SOURCE APPORTIONMENT OF PRIMARY
CARBONACEOUS AEROSOL USING THE
COMMUNITY MULTISCALE AIR QUALITY MODEL
Prakash V. Bhave**, George A. Pouliot*, and Mei Zheng"	EPA/600/A-04/074
INTRODUCTION
A substantial fraction of fine particulate matter (PM) across the United States is
composed of carbon, which may be either emitted in particulate form (i.e., primary) or formed
in the atmosphere through gas-to-particle conversion processes (i.e., secondary). Primary
carbonaceous aerosol is emitted from numerous sources including motor vehicle exhaust,
residential wood combustion, coal combustion, forest fires, agricultural burning, solid waste
incineration, food cooking operations, and road dust. Quantifying the primary contributions
from each major emission source category is a prerequisite to formulating an effective control
strategy for the reduction of carbonaceous aerosol concentrations. A quantitative assessment
of secondary carbonaceous aerosol concentrations also is required, but falls outside the scope
of the present work.
A common method of primary carbonaceous aerosol source apportionment involves a
molecular characterization of emission source effluents and ambient aerosol samples followed
by a determination of the linear combination of source signatures that best matches the
measured composition of the ambient sample. This method, referred to as an organic tracer-
based chemical mass balance (CMB), has been demonstrated using atmospheric aerosol
samples collected at a number of receptor sites across the United States (Schauer et al., 1996;
Fujita et al., 1998; Zheng et al., 2002; Fine 2002; Fraser et al., 2003). An alternative source
apportionment methodology makes use of source-specific emission rates and atmospheric
transport calculations in a source-oriented air quality modeling framework. The PM emitted
from each major source category is tagged at the point of emission and tracked numerically as
it is transported through the study region. In this manner, the ambient pollutant concentration
Atmospheric Sciences Modeling Division, Air Resources Laboratory, National Oceanic and Atmospheric
Administration, Research Triangle Park, NC 27711, U.S.A. On Assignment to the National Exposure Research
Laboratory, U.S. Environmental Protection Agency - Office of Research and Development.
^ Corresponding author, e-mail: bhave.prakash@epa.gov. tel. (919) 541-2194. fax. (919) 541-1379.
^ School of Earth and Atmospheric Sciences, Georgia Institute of Technology, Atlanta, GA 30332, U.S.A.

-------
increments due to each source of primary carbonaceous aerosol can be estimated at any time
and location within the modeling domain. Applications of this method have been limited in
large part due to the input requirement of a detailed emission inventory that includes the
strengths, temporal distributions, and spatial allocations of each major emission source of
carbonaceous PM. Moreover, it is difficult to evaluate results of this method without
atmospheric measurements of source-specific chemical tracers. For these reasons, most
applications of the source-oriented approach reported to date are for the Los Angeles
metropolitan area during intensive field measurement campaigns (Hildemann et al., 1993;
Rogge et al., 1996; Fraser et al., 2000). Recently, a global-scale three dimensional model
(GEOS-CHEM) was used to track the carbonaceous aerosol contributions from three primary
source categories (fossil fuel combustion, biofuel combustion, and biomass burning) across
the U.S. in 1998 (Park et al., 2003). That model application was intended to apportion
sources at a coarse spatial resolution (2° latitude by 2.5° longitude) for regional visibility
calculations and the evaluation was limited by bulk compositional data.
Over the past decade, the U.S. Environmental Protection Agency (EPA) in cooperation
with state and local agencies has developed a National Emission Inventory (NEI) for fine PM
(EPA, 2001). In addition, the EPA has been developing the Community Multiscale Air
Quality (CMAQ) model for the mechanistic prediction of gas and aerosol-phase pollutant
concentrations (Byun and Ching, 1999). In the present work, an extension to the CMAQ
model is described that allows the user to track the emissions from an arbitrary number of
primary aerosol sources as they are transported through the atmosphere. The model is
coupled with the NEI to estimate primary carbonaceous aerosol concentration increments
contributed by nine major emission categories over the continental U.S. from June 15 -
August 31, 1999. Model results are evaluated against source-specific molecular
measurements collected at eight receptor sites in the southeastern U.S.
EMISSION INVENTORY AND MODEL DESCRIPTIONS
Gaseous and particle-phase emissions in the NEI are categorized by geographic region
and source classification code (SCC). For typical CMAQ modeling applications, the NEI is
processed using the Sparse Matrix Operator Kernel Emissions (SMOKE) model to yield
model-ready input files that contain chemically, spatially, and temporally resolved pollutant
emissions. These gridded emission files include particulate elemental carbon (coded as PEC),
organic aerosol (POA), sulfate (PS04), nitrate (PN03), and other unspecified fine PM. In the
NEI, POA mass is defined implicitly as primary organic carbon times 1.2, to account for the
masses of H, O, N, and S atoms that are associated with organic carbon emissions. The
temporal resolution of the emission file is hourly and, for the present application, the grid
spacing is 32 km. In order to track different sources of carbonaceous aerosol, the fine particle
speciation profiles used in the SMOKE model are duplicated to create a source-specific
profile for each emission category of interest. In the source-specific profiles, PEC and POA
emitted from the first source category are designated respectively as PEC1 and POA1, those
from the second source category as PEC2 and POA2, and so forth. Also, the SCC-to-
speciation profile reference table used in the SMOKE model is modified to appropriately map
each SCC to the newly-created source-specific speciation profiles.

