EPA/600/A-95/00:
Receptor Methods for VOC Source Apportionment in Urban
Environments
Charles W. Lewis
Atmospheric Research and Exposure Assessment Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, North Carolina
U.S.A.
Introduction
Receptor modeling provides quantitative estimates of the impacts
of sources on ambient air, primarily from inter-relationships of
the ambient concentrations of individual air pollutant species
measured at a receptor site. In contrast to dispersion modeling,
receptor modeling employs only minimal meteorological and
emissions inventory information. Its ultimate use is to arrive
at minimum cost emissions control strategies that will be
effective in reducing the air pollutants of interest. The
development of receptor modeling methods began about 20 years
ago, to provide answers to questions about the quantitative
contributions made by potential sources to particulate air
pollution. Since that time particulate receptor modeling has
expanded from its initial focus on determining the sources of
measured ambient particulate mass concentration, to the sources
of particulate-related parameters such as visibility degradation
(Lewis et al., 1986), mutagenicity (Lewis et al., 1988) and
tumorigenicity (Lewtas et al., 1992). The success of these
methods for particulate problems has naturally led to their
consideration in the volatile organic compound (VOC) area. By
definition VOC excludes CH4 and the "inorganic" compounds CO and
C02.
There are two general reasons why determining the source
contributions of VOC is of interest in urban air pollution: (1)

-------
the toxicity of some VOC species, and (2) their contribution to
tropospheric ozone formation through photochemical reaction with
oxides of nitrogen (NOJ . Although both are of importance, the
work discussed in this article is motivated more by the second
reason than the first. The article will summarize current work
on the source apportionment of VOCs which is being sponsored by
the United States Environmental Protection Agency's Atmospheric
Research and Exposure Assessment Laboratory. The work has three
components: (1) chemical mass balance, (2) extracting source
profiles from ambient data, and (3) 14C determination of biogenic
VOC emissions. Each is illustrated by recent results derived
from ambient VOC data collected in Atlanta, Georgia, USA, an
ozone non-compliance area.
Chemical Mass Balance (CMB)
The CMB method has become the principal receptor modeling tool
for mathematically analyzing measured air pollutant species
concentrations for the purpose of determining their source
contributions. In this model the concentration cs of species i
measured at the receptor site is expressed as
Ci •= Z A^ Sj + ei	(1)
where Ay is the abundance of species i in the profile of source
j, and Sj is the mass concentration at the receptor site due to
source j. This series of equations (one for each fitting species
i) is simultaneously solved for each Sj by minimizing the reduced
chi square
X2 = (N, - NJ"1 I (ej/EJ2	(2)
where e; is a residual, N, is the number of fitting species, Ne is
the number of sources, and E? is an "effective variance" which
includes uncertainties in both A^ and Cj.

-------
The U.S. EPA distributes a software module (CMB7) which embodies
this approach and has been useful in standardizing CMB
applications. It is we11-documented (Watson et al., 1990) and
contains a number of useful diagnostics. The software was
developed for aerosol applications, but is usable for VOC with
only modest changes in the input files.
The selection of fitting species is an important part of a CMB
calculation. Generally, fitting species should satisfy two
selection criteria:
(1)	be "unreactive," i.e., atmospheric lifetimes longer
than eight hours;
(2)	be emitted only by the selected sources, i.e., its
measured ambient concentration can be satisfactorily
reproduced by those sources included in the
calculation.
Figure 1 shows a result of applying CMB7 to ambient VOC species
concentrations measured in Atlanta, Georgia, U.S.A., during the
1990 summer (Purdue, 1991). The measurements were obtained on
the campus of the Georgia Institute of Technology, located near
the urban center of Atlanta. The measurements were made on an
hourly basis with an automated gas chromatographic system. The
"weekday average" result shown is for an average over five
weekdays for which nearly every hourly period was successfully
measured (a total of 114 measurement periods).
Sixteen fitting species (of the 47 measured) were used in the CMB
procedure. In Figure 1 the "road" source (representing the sum
of vehicle tailpipe emissions and running losses) is seen to
account for about half of the total measured VOC (defined here as
the sum of all chromatographic peaks in the C, to Cl0 range). The
"gasoline" and "vapor" sources represent two distinguishable
kinds of gasoline evaporation, whole gasoline and headspace
gasoline vapor, respectively. It is noteworthy that the whole
gasoline contribution is 2-3 times greater than the headspace

