Comparison of Atlanta Emission Inventory with Ambient Data Using
Chemical Mass Balance Receptor Modeling
EPA/600/A-94/244
Teri L. Conner1, John F. Collins2, William A. Lonnemair, ana Kooerc l. »eiia
'Atmospheric Research and Exposure Assessment Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, North Carolina, 27711
2Civil Engineering Department, Environmental Engineering Program
University of Southern California
3620 South Vermont Avenue
Los Angeles, California, 90089-2531
ABSTRACT
A comprehensive hydrocarbon database was obtained at 6 sites in the Atlanta metropolitan
area during the summer of 1990. Samples were collected in stainless steel canisters and analyzed
for 54 hydrocarbon species plus total non-methane organic compounds (TNMOC). The
contributions of the major sources of TNMOC at each of the 6 sites were estimated through a
procedure called Chemical Mass Balance (CMB) receptor modeling. Spatial variability of the
source contributions is discussed. Results of the CMB analysis for one of the sites are compared
with the emission inventory for Atlanta using several different approaches. The inventory highway
mobile source estimate tends to be smaller than the minimum ambient data-derived highway mobile
source estimate, and the inventory area plus point source estimate tends to be larger than the
maximum ambient data-derived estimate for the data set examined. However, these source estimates
are interdependent to some extent. Limitations of these comparisons are discussed.
INTRODUCTION
During the summer of 1990, the U.S. Environmental Protection Agency conducted a 2-
month, 6-site air quality monitoring study in the Atlanta metropolitan area referred to as the "1990
Atlanta Ozone Precursor Study". The goal of the study was to obtain hourly data for ozone and its
precursors at multiple sites in an area which is out of compliance for the ozone NAAQS. Earlier
reports on this work focussed on hydrocarbon data obtained from hourly on-site measurements made
at one of the sites. In this report, we examine data obtained at all 6 sites, but on a less frequent
basis, from samples collected in stainless steel canisters. The data set obtained provides an
opportunity for reconciling ambient data with emission inventories. Previous efforts to compare
ambient data with emission inventories have made use of a receptor modeling technique called
Chemical Mass Balance (CMB) and other techniques employing the ambient data to determine the
sources of the measured species.1'3 Source contributions obtained in this way - from ambient data
with no explicit use of meteorology or emission inventories - are compared in a relative sense with
emissions inventories derived by traditional methods. In this effort, CMB is applied to the
hydrocarbon data obtained at 6 sampling sites. Spatial variability of the results is examined, and
new approaches for comparing CMB results with emission inventories are demonstrated for one of
the sampling sites. While this analysis emphasizes results for motor vehicle-related sources, other
sources are considered as needed for the CMB calculations.
AMBIENT DATA
Sample Collection and Analysis
Whole-air samples were collected in 6-liter Summa polished stainless steel canisters.
Sampling was conducted approximately every other day at each of 6 sampling sites throughout the
Atlanta metropolitan area. Sites 2 and 3 were located near downtown Atlanta. Sites 4, 5, and 6

-------
were located in a variety of surroundings away from the city. Site 4 was located about 5 km south
of downtown Atlanta. Sites 5 and 6 were at typical downwind locations from Atlanta approximately
15 and 10 km, respectively. Site 1, located northwest of Atlanta, served as a background site.
Start times for the 30-minute samples were rotated through the hours 0 (midnight), 8, 10, 12, 15,
18. A total of 163 samples were collected in this manner. Samples were analyzed by a well-
characterized gas chromatograph with flame ionization detection (GC/FID) system operated by the
Gas Kinetics and Photochemistry Research Branch of AREAL.6 This is the same system used to
analyze the mobile source profile samples prepared for this study.7 Detailed descriptions of the
sampling and analytical procedures are presented elsewhere.8
Data Screening
The primary data screening tool used was to examine plots of each species versus every other
species. This screening method has been demonstrated to be quite effective in identifying outliers
and other problems with the data.5,9 Particular attention was paid to those pairs of species which are
expected to track each other very well. Notable outliers were observed for the butane and pentane
isomers, resulting in elimination of the affected samples from further consideration. The problem
was most severe at site 5, so this site was occluded from further analysis. The reason for the
unusual butane and pentane compound concentrations is not known at this time. However, the
objective of this exercise is to look at the usual situation, so screening out unusual data is
appropriate in this case. Any offsets observed in comparing one species with another resulted in
exclusion of the affected species from consideration as a fitting species in the CMB analysis.
Species affected by offsets were 2-methylpentane and 2-methylhexane. One hundred of a possible
135 samples wore retained (site 5 excluded) after data screening.
Data Averaging
Chemical Mass Balance (CMB) calculations were performed on the average of samples from
each site. Averages provide a snapshot of typical conditions at a sampling site and cancel out some
of the random variations associated with individual samples. Only those samples which survived the
screening process were included in die averages. Species with missing values were assigned a value
of zero rather than excluding those samples from the average. This approach will tend to give a
lower estimate of the species average concentration. In most, if not all cases, the missing values
were assigned when concentrations were below detection; thus, assigning a value of zero to these
species concentrations is a reasonable approach. It is unlikely such species will figure prominently
into a CMB calculation, due to their high relative uncertainty.
CHEMICAL MASS BALANCE CALCULATION PROCEDURE
Introduction
The U.S. EPA/DRI Chemical Mass Balance model, version 7.0J0,n was used to quantitatively
apportion chemical species measured at the sampling sites to the major sources contributing to the
total non-methane organic compounds, TNMOC (defined in this paper as the sum of all GC peaks
excluding ethane) at those sites. The Chemical Mass Balance model, version 7 (CMB7) consists of
an effective variance least squares solution to a set of linear equations which express each measured
chemical species concentration as a linear sum of the contributions of each source to the chemical
species. The effective variance solution gives the most weight to source and ambient measurements
with the lowest relative uncertainty estimates. Source contributions are expressed as the product of
the abundance of the species as emitted by the source and the total mass concentration contributed
by the source. The set of abundances of all species as emitted by each source represents the "source
profile" or "source fingerprint".
In practice, it is not possible to apportion mass to each individual contributing source.
Individual sources may be too similar to one another, too numerous, or may not contribute
significantly to the total mass loading. Sources are generally grouped together to represent a single