-------
Version 1 of the 1999 NEI serves as the base inventory for the present model application.
It was developed by applying growth factors to the 1996 National Emission Trends criteria air
pollutant inventory, which is described in detail elsewhere (EPA, 2001). When preparing
model-ready emission files for the present study, all NEI estimates of fugitive dust emissions
(e.g., from paved and unpaved roads, agricultural tilling, construction activities, etc.) are
reduced by a factor of four to account approximately for the dust removal processes that occur
within several hundred meters of their sources (Watson and Chow, 2000). Commercial
cooking emissions were reported by very few states in the NEI, so emissions from that
category are replaced by a more comprehensive 2002 commercial cooking emission inventory
(Roe et al., 2004). The resulting fine PM emission inventory is categorized into 2,890 SCCs.
To reduce the computational burden that would be associated with tracking each of these
sources throughout the modeling domain, emissions associated with each SCC are lumped
into nine major source categories plus a tenth miscellaneous category. These nine categories
constitute nearly 95% of the total POA and PEC emissions on an annual basis, as shown in
Table 1. It should be noted that vegetative detritus, fungal spores, natural windblown dust,
and cigarette smoke, are in neither the base nor the model-ready emission inventories.
Table 1. Fine particle mass emission totals (tons/yr) in the model-ready inventory.

EC
OCxl.2
S04
N03
OTHER
TOTAL
Diesel Exhaust
350000
105000
10600
760
2800
470000
Gasoline Exhaust
18600
100000
2400
600
23000
145000
Biomass Combustion
89000
620000
15500
5700
300000
1030000
Coal Combustion
3000
1770
2800
0
141000
149000
Oil Combustion
5700
6600
860
49
19700
33000
Natural Gas Combustion
0
23000
7400
330
6600
37000
Food Cooking
870
68000
148
14
920
70000
Paved Road Dust
1880
30000
1170
370
135000
168000
Crustal Material
2500
37000
270
750
640000
680000
Other Sources
33000
60000
72000
2400
520000
690000
Grand Total
500000
1050000
113000
11100
1790000
3500000
Version 4.3 of the CMAQ model (2003 public release) is used as the base model
configuration for this study. Aerosol components of the CMAQ model are described in detail
elsewhere (Binkowski and Roselle, 2003). In the base configuration, primary carbonaceous
aerosols are tracked as four model species to distinguish their size and composition
distribution: Aitken mode organic aerosol (AORGPAI), Aitken mode elemental carbon
(AECI), accumulation mode organic aerosol (AORGPAJ), and accumulation mode elemental
carbon (AECJ). For each carbonaceous aerosol source category tracked in the extended
CMAQ model, four species are added. For example, AORGP4I, AEC4I, AORGP4J, and
AEC4J, represent primary carbonaceous species originating from coal combustion in the
present application. The 40 additional model species are internally mixed within their
designated aerosol mode (i.e., Aitken or accumulation) and participate in advection, diffusion,
deposition, condensational growth, and coagulation processes in a manner identical to the
treatment of AORGPA and AEC species in the base model configuration.