-------
contribution (9.9 % vs. 3.7 %). The relative uncertainties for
the source contributions range from 9 - 14 * for the "road,"
"natural gas," "propane-rich," and "isoprene-rich" sources, 30 -
40 % for the evaporative gasoline sources, and 54 % for the
solvent source. The poor accuracy of the solvent source is
minimized by its small contribution. The unexplained portion in
Figure 1 is likely associated vith the cumulative effect of
multiple small sources, as veil as secondary products of
atmospheric photochemical transformation. Additional details and
results of this work may be found in Lewis et al. (1993).
Figure 1. CMB estimate of source contributions to total ambient
VOC in Atlanta during 1990 summer (5-day 24-h average).
Extracting Source Profiles from Ambient Data
The most difficult obstacle to applying the CMB method is the
frequent unavailability of source profiles which are both
accurate and representative of the source emissions in the
airshed under consideration. A new procedure has been developed
WEEKDAY AVERAGE
(7/26,8/3,8/9,8/14,8/21)
UNEXPLA	IAD (49.4%)

PROPANE-rich
GASOLINE (9.9%)
VAPOff(3T%) /
/ 1 \ ISOPRENE-rlch (2.2%)
SOLVENT (2.6%) \ PROPANE-rich (4.9%)
NATURAL GAS (2.9%)

-------
to extract profiles from the ambient data themselves. The
procedure combines two graphical/mathematical components, GRACE
(Graphical Ratio Analysis for Composition Estimates) and SAFER
(Source Apportionment by Factors vith Explicit Restrictions).
The basis for the GRACE/SAFER approach is illustrated in Figures
2 and 3. Figure 2 shows a scatterplot of simultaneously measured
concentrations of acetylene and ethylene, from the same dataset
discussed in the CKB section above. For summertime Atlanta there
is good reason to believe that acetylene should be mostly from
motor vehicle exhaust emissions (Henry et al, 1994)• The
approximately linear relationship between ethylene and acetylene
then suggests that most of the ethylene has the same source, and
that the ethylene/acetylene emissions ratio for the source is
nearly unity, as given by the slope of the best-fit line in
Figure 2.
40
35
30
25
20
15
10
*•••
5
0
0
10
15
20
30
5
25
35
40
Acetylene (ppbC)
Figure 2. Ethylene vs. acetylene (Atlanta).

