PMF and CMB Source Apportionments of PM2.5: A comparison
Wei Liu1,1, Yuhang Wang1, Sangil Lee2, Armistead Russell2 and Eric S. Edgerton3
1. Georgia Institute of Technology, School of Earth and Atmospheric Sciences, Atlanta, GA 30332
2. Georgia Institute of Technology, Civil and Environmental Engineering, Atlanta, GA 30332
3. Atmospheric Research & Analysis, Inc., Cary, North Carolina 27513
ABSTRACT: Two commonly used receptor models, Positive Matrix Factorization (PMF) and
Chemical Mass Balance (CMB), are applied to 3-year SEARCH measurements at two urban sites
(Atlanta, GA and Birmingham, AL) and two rural sites (Yorkville, GA and Centreville, AL). The
measurements include fine particle (PM2.5) concentrations of inorganic ionic species, trace
metals, elemental and organic carbon from January 2000 through December 2002. Source
apportionment results using the two methods are analyzed and compared. Conditional probability
functions (CPFs) of wind directions are calculated for source factors identified by PMF and
CMB in order to investigate the impact of point sources and elucidate differences between the
two methods. Source contributions and profiles of secondary sulfate and nitrate factors derived
by the two methods agree well, as do corresponding CPF distributions. Secondary OC factor
calculated from PMF correlates well with that found using the EC tracer approach in the CMB
analysis, while the source contribution of the former is lower. Derived characteristics of other
(mostly primary) source factors can be quite different between the two methods. CMB source
profiles for motor vehicles are quite different from the PMF factor, although the corresponding
source contributions and CPF results show generally good correlations. The coal combustion
factors derived by the two methods are poorly correlated and have very different CPF
1 Address correspondence to Wei Liu, Georgia Institute of Technology, School of Earth and Atmospheric Sciences, Atlanta, GA 30332-0340, Tel: (404) 894-1624, Fax: (404) 894-5638, E-mail: [email protected].
1
distributions. The wood smoke factor also shows large differences, especially at the two urban
sites. PMF resolves two dust factors, one for long-range transport and the other for local/industry
sources, while the CMB resolves only one dust source since long-range transport was not
considered. PMF was able to resolve an industry source with high zinc concentrations while
CMB included a cement source, which was unresolved by PMF. There are several reasons that
may explain the disagreements between the results using the two methods: (1) the lack of proper
markers for additional sources; (2) errors in estimation of source profiles in CMB due to a lack
of local-specific profiles or the incompleteness of profiles; (3) the uncertainties in the estimation
of secondary organic carbon using EC tracer method; and (4) the tendency of PMF to mix
primary and secondary aerosols due in part to atmospheric mixings.
1. Introduction Particulate matter has been linked with cardiovascular and respiratory problems,
including problems leading to premature mortality (Peters et al., 2001; Peel et al., 2002; Pope et
al., 2002; Metzger et al., 2004). PM2.5 is also the main anthropogenic cause of pollution-related
visibility impairment (Milne et al., 1982), and can contain constituents leading to acid deposition
(NADP, 1993). PM2.5 is emitted directly from the sources or forms in the atmosphere through
reactions of precursor gases, such as nitrogen oxides (NOx), sulfur oxides (SOx), and certain
organic vapors. Cost effective air quality management planning relies on identifying how
different sources impact pollutant concentrations.
Receptor models, which attribute observed concentrations to sources through statistical
and/or meteorological interpretation of data often yield useful insights on the sources of aerosols
(Hopke, 2003). Such models are generally based on the observed mass concentrations and
appropriate use of mass balance. Various methods have been developed for this purpose.
2
If the number and nature of the sources in the region are known, the mass contribution of
each source to each sample can be estimated using the Chemical Mass Balance (CMB) model
(Chow et al., 1991; Schauer et al., 1996). When source information is largely unknown, factor
analysis is generally applied. Three factor analysis approaches that have been often used in
recent years are UNMIX, Multilinear Engine (ME) and Positive Matrix Factorization (PMF). We
use the PMF approach in this work. It has been applied to a number of aerosol data sets (e.g.,
Polissar et al., 1998 , Lee et al., 1999; Liu et al., 2003). Similarities between PMF and CMB
include using least-squares fitting to minimize the differences between measured and estimated
concentrations. CMB makes use of an effective variance least-squares fitting by using source
composition data as independent variables and ambient air quality data as dependent variables.
PMF uses an alternating least-square method and only needs ambient measurement data without
a priori assumptions of the factor profiles. CMB incorporates uncertainties from both ambient
measurement and source profile data while PMF estimates the profile uncertainties based on
those from the ambient measurement data.
In this work, both source apportionment methods were applied to four data sets, two
urban (Atlanta, GA and Birmingham, AL) and two rural (Yorkville, GA and Centreville, AL).
Data used include 3-year measurements from January 2000 to December 2002. Source profiles
(as used by CMB) and factors (as developed by PMF), as well as source contributions from the
two analyses were compared in order to investigate the usefulness and limitations of each
method.
