PM_2.5 source apportionment with organic markers in the Southeastern Aerosol Research and Characterization (SEARCH) study

Abstract

Positive matrix factorization (PMF) and effective variance (EV) solutions to the chemical mass balance (CMB) were applied to PM_2.5 (particulate matter with an aerodynamic diameter <2.5 μm) mass and chemically speciated measurements for samples taken from 2008 to 2010 at the Atlanta, Georgia, and Birmingham, Alabama, sites. Commonly measured PM_2.5 mass, elemental, ionic, and thermal carbon fraction concentrations were supplemented with detailed nonpolar organic speciation by thermal desorption-gas chromatography/mass spectrometry (TD-GC/MS). Source contribution estimates were calculated for motor vehicle exhaust, biomass burning, cooking, coal-fired power plants, road dust, vegetative detritus, and secondary sulfates and nitrates for Atlanta. Similar sources were found for Birmingham, with the addition of an industrial source and the separation of biomass burning into open burning and residential wood combustion. EV-CMB results based on conventional species were qualitatively similar to those estimated by PMF-CMB. Secondary ammonium sulfate was the largest contributor, accounting for 27–38% of PM_2.5, followed by biomass burning (21–24%) and motor vehicle exhaust (9–24%) at both sites, with 4–6% of PM_2.5 attributed to coal-fired power plants by EV-CMB. Including organic compounds in the EV-CMB reduced the motor vehicle exhaust and biomass burning contributions at both sites, with a 13–23% deficit for PM_2.5 mass. The PMF-CMB solution showed mixing of sources within the derived factors, both with and without the addition of speciated organics, as is often the case with complex source mixtures such as those at these urban-scale sites. The nonpolar TD-GC/MS compounds can be obtained from existing filter samples and are a useful complement to the elements, ions, and carbon fractions. However, they should be supplemented with other methods, such as TD-GC/MS on derivitized samples, to obtain a wider range of polar compounds such as sterols, sugars, and organic acids. The PMF and EV solutions to the CMB equations are complementary to, rather than replacements for, each other, as comparisons of their results reveal uncertainties that are not otherwise evident.

Implications: Organic markers can be measured on currently acquired PM_2.5 filter samples by thermal methods. These markers can complement element, ion, and carbon fraction measurements from long-term speciation networks. Applying the positive matrix factorization and effective variance solutions for the chemical mass balance equations provides useful information on the accuracy of the source contribution estimates. Nonpolar compounds need to be complemented with polar compounds to better apportion cooking and secondary organic aerosol contributors.

Introduction

Particulate matter (PM) source apportionment by receptor modeling relies on ambient measurements and source profiles for size-specific PM mass and chemical constituents (Watson et al., 2008). The chemical mass balance (CMB) receptor model describes the ambient concentration (C_it) of the ith chemical species measured at sample t as the linear sum of contributions from source types, j, with different emission compositions (Watson and Chow, 2015):

(1)

where F_ij is the fractional abundance (source profile) of the ith species in emissions from the jth source, S_jt is the source contribution estimate (SCE) of the jth source for the receptor sample t, and E_it represents the difference between the measured and CMB-calculated (i.e., estimated) receptor concentration of the ith species in sample t. E_it reflects measurement uncertainty and source profile variability.

A solution to eq 1 uses one of several forms of factor analysis (Paatero and Tapper, 1994; Roscoe et al., 1982; Thurston and Spengler, 1985) to obtain source factors from the measured receptor concentrations. These factors are then associated with measured F_ij and calculated SCEs (S_jt). The Positive Matrix Factorization (PMF) software (U.S. Environmental Protection Agency [EPA], 2014b) is most commonly used for implementing factor analysis solutions (PMF-CMB). Although it has been stated that factor analysis solutions do not require measured source profiles, there is no way to associate the factors with a source type when measured profiles are lacking. Since many receptor concentrations are correlated over time for reasons other than origination from the same emitter, the derived factors may represent mixtures from a variety of source types with this solution. Evaluations of the effects of correlation and uncertainty of input data on the factors (Balachandran et al., 2012; Brinkman et al., 2006; Chen et al., 2010; Christensen and Schauer, 2008a, 2008b; Habre et al., 2011; Hemann et al., 2009; Henry and Christensen, 2010; Kim and Hopke, 2007; Lingwall and Christensen, 2007; Lowenthal et al., 2010; Miller et al., 2002; Reff et al., 2007; Zhang et al., 2009) indicate that some of the samples in the data set need to be source dominated, i.e., representing single-source profiles. Source-dominated samples are more easily obtained at sampling sites located close to sources (Chen et al., 2007, 2011; Mooibroek et al., 2011) than at urban- to regional-scale monitors that measure complex source mixtures. The effective variance (EV)-CMB solution (Watson et al., 1984) to eq 1 uses measured F_ij along with the C_it to estimate the S_jt, inversely weighting each species, i, by its variability in both C_it and F_ij. This solution has also been implemented as open-source software (EPA, 2004). Chen et al. (2007, 2011) demonstrate the value of applying both the PMF and EV solutions to a data set in order to reveal excessive deviations from assumptions, and that is the approach taken here.

