Volume /.- Stationary Point and Area
Sources, AP-42, Fourth Edition (GPO 055-000-00251-7); U.S. Environmental Protection
Agency, Office of Air Quality Planning and Standards: Research Triangle Park, September
1985; pp. 2, 4.2.1-1.
9. Joyner, W.; Supplement D to Compilation of Air Pollution Emission Factors, Volume
I. Stationary Point and Area Sources, AP-42, Fourth Edition (GPO 055-00-00391-2); U.S.
Environmental Protection Agency, Office of Air Quality Planning and Standards: Research
Triangle Park, September 199 1; pp 2, 4.2. 1-1.
10. Procedures for the Preparation of Emission Inventories for Carbon Monoxide and
Precursors Of Ozone, Volume /.- General Guidance For Stationary Sources; EPA-450/4-
91-016 (NTIS PB92112168); U.S. Environmental Protection Agency: Research Triangle
Park, May 1991; pp. 4-24, 4-47.
11. Kimbrough, E.; Documentation of AIRS AMS National Methodologies; EPA-600-R-92-
001 (NTIS PB92-132869); U.S. Environmental Protection Agency, Air and Energy
10
-------
Engineering Research Laboratory: Research Triangle Park, January 1992.
Aerometric Information Retrieval System (AIRS): "Short List" of AMS SCCs and Emission
Factors; U.S. Environmental Protection Agency, Office of Air Quality Planning and
Standards: Research Triangle Park, July 1992
County Business Patterns; U.S. Department of Commerce, Bureau of the Census:
Washington; Annual Publication.
Goldberg, D.E.; Genetic Algorithms in Search Optimization and Machine Learning;
AddisonWesley Publishing Company: Reading, 1989.
Holland, J.H.; Adaptation in Natural and Artificial Systems; University of Michigan Press:
Ann Arbor, 1973.
11
-------
Table 1. Recommended emission factors for the automobile refinishing area source.
National VOC VOC emission factors N'imbcr of U.S.
emissions (pounds/year) U.S. population automobile
Reference
-------
u
o
>
•8
o
O.
.a
9000000
8000000
7000000
6000000
5000000
4000000
3000000
2000000
1000000
0
-1000000
Difference in SIP Guidance ettimatei
tl
W V2
26 California Counties
Figure 2. The difference in automobile refmishing VOC emission estimates, using the two SIP
Guidance emission factors, per employee and per capita.
Figure 3. Preliminary targets of nationwide automobile refmishing survey: six highly populated
areas and 60 primaiy sample units (PSUs).
13
-------
o
p
O
EXPLANATORY
VARIABLES
Resident
Population "
Licensed
Drivers
Registered Motor
Vehicles
Civilian Labor
Force, Total |
Motor Vehicle
Deaths
_o_
o __
0
Use of Control
Equipment
L
Network learns inter-vsuriable relationships
without mathematical models
UNKNOWN
£voc
| Emissions
Survey data will be used to train the network
Figure 4. A use of neural networks in the estimation of VOC emissions.
14
-------
VOCEES SIP Generation
VOC Emissions
(tons) ~ j 4
¦ 96
¦ 226
Raleigh-Duitiam-Chapet Hill NC,Nonattainment Area
c
o
u
n
t
y
Durham
Granvffle
Wake
100 150
1993 VOC Emissions (tons)
250
Figure 5. Example emission reporting format using VOCEES for a three-county
nonattainment area in North Carolina.
15
------- |