-------
MODEL RESULTS
The extended CMAQ model is used to simulate gaseous and aerosol-phase pollutant
concentrations while tracking the contributions from nine major primary PM source
categories across the continental U.S. from June 15 - August 31, 1999. The modeling
domain, meteorological inputs, boundary conditions, and initial spin-up period are identical to
those used by Yu et al. (2004). Figure 1 displays model results in the lowest vertical layer
averaged over the 78 day simulation period for a subset of source categories. Carbon
concentrations are calculated as (AEC + AORGPA/1.2) and summed over the Aitken and
accumulation modes.
Primary Carbon trow Diesel Exhaust
J
075
0,50
Primary Carbon from Coat Oil & Natuiai Gas Combustion
Primary Carbon from Crustat
o.o8 m
0.08
0.015
war ¦ > \
0.000 1
ugC/m3
1
178
Figure 1. Model predictions of fine particle primary carbon concentrations [jag C m"3] from select source categories
averaged over the June 15 - August 31, 1999 period. Note the differences in scale. Plots are prepared using PAVE
by MCNC.
The spatial pattern of diesel exhaust concentrations resembles closely the U.S. population
density distribution, with highest concentrations found over urban areas (see Figure la).
Concentrations of gasoline exhaust, paved road dust, and food cooking, also exhibit spatial
patterns similar to the population density distribution, and therefore, are not displayed in
Figure 1. Over most urban areas during the 1999 summer, model results indicate that diesel
exhaust makes a larger contribution to primary carbonaceous fine PM than any other source
category. The highest seasonal average concentration of diesel exhaust carbon is calculated
as 5.2 |ag C m"3 in the grid cell surrounding New York City. The next highest concentrations