Any number or type of species, i, can be used as input data, but these are most useful when they are present in some source types and absent in others and when their ratios to PM mass and other species remain reasonably constant within a source type. Early source apportionment studies measured combinations of PM mass, elements, ions, and carbon fractions, typical of those available from long-term speciation networks (Dabek-Zlotorzynska et al., 2011; IMPROVE, 2015; EPA, 2014a; Zhang et al., 2012) that contain these markers. However, as pollution controls, fuels and feedstocks, and operating practices have improved, many of these markers, especially elements, have been removed from emission profiles. Primary PM emissions from well-controlled stationary sources, such as power stations and industrial processes, are small, and the detection of their primary S_jt values are more useful for attributing secondary aerosols (Eatough et al., 2000; Iyer et al., 1987; Watson et al., 2002) than to account for ambient PM concentrations. Carbonaceous aerosols constitute an important fraction of PM, and this fraction contains a wide variety of organic source markers (Daisey et al., 1986; Hildemann et al., 1991; Schauer et al., 1996; Chow et al., 2007a). Although measuring organic markers once required large sample deposits and complex extractions (Mazurek et al., 1987), modern thermal desorption (TD) methods (Chow et al., 2007b; Hays and Lavrich, 2007; Ho and Yu, 2004; Orasche et al., 2011) allow such measurements to be made on portions of quartz-fiber filter samples.

The Southeastern Aerosol Research and Characterization (SEARCH) study is a long-term air pollutant monitoring network at eight urban and rural locations in the southeastern United States (Edgerton et al., 2005, 2006; Hansen et al., 2003; Watson et al., 2009). Concerns over air quality remediation and health effects prompted studies to identify and quantify atmospheric PM source contributions in this region and track their trends over time (Blanchard et al., 2013, 2014; Hidy et al., 2014; Ke et al., 2008; Lee et al., 2008; Lee and Russell, 2007; Liu et al., 2006; Marmur et al., 2006; Tian et al., 2009; Zheng et al., 2002, 2006, 2007). In addition to PM_2.5 (particles with aerodynamic diameters less than 2.5 μm) mass, elements, ions, and thermal carbon fractions, a set of nonpolar organic compounds was measured at the Atlanta, Georgia (JST), and Birmingham, Alabama (BHM), sites from 2008 to 2010. In contrast to previously applied solvent extraction approaches that need to aggregate samples from multiple days, the thermal desorption-gas chromatography/mass spectrometry (TD-GC/MS; Chow et al., 2007b) method allows for rapid measurements on every sample. This study examines the utility of the TD-GC/MS measurements for quantitative source apportionment using the PMF and EV solutions to the CMB equations.

Materials and Methods

Ambient PM_2.5 chemical speciation

The species measured and concentration summaries are detailed in the Supplemental Material (Tables S-1 to S-4), as are maps (Figure S-1) showing the sampling locations, which are urban-scale monitors (Chow et al., 2002a) in commercial/residential areas. Samples for mass, elements, ions, and carbon fractions were collected at the JST and BHM sites with a multichannel, sequential filter-based particle composition monitor (PCM) operated at 16.7 L/min per channel through WINS (well impactor ninety-six; Peters et al., 2001) PM_2.5 inlets (Edgerton et al., 2005) preceded by a URG (Chapel Hill, NC) PM₁₀ cyclone. Acidic and basic annular denuders preceded the Teflon-membrane and nylon-membrane filter channels to remove nitric acid and ammonia gases, respectively. An activated carbon–coated honeycomb denuder preceded the quartz-fiber filter channel to remove adsorbed organic vapors (Chow et al., 2010; Watson et al., 2009). PM_2.5 mass was measured gravimetrically on Teflon-membrane filters prior to elemental analysis by nondispersive X-ray fluorescence (XRF; Watson et al., 1999). These filters were then extracted in deionized-distilled water (DDW) to quantify water-soluble metals using inductively coupled plasma mass spectrometry (ICP-MS; EPA, 1999). The nylon-membrane filter was extracted in DDW for anion and cation analyses by ion chromatography (IC; Chow and Watson, 1999). The IMPROVE_A thermal/optical reflectance (TOR) protocol (Chow et al., 2007c, 2011) was applied to 0.5-cm² portions of the quartz-fiber filter to obtain organic and elemental carbon (OC and EC) as well as seven thermally defined carbon fractions.