-------
"source category" or "source type". Furthermore, these source types are distinguished chemically
(e.g., gasoline head space vapor) rather than by the mechanism of emission (e.g., fuel storage tank
vapor displacement versus vehicle refueling vapor displacement.) This distinction will be important
for comparison of CMB results with emission inventories.
Assumptions made in performing a chemical mass balance include: 1) the abundance of each
species used in the fitting procedure is known for each source type, 2) all major sources of each
species used in the fitting procedure must be included in the CMB, and 3) chemical species do not
react with each other. The third assumption listed here is of particular concern for NMOC's, as
virtually all of these species are reactive towards the OH radical to some degree. Thus, choice of
fitting species must be restricted to the least reactive compounds. Other assumptions made for the
CMB model are listed and explained elsewhere.10"12
The CMB7 model requires both ambient concentrations and uncertainties as input. Ambient
measurement uncertainties were calculated from the following expression3 applied to die average
species concentrations at each site:
AQ(ppbC) = [(0.2)2 + (O.Q5*Ci)2]1/2
where C; is the measured ambient concentration of species i. This method of estimating the
uncertainty is used rather than calculating the standard deviation of the average, because the latter
might give low-concentration species more weight in the CMB than the well-measured higher
concentration species.
Source Profiles
Local source profiles were developed for this study to represent roadway, whole gasoline,
and gasoline head space vapor.7 The roadway profile (ROAD) was obtained from samples collected
in an extended underpass in downtown Atlanta and captures the emissions of vehicles in motion
(tailpipe plus running losses). This profile was not corrected for background concentrations and is
dominated by emissions of vehicles in motion.7 The whole gasoline (GAS) and gasoline head space
vapor (HS) profiles are each a composite of profiles of 3 octane grades from the 6 major vendors in
the Atlanta area. The gasoline head space vapor profiles used in this analysis were determined for
24°C. These mobile source related profiles were all analyzed on the same GC/FID system used to
analyze the ambient samples.
In a previous analysis of Atlanta continuous monitoring GC data,3 natural gas plus an
additional propane-rich source were found to account for most of the propane and about 10% of the
TNMOC (defined in that study as the sum of all GC peaks with ethane included). In that study, a
propane-rich profile derived from the ambient data itself9 was used to account for the excess
propane. The abundance of propane in that profile was 45%, but it is not clear what made up the
remaining 55%. In this study, a pure propane profile (PROPANE) was used to account for the
propane not explained by the natural gas profile. Using a single-constituent profile in this way will
underestimate the propane source if there are other species emitted by that source. However, this
approach avoids the uncertainty associated with using a profile derived from a data set different
from that to which the profile will be applied. The natural gas profile (NG) used in this exercise
was obtained from existing measurements on Atlanta utility natural gas.3
For some sites and samples, toluene could not be folly explained by the mobile source
profiles, suggesting the possibility of some solvent-type of activity. The existence of solvent-type
activities is confirmed by the emission inventory.14 Scheff et al. (1989)13 compiled profiles for CMB
apportionment of volatile organic compounds. Included in that compilation are profiles for surface
coatings and other solvent-related activities. The autopainting profile (AUTOCOAT) was found to
best explain the data. The profiles are given as weight % while the other profiles used here are in
ppbC %. The two profile representations are nearly the same, so no conversion was done for the
AUTOCOAT profile.

-------
All profiles are normalized to the TNMOC defined in this paper as occluding ethane. Ethane
is excluded because it is excluded from the SEP guidance for emission inventories for ozone non-
compliance.16 The only profiles affected by this exclusion are roadway (ROAD) and natural gas
(NG). The emission inventory also excludes many chlorinated compounds, but these are not
expected to figure prominently into the sources apportioned here. The emission inventory includes
formaldehyde and acetaldehyde which are, at best, only partially measured by the FID. These
compounds could potentially figure prominently into the roadway emissions, but data from a
previous roadside sampling study" suggests that these aldehydes are only a small part (< 2%) of
the TNMOC.
Fitting Species
One of the requirements of the CMB is that chemical compounds do not react with other
species. In this exercise, fitting species will be those with a reactivity towards OH radical less than
that of ethene. In a polluted atmosphere (*OH = 5x10* molecules/cm3), this would correspond to a
lifetime greater than 6.S hours. In the canister data set, all alkenes, plus n-alkanes greater than or
equal to C-8, xylenes, trimethylbenzenes, and methylheptanes are all excluded from being fitting
species on the basis of reactivity. In addition, 2-methylpentane and 2-methylhexane were excluded
from being fitting species because of offsets observed in the scatter plots. Methylcyclopentane, 3-
methylhexane, methylcyclohexane, and n-heptane woe eliminated from the fitting species because
they were frequently missing or underpredicted. The list of measured species and their fitting
species status is shown in Table 1.
CHEMICAL MASS BALANCE RESULTS AND DISCUSSION
Results of Average - Fitting Diagnostics and Colllnearities
Table 2 shows the results of CMB7 calculations applied to the site averages as a percent of
the total mass apportioned (sum of source estimates). Each calculation had at least 12 degrees of
freedom, which should be ample to give statistically meaningful results. The goodness of fit
indicators R2 and Chi2 were excellent. However, each fit was accompanied by one or more
uncertainty/similarity clusters. These clusters are indicated in the CMB7 output display and are
formed when the standard error of any of the source estimates involved is 50% or higher and there
is excessive similarity among the source profiles, as indicated by an established diagnostic.52 The
standard error of the combined source contributions indicated in the cluster display may be smaller
than the standard error of the source contribution estimate of any single source in the cluster if
collinearity is the cause of the high standard error. Source combinations typically affected by
collinearity were whole gas (GAS) and roadway (ROAD), and propane source (PROPANE) and
natural gas (NG).
There are several ways of dealing with uncertainty/collinearity clusters in the CMB.12 One
approach is to use additional species in the fit. Unfortunately, species most likely to further
distinguish ROAD and GAS from one another are fairly reactive (e.g., ethene, propene). However,
ignoring the reactivity rules established for this exercise and including some of the species likely to
be associated with exhaust did not eliminate the collinearity. Another approach would be to prepare
a composite profile for the collinear sources. However, this would require having some independent
estimate of the relative importance of the profiles involved.
A test can also be performed which may indicate if the resulting cluster is due to high profile
uncertainty rather than true collinearity. The test consists of reducing uncertainties in the affected
profile(s) and thai rerunning the CMB. If the clusters axe no longer listed (i.e., the standard error
is reduced to less than 50%), then the apparent collinearity was likely due to profile uncertainties.
This test was applied to the GAS/ROAD cluster for the average ate 2 sample. Uncertainties in the
GAS and ROAD profiles were arbitrarily reduced by half. When the source estimates were
recomputed, the uncertainty/similarity clusters and high standard errors remained. Even when the
GAS profile standard error was reduced to an unreasonably low level (0.00001, in terms of ratio of