-------
Figure 3 shows an analogous scatterplot for toluene and
acetylene, but with a very different character from that of
Figure 2. The principal feature of Figure 3 is the quite sharp
linear lover boundary to an otherwise rather diffuse
distribution. The interpretation of this is that acetylene,
assumed to be primarily from motor vehicle exhaust, is always
accompanied by toluene exhaust emissions at a fixed
toluene/acetylene ratio of about 1.8, as given by the slope of
the lover boundary. Data points above the lower boundary reflect
the additional contribution of other independent sources of
toluene.
160
140
120-
40
•• • -«
-------
which generates the final source profiles.
A validation of the GRACE/SAFER methodology has been carried out
by comparing the ambient-derived source profiles with profiles
directly measured in Atlanta at the same time the ambient data
were collected (Conner et al., 1995). For the three vehicle-
related profiles — vehicle exhaust, whole gasoline evaporation
and headspace gasoline evaporation — for which the comparison
could be made, the agreement was excellent. Details may be found
in Henry et al. (1994).
14C Determination of Biogenic VOC Emissions
A fundamental uncertainty in understanding the origin of urban
tropospheric ozone is the lack of reliable knowledge on the size
of the biogenic contribution to precursor VOC. Biogenic VOC is
known to be very photochemical ly active and is basically
uncontrollable, so that if it constitutes an appreciable fraction
of total VOC emissions, control of the remaining anthropogenic
VOC may be ineffective (relative to NO, control) in ozone
abatement.
In principle the CMB approach described above is applicable to
all source types. In practice however it is not well suited to
the quantification of biogenic sources, because of the lack of
non-reactive species that are suitable tracers of biogenic
emissions. For that reason an independent effort, based on 14C
methodology, is being developed. The biogenic fraction fk of an
ambient VOC sample is given by
- f. / fco2	(3)
where f, and f^ are the MC/,2C ratios for the VOC sample and
atmospheric C02, respectively. The use of l4C for VOC is an
extension of a methodology that has been successfully used to
determine the fraction of ambient aerosol resulting from
-------
wintertime voodburning (Levis et al., 1988).
The most critical problems in applying 14C methodology to ambient
VOC samples are (1) collecting sufficient sample quantity to
allow the 14C measurement to be performed with adequate precision,
and (2) insuring that the VOC sample contains minimal, quantified
amounts of carbonaceous contaminants that would otherwise render
the MC measurements misleading or uninterpretable.
Problem (1) has been addressed by the use of large-volume (32-L)
SUMMA® stainless steel canisters to collect whole air samples.
Pressurizing the samples to 30 psig during sample collection
results in 0.1 m3 of air per canister. At the current state of
14C measurement technology the 14C fraction can be routinely
measured to a few percent precision on samples containing as
little as 50 jxg carbon. This amount is achievable for most
airsheds of interest by compositing several such canister
samples.
Problem (2) was initially addressed by a purely cryogenic method
(Klouda et al., 1993) to separate the VOC fraction from those
whole air constituents which would otherwise be interferents in
the >4C measurement (CH4, CO and C02). While this was effective
for CH4 and CO, the separation of C02 from the VOC was not as
complete as desired. The solution was to introduce an additional
step of passing the whole air sample through a bed of LiOH prior
to cryogenic elimination of CH4 and CO. A 104 reduction factor
for C02 has been achieved, with only modest loss (10-15 % by
mass) of VOC.
The full analytical measurement scheme has been demonstrated on
canister samples collected during late summer 1992 at the same
Atlanta site indicated above. Measurements of the VOC 14C content
have been performed on two composited samples, an "AM" composite
of mid-morning samples, and a "PM" composite of late evening
samples. The AM and PM samples showed biogenic fractions of 12
and 14 %, respectively (but with relative uncertainties of about
-------
50 % that are mostly associated with a large and incompletely
characterized processing blank). The results are plausible when
compared with the VOC species composition of the samples.
Conclusion
Three projects that are part of a coordinated VOC source
apportionment research program at the U.S. EPA have been
described. Work is continuing in all three areas. The general
goal of emissions inventory validation is being pursued with the
CMB approach on additional datasets (Conner et al., 1994). The
GRACE/SAFER methodology for source profile extraction is being
refined to include estimates of profile uncertainty. Anticipated
improvements in the 14c methodology for biogenic estimation
include equipment modifications to reduce the blank, and high
pressure canister sample collection to increase the amount of
sample material.
Acknowledgements
The work summarized in this article represents the collective
efforts of a number of individuals and organizations: Robert K.
Stevens, Teri L. Conner, William A. Lonneman, Robert L. Seila
(U.S. Environmental Protection Agency); Ronald C. Henry, John F.
Collins (University of Southern California); George A. Klouda
(National Institute of Standards and Technology); Reinhold A.
Rasmussen (Oregon Graduate Institute); William D. Ellenson and
David C. Stiles (ManTech Environmental Technology, Inc.).
Disclaimer
The information in this document has been funded wholly or in
part by the United States Environmental Protection Agency under
Cooperative Agreement Mo. 818410 to the University of Southern
California, Interagency Agreement No. DW13935098 to the National
-------
Institute of Standards and Technology, and Contract No. 68-DO-
0106 to ManTech Environmental Technology, Inc. It has been
subjected to Agency review and approved for publication. Mention
of trade names or commercial products does not constitute
endorsement or recommendation for use.
References
Conner TL, Lonneman WA, Sella RL (1995) Transportation-related
volatile hydrocarbon source profiles measured in Atlanta.
J Air Waste Manage Assoc (to be published)
Conner TL, Collins JF, Lonneman WA, Sella RL (1994) Comparison of
Atlanta emission inventory with ambient data using chemical
mass balance receptor modeling. Proceedings of the AiWMA/EPA
Symposium, The emission inventory: application and improvement
Henry RC, Lewis CW, Collins JF (1994) Vehicle-related hydrocarbon
source compositions from ambient data: the GRACE/SAFER method.
Environ Sci and Technol 28:823-832
Klouda GA, Norris JE, Currie LA, Rhoderick GC, Sams RL, Dorko WD,
Lewis CW, Lonneman WA, Sella RL, Stevens RK (1993) A method
for separating volatile organic carbon from 0.1 ms of air to
Identify sources of ozone precursors via isotope (14C)
measurements. Proceedings of the 1993 A&WMA/EPA symposium,
Measurement of toxic and related air pollutants 585-603
Lewis CW, Conner TL, Stevens RK, Collins JF, Henry RC (1993)
Receptor modeling of volatile hydrocarbons measured in the
1990 Atlanta ozone precursor study. Paper No. 93-TP-58.04, in
Proceedings of the 86th annual meeting of A&WMA
Lewis CW, laumgardner RE, Stevens RK, Claxton LD, Lewtas J (1988)
Contribution of woodsmoke and motor vehicle emissions to
ambient aerosol mutagenicity. Environ Sci Technol 22:968-971
Lewis CW, Baumgardner RE, Stevens RK, Russwurm GM (1986) Receptor
modeling study of Denver winter haze. Environ Sci Technol
20:1126-1136
Lewtas J, Lewis C, Zweidinger R, Stevens R, Cupitt L (1992)
Sources of genotoxicity and cancer risk in ambient air.
Pharmacogenetics 2:288-296.
Purdue LJ (1991) Summer 1990 Atlanta ozone precursor study.
Paper No. 91-68.8, in Proceedings of the 84th annual meeting
of A&WMA
Watson JG, Robinson NF, Chow JC et al (1990) CMB7 user's manual.
Report No. EPA-450/4-90-004, US Environmental Protection
Agency, Research Triangle Park, NC, USA
-------
4.
TECHNICAL REPORT DATA
1. REPORT NO.
EPA/600/A-95/007
2.