2. Method
2.1 Measurement data
3
PM2.5 composition data analyzed in this study consist of measurements taken at two
urban-rural pairs in Alabama (North Birmingham [BHM] and Centreville [CTR]), and Georgia
(Atlanta [JST] and Yorkville [YRK]). These sites are operated by the Southeastern Aerosol
Research and Characterization Study (SEARCH) (Hansen et al., 2003). Twenty-four hour
integrated PM2.5 samples were collected daily at the JST site. PM2.5 samples were collected every
third day at the other sites. Samples were collected using particulate composition monitors
(PCM, Atmospheric Research and Analysis, Inc., Durham, NC) that have three sampling lines
(air flow rate 16.7 l/min) with inlets 5 m above ground. More detailed descriptions can be found
elsewhere (e.g., Hanson et al., 2003; Liu et al., 2005a).
A total of 932 samples for the JST site, 336 samples for the BHM site, 347 samples for
the YRK site and 338 samples for the CTR site were obtained and analyzed, covering the time
period from January 2000 through December 2002. For each sample, concentrations of the
following 19 chemical species were usually available: SO42-, NO3
-, NH4+, EC, OC (OC was
calculated as OC1+OC2+OC3+OC4+OP and EC as EC1+EC2+EC3-OP), As, Ba, Br, Cu, Mn,
Pb, Se, Ti, Zn, Al, Si, K, Ca, and Fe, although there are occasional “missing data” (no reported
measurements) for one or more species. Total PM2.5 mass concentrations for each day, analytical
uncertainty and detection limit for each chemical species were also obtained.
2.2 PMF
PMF (Paatero and Tapper, 1994; Paatero, 1997) was used to analyze PM2.5 data at the
four sites. In this work, missing data were replaced by the geometric mean of corresponding
species and four times of geometric mean as the corresponding error estimates (Polissar et al.,
1998). Half of the detection limit was used for the values below the detection limit and 5/6 of the
detection limit was used for the corresponding error estimate (Polissar et al., 1998).
4
With the total PM2.5 mass concentration measured for each sample, multiple linear
regression (MLR) was performed to regress the mass concentration against the factor scores
obtained from PMF. Because of the uncertainties introduced by the measurement matrix, PMF
results always have a portion of unexplained variation. Mass concentrations excluding the
unexplained variation portion from G factors (factor contributions) were used to regress the
factor scores to obtain the quantitative factor contributions for each resolved factor.
2.3 CMB
EPA’s CMB 8.0 model, using the effective variance weighted least-squares fitting, was
applied to calculate source contributions to PM2.5 on a daily basis (Watson, 2001). All
performance diagnostic criteria were met for each calculation. For example, the ratios of the
calculated to measured concentrations of fitting species, R-square, Chi-square, percent of
attribute mass are 0.5 to 2.0, 0.8 to 1.0, 0 to 4.0, and 100±20% respectively, each time.
The same data treatments of missing and below detection limit data in PMF analysis were
used in the CMB analysis. Source profiles for the primary sources selected in the CMB analysis
included motor vehicle (Watson et al., 1998), wood burning (Watson et al., 1998), coal
combustion from power plants (Chow et al., 2004), Alabama road dust (Cooper et al., 1981), and
cement kilns (Chow et al., 2004). The composites of light duty gasoline and heavy duty diesel
vehicles are created by weighing emission rates and four different profiles using weights derived
from the emission inventories for Georgia and Alabama. The wood burning source profile is also
created by weighting emission rates of three wood burning source measurements, soft-wood
fireplace, hard-wood fireplace, and woodstove (Zielinska et al., 1998). Ammonium bisulfate
(NH4HSO4), ammonium sulfate ((NH4)2SO4), and ammonium nitrate (NH4NO3) were used as
inorganic secondary sources.
5
Primary organic carbon in the CMB analysis was calculated by using the EC tracer
approach (Turpin et al., 1991). First, the primary OC to elemental carbon (OC/EC) ratio was
estimated and then primary OC was calculated by multiplying OC/EC ratio to EC assuming that
all EC is primary. If the estimated primary OC was higher than measured OC, measured OC was
used as primary OC.
2.4 Conditional probability function
CPF (Ashbaugh et al., 1985; Kim et al., 2003) estimates the likelihood for a source factor
identified by CMB or PMF to originate from a given wind direction. Hourly wind data are used.
The CPF is defined as,
θ
θθ
∆
∆∆ =
nm
CPF (1)
where m∆θ is the number of occurrences of the source contribution exceeding a threshold
criterion from wind sector ∆θ, and n∆θ is the total number of data from the same wind sector. In
this study, 36 sectors were used (100 in each sector). The threshold criterion of the upper 25th
percentile in the source contribution was chosen to define the directionality of a source factor.
Sources are likely to originate from the directions that have high conditional probability values.
Calm wind (<1ms-1) periods were excluded from this analysis.