A separate medium-volume sequential filter sampler (Chow et al., 1993; EPA, 1989) was collocated at each site equipped with a Bendix 240 PM_2.5 cyclone (Chan and Lippmann, 1977) and obtained quartz-fiber filter deposits at 20 L/min drawn into a plenum; the remaining 93 L/min was exhausted as makeup air to achieve the PM_2.5 cut-point. One hundred and fifteen organic compounds were determined (i.e., 28 each for polycyclic aromatic hydrocarbons [PAHs] and n-alkanes, 18 each for iso/anteisoalkanes and hopanes, 12 steranes, 2 methylalkanes, 3 branched alkanes, 5 cycloalkanes, and 1 alkene) following the TD-GC/MS procedure of Ho et al. (2004, 2008, 2012). Four PAHs, 1-methylnaphthalene, 2-methylnaphthalene, 2,6-dimethylnaphthalene, and anthraquinone, were below the minimum detectable limits (MDLs) for 995 of 997 samples and were excluded from the data set.

Source profiles

Real-world source profiles should be measured to represent emissions for the study period and region (Wang et al., 2012a; Watson et al., 2012). Area-source emission profiles were not available for Atlanta or Birmingham, so applicable profiles were sought from other studies, as detailed below; this is consistent with prior source apportionment studies in the region (Zheng et al., 2002). A relatively small number of source profiles in the EPA’s (2014c) SPECIATE database contain abundances for organic compounds, and there is an incomplete overlap between the compounds reported and the ones measured by TD-GC/MS. To assure compatibility between the source profile and ambient species, 20 archived (under refrigeration) quartz-fiber filter remnants from prior real-world emission tests were submitted to the same TD-GC/MS analyses applied to the SEARCH samples for (1) roadside motor vehicle exhaust, biomass burning, and cooking emissions collected during the Big Bend Regional Aerosol and Visibility and Observational (BRAVO) study (Chow et al., 2004); and (2) biomass burning and roadside emission samples from the Lake Tahoe Source Characterization Study (Kuhns et al., 2004). These samples had been previously analyzed for conventional species including elements, ions (i.e., Cl⁻, SO₄²⁻, NO₃⁻, NH₄⁺, Na⁺, and K⁺), and carbon (OC and EC and their thermal fractions).

As a complement to this study, primary PM samples were collected from five coal-fired power plant boiler stacks (CFPPs; four in Georgia and one in Alabama) using the heated front filter from EPA (2000) Method 5. The normal glass-fiber filter was substituted with Teflon-membrane (R2PJ047; Pall Corporation, Ann Arbor, MI) and prefired quartz-fiber (Tissuquartz; Pall Corporation) filters. Because hot-stack gases are measured with Method 5, some of the volatile components that would condense upon cooling are underestimated by hot-filter measurements (England et al., 2007; Hildemann et al., 1989; Sheya et al., 2008), especially for semivolatile organic compounds. However, these organics are expected to be a small portion of PM emissions for well-controlled power plants such as those tested and operating in the region. Several samples were obtained from each source after PM and sulfur dioxide (SO₂) scrubbers with sample durations ranging from 1 to 6 hr, and Teflon-membrane and quartz-fiber samples were taken sequentially at each source rather than in parallel. Mass and elements were measured on the Teflon-membrane filters, whereas ions, carbon fractions, and organic species were measured on the quartz-fiber filters. The average sum of species to mass ratio was less than about 0.5, lower than for the ambient data set (0.75 ± 0.13). The CFPPOUT profiles (Table S-5) were calculated as the average of all chemical concentrations divided by the corresponding sums of species. A similar approach was taken by Chow et al. (2007a) for source profiles where there was disagreement between measured mass and the sum of species. Wang et al. (2012b) found that much of the missing mass was due to liquid water associated with sulfuric acid in the samples, which is retained at the standard mass equilibration conditions of 35% relative humidity (RH). Composite source profiles were calculated for each source type, e.g., motor vehicle exhaust, biomass combustion, cooking, and coal-fired power plant boiler stack emissions (i.e., coal combustion) and are summarized in Table S-5.