-------
species to total), the clusters and high uncertainties remained. It is therefore likely that a true
collinearity as defined by CMB7 exists between the GAS and ROAD profiles and the two cannot be
resolved reliably based on the collinearity criteria of the CMB7 model. Thus, the separate reporting
of ROAD and GAS should be regarded with caution. The combined source contribution of ROAD
and GAS as reported in the CMB7 cluster display are reported in Table 2, as well as their separate
contributions. It is important to point out that the collinearity criteria of CMB7 are somewhat
arbitrary and were originally established for typical particulate data. Furthermore, no ROAD plus
GAS cluster was observed for site 3. The uncertainty estimated for GAS was below the 50%
threshold for that site, but otherwise uncertainties for ROAD and GAS were similar to estimates at
other sites.
The PROPANE and NG sources are not likely to be resolved with any standard remedies
because of the small number of species associated with these sources. In this case, their source
contributions are reported as the sum presented in the cluster display.
The surface coatings source (AUTOCOAT) was not collinear with other sources, but
generally was small with a very large standard error, indicating it is probably not an important
source at most sites. The standard error was less than 50% at sites 2 and 4, but greater than 50%
and sites 1, 3, and 6. The AUTOCOAT profile was removed from the calculations for the latter
sites without any significant effect on the other source estimates.
Site Comparisons - Spatial Variability
Examination of the CMB results presented in Table 2 for site averages reveals both
similarities and differences among the sites. The combined ROAD and GAS sources comprise by
far the largest fraction of the total apportioned mass (83-91%). The HS accounts for only 4-15% of
the total apportioned mass and was smallest at the downtown sites (sites 2 and 3). The combined
fuel gases (PROPANE+NG) accounted for only a small fraction of the total (2-5%). AUTOCOAT
was only found at sites 2 and 4, with site 2 having the largest percentage (5%). An automotive
assembly plant, one of the largest VOC point sources in the Atlanta area, is located less than 10 km
sites 2 and 4 (though not in the same direction from each site). The AUTOCOAT source impacts
seen at these sites may also have contributions from one or more small but nearby sources.
ROAD and GAS were not resolvable based on the CMB7 criteria for resolving sources.
However, it is worth noting their individual results, keeping in mind that the uncertainties are much
larger than for the combined source estimate. While the sum of the ROAD and GAS percent varied
little from site to site, the ratio of ROAD to GAS varied considerably (although uncertainties were
sometimes quite large). The smallest ratios of ROAD to GAS (4-5) were observed at the downtown
sites (sites 2 and 3), whereas ratios were much larger (10-13) at the outlying sites, with the largest
ratio occurring at the background site (site 1). Whole gasoline emissions have several potentially
significant sources, including spillage from refueling and fuel transfer and storage and hot soak
evaporative emissions. Hot soak emissions are those evaporative emissions which escape from an
automobile during the hour or so after it has been turned off but while it is still hot. These
emissions are the most chemically similar to whole gasoline emissions. Differences in daytime
traffic patterns in the vicinity of each site mav explain the observed trend. Automobiles converge
on the downtown area, mostly during the morning, and remain there only for the duration of the
workday. One would expect a greater proportion of automobiles parked for a short time in the
downtown area compared with the outlying areas where much of the traffic volume occurs (such as
on the major interstate loop around the city). This distribution of automobile activity may explain
the distribution of ROAD and GAS emissions suggested by the CMB7 results. Furthermore,
automobiles are parked at residences in the outlying areas for longer periods of time and in greater
numbers than in the downtown area. This longer period of inactivity would lead to a larger
percentage of diurnal evaporative emissions and hence the larger percentage of gasoline head space
evaporative emissions in the outlying areas as suggested by the CMB results. Hourly results may
shed some light on these hypotheses.