3. Results
The aforementioned seven source profiles were used in the CMB analysis to derive the
source contributions at the four sites. PMF was able to resolve eight and seven factors for the two
urban sites and the two rural sites, respectively. PMF factors common to all locations included:
(1) secondary sulfate dominated by high concentrations of sulfate and ammonium with a strong
seasonal variation peaking in summer; (2) nitrate and the associated ammonium with a seasonal
6
maximum in winter; (3) “coal combustion/other” factor with the presence of sulfate, EC, OC,
and Se; (4) soil, with high levels of Al, Ca, Fe, K, Si and Ti; (5) wood smoke with high
concentrations of EC, OC and K; and (6) an industry/dust factor with elevated levels of Ca, Fe,
K, Si and Ti. A motor vehicle factor with high concentrations of EC and OC and the presence of
some soil dust components was found at the urban sites, but not at the two rural sites. One
similar industry factor with high zinc concentrations was found in each site in the PMF analysis.
Secondary sulfate was a dominant source factor in urban and rural areas. CMB had two
source profiles to represent secondary sulfate (NH4HSO4 and (NH4)2SO4). PMF resolved factor
profiles are compatible with the CMB profiles at all four sites (Figure 1) except that OC and
small amounts of EC are associated with this factor in the PMF results. In PMF, the OC and EC
components reflect the effects of mixing of primary and secondary pollutants while CMB by
definition is not affected by mixing. The resolved PMF factor therefore does not represent a
single pure source. The OC association implies that secondary organic aerosol formation
coincides with the secondary sulfate formation while the small EC content likely reflects an
increase of sulfate and EC concentrations during stagnant conditions. In the PMF sulfate factor,
molar ratios of ammonium to sulfate were 2.3, 2.0, 2.1 and 1.6 for the JST, YRK, BHM, and
CTR site, respectively. The ratios suggest that sulfate is present primarily as ammonium sulfate
at these four receptor sites, although sulfate at CTR is probably not fully neutralized. In the CMB
results, NH4HSO4 had low source contributions at the JST and YRK sites (<10% of secondary
sulfate) and relatively high source contributions at the BHM (30%) and CTR (40%) sites.
Therefore, the two approaches agreed except at the BHM site. The resolved source contributions
from PMF and CMB model are well correlated (Figure 1); the corresponding CPF results are
7
also in agreement (Figure 2). Average source contributions from both models agree well,
although the PMF value is lower at JST (Table 2).
PMF resolved a nitrate factor, which corresponds to the CMB NH4NO3 profile. The
source contributions from the two methods are generally in good agreement (Figure 3). Sulfate
and some OC are mixed in the PMF profile, likely arising from concurrent oxidations of NO2,
SO2, and VOCs. The PMF and CMB source contributions are well correlated and CPF
distributions are similar (Figure 4). However, the PMF source contributions tend to be lower
except at the BHM site (Table 2).
The CMB source profiles and the PMF factor associated with wood burning are in
relatively good agreement, with high concentrations of OC, EC, and K. However, PMF factor
profiles are mixed with some sulfate at the two urban sites. For the two rural sites, source
contributions agree better between the models (Figure 5). Again, the non-locally measured
source profiles (the CMB wood burning source profiles were obtained in Colorado) may, in part,
contribute the differences between CMB and PMF. Furthermore, Liu et al. [2005b] found that
the PMF resolved wood burning factors at urban sites peaked in winter while those at the rural
sites peaked in spring. Winter residential burning appears to be a major contributor at urban sites
while agricultural burning in spring appears to dominate at rural sites. The PMF-CPF and CMB-
CPF distributions also reflect the urban-rural differences between the two models (Figure 6). At
the JST site, the PMF-CPF distribution shows the impact mainly coming from the south while
the CMB-CPF distribution shows the impact coming from the south, northeast and northwest.
For YRK, source locations are similar. Average source contributions were in good agreement at
JST and YRK, while they are very different for the BHM and CTR sites. This “source” has
larger contributions in rural areas than in urban areas (Table 2).
8
The PMF coal combustion factor has strong signals of sulfate, ammonium, EC, OC, and
Se at all sites (Figure 7). The CMB source profile has higher levels of elements associated with
dust. This difference may be due, in part, to a lack of locally appropriate source profile for use in
the CMB, or PMF is “losing” the other elements to a dust factor. Compared to CMB results,
PMF has higher source contributions at all four sites. This is due largely to the OC content mixed
in this factor. There is increasing evidence that highly acidic particulate matter can catalyze the
formation of secondary organic aerosol, and soluble organics may be absorbed in water-laden
aerosols (Jang et al., 2002). Thus, the increased OC associated with these particles could arise
either by increased condensation of urban OC onto the particles or it could represent acid
catalyzed conversion of VOCs including isoprene. The PMF-CPF distributions at YRK and JST
sites suggest that this factor has a common source for the two sites, northwest of JST and north
of Yorkville. For the CMB-CPF results, this source comes from different locations for the two
sites, south-north direction for JST and northwest and south for YRK (Figure 8).