Results

Ambient PM_2.5 data

Complete data for mass, elements, ions, carbon, and speciated organics were available for 279 JST samples and 718 BHM samples. Uncertainties for each concentration were estimated as a function of analytical replicates, field blank standard deviations, and sample volume performance tests (Watson et al., 2001). More than 75% of concentrations exceeded twice their uncertainties (Chow et al., 2007a). This criterion was used to select species for inclusion in the receptor modeling. Since uncertainties are required as input to the PMF-CMB and EV-CMB software, a few zero values found in the ambient data set were replaced by the root-mean-square error (RMSE) of the remaining concentrations for a given species.

Average PM_2.5 mass and chemical species concentrations are summarized in Table S-4. Similar averages were found at the JST and BHM sites for mass (12.9 vs. 12.8 µg/m³), SO₄²⁻ (3.7 vs. 3.1 µg/m³), NO₃⁻ (0.20 vs. 0.19 µg/m³), NH₄⁺ (1.26 vs. 1.05 µg/m³), OC (2.8 vs. 3.0 µg/m³), and EC (0.86 vs. 1.17 µg/m³), respectively. Average crustal element concentrations (e.g., Al, Si, Ca, and Fe) were low (≤0.2 µg/m³), but these were up to 3 times higher at BHM compared with JST. Average Se concentrations, a coal combustion marker (Andren et al., 1975), were ~1 ng/m³ at both sites. PM_2.5 Mn, Zn, As, and Pb were factors of 5.8, 6.6, 2, and 6.8 times higher, respectively, at BHM compared with JST.

The importance of noncrustal sources of trace metals can be evaluated using crustal enrichment factors (EFs) with respect to Si (Duce et al., 1975) based on the crustal composition of Mason (1966). Crustal EFs at the JST and BHM sites for average concentrations were 1.5 and 1.2 for Al; 2.5 and 3.7 for Ca; 3.4 and 10.3 for Mn; 407 and 1393 for Zn; 1056 and 1091 for As; and 585 and 2037 for Pb, respectively. PM_2.5 Al was not enriched at either site, whereas Ca was weakly enriched at both sites. PM_2.5 Mn was weakly enriched (3.4) at JST but moderately enriched (10.3) at BHM, consistent with influence from industrial source(s). PM_2.5 Zn, As, and Pb were highly enriched at both sites. EFs for As were similar at the JST and BHM sites, consistent with similar industrial sources for this species. Conversely, EFs for Zn and Pb were 3–4 times higher at BHM than JST, consistent with more industrial emissions of these species at BHM.

The sum of all TD-GC/MS species was 18.3 ng/m³ at JST and 33.7 ng/m³ at BHM, representing only 0.46% and 0.79% of organic mass (OM), respectively (where OM/OC = 1.4 to account for unmeasured H, O, N, and S associated with organic compounds; Malm et al., 1994). This difference was mainly due to the PAHs, as the sum of average PAHs was 11.2 times higher at BHM compared with JST. Except for the average alkenes (1-octadecene) of 1.84 and 3.2 ng/m³ at the JST and BHM sites, respectively, average concentrations for other compound classes were similar at both sites.

The CMB model assumes that the composition of source emissions is conserved between source and receptor. This is a reasonable assumption for many species, but not for those formed by gas-to-particle conversion between source and receptor, as is the case for secondary SO₄²⁻, secondary NO₃⁻, and secondary organic carbon (SOC). Some PAHs are semivolatile, i.e., they are found in both gas and particle phases in the atmosphere. Phase partitioning for these compounds can change during transport from an emission source to a receptor, and the heavier PAHs with lower volatility are better markers than the light PAHs (Chow et al., 2007a).