-------
It is interesting to compare the percent of the TNMOC actually accounted for by the sources
included in the CMB calculations, listed as SUM/TNMOC in Table 2 (recall that ethane must be
subtracted from TNMOC to be consistent with the emission inventory). The smallest fraction of the
TNMOC (49%) accounted for was found for site 1, the background site. This may be explained by
the larger proportion of biogenic emissions likely to be found at the background site. Biogenic
sources were not included in the CMB calculations. A much greater portion of the TNMOC (64-
69%) was explained at the other sites with the exception of site 2, where only 54% of the TNMOC
was accounted for. Hourly results may help explain these results.
COMPARISON OF CMB RESULTS WITH EMISSION INVENTORY
Discussion of limitations of Comparison and Possible Approaches
The CMB model distinguishes among source types by their chemical differences, while
emission inventories are generated based on a specific emission source or mechanism. Different
emission mechanisms can produce emissions which are chemically similar and therefore would be
treated as a single source type by the CMB model. Thus, the ability to directly compare CMB
results with emission inventories is limited.
There are a number of different approaches that can be taken to maximize the information
obtained from such comparisons. The inventory estimate for Highway Mobile Sources accounts for
some, but not all, of the whole gas and head space evaporative emissions. Therefore, a range of
estimates for the Highway Mobile Sources should be calculated from the CMB estimates which
would have a minimum equal to the ROAD estimate and a maximum equal to the sum of the
ROAD, GAS, and HS estimates. One could get at the distribution of evaporative emissions possibly
by using a typical MOBILE model1* output to apportion the HS and GAS emissions that should be
included in the Highway Mobile Source estimate. However, that approach would involve making
some assumptions about the meaning of the ROAD estimate compared to the MOBILE model
exhaust (plus running loss) estimate, the distribution of whole gasoline and gasoline vapor among
the different evaporative emissions, and the accuracy of the ratio of evaporative to exhaust
emissions, and would remove some of the independence of the ambient estimate.
The area and point sources are similarly difficult to deal with, in part because of the
distribution of GAS and HS between mobile and non-mobile sources. A minimum estimate might
equal the sum of NG, PROPANE, and AUTOCOAT while a maximum estimate could equal the
sum of these source estimates plus GAS and HS. This range could be compared to either area
sources or the sum of area and point sources. The CMB may be missing or underestimating point
and area sources, both because the CMB estimates are based only on FID measurements (and
therefore nay miss some chemical species important to architectural coatings, solvents, and other
area sources) and because the impact of point sources on any particular sampling site is meteorology
dependent.
Comparisons can include or exclude biogenic emissions. The ambient isoprene
concentrations could be used to represent the lower-limit estimate of the biogenic fraction.
Unfortunately, some suspiciously low values of isoprene are found at each site (as revealed by plots
of isoprene versus 2,2-dimethylbutane). Another way to handle the biogenic fraction is to use the
inventory value (16.8% biogenics) and then scale down the other source percent contributions
estimated from the CMB so that the sum is 100%. This approach is probably closer to estimating
the true biogenic emissions than using a minimum estimate and no worse than guessing species
which should be included with biogenics and trying to estimate their reaction losses. Some of the
species that should be included as biogenic emissions are aldehydes, which are not measured
quantitatively by FID. Yet another approach would be to attribute all of the TNMOC unexplained
by other sources to the maximum biogenic contribution. This approach could significantly
overestimate the biogenic contribution, as species from other unapportioned sources plus secondary
reaction products as well as biogenic species are all potential contributors to the unexplained
TNMOC. A fourth alternative is to exclude biogenics altogether and renormalize the inventory

-------
estimates to exclude biogenic emissions.
Results and Discussion
The CMB results for the average site 2 sample (weekends excluded*) were used to
demonstrate the different approaches to comparing ambient data with emission inventories. A
summary of these comparisons are presented in Table 3. The VOC emission inventory for Atlanta
was obtained from the Georgia Department of Natural Resources." He inventory was constructed
to represent a typical ozone season 24-hour average weekday in 1990. The inventory developed for
Fulton County, which encompasses most of downtown Atlanta, was judged to be the most
appropriate for comparing with data from the ate 2 site.5 Inventory values are presented as a
percent for all source categories and for source categories excluding biogenic sources (inventory
estimates were normalized to sum to 100% in each case).
Starting with the comparisons which exclude biogenic emissions altogether, the inventory
highway mobile source estimate is smaller than the minimum ambient data-derived highway mobile
source estimate, and the inventory area plus point source estimate is larger than the maximum
ambient data-derived estimate. A similar pattern is observed for the comparisons which
accommodate biogenic emissions in a variety of ways, with the exception of the approach which sets
the biogenic emissions equal to the total unexplained NMOC. In that comparison, the inventory-
derived estimate for highway mobile sources exceeds the maximum ambient-derived estimate, and
the difference between the inventory-derived and ambient-derived point plus area source estimate is
even greater than for the other approaches. This observation is likely a direct result of attributing
all the mass unexplained by the CMB to the biogenic fraction, an approach likely to overestimate the
biogenic fraction while reducing the percent estimates of other sources because of renormalization.
The inventory does not have a mechanism for determining whether mass has been unaccounted for,
as does CMB,
The comparisons presented here are expressed in relative terms (percent of total emissions)
rather than absolute terms (mass emitted per source category per unit time), resulting in a less
sensitive comparison.19 Nevertheless, the trend for most of the comparison approaches was that the
inventory underestimates the highway mobile sources and overestimates the combined area and point
sources. The comparison is complicated by the fact that the CMB results reflect mostly daytime
(hours 8-18) conditions, while the inventory represents 24-hour average emissions.
CONCLUSIONS
Data collected in Atlanta during the summer of 1990 are used to demonstrate some new
approaches to comparing emission inventories with ambient data using CMB calculations. These
new approaches make use of ranges of source estimates obtained from the ambient data. Ranges of
source estimates can be used to deal with some of the inherent difficulties of comparing inventories
with ambient data.
For most comparison approaches, the inventory highway mobile source estimate tended to be
smaller than the minimum value estimated from the CMB results. However, the inventory estimate
was derived from version 4.1 of the MOBILE model. Newer versions of that model may narrow
the gap. Furthermore, the CMB estimate for mobile sources may include some non-automotive
source emissions. An underestimation of the point plus are source category would also force the
highway mobile sources to be overestimated (see below).
In the comparison in which the biogenic emissions were set equal to the total unexplained
NMOC, the inventory highway mobile source estimate was larger than the CMB maximum estimate,
most likely because this approach tends to overestimate biogenics, and thus underestimate the other
sources on a percent basis. The results of Lewis et al., 19935 compared in the same manner (using
maximum and minimum values and including unexplained TNMOC) yield an emission inventory
estimate for highway mobile sources which falls in the middle of the minimum and maximum range
of the CMB estimates. Differences between the Lewis et al., 1993 results and those of this analysis