A motor vehicle factor is resolved only at the two urban sites by PMF. Motor vehicle
contributions estimated by CMB at the two rural sites are much smaller (~1/10) than the urban
sites. We compare the PMF and CMB motor vehicle factors at the two urban sites. This factor
has high concentrations of EC and OC. However, the OC/EC ratios in the source profiles
assigned in CMB and the factors calculated by PMF are markedly different. The ratios are about
2 in PMF as compared to 0.5 in CMB (Figure 9). The higher OC concentrations in the PMF
factor could arise from condensation of secondary OC onto the particles. It is also possible that
the EC/OC ratio specified in the CMB profile is in error. Liu et al. [2005b] found that the PMF
resolved gasoline and diesel factor profiles agreed with those measured by Cao et al., [2005] but
differed from an earlier study by Watson et al., [1994]. Changes in fuel composition could be
9
part of the reason, though there is considerable variability in emission profile measurements as
well (e.g. Cocker et al., 2005). Source contributions from both models are well correlated; the
CPF results also agree well. However, PMF estimated higher source contributions for this factor
due to the calculated higher OC/EC ratio.
PMF resolved two dust factors. We refer to the first one as “dust” and the second one as
“industry/dust”. The dust factor is rich in Si, Al, K, Ca, and Fe, and is associated with some
secondary OC and sulfate. There is evidence that this factor comes from long range transport of
desert dust. The source contribution peaked in April 2001, July 2001 and July 2002. These are
likely intercontinental dust transport events. The April 2001 event is due to transport from Asia
(EPA, 2003), while the July episodes in 2001 and 2002 are probably transported from Saharan
deserts (Prospero, 2001). The Industry/dust factor has high concentrations of Si, K, Ca, and Fe
coupled with secondary OC and sulfate. The EC content in this factor may come from local
industrial sources. CMB includes only one dust profile for paved road dust. Compared to the
PMF dust factor, the CMB profile has higher concentrations of Si. A cement source is also
resolved by CMB at all of the four sites. From the time series comparison (Figures 11 and 12)
and the CPF distributions (Figures 13 and 14), the long-range transport dust resolved by PMF is
probably distributed to the CMB dust and cement sources.
Major OC sources are wood smoke, motor vehicle and secondary production at urban
sites, and wood smoke and secondary production at rural sites. The average OC concentrations
apportioned to each source are compared between the two models (Figure 15). At the two urban
sites, PMF apportioned more OC to the motor vehicle factor while CMB apportioned more OC
to the wood smoke source. At the two rural sites, PMF apportioned more OC to the wood smoke
factor. PMF SOC is lower than that found using the EC tracer approach used in CMB analysis.
10
PMF secondary OC was calculated from the sum of OC fractions mixed in the PMF sulfate, and
nitrate factors and unexplained variations. Yuan et al. (2005) previously applied PMF to estimate
secondary OC production in Hong Kong and found that PMF estimated secondary OC is lower
than obtained using the EC tracer method. The large uncertainty of the EC tracer method arises
from the assumption that a single OC/EC ratio can represent a mixture of primary sources
varying in time and space, while secondary OC mixed into the primary source factors in the PMF
analysis may lead to underestimation. There is also evidence that meat cooking and natural gas
contribute significant portion of OC in the Southeast (Zheng et al., 2002, Jaemeen et al., 2005).
However, both PMF and CMB can not resolve such sources because of the lack of additional
markers and source profiles. These sources must have been distributed among the sources
resolved above to an unknown degree. This may also cause some differences between the two
models.
4. Conclusions
PMF and CMB methods were applied to 3-year measurements at four monitoring sites in
GA and AL to identify major source factors and contributions to PM2.5. Primary organic carbon
concentrations calculated by the EC tracer approach are used in the CMB analysis. For
comparison purposes, corresponding CPF values were calculated using source contributions
estimated by PMF and CMB coupled with wind direction measurements at these sites. There is
relatively good agreement between the PMF and CMB-derived secondary aerosol source
contributions and the resulting CPFs. Both approaches found that secondary sulfate and
secondary nitrate account for the majority of measured PM2.5 mass at the Southeast, with similar
day to day variations. Secondary OC calculated from PMF is well correlated with that from
CMB using the EC tracer approach, although the values are typically lower than the latter.
11
Differences between PMF computed primary source factor contributions and CMB
source profiles are much larger than secondary source factors, resulting in the differences in
calculated source attributions. The OC/EC ratios in the motor vehicle profiles assigned in CMB
and calculated by PMF are markedly different, 2 in PMF as compared to 0.5 in CMB. As a
result, higher source contributions are calculated in PMF. PMF and CMB wood burning profiles
are in relatively good agreement, with high concentrations of OC, EC, and K. The PMF coal
combustion factor has strong signals of sulfate, ammonium, EC, OC, and Se at all sites. The
CMB source profile has higher levels of elements associated with dust. PMF was able to resolve
two dust factors, one due to long-range transport from Sahara and Asia and the other
representing local sources. CMB resolved only one dust source factor since long-range transport
is not considered in the CMB source profiles. PMF resolved an industry factor with high zinc
concentrations while CMB resolved a cement source.