Source profiles

Table 1 documents the 10 source profiles applied to EV-CMB, with Table S-5 providing detailed species abundances. Profiles for paved road dust (PVRD), ammonium sulfate (AMSUL), ammonium nitrate (AMNIT), and salt (MAR100) did not have organic compound abundances, so these were set to zero (Chow et al., 2007a). This is justified based on the nonexistent or low-OC fractions in these profiles and that their SCEs are sufficiently sensitive to elemental and ion abundances. The average molar ratios of NH₄⁺ to SO₄²⁻, corrected for NH₄⁺ associated with NO₃⁻ (i.e., 0.29 NO₃⁻), was 1.82 ± 0.33 at JST and 1.74 ± 0.34 at BHM, within 1 standard deviation of 2 molar ratio for pure ammonium sulfate [(NH₄)₂SO₄]. Therefore, only the completely neutralized form of SO₄²⁻ (i.e., AMSUL; (NH₄)₂SO₄) was used in the EV-CMB solution. Salt (typically from de-icing material at inland sites) is not expected to be an important contributor to PM_2.5 at either site, but a pure sea salt profile (MAR) was used to estimate and verify the salt contribution (Chow et al., 2007a).

Table 1. Descriptions of source profiles available for effective variance–chemical mass balance (EV-CMB) modeling

ID^a	Source Mnemonic	Source Profile Description	Number of Samples
1	BRVEH	BRAVO roadside sampling	2^b
2	TAVEH	Tahoe roadside sampling	2^c
3	MXTUN	Mexico City tunnel sampling	3^d
4	BRBURN	BRAVO biomass burning	4^e
5	BRCOOK	BRAVO cooking	4^f
6	CFPPOUT	Coal-fired power plant emissions	5^g
7	PVRD^h	Paved road dust	—
8	AMSUL	Pure ammonium sulfate	—
9	AMNIT	Pure ammonium nitrate	—
10	MAR	Pure sea salt	—

Notes: ^aProfiles 1–6 include organic speciation by thermal desorption-gas chromatography/mass spectrometry (TD-GC/MS). Profiles 7–10, taken from Chow et al. (2007a), contain no TD-GC/MS species. See Table S-5 for the 10 source profiles included in the EV-CMB source apportionment.^bRoadside sampling (Chow et al., 2004).^cVehicle emissions (Kuhns et al., 2004).^dTunnel sampling (Chow et al., 2002b).^eOpen burning of Mesquite, Tamarisk, Huisache, and grass (Chow et al., 2004).^fSmoked, charcoal, and propane chicken, as well as charcoal hamburger.^gSampling after sulfur dioxide scrubbers (outlet) at four coal-fired power plant boiler stacks in Georgia and one in Alabama.^hChow et al. (2007a).

Source contributions from the PMF-CMB solution

PMF 3.0 software (EPA, 2010) was applied separately to the entire data sets from the JST and BHM sites, with species showing signal-to-noise ratios (SNRs) <2 excluded from the final PMF-CMB solutions. Additional uncertainties (i.e., a factor of 3) were introduced to species that produced “singlet” factors (i.e., factors consisting of nearly 100% of a single species), as uncertainties of these species were likely underestimated (Chen et al., 2010; Green et al., 2012). Chen et al. (2010) found that the number of factors should adequately explain the variance of the input data (i.e., R² > 0.95) for all species with relatively high SNRs. These selections produced converged solutions in all PMF runs. Factor rotations were attempted by adjusting the FPEAK parameter (−1 to 1), but no important changes to the solutions resulted. Source factors and their uncertainties were derived using bootstrapping as the averages and standard deviations of the profiles from 100 runs with different starting points and 5% of samples randomly removed in each run. Consistency of the source factors over 100 runs indicates a stable solution.

As shown in Table 2, the PMF-CMB solution resolved 8 and 10 sources for the JST and BHM sites, respectively. PMF 3.0 software does not contain a diagnostic for identifying influential source markers. Marker species for each source in Table 2 were determined using the source factors from the PMF analysis (Tables S-6 and S-7) as input to the EV-CMB software (EPA, 2004), applying them to the respective average concentrations at each site, and generating the MPIN diagnostics (Tables S-8 and S-9). PMF factors were subjectively associated with source types based on the most prevalent species in the source profiles by comparing the relative abundances of species with measured source types and from the MPIN diagnostics.