-------
are found over northern Ohio, Los Angeles, New Orleans, Phoenix, and Atlanta, ranging from
2.6 - 3.5 |-ig C m~3. The highest seasonal average concentrations of gasoline exhaust carbon
are 1.0 \x.g C m"3 over the Los Angeles area, 0.7 over New York City, and 0.5 over Chicago,
resulting from large volumes of automobile traffic in these cities. Carbon concentrations from
paved road dust are roughly twenty times lower than the motor vehicle exhaust contributions.
The highest seasonal average primary carbon concentrations from food cooking are 1.4 and
1.2 |_ig C m"3 over New York City and Los Angeles, and 0.5 - 0.6 |.ig C m"3 over Chicago, San
Francisco, and Washington, D.C. It is of interest to note that model calculations of food
cooking carbon concentrations exceed those of gasoline exhaust in most urban areas across
the U.S., even though total carbon emissions from the latter source category are greater on a
national scale (see Table 1).
In the model-ready inventory, 90% of biomass combustion carbon emissions during the
summer months are from wildfires. Hence, the spatial distribution of primary carbonaceous
aerosol concentrations originating from biomass combustion (see Figure lb) is roughly
proportional to the number of acres that burned in 1999. The highest modeled concentrations
of biomass combustion carbon are over Florida, Montana, New Mexico, and California. As
shown in Figure lc, carbon concentrations from crustal material are highest over the Midwest
and central states. In the inventory, summertime emissions of fine crustal material are
dominated by unpaved road dust (51% of total) and agricultural tilling (31%), followed by
smaller contributions from construction activities (14%) and beef cattle feedlots (3%). Hence,
the spatial patterns of crustal carbon are concentrated over rural and agricultural areas. The
inclusion of natural windblown dust emissions (e.g., from desert dust storms) in future
inventories would likely increase crustal aerosol concentrations over the arid Southwest.
Although tracked separately in the present model application, the aggregate of coal, oil,
and natural gas combustion contributions are displayed in Figure Id. Coal combustion carbon
is highest over the Ohio River valley but exhibits surprisingly low concentrations (max = 40
ng C m"3). The speciation profile for coal combustion emissions designates only 2.7% of the
fine particle mass as carbon, based on measurements taken at a Philadelphia power plant over
20 years ago. A number of recent studies estimate the carbonaceous fraction to be over 15%,
indicating a need to update this particular speciation profile (Ryan, 2003). Domain-wide
maximums from oil and natural gas combustion are found in New Jersey (0.74 and 0.97 |.ig C
m"3, respectively) due to very high emissions from a single utility company. Excluding the
New Jersey plumes, the domain-wide maximum concentrations from oil and natural gas
combustion are 0.23 and 0.09 |ag C m"3, respectively.
MODEL EVALUATION
Atmospheric concentrations of about 100 individual organic compounds were measured
from fine particle samples collected in 1999 at eight receptor sites across the southeastern
U.S. (Zheng et al., 2002). This is the first available set of source-specific carbonaceous
concentration data that spans a multi-state geographic region. To obtain model estimates of
individual organic compound concentrations at each receptor site, model calculations of
source-specific carbon concentrations are multiplied by organic molecular speciation profiles.
The source profiles used for diesel exhaust, gasoline exhaust, food cooking, biomass
combustion, natural gas combustion, and paved road dust are identical to those described by

-------
Zheng et al. (2002). The oil combustion profile is an average of two source tests reported by
Rogge et al. (1997). Organic molecular profiles are not available for coal combustion, crustal
material, and the numerous miscellaneous sources, so model calculations of carbonaceous
aerosol from these source categories are not speciated in the present study.
Bulk Cornp.
Motor Vehicle Exhaust
Biomass Comb.
Cooking
Oil & NG
dO

; ~


£
f #

~: ^
M
	m	J_		
0
*
o
Centreville, AL
©
N. Birmingham, AL
~
Oak Grove, MS
o
Gulfport, MS
0
Yorkville, GA
0
Atlanta, GA
A
OLF#8, FL
-A
Pensacola, FL

A
O
f
A
m
~ •
-B	B-
un
cc
cc
o
S3
Figure 2. Ratios of CM AQ model results to ambient measurements of EC, OC, and individual organic compounds at
eight southeastern U.S. sites in July 1999. Horizontal lines bound the region in which model-observation agreement
is within a factor of two. Vertical dashed lines distinguish molecular markers specific to different source categories.
Figure 2 displays a model evaluation summary comparing the extended CMAQ model
results for July 1999 against atmospheric measurements at all eight receptor sites in the
Southeast. The ratios of model predictions to observations are displayed along the vertical
axis for all cases where the given species was detected above quantifiable limits at the given
site. Symbols lying between the two horizontal lines represent cases where model predictions
are within a factor of two of the observed concentrations. Shaded symbols represent urban
monitoring sites, whereas unfilled symbols correspond to rural (Centreville, Oak Grove, and
Yorkville) or suburban (OLF#8) locations. Seventeen organic species and bulk elemental
carbon (EC) and organic carbon (OC) are arranged in sections along the horizontal axis,
separated by vertical dashed lines that delineate conserved tracers emitted from different
source categories. Conserved organic markers unique to paved road dust are unavailable, so it
is not possible to directly evaluate model results from that source in the present study.