Several factors may have contributed to the different apportionment resulting using PMF
and CMB. First, there are no appropriate markers to apportion the emissions from meat cooking
and natural gas. Secondly, the source profiles in CMB may be inaccurate due to (a) a lack of
locally available source profiles (the vegetative burning and coal combustion source profiles
used in CMB were based on measurements in Colorado and Texas, respectively), (b) the
uncertainties in the mobile source profiles, and (c) a lack of long range transport dust profile.
Thirdly, there appears to be a high bias in the secondary organic carbon contributions estimated
using the EC tracer method in the CMB analysis compared to the PMF results. Lastly, PMF has a
tendency to mix primary and secondary aerosols due in part to atmospheric mixing. In addition,
PMF and CMB results are affected by measurement uncertainties and missing data. Neither PMF
nor CMB accounts for the seasonal variations of source profiles.
12
Acknowledgement. This study was supported by the Southern Company and US EPA under
grant RD-83215901.
References: Baek, J., Park, S. K., Hu, Y., Russell, A.G., Source apportionment of fine organic aerosol using
cmaq tracers. Proceeding of 2005 Models-3 conference, Research Triangle, NC. Chow, J. C., Watson, J.G., Kuhns, H., Etyemezian, V., Lowenthal, D.H., Crow, D., Kohl, S.D.,
Engelbrecht, J.P., Green, M.C. (2004). Source profiles for industrial, mobile, and area sources in the Big Bend Regional Aerosol Visibility and Observational study. Chemosphere 54(2): 185-208.
Chow, J. C., Watson, J.G., Richards, L.W., Hasse, D.L., McDade, C., Dietrich, D.L., Moon D. and Sloane, C. (1991). The 1989-90 Phoenix PM10 study. Volume II: Source apportionment.
Cooper, J. A., et al. (1981). Determination of source contributions to fine and coarse suspended particulate levels in Petersville, Alabama. Report to Tennessee Valley Authority by NEA, Inc.
Hansen, D.A., Edgerton, E.S., Hartsell, B. E., Jansen, J. J., Kandasamy, N., Hidy, G. M., Blanchard, C. L., 2003. The Southeastern Aerosol Research and Characterization Study: Part 1—Overview. Journal of the Air and Waste Management Association 53, 1460–1471.
Hopke, P. K. (2003). Recent developments in receptor modeling. Journal of Chemometrics 17: 255-265.
Jang, M., Czoschke, N.M., Lee, S., Kamens, R.M. (2002). Heterogeneous Atmospheric Aerosol Production by Acid-Catalyzed Particle-Phase Reactions. 298: 814-817
Lee, E., Chan C.K. and Paatero P. (1999). Application of Positive Matrix Factorization in source apportionment of particulate pollutants in Hong Kong. Atmospheric Environment. 33: 3201-3212.
Liu, W., Hopke, P. K., Han, Y., Yi, S.M., Holsen, T. M., Cybart, S. and Milligan, M. (2003). Application of receptor modeling to atmospheric constituents at Potsdam and Stockton, NY. Atmospheric Environment 37: 4997–5007.
Metzger, K. B., Tolbert, P. E., Klein, M., Peel, J. L., Flanders, W. D., Todd, K., Mulholland, J. A., Ryan, P. B., Frumkin, H. (2004). Ambient air pollution and cardiovascular emergency department visits. Epidemiology 15(1): 46-56.
Milne, J. W., Roberts, D.B., Walk, S.J., William, D.J. (1982). Source of Sydney brown haze. The urban atmosphere-Sydney, A case study.
NADP (1993) National Acid Deposition Program. Annual Data Summary. Precipitation Chemistry in the United States. 1992. National Research Ecology Laboratory, Colorado State University, Fort Collins, Colorado, pp. 480. Peel, J., Tolbert, P., Klein, M., Metzger, K., Flanders, W. D., Todd, K., Mulholland, J., Ryan, P.
B., Frumkin, H. (2002). Ambient air pollution and respiratory emergency department
13
visits in Atlanta, August 1998-August 2000 (ARIES/SOPHIA). Epidemiology 13(4): S124-S124.
Peters, A., Dockery, D.W., Muller, J.E., Mittleman, M.A. (2001). Increased particulate air pollution and the triggering of myocardial infarction. Circulation. 103: 2810-2815.
Polissar, A. V., Hopke, P.K., Malm, W.C., Sisler, J.F. (1998). Atmospheric Aerosol over Alaska: 2. Elemental Composition and Sources. Journal of Geophysical Research 103: 19,045-19,057.