Table 2. Average source contributions (µg/m³) to PM_2.5 mass estimated with the PMF-CMB solution

Source	Marker Species^a	Average Source Contribution Estimate (SCE; µg/m^3)	SCE (%)^b
Atlanta (JST site)
Motor vehicle exhaust	Ba, Fe, Cu, and EC	2.45 ± 1.00	19.4
Biomass burning	K and OC	2.45 ± 0.86	19.4
Vegetative detritus	n-alkanes (n-C29 to n-C35)	0.19 ± 0.25	1.5
Coal combustion	Se and As	0.38 ± 0.74	3.0
Road dust	Al, Si, and Fe	0.52 ± 0.16	4.1
Secondary sulfate	SO4^2−	5.69 ± 0.81	45.1
Secondary nitrate	NO3^−	0.46 ± 0.22	3.6
Salt	Na^+	0.49 ± 0.20	3.9
Average measured PM2.5 mass		12.86 ± 0.62
Average estimated PM2.5 mass		12.63 ± 1.77
Birmingham (BHM site)
Motor vehicle exhaust	EC and OC	2.64 ± 0.54	21.0
Biomass burning 1 (open burning)	K^+ and K	1.42 ± 0.32	11.3
Biomass burning 2 (RWC^c)	Retene^d	0.81 ± 0.81	6.5
Vegetative detritus	n-alkanes (n-C31 to n-C35)	0.48 ± 0.34	3.8
Industrial emissions	Cu and Fe	0.51 ± 0.27	4.1
Unknown	Hopanes	0.19 ± 0.30	1.5
Road dust	Al and Si	0.78 ± 0.18	6.2
Secondary sulfate	SO4^2−	5.37 ± 0.41	42.7
Secondary nitrate	NO3^−	0.28 ± 0.11	2.2
Salt	Na^+	0.10 ± 0.04	0.8
Average measured PM2.5 mass		12.85 ± 0.62
Average estimated PM2.5 mass		12.57 ± 1.24

Notes: ^aDetermined from EV-CMB “MPIN” diagnostics using PMF-derived source factors and average concentrations at each site.^bAverage percent of estimated PM_2.5 mass accounted for by estimated source contributions.^cResidential wood combustion.^dRetene was first recognized as a biomass burning marker by Ramdahl (1983).

The PMF-CMB SCEs accounted for nearly all of the PM_2.5 mass at both sites (Table 2). For JST, the major contributors to PM_2.5 were secondary sulfate (45.1%), motor vehicle exhaust (19.4%), and biomass burning (19.4%). The remaining sources each contributed less than 4%. Marker species for motor vehicle exhaust were Ba, Fe, Cu, and EC. However, the major constituents in the PMF factor associated with motor vehicle exhaust (Table S-6) were SO₄²⁻ (22 ± 16%), OC (19.8 ± 13.7%), and EC (16.8 ± 9.4%), indicating source mixing within the factor. The relatively high EC/OC ratio is consistent with diesel vehicle emissions, although the EC and OC abundances appear low for real-world diesel- and gasoline-engine profiles (Fujita et al., 2007). There is a low correlation (R² = 0.32) between ambient EC and SO₄²⁻ concentrations. The abundant (22 ± 16%) SO₄²⁻ fraction in the PMF-derived motor vehicle exhaust profile represents the majority of uncertainty in the SCE. The loadings of Ba, Cu, and Fe in the MPIN matrix based on EV-CMB (Table S-8) were 1.0, 0.69, and 0.57, respectively, and none of these species’ ambient concentrations were correlated with SO₄²⁻. However, the correlations between Ba and Fe with EC were 0.60 and 0.56, respectively. The correlation between EC and OC was 0.83, consistent with a common source. Ba in motor vehicle exhaust along with Cu and Fe may be associated with brake wear or road dust, although Ba has been used as a diesel fuel additive in the past (Truex et al., 1980; Turley et al., 1973).

Biomass burning markers were K and OC (Table 2). Water-soluble K⁺, the expected marker for biomass burning, had the third largest MPIN loading (0.2), and its correlation with K was 0.91. Vegetative detritus (1.5%) was characterized by n-C29 to n-C35 alkanes (NONCOS to PTRICONT, Table S-3), although two hopanes (HOP6 and HOP10, Table S-3) were also associated with this source. The primary PM coal combustion source contribution was small (0.38 ± 0.74 µg/m³), with MPIN loadings of 1 for Se and 0.83 for As, consistent with expectations. Table S-6 shows that the coal factor is composed mainly of SO₄²⁻ (65.9 ± 116%) and NH₄⁺ (23.2 ± 53%), again indicating a mixing of primary and secondary contributions. Since the primary contribution to PM_2.5 mass was small, its contribution to SO₄²⁻ was also small. The Se abundance in this profile (0.14 ± 0.29%) was nearly 2 orders of magnitude higher than that in the other PMF factors. A road dust contribution (4.1%) resulted from a factor with Al and Si markers (with abundances of 7.7 ± 2.9% and 16.7 ± 6%, respectively; Table S-6). The salt contribution (3.9%) was influenced by a factor with an 11.4 ± 4.9% Na⁺ abundance.