4. TITLE AND SUBTITLE
Receptor Methods for VOC Source Apportionment in
Urban Environments
5.REPORT DATE
6.PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
Charles W. Lewis
8.PERFORMING ORGANIZATION REPORT
NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Atmospheric Research and Exposure Assessment Lab.
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
10.PROG RAM ELEMENT NO.
11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME AND ADDRESS
Atmospheric Research and Exposure Assessment Lab.
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
13 .TYPE OF REPORT AND PERIOD COVERED
Symposium Paper
14. SPONSORING AGENCY CODE
EPA/600/09
IS. SUPPLEMENTARY NOTES
To be submitted for publication in the Proceedings of the NATO workshop,
"Monitoring and Control Strategies for Urban Air Pollution," Erice, Sicily, October
10-14, 1994.
16. ABSTRACT ?
Receptor methods provide estimates of the impacts of sources on ambient air,
primarily from the inter-relationships of measured ambient concentrations of
individual air pollutant species, using only minimal meteorological and emissions
inventory information. The article summarizes recent resultsVfrora .applying
receptor methods (principally Chemical Mass Balance) to antoientj^VOCj data from
Atlanta, Georgia, USA, an ozone non-compliance area. A novel feature of the
analysis is the extraction of source profiles from the ambient data themselves.
Recent results on applying radiocarbon measurements to determine the biogenic
fraction of ambient VOC are also being discussed.
17. KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
b.IDENTIFIERSI OPEN ENDED
TERMS
c.COS ATI



18. DISTRIBUTION STATEMENT
RELEASE TO PUBLIC

19. SECURITY CLASS (This Report)
UNCLASSIFIED
21.NO. OF PAGES

20. SECURITY CLASS (This Page)
UNCLASSIFIED
22. PRICE
-------