-------
Daily fine particle EC and OC measurements are obtained from the Southeastern Aerosol
Research and Characterization network (Hansen et al., 2003) and averaged over the month of
July at each site for comparison against monthly-averaged model predictions. Model
calculations of EC fall within a factor of two of observations at more than half of the sampling
locations. This level of agreement builds confidence in the transport algorithms used in
CMAQ and in the inventory of diesel emissions, because diesel exhaust is the dominant
source of atmospheric EC in the U.S. Model predictions of total OC (primary + secondary)
are in reasonable agreement with observations in Atlanta, but fall below measurements by a
factor of three, on average, at the remaining sites. This indicates that total OC is
underestimated across the southeastern U.S. From bulk EC and OC measurements alone, it is
impossible to determine which source contributions have been underestimated.
Out of 34 quantitative measurements of motor vehicle exhaust tracers, 21 model-
observation pairs agree within a factor of two (see Figure 2). Considering that the observed
concentrations of these species span a broad range (15 to 500 pg m"3), the displayed level of
agreement is quite good and indicates that the contributions of gasoline and diesel exhaust to
primary carbon concentrations are captured reasonably well in the model. A systematic
model underestimation of biomass combustion throughout the Southeastern U.S. is apparent
from comparisons shown in Figure 2. Model estimates of pimaric acid and sandaracopimaric
acid at Oak Grove fall short of measurements by more than a factor of 500, indicating that a
biomass combustion event near that site is not captured in the model simulation.
Measurements of levoglucosan, a well-established tracer for wood smoke, exceed model
results by a factor of six on average when the Oak Grove data are excluded (obs. range = 26 -
80 ng m"3). A close examination of the inventory reveals that none of the annual emissions
from agricultural and prescribed forest burning is allocated to the summer months. Emission
allocation refinements from these two sources and from wildfires in future inventories will
likely increase modeled concentrations of biomass combustion carbon during summer months
in the Southeast. Cholesterol is a reliable tracer for meat cooking, but unfortunately, was not
quantified in the July 1999 samples due to a limitation of the analytical techniques used in
that study. As a substitute for cholesterol, particle-phase nonanal is used as a meat cooking
tracer in the present study because it is absent from the remaining source profiles. The
nonanal comparisons indicate that emissions of food cooking carbon in the Southeast are
underestimated by a factor of 2 to 6 in the present inventory (obs. range = 0.9 - 2.9 ng m"3).
The other food cooking tracers listed in Figure 2 (9-hexadecenoic acid and 9-octadecenoic
acid) are emitted from multiple sources in addition to meat cooking, which may explain much
of the scatter shown in Figure 2. Model comparisons against measurements of chemical
tracers unique to oil and natural gas combustion are plotted in the rightmost section of Figure
2. The agreement between model calculations and observations for these two species is
reasonable, but very few data points are available for comparison.
CONCLUSIONS
We have extended the CMAQ model to provide a first estimate of the source
contributions to fine particle primary carbon concentrations across the United States using a
detailed emission inventory and mesoscale meteorological inputs. Spatial distributions of the
various source contributions agree qualitatively with our knowledge of emission patterns.