-------
may be due not only to differences in the dates and times the samples were collected but also the
differences in the total NMOC reported by the different GC/FID systems used in the 2 studies.20
The differences in the results of these 2 studies need to be investigated further.
The combination of point and area sources tended to be underestimated by the CMB
compared with the emission inventory. Some chemical species which may be important components
of solvents, coatings, and similar sources may not be measured adequately with a GC/FID and thus
would not be included in estimates derived from the ambient data. Furthermore, chemical profiles
used to represent the bulk of the point and area sources may be inadequate.
In general, comparisons of emission inventories with ambient data using CMB calculations
are limited. Use of a more detailed highway mobile source inventory estimate in the comparisons
may be helpful, as would uncertainty estimates for the emission inventory. The inventory used in
these comparisons represents a 24-hour average of the entire summer, while the ambient data
represented limited days and hours.
ACKNOWLEDGEMENTS
The authors thank Bobby Edmonds, John Wall, and Jerry Burger for collecting the canister
samples used in this analysis.
DISCLAIMER
This paper has beat reviewed in accordance with the U.S. Environmental Protection
Agency's peer and administrative review policies and approved for publication. Mention of trade
names or commercial products does not constitute endorsement or recommendations for use.
REFERENCES
1.	P.F. Nelson, S.M. Quigley, M.Y. Smith, Atmospheric Environment 1983 12, 439-449.
2.	R.A. Harley, M.P Hannigan, G.R. Cass, EflyjfPB, Sci,	1992 2& 2395-2408.
3.	D.M. Kenski, R.A. Wadden, P.A. Scheff, W.A. Lonneman, "A receptor modeling approach
to VOC emission inventory validation in five U.S. cities," in Proceedings of the 86* Annual
Meeting of the Air and Waste Management Association, Denver, Colorado, 1993, Paper 93-
WP-100.04.
4.	P.A. Scheff and R.A. Wadden, Environ. Sci. Technol. 1993 21, 617-625.
5.	C.W. Lewis, T.L. Conner, R.K. Stevens, J.F. Collins, R.C. Henry, in "Receptor modeling
of volatile hydrocarbons measured in the 1990 Atlanta ozone precursor study," in
Proceedings of the Annual Meeting of the Air and Waste Management Association,
Denver, CO, 1993, Paper 93-TP-58.04.
6.	R.L. Sella, W.A. Lonneman, S.A. Meeks; Determination ofC3 to Ambient Air
Hydrocarbons in 39 U.S. Cities, From 1984 Through 1986, EPA/600/3-89/058; U.S.
Environmental Protection Agency: Research Triangle Park, 1989.
7.	T.L. Conner, W.A. Lonneman, R.L. Seila "Transportation-related volatile hydrocarbon
source profiles measured in Atlanta," accepted for publication in J. Air Waste Management.
Assoc..
8.	L.J. Purdue et al.; Atlanta Ozone Precursor Monitoring Study Data Report, EPA-600/R-
92/157; U.S. Environmental Protection Agency: Research Triangle Park, NC, 1992.

-------
9.	R.C, Henry, C.W. Lewis, J.F. Collins, Environ. Sri, Techno!. 1994 28, 823-832.
10.	J.G. Watson, N.F. Robinson, J.C. Chow, R.C. Henry, B.M. Kim, T.G. Pace, E.L. Meyer,
Q. Nguyen, Environ. Software 1990 $, 38-49.
11.	J.G. Watson et al.; Receptor Model Technical Series, Volume III (1989 Revision): CMB7
User's Manual, EPA-450/4-90/004; U.S. Environmental Protection Agency: Research
Triangle Park, NC, 1990.
12.	T.G Pace and J.G. Watson; Protocol for Applying and Validating the CMS Model, EPA-
450/4-87/010; U.S. Environmental Protection Agency: Research Triangle Park, NC, 1987.
13.	W.E. Liss, W.H. Thrasher, G.F. Steinmetz et al., Variability of Natural Gas Composition in
Select Major Metropolitan Areas of the United States, Report No. GRI-92/0123; Gas
Research Institute: Chicago, IL, 1992, p. 58.
14.	R. Methier, "Georgia's experience in preparing the 1990 baseline inventory," in Proceedings
of the EPA/A&WMA International Conference on Emission Inventory Issues, 1992.
15.	P.A. Scheff, R.A. Wadden, B.A. Bates, P.F. Aronian, JAPCA 1989 22, 469-478,
16.	Emission Inventory Requirements for Ozone State Implementation Plans, EPA-450/4-91-G10;
United States Environmental Protection Agency: Research Triangle Park, NC, 1991.
17.	R.B. Zweidinger, J.E, Sigsby, Jr., S.B. Tejada, F.D. Stump, D.L, Dropkin, W.D, Ray,
J.W. Duncan, Environ. Sci. Technol. 1988 22, 956-962.
18.	U.S. EPA, "User's guide to MOBILES (mobile source emission factor model)," U.S.
Environmental Protection Agency, Office of Air and Radiation, Test and Evaluation Branch,
2625 Plymouth Road, Ann Arbor, MI, 48105, 1993.
19.	C.W. Lewis, U.S. Environmental Protection Agency, Research Triangle Park, NC, personal
communication, 1993.
20.	J.H. Shreffler, JAWMA 1993 42, 1576-1584.