The PMF-CMB results for the BHM site were qualitatively and quantitatively similar to those from JST, with 10 instead of 8 source factors. The major contributors to PM_2.5 were secondary sulfate (42.7%), motor vehicle exhaust (21%), and biomass burning (17.8%). The MPIN matrix in Table S-7 shows that biomass burning was separated into two source subtypes. Biomass burning 1, associated with open burning, was characterized by potassium (K⁺ and K), and biomass burning 2, associated with residential wood combustion (RWC), was characterized by retene (Ramdahl, 1983; Shen et al., 2012). No correlation was found between K and retene in the ambient data, consistent with these coming from different source subtypes. A coal combustion factor was not found for BHM ambient data. None of the PMF source factors contained more than 0.009% Se, and the MPIN matrix did not identify Se as “influential” for any of these factors. A vegetative detritus source contribution (3.8%) was characterized by n-C31 to n-C35 alkanes (HTRICONT to PTRICONT, Table S-3). An industrial dust source contribution (4.1%) was characterized by a factor with Cu and Fe abundances. A source factor that couldn’t be identified (“unknown”) accounted for 0.1% of PM_2.5 mass and was characterized in the MPIN by hopanes (HOP3 and HOP6, Table S-3).

No PMF-CMB factor was associated with SOC, which is expected to have loadings on oxygenated compounds such as carboxylic acids, alcohols, and ketones (Couvidat et al., 2013; Jaoui et al., 2006; van Eijck et al., 2013). TD-GC/MS species are mostly hydrocarbons, so there was a dearth of SOC markers available as input data. Since nearly all of the PM_2.5 mass was accounted for by the “primary” PMF factors, any SOC present must be spread across the OC abundances in sources associated with them. This does not necessarily mean that the SOC originated from precursors emitted by these sources. As seen in Tables S-6 and S-7, the PMF-derived secondary sulfate source profiles contain 10.2 ± 4 and 8 ± 2% OC for the JST and BHM sites, respectively. The PMF thus apportions 0.60 and 0.43 µg/m³ of OC to this source at the JST and BHM sites, respectively. Note that although SO₄²⁻ associated with this PMF source may be secondary, this is not necessarily the case for OC.

Temporal variation of the monthly averaged PMF-CMB SCEs (standardized to a mean of 1) shown in Figure 1 indicates an increase in motor vehicle exhaust, road dust, and secondary sulfate contributions during the summer months at JST. Both biomass burning and vegetative detritus displayed a maximum in November. Coal combustion appeared relatively constant during the study. Although secondary nitrate contributions were minimal during the summer, they began to increase during the colder months when the thermodynamic equilibrium for ammonium nitrate favors the particle phase (Watson et al., 1994).

Figure 1. Monthly average PMF-CMB source contributions for Atlanta, Georgia (JST site). Values are normalized to the average contribution for each source for all samples. Motor vehicle exhaust contributions are dominated by engine exhaust. Complete data were available from January through April on only 26 days. Therefore, averages are presented only for May through December.

For BHM (Figure 2), the highest vehicle contributions occurred during late summer and fall. The highest biomass burning contribution was seen in November, as was the case at JST. RWC reached a minimum during summer and was elevated during winter. Dust contributions peaked during summer (similar to JST), whereas salt contribution went through a minimum during summer. This may reflect large-scale meteorological patterns affecting transport from the Gulf of Mexico or the Atlantic Ocean. As seen in JST, secondary sulfate peaked during the warmer months, whereas secondary nitrate peaked during winter.

Figure 2. Monthly average PMF-CMB source contributions for Birmingham, Alabama (BHM site). Values are normalized to the average contribution for each source for all samples. Motor vehicle exhaust is dominated by engine exhaust.

Source contributions from the EV-CMB solution

The EV-CMB applications and validation protocol (Watson, 2004) guided this software application. Table 3 summarizes sensitivity tests conducted on the average concentrations from JST to determine the effect of source profiles and fitting species on the source mix and SCEs. Fitting species included and excluded organic species using the BRVEH, TAVEH, or MXTUN motor vehicle exhaust profiles (Table 3). The BRBURN, BRCOOK, and CFPPOUT profiles were found to best represent biomass burning, cooking, and coal combustion emissions, respectively. The thermal carbon fractions were not included, as they were not available for all profiles. Although Zn and Pb were well detected in the ambient samples, the EV solution did not account for their concentrations using the available source profiles, so they were excluded. Both of these metals are enriched in industrial emissions, particularly from municipal incinerators (Sakata et al., 2000).