Prospero, J.M., African Dust in America, Geotimes, 2001, pp.24-27, November. Pope, C. A. I., Burnett, R.T., Thun, M.J., Calle, E.E., Krewski, D., Ito, K., Thurston, G.D.
(2002). Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. Journal of the American Medical Association 287: 1123-1141.
Ramadan, Z., Song, X.-H., Hopke, P.K. (2000). Identification of sources of Phoenix aerosol by Positive Matrix Factorization. Journal of the Air and Waste Management Association. 50: 1308-1320.
Schauer, J. J., G.R. Cass (2000). Source apportionment of wintertime gas-phase and particle-phase air pollutants using organic compounds as tracers. Environmental Science and Technology 34: 1821-1832.
Schauer, J. J., Kleeman, M.J., Cass, G.R., Simoneit, B.R.T. (2001). Measurement of emissions from air pollution sources. 3. C-1-C-29 organic compounds from fireplace combustion of wood. Environmental Science & Technology 35(9): 1716-1728.
Schauer, J. J., W.F. Rogge, L.M. Hildemann, M.A. Mazurek, G.R. Cass, B.R.T. Simoneit (1996). Source apportionment of airborne particulate matter using organic compounds as tracers. Atmospheric Environment 30: 3837-3855.
Turpin, B. J., J.J. Huntzicker, S.M. Larson, G.R. Cass (1991). Los Angeles summer midday particulate carbon: Primary and secondary aerosol. Environmental Science and Technology 25: 1788-1793.
Watson, J.G, E.M. Fujita, J.C. Chow, B. Zielinska, L.W. Richards, W. Neff, D. Dietrich (1998). Northern Front Range Air Quality Study Final Report.
Watson, J.G.; Robinson, N.F.; Chow, J.C.; Fujita, E.M.; Lowenthal, D.H. CMB8 User’s Mannual Draft; U.S. Environmental Protection Agency: Washington, DC, 2001
Yuan, Z. B.; Yu, J. Z.; Lau, A. K. H. ; Louie, P. K. K.; and Fung, J. C. H. (2005). Application of positive matrix factorization in estimating aerosol secondary organic carbon in HongKong and insights into the formation mechanisms. Atmospheric Chemistry and Physics Discussions 5299-5324.
Zielinska, B.; McDonald, J.; Hayes, T.; Chow, J.C.; Fujita, E.M.; Watson, J.G. (1998) Northern Front Range Air Quality Study Final Report Volume B: Source Measurements. Desert Research Institute.
Zheng, M., Cass, G.R., Schauer, J.J., Edgerton, E.S. (2002) Source Apportionment of PM2.5 in the Southeastern United States using Solvent-Extractable Organic Compounds as Tracers. Environment. Science and Technology. 36, 2361-2371.
14
Table 1. PMF Factor- CMB source contribution correlation coefficients (R2) at the four sites
JST BHM YRK CTR Sulfate 0.99 0.93 0.97 0.97 Nitrate 1 0.91 0.99 0.98 Coal 0.02 0.04 0.01 0.0003 Wood smoke 0.27 0.35 0.73 0.72 Motor vehicle 0.88 0.89 N/A N/A Dust 0.15 0.2 0.42 0.5 SEOC 0.64 0.45 0.49 0.37 CMB cement vs. PMF industry / dust 0.12 0.42 0.17 0.19
15
Table 2. The comparison of average factor contributions (%) to PM2.5 mass concentrations
between CMB and PMF at the four sites
PMF CMB
JST YRK BHM CTR JST YRK BHM CTR
(NH4)2SO4 39.4 41.6 24.5 26.5
Sulfate 32.3 47.0 31.0 37.4 NH4HSO4 2.0 4.6 9.0 16.1
Nitrate 5.3 6.8 8.4 2.1 NH4NO3 7.3 8.6 6.9 3.3
Motor vehicle 19.0 16.4 Motor vehicle 9.8 2.0 12.2 1.2
Wood smoke 12.8 18.5 8.7 28.3 Wood smoke 13.6 15.3 14.2 18.5
Coal 5.9 5.5 8.1 4.6 Coal 2.4 1.2 5.5 1.2
Dust 2.7 2.4 3.1 2.0 Dust 3.8 2.1 5.7 2.4
Industry / dust 8.0 6.0 7.0 9.6 Cement 0.3 0.2 1.4 0.5
Industry factor Zn 5.2 2.3 2.4 2.3
SEOC 5.2 4.9 6.9 4.2 SEOC 11.2 8.9 9.3 8.4
undetermined 3.6 6.6 8.0 9.5 undetermined 10.2 15.5 11.3 21.9
16
Figure captions
Figure 1. Comparison of source profiles and contributions of secondary sulfate resolved from
PMF and CMB at the four sites. The top panel shows the source profiles and the bottom panel
shows the correlations the source contributions resolved at the four sites.
Figure 2. Comparison of sulfate CPF distributions between PMF and CMB results. The top panel
shows the CPF distributions for JST site and the bottom panel shows the CPF distributions for
YRK site.