-------
Table 1. Measured hydrocarbon species.
Hydrocarbon Species
Status
Hydrocarbon Species
Status
ethene
r
t-2-hexene
r
acetylene
*
c-2-hexene
r
ethane
n
methylcyclopentane
-
propene
r
2,4-dimethylpentane
~
propane
*
benzene
*
i-butane
*
cyclohexane
*
1-butene
r
2-methylhexane
0
n-butane
*
2,3-dimethylpentane
*
t-2-butene
r
3-methylhexane
-
c-2-butene
r
2,2,4-trimethylpentane
*
3-methyl-1-butene
r
n-heptane
-
| i-pentane
*
methylcyclohexane
_
I 1-pentene
r
2,3 »4-trimethylpentane
*
| n-pentane
*
toluene
*
1 isoprene
x,r
2-methylheptane
X j
| t-2-pentene
r
3-methylheptane
r |
I c-2-pentene
r
n-octane
r |
| 2-methyl-2-butene
r
ethylbenzene
* I
1 2,2-dimethylbutane
X
m/p-xylene
r
I cyclopentene
r
o-xylene
r
4-methyl-1-pentene
r
n-nonane
r
cyclopentane

i-propylbenzene
-
2,3-dimethylbutane
*
a-pinene
n,r
2-methylpentane
o
n-propylbenzene
*
3-methylpentane
*
1,3,5-trimethylbenzene
r
2-methyl-1 -pentene
r
b-pinene
n,r
n-hexane
*
1,2,4-trimethylbenzene
r
KEY: x = unusual behavior in scatter plots
n = not available or not included in profil*
r = reactive	- = frequently underpredicted
~ « fitting species	o = offset

-------
Table 2. Results of CMB analysis as percent of total mass apportioned (sum of source estimates), unless otherwise indicated.
| Source Category
Site 1
Site 2 (weekends
included)
Site 2 (weekends
excluded)
Site 3
Site 4
Site 6
[ROAD
79.1 ± 11.1%
69.3 ± 9.2%
72.7 ± 9.2%
75.0 ±9.1%
78.1 ±9.4%
75.1 ± 9.2%
Igas
6.3 ± 10.0%
16.9 ±9.1%
14.0 ± 8.9%
16.1 ± 8.0%
7.6 ± 8.7%
7.6 ± 8.0%
ROAD+GAS
85.4 ± 4.4%
86.6 ± 4.9%
86.8 ± 4.6%
91.1 ± 12.1%
85.8 ± 4.5%
82.7 ± 3.6%
HS
9.4 ± 2.5%
4.3 ± 2.2%
4.0 ± 2.2%
6.4 ± 2.2%
10.7 ± 2.5%
14.7 ± 2.5%
PROP+NG
5.2 ± 0.7%
4.0 ± 0.6%
3.7 ± 0.6%
2.4 ± 0.4%
1.8 ± 0.4%
2.6 ± 0.4%
AUTOCOAT
0%
5.1 ± 1.1%
5.4 ± 1.1%
0%
1.8 ± 0.9%
0%
SUM
110.7 ± 5.7
ppbC
182.1 ± 10.0
ppbC
204.9 ± 10.8
ppbC
259.2 ± 31.9
ppbC
238.0 ± 12.4
ppbC
221.7 ± 9.7
ppbC
Jtnmoc
225.2 ± 19.5
ppbC
340.1 ± 19.7
ppbC
370.8 ± 18.8
ppbC
374.9 ± 19.6
ppbC
373.7 ± 19.6
ppbC
327.6 ± 19.5 1
PPbC |
JsUM/TNMOC
49.2 ± 5.0%
53.6 ± 4.3%
55.2 ± 4.0%
69.2 ± 9.3%
63.7 ± 4.7%
67.7 ±5.0% |
I ROAD/GAS
12.6 ± 20.1
4.1 ± 2.3
5.2 ± 3.3
4.7 ± 2.4
10.3 ± 11.8
9.8 ± 10.4
1 Degrees of Freedom
13
12
12
13
12
13
|r2
0.99
1.00
1.00
1.00
1.00
0.99
|chi2 1 0.69
0.32
0.30
0.26
0.44
0.77

-------
Table 3. Comparison of emission inventory with relative source estimates derived from the site 2 ambient data (weekends excluded) using
CMB calculations.
j Summary of CMB Results*
Summary of Emission Inventory
Assume biogenic % = % reoorted in inventory (16.8%):
Including Biogenics:
MIN. MAX.
HIGHWAY MOBILE SOURCES: 61% 75%
POINT+AREA SOURCES: 8% 23%
Assume biogenic min. % = isoprene % fl.9%):
MIN. MAX.
HIGHWAY MOBILE SOURCES: 71% 89%
POINT+AREA SOURCES: 9% 27%
HIGHWAY MOBILE SOURCES: 56.0%
POINT+AREA SOURCES; 27.2%
| Assume biogenic max. % - % unexplained (47.4%Y
| MIN. MAX.
[ HIGHWAY MOBILE SOURCES: 41% 51%
| POINT+AREA SOURCES: 5% 15%
I Assume no biogenics:
1 MIN. MAX.
I HIGHWAY MOBILE SOURCES: 73% 91%
1 POINT+AREA SOURCES: 9% 27%
Excluding Biogenics:
HIGHWAY MOBILE SOURCES: 67.3%
POINT+AREA SOURCES: 32.7% |
* Results presented as percent of total apportioned NMOC (sum of source estimates).