-------
Molecular speciation of the source apportioned model results allows the calculation of
individual organic compound concentrations at selected receptor sites. Model evaluation
against measurements of individual organic compounds reveals that fine particle emission
estimates of motor vehicle exhaust and natural gas combustion are reasonably accurate over
the southeastern U.S., whereas carbonaceous emissions from biomass combustion and food
cooking are biased low by more than a factor of two.
REFERENCES
Binkowski, F. S., Roselle, S. J., 2003, Models-3 Community Multiscale Air Quality (CMAQ) model aerosol
component 1. Model description, J. Geophys. Res., 108(D6), 4183, doi:10.1029/2001JD001409.
Byun. D. W.. Ching. J. K. S., 1999, Science Algorithms of the EPA Models-3 Community Multiscale Air Quality
(CMAQ) Modeling System, EPA report 600, R-99 030, Washington DC.
EPA, 2001, Procedures Document for National Emission Inventory, Criteria Air Pollutants 1985-1999, EPA report
454 R-01-006, Research Triangle Park, NC.
Fine, P. M., 2002, The Contribution of Biomass Combustion to Ambient Fine Particle Concentrations in the United
States, Ph. D. thesis, Calif. Inst, of Technol., Pasadena, CA.
Fraser, M. P.. Kleeman. M. J.. Schauer. J. J.. Cass. G. R.. 2000, Modeling the atmospheric concentrations of
individual gas-phase and particle-phase organic compounds, Environ. Sci. Technol, 34:1302.
Fraser, M. P.. Yue. Z. W.. Buzcu. B.. 2003. Source apportionment of fine particulate matter in Houston, TX, using
organic molecular markers, Atmos. Environ., 37:2117.
Fujita, E. M., Watson, J. G., Chow, J. C., Robinson, N., Richards, L., Kumar, N., 1998, Northern Front Range Air
Quality Study. Volume C: Source Apportionment and Simulation Methods for Evaluation, Final report to
Colorado State University, Fort Collins, CO.
Hansen, D. A., Edgerton, E. S., Hartsell, B. E., Jansen, J. J., Kandasamy, N., Hidv, G. M., Blanchard, C. L., 2003,
The southeastern aerosol research and characterization study: Part 1 - Overview, J. Air & Waste Manage.
Assoc., 53:1460.
Hildemann, L. M., Cass. G. R.. Mazurek, M. A., Simoneit, B. R. T., 1993. Mathematical modeling of urban organic
aerosol: properties measured by high-resolution gas chromatography. Environ. Sci. Technol, 27:2045.
Park. R. J.. Jacob. D. J.. Chin. M.. Martin, R. V.. 2003. Sources of carbonaceous aerosols over the United States and
implications for natural visibility, J. Geophys. Res., 108(D12), 4335, doi:10.1029/2002JD003190.
Roe. S. M., Spivev. M. D., Lindquist, H. C.. Hemmer, P., Huntley. R., 2004, National emissions inventory for
commercial cooking. 13th Annual Emission Inventory Conference, Clears ater, FL.
Rogge, W. F., Hildemann, L. M., Mazurek, M. A., Cass, G. R., Simoneit, B, R. T., 1996, Mathematical modeling of
atmospheric fine particle-associated primary organic compound concentrations, J. Geophys. Res., 101:19379.
Rogge, W. F.. Hildemann. L. M., Mazurek. M. A.. Cass. G. R., Simoneit, B. R. T.. 1997, Sources of fine organic
aerosol. 8. Boilers burning No. 2 distillate fuel oil, Environ. Sci. Technol, 31:2731.
Ryan, R.. 2003, personal communication.
Schauer, J. J., Rogge, W. F., Hildemann, L. M., Mazurek, M. A., Cass, G. R., Simoneit, B. R. T., 1996, Source
apportionment of airborne particulate matter using organic compounds as tracers, Atmos. Environ., 30:3837.
Watson, J. G.. Chow, J. C., 2000, Reconciling Urban Fugitive Dust Emissions Inventory and Ambient Source
Contribution Estimates: Summary of Current Knowledge and Needed Research, DRI document 6110.4F, Reno,
NV.
Yu, S., Dennis, R. L,, Bhave, P. V., Eder, B, K., 2004, Primary and secondary organic aerosols over the United
States: estimates on the basis of observed organic carbon (OC) and elemental carbon (EC), and an ]uilit\
modeled primary OC/EC ratios, Atmos. Environ., in press.
Zheng, M., Cass, G. R., Schauer, J. J., Edgerton, E. S., 2002, Source apportionment of PM2.5 in the souineasiern
United States using solvent-extractable organic compounds as tracers, Environ. Sci. Technol, 36:2361.
The research presented here was performed under the Memorandum of Understanding between the U.S.
Environmental Protection Agency (EPA) and the U.S. Department of Commerce's National Oceanic and Atmospheric
Administration (NOAA) and under agreement number DW13921548. Although it has been reviewed by EPA and
NOAA and approved for publication, it does not necessarily reflect their policies or views.

-------