Figure 3. Same as Figure 1 but for secondary nitrate.
Figure 4. Same as Figure 2 but for secondary nitrate.
Figure 5. Same as Figure 1 but for wood smoke.
Figure 6. Same as Figure 2 but for wood smoke.
Figure 7. Same as Figure 1 but for coal combustion.
Figure 8. Same as Figure 2 but for coal combustion.
Figure 9. Comparison of source profiles and contributions of motor vehicle from PMF and CMB
at the JST and BHM sites. The top panels shows the source profiles, the middle panel shows the
correlations of the source contributions, and the bottom panel shows the CPF distributions at
JST.
Figure 10. Comparison of source profiles of dust and “Industry /dust” from PMF and dust and
cement from CMB at the four sites.
Figure 11. Comparison of source contributions of dust and “industry /dust” from PMF and dust
and cement from the CMB at JST and BHM sites.
Figure 12. Comparison of source contributions of dust and “industry /dust” from PMF and dust
and cement from CMB at the YRK and CTR sites.
17
Figure 13. Comparison of the dust factor CPF distributions between PMF and CMB. The top
panel shows the CPF distributions for JST and the bottom panel shows the CPF distributions for
YRK.
Figure 14. Comparison of industry/dust factor PMF-CPF and cement PMF-CPF distributions.
The top panel shows the CPF distributions for JST and the bottom panel shows the CPF
distributions for YRK.
Figure 15. Comparison of PMF and CMB averaged OC source contributions for the four sites.
18
SO4
NO3
NH4
EC OC As Ba Br Cu Mn
Pb Se Ti Zn Al Si K Ca Fe
0.00.20.40.60.81.0
(NH4)2SO4NH4HSO4
0.00.20.40.60.81.0
0.00.20.40.60.81.0
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
Con
cent
ratio
ns (礸
/礸)
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
JST
BHM
CTR
YRK
CMB (礸 /m3)
0 5 10 15 20 25
PMF
(礸/m
3 )
0
5
10
15
20
25
CMB (礸 /m3)
0 5 10 15 20 25
PMF
(礸/m
3 )
0
5
10
15
20
25
CMB (礸 /m3)
0 5 10 15 20 25
PMF
(礸/m
3 )
0
5
10
15
20
25
CMB (礸 /m3)
0 5 10 15 20 25
PMF
(礸/m
3 )
0
5
10
15
20
25
JST BHM
YRK CTR
Figure 1
19
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
JST PMF JST CMB
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
YRK PMF YRK CMB
Figure 2
20
SO4
NO3
NH4
EC OC As Ba Br Cu Mn
Pb Se Ti Zn Al Si K Ca Fe
0.00.20.40.60.81.0
0.00.20.40.60.81.0
0.00.20.40.60.81.0
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
Con
cent
ratio
ns (礸
/礸)
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
JST
BHM
CTR
YRK
CMB (礸 /m3)
0 2 4 6 8
PMF
(礸/m
3 )
0
2
4
6
8
CMB (礸 /m3)
0 2 4 6 8
PMF
(礸/m
3 )
0
2
4
6
8
CMB (礸 /m3)
0 1 2 3 4 5 6
PMF
(礸/m
3 )
0
1
2
3
4
5
6
CMB (礸 /m3)
0 1 2 3 4
PMF
(礸/m
3 )
0
1
2
3
4
JST BHM
YRK CTR
CMB source profile (NH4NO3)
Figure 3
21
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
JST PMF JST CMB
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
YRK PMF YRK CMB
Figure 4
22
SO4
NO3
NH4
EC OC As Ba Br Cu Mn
Pb Se Ti Zn Al Si K Ca Fe
0.00.20.40.60.81.0
0.00.20.40.60.81.0
0.00.20.40.60.81.0
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
Con
cent
ratio
ns (礸
/礸)
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
JST
BHM
CTR
YRK
CMB (礸 /m3)
0 2 4 6 8 10 12 14
PMF
(礸/m
3 )
0
2
4
6
8
10
12
14
CMB (礸 /m3)
0 2 4 6 8 10 12 14
PMF
(礸/m
3 )
0
2
4
6
8
10
12
14
CMB (礸 /m3)
0 2 4 6 8 10
PMF
(礸/m
3 )
0
2
4
6
8
10
CMB (礸 /m3)
0 2 4 6 8 10 12 14
PMF
(礸/m
3 )
0
2
4
6
8
10
12
14
JST BHM
YRK CTR
CMB source profile (Wood smoke)
Figure 5
23
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