-------
Comparison of Atlanta Emission Inventory with Ambient
Data Using Chemical Mass Balance Receptor Modeling
Teri L. Conner
John F. Collins
William A. Lonneman
Robert L. Seila
USEPA/AREAL
use
USEPA/AREAL
USEPA/AREAL

-------
STUDY OBJECTIVES
•	Independent assessment of VOC source emissions
•	Use ambient data and CMB receptor modeling
•	Compare with emission inventory
•	Emphasis on highway motor vehicle emissions

-------
What is Chemical Mass Balance Receptor Modeling?
•	Ambient Concentrations = Sum of Contributions from
Different Source Types
•	Need Ambient Data and Chemical Profiles of Sources

-------
OUTDOOR AMBIENT SAMPLE COLLECTION
•	Whole-air samples in stainless steel canisters
•	6 sampling sites (urban, suburban, background)
•	30 minute samples approx. every other day
•	Start times rotated through hours 0, 8, 12, 15, 18

-------
SAMPLE ANALYSIS
• P _ O 4. nnn.mofhano hi/Hrnr^arkone
w£ 12 viijvi VTit#Liiciiit# iiyuriJvcirKJiJiio
•	Two GC columns used to improve separation of C2's
•	Flame-ionization detection

-------
DATA SCREENING
Necessary for any GC/FID data set
Examine plots of each species vs other species
Identify outliers and other problems
Use procedure to guide selection of samples/species

-------
CANISTER DATA
ACETYLENE

-------
CANISTER DATA
120
100-
80-
w
z
J5
z
60-
a.
c
40-
20-
23
	r
0 5
25
30
10
20
35
40
45
15
ACETYLENE

-------
REASONS FOR EXCLUDING SPECIES FROM
CMB CALCULATIONS
•	Reactive
•	Not available/not included in profiles
•	Frequently underpredicted by CMB (missing source?)
•	Offset or other unusual behavior in scatter plots

-------
CMB FITTING SPECIES
(18 out of possible 54 species)
acetylene
propane
i-butane
n-butane
i-pentane
n-pentane
2.3-dimethylbutane
3-methylpentane
n-hexane
2.4-dimethylpentane
benzene
i^v^lfihoysino
vjf vIUl ICACll IC
2,3-dimethylpentane
2,2,4-trimethylpentane
2,3,4-trimethy Ipentane
toluene
ethylbenzene
n-propylbenzene

-------
SOURCE PROFILES
Derive from ambient data (SAFER)
"Off-the-shelf" source measurements
Concurrent, on-location source measurements

-------
PROFILES USED
Measured:
•	Emissions from a busy roadway
•	Whole gasoline composite
•	Headspace gasoline composite
Literature:
•	Auto painting profile
•	Natural gas
Other:
•	Propane

-------
Table 2. Results of CMB analysis as percent of total mass apportioned (sum of source estimates), unless otherwise indicated.
Source Category
Site 1
Site 2 (weekends
included)
Site 2 (weekends
excluded)
Site 3
Site 4
Site 6
Iroad
79.1 ± 11.1%
69.3 ± 9.2%
72.7 ± 9.2%
75.0 ±9.1%
78.1 ± 9.4%
75.1 ± 9.2%
Igas
6.3 ± 10.0%
16.9 ±9.1%
14.0 ± 8.9%
16.1 ± 8.0%
7.6 ± 8.7%
7.6 ± 8.0%
|road+gas
85.4 ± 4.4%
86.6 ± 4.9%
86.8 ± 4.6%
91.1 ± 12.1%
85.8 ± 4.5%
82.7 ± 3.6%
Ins
9.4 ± 2.5%
4.3 ± 2.2%
4.0 ± 2.2%
6.4 ± 2.2%
10.7 ± 2.5%
14.7 ± 2.5%
jpROP+NG
5.2 ± 0.7%
4.0 ± 0.6%
3.7 ± 0.6%
2.4 ± 0.4%
1.8 ±0.4%
2.6 ± 0.4%
|AUTOCOAT
0%
5.1 ± 1.1%
5.4 ± 1.1%
0%
1.8 ± 0.9%
0%
SUM
110.7 ± 5.7
ppbC
182.1 ± 10.0
PPbC
204.9 ± 10.8
ppbC
259.2 ±31.9
PpbC
238.0 ± 12.4
ppbC
221.7 ± 9.7 |
ppbC J
TNMOC
225.2 ± 19.5
ppbC
340.1 ± 19.7
ppbC
370.8 ± 18.8
ppbC
374.9 ± 19.6
ppbC
373.7 ± 19.6
PPbC
327.6 ± 19.5 |
ppbC |
SUM/TNMOC
49.2 ± 5.0%
53.6 ± 4.3%
55.2 ± 4.0%
69.2 ± 9.3%
63.7 ± 4.7%
67.7 ±5.0% I
ROAD/GAS
12.6 ± 20.1
4.1 ± 2.3
5.2 ± 3.3
4.7 ± 2.4
10.3 ± 11.8
9.8 ± 10.4 I
Degrees of Freedom
13
12
12
13
12
13
R2
0.99
1.00
1.00
1.00
1.00
0.99
J Chi2
0.69
0.32
0.30
0.26
0.44
0.77

-------
PROBLEMS TO OVERCOME
Compare ambient data and emission inventory on
equal basis (e.g., species represented, definition of
total emissions)
Sources distinguished chemically by CMB vs
mechanistically by inventory
Representativeness of source profiles for CMB
calculations
Meet other CMB requirements

-------
Source Profiles and MOBILE5 Outputs
Reconcile ambient measurements with emissions
inventories using CMB and profiles.
•	MOBILE5 - distinguishes emissions by mechanism.
*
•	CMB/profiles - distinguishes emissions by composition.

-------
COMPARISONS OF CMB RESULTS WITH
EMISSION INVENTORY
Calculate % of total mass apportioned (sum of source
estimates) rather than percent of total NMOC
Exclude ethane from TNMOC in CMB calculations and
profile normalization
Compare CMB with inventory on a relative basis
Report minumum and maximum source estimates
from CMB calculations

-------
1990 ATLANTA EMISSION INVENTORY

Fulton Co.
13 Co.
Highway Mobile Sources
56.0%
39.8%
Area Sources
20.6%
19.4%
Point Sources
6.6%
3.9%
Biogenic Sources
16.8%
36.9%

-------
HIGHWAY MOBILE SOURCES
.a,	¦	m
•	Minimum = ROAD
•	Maximum = ROAD+GAS+HS

-------
POINT + AREA SOURCES
•	Minimum = NG + PROPANE + AUTOCOAT
•	Maximum = NG + PROPANE + AUTOCOAT + GAS + HS

-------
BIOGENICS - DEAL WITH THEM INDIRECTLY
Assume inventory % is correct; renormalize other
source estimates accordingly
Assume biogenic % = isoprene % (minimum
estimate)
Assume biogenic % = total unexplained NMOC
(maximum estimate)
Assume dq biogenics; renormalize inventory to
exclude biogenics

-------
Table 3. Comparison of emission inventory with relative source estimates derived from the site 2 ambient data (weekends excluded) using
CMB calculations.
Summary of CMB Results*	Summary of Emission Inventory
Assume biogenic % = % reported in inventory (16.8%):
MSN. MAX.
HIGHWAY MOBILE SOURCES: 61% 75%
POINT+AREA SOURCES: 8% 23%
Including Biogenics:
Assume biogenic min. % = isoorene % (1.9%):
MIN. MAX.
HIGHWAY MOBILE SOURCES: 71% 89%
POINT+AREA SOURCES: 9% 27%
HIGHWAY MOBILE SOURCES: 56.0%
POINT+AREA SOURCES: 27.2%
J Assume biogenic max. % - % unexolained (47.4%):
MIN. MAX.
HIGHWAY MOBILE SOURCES: 41 % SI %
POINT+AREA SOURCES: 5% 15%
Assume no biogenics:
MIN. MAX.
HIGHWAY MOBILE SOURCES: 73% 91%
POINT+AREA SOURCES: 9% 27%
Excluding biogenics:
HIGHWAY MOBILE SOURCES: 67.3%
POINT+AREA SOURCES: 32.7%
* Results presented as percent of total apportioned NMOC (sum of source estimates).

-------
100%-
90%-
80%-
70%-
CMB RESULTS FOR AVERAGE SAMPLES
60%-
¦k
mmm
<0
y.
O
H 50%-
q 40%-
(E
£ 30%-
20%-
10%-
0%-
ROAD+GAS
	UTIkth	r
PROP+NG
HS24
SOURCE CATEGORY
H
SZ2L
AUTOCOAT
MH1
FM4
] GT2
DK6
MK3


-------
CMB RESULTS FOR AVERAGE SAMPLES

-------
SUMMARY
Method causes source %'s to be interdependent
Representativeness of solvent-type source uncertain
No breakout of components of highway mobile source
Inventory and ambient data do not represent exactly
the same dates and times
Collinear sources in CMB
No uncertainty estimates for emission inventory

-------
TECHNICAL REPORT DATA
1
1. REPORT NO.
EPA/600/A-94/244
2.
3.RECIPI

4. TITLE AND SUBTITLE
Comparison of Atlanta Emission Inventory with
Ambient Data Using Chemical Mass Balance Receptor
Modeling
5.REPORT DATE
6.PERFORMING ORGANIZATION CODE
7. AUTHOR®
Teri L. Conner1, John F. Collins2, William A.
Lonneman1, Robert L. Seila1
8.PERFORMING ORGANIZATION REPORT
NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
'U.S. Environmental Protection Agency/AREAL
MD-47
Research Triangle Park, NC, 27711
'University of Southern California
Civil Engineering Dept., Env. Engineering Pgm.
3620 South Vermont Avenue
Los Angeles, CA, 90089-2531
10.PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
CR-818410 (Coop. Agr. with
use)
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency/AREAL
MD-47
Research Triangle Park, NC, 27711
13.TYPE OF REPORT AND PERIOD COVERED
Symposium Paper and
Presentation
14. SPONSORING AGENCY CODE
EPA/600/09
15. SUPPLEMENTARY NOTES
For presentation at EPA/A&WMA conference "The Emission Inventory: Applications and
Improvement" to be held Nov. 1-3 in Raleigh, NC.
16. ABSTRACT
A con$>rehensive hydrocarbon database was obtained at 6 sites in the Atlanta
metropolitan area during the summer of 1990. Samples were collected in stainless
steel canisters and analyzed for 54 hydrocarbon species plus total non-methane
organic confounds (TNMOC). Hie contributions of the major sources of TNMOC at each
of the 6 sites were estimated through a procedure called Chemical Mass Balance
(CMB) receptor modeling. Spatial variability of the source contributions is
discussed. Results of the CMB analysis for one of the sites are cooqparad with the
emission inventory for Atlanta using several different approaches. -"The inventory
hiahwav mobile source estimate tends to be smaller than the minimum ambient data-
derived highway mobile source estimate, and the inventory area plus point source
estimate tends to be laraer than the maximum ambient data-derived estimate for the
data set examined. However, these source estimates are interdependent to some
extent. Limitations of these comparisons are discussed.
17. KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
b.IDENTIFIERS/ OPEN ENDED
TERMS
c.COSATI



18. DISTRIBUTION STATEMENT
Release to Public

19. SECURITY CLASS (Hut Report)
Unclassified
21 .NO. OF PAGES

20. SECURITY CLASS (This Panel
Unclassified
22. PRICE

-------