JST PMF JST CMB
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
YRK PMF YRK CMB
Figure 6
24
SO4
NO3
NH4
EC OC As Ba Br Cu Mn
Pb Se Ti Zn Al Si K Ca Fe
0.001
0.01
0.1
1
0.001
0.01
0.1
1
0.001
0.01
0.1
1
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.001
0.01
0.1
1
Con
cent
ratio
ns (礸
/礸)
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.001
0.01
0.1
1
JST
BHM
CTR
YRK
CMB (礸 /m3)
0 2 4 6 8
PMF
(礸/m
3 )
0
2
4
6
8
CMB (礸 /m3)
0 2 4 6 8
PMF
(礸/m
3 )
0
2
4
6
8
CMB (礸 /m3)
0 1 2 3 4 5 6
PMF
(礸/m
3 )
0
1
2
3
4
5
6
CMB (礸 /m3)
0 1 2 3 4
PMF
(礸/m
3 )
0
1
2
3
4
JST BHM
YRK CTR
CMB source profile (Coal)
Figure 7
25
0.00
0.15
0.30
0.00
0.15
0.30
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
JST PMF JST CMB
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
YRK PMF YRK CMB
Figure 8
26
SO4
NO3
NH4
EC OC As Ba Br Cu Mn
Pb Se Ti Zn Al Si K Ca Fe
0 .00 .20 .40 .60 .81 .0
0 .00 .20 .40 .60 .81 .0
SO4NO3
NH4 EC OC As Ba Br Cu M n Pb Se Ti Zn Al Si K Ca Fe0 .00 .20 .40 .60 .81 .0
Con
cent
ratio
ns (礸
/礸)
JS T
B H M
C M B (礸 /m 3)
0 5 1 0 1 5 2 0
PMF
(礸/m
3 )
0
5
1 0
1 5
2 0
C M B (礸 /m 3)
0 2 4 6 8 1 0 1 2 1 4
PMF
(礸/m
3 )
0
2
4
6
8
1 0
1 2
1 4JS T B H M
C M B so u rce p ro file (M o to r V eh ic le )
Figure 9
27
SO4
NO3
NH4
EC OC As Ba Br Cu Mn
Pb Se Ti Zn Al Si K Ca Fe
0.00.20.40.60.81.0
0.00.20.40.60.81.0
0.00.20.40.60.81.0
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
Con
cent
ratio
ns (礸
/礸)
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
PMF JST (dust)
PMF BHM (dust)
PMF CTR (dust)
PMF YRK (dust)
CMB source profile (Dust)
SO4
NO3
NH4
EC OC As Ba Br Cu Mn
Pb Se Ti Zn Al Si K Ca Fe
0.00.20.40.60.81.0
0.00.20.40.60.81.0
0.00.20.40.60.81.0
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
Con
cent
ratio
ns (礸
/礸)
SO4NO3
NH4 EC OC As Ba Br Cu Mn Pb Se Ti Zn Al Si K Ca Fe0.00.20.40.60.81.0
PMF JST (Industry / dust)
PMF BHM (Industry / dust)
PMF CTR (Industry / dust)
PMF YRK (Industry / dust)
CMB source profile (Cement)
Figure 10
28
1/00 4/00 7/00 10/00 1/01 4/01 7/01 10/01 1/02 4/02 7/02 10/02
0
3
6CMB JST (dust)
0.00.30.60.91.2 CMB JST (Cement)
0369
1215
PMF JST (dust)
0.03.06.09.0 PMF JST (Industry / dust)
0369
12CMB BHM (dust)
0.00.51.01.52.0 CMB BHM (Cement)
0369
12PMF BHM (dust)
1/00 4/00 7/00 10/00 1/01 4/01 7/01 10/01 1/02 4/02 7/02 10/0202468 PMF BHM (Industry / dust)
Con
cent
ratio
ns (礸
/m3 )
Date
Figure 11
29
1/00 4/00 7/00 10/00 1/01 4/01 7/01 10/01 1/02 4/02 7/02 10/02
0.00.51.01.52.02.5
CMB YRK (dust)
0.00.20.40.6 CMB YRK (Cement)
02468
PMF YRK (dust)
0.00.51.01.52.02.5 PMF YRK (Industry / dust)
0.00.51.01.52.02.53.0
CMB CTR (dust)
0.0
0.2
0.4
0.6 CMB CTR (Cement)
1/00 4/00 7/00 10/00 1/01 4/01 7/01 10/01 1/02 4/02 7/02 10/020
2
4
6 PMF CTR (Industry / dust)
0123456
Date
PMF CTR (dust)
Figure 12
30
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
JST PMF JST CMB
315
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
270
225
180
135
90
45
YRK PMF YRK CMB
Figure 13
31
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
JST PMF Industry/ dust JST CMB Cement
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
0.00
0.15
0.00
0.15
360
315
270
225
180
135
90
45
YRK PMF Industry /dust YRK CMB Cement
Figure 14
32
JSTPMFJSTCMB
BHMPMF
BHMCMB
YRKPMF
YRKCMB
CTR PMF
CTRCMB0
1
2
3
4
5
Wood SmokeCoal combustionDustMotor VehicleIndustry/dust (PMF) | Cement (CMB) Industry(Zn) (PMF)SEOC
Con
cent
ratio
ns ( 礸
/m3 )
Figure 15
33