Table 3. Sensitivity tests for EV-CMB source contribution estimates^a for average concentrations^b at Atlanta, Georgia (JST site)

	EV-CMB Source Contribution Estimates (SCEs; µg/m^3)
Case	BRVEH	TAVEH	MXTUN	BRBURN	BRCOOK	CFPPOUT	PVRD	AMSUL	AMNIT	MAR	PCT^c	R^2d	Chi^2e
1A^f	2.1 ± 0.9	—	—	0.89 ± 0.44	2.5 ± 0.6	0.56 ± 0.06	0.04 ± 0.10	4.3 ± 0.4	0.14 ± 0.11	0.15 ± 0.06	82%	0.98	1.55
1B^g	0.06 ± 0.05	—	—	1.11 ± 0.33	3.3 ± 0.3	0.58 ± 0.06	0.15 ± 0.10	4.4 ± 0.4	0.24 ± 0.03	0.15 ± 0.07	78%	0.96	0.94
2A^f	—	0	—	0.85 ± 0.44	3.7 ± 0.4	0.57 ± 0.06	0.13 ± 0.10	4.5 ± 0.4	0.24 ± 0.03	0.16 ± 0.06	79%	0.97	1.48
2B^g	—	0.33 ± 0.44	—	1.04 ± 0.34	3.2 ± 0.4	0.59 ± 0.06	0.16 ± 0.10	4.4 ± 0.4	0.24 ± 0.03	0.15 ± 0.06	79%	0.96	0.98
3A^f	—	—	1.85 ± 0.57	1.55 ± 0.68	2.3 ± 0.6	0.49 ± 0.07	0	4.2 ± 0.4	0	0.13 ± 0.09	82%	0.98	1.16
3B^g	—	—	1.03 ± 0.46	1.08 ± 0.33	2.7 ± 0.4	0.54 ± 0.06	0.06 ± 0.10	4.4 ± 0.4	0.08 ± 0.10	0.15 ± 0.06	78%	0.96	0.92

Notes: ^aSee Table 1 for source profile descriptions and Table S-5 for source profiles. ^bThe uncertainties of the average concentrations are estimated as the root-mean-square error (RMSE) of the individual samples. ^cPercent of measured mass accounted for, by estimated source contribution. ^dR², squared correlation coefficient. ^eChi-square. ^fEV-CMB with conventional species (i.e., elements, ions, and organic and elemental carbon) only. ^gEV-CMB with conventional and TD-GC/MS organic species.

As shown in Table 3, SCEs for Case A are based only on elements, ions, and carbon species, whereas Case B represents SCEs with the addition of organic species for the three motor vehicle exhaust profiles (i.e., BRVEH, TAVEH, and MXTUN). The R-square (R²) of ≥0.96 (performance target: 0.8–1) and chi-square (Chi) of ≤1.55 (performance target <4) suggest a good fit to ambient concentrations. The percent of mass (PCT) varied from 78% when organic species were included to 82% without organic markers (performance target: 100 ± 20%). Approximately 20% of PM_2.5 mass (~2 µg/m³) was not represented by SCEs. PM_2.5 OC was estimated to within 7% in all cases in Table 3, with the cooking profile (BRCOOK) constituting 69% of OC, a key marker in MPIN diagnostics, or 97% of OM, assuming OM = OC × 1.4. It is possible that a SOC profile (composed only of OC) could be part of the cooking source contribution.

The average ambient Si concentration at JST was 0.1 µg/m³. Given the Si abundance (28 ± 9%) in the PVRD (road dust) profile, it is unlikely that geological dust accounts for >0.5 µg/m³ of PM_2.5. According to the MPIN diagnostic, Si and Al are the key marker species for the road dust profile. While no industrial source profiles other than coal combustion were included, such sources are not expected to significantly impact PM_2.5 mass, although they do emit trace elements such as Zn and Pb. Because the chemical species concentrations are fit well by the model, the 20% underprediction of mass is probably associated with systematic errors (bias) in the source profiles.

Including organic species, Case 1B (Table 3) shows a 35-fold decrease in the motor vehicle exhaust SCE, from 2.1 to 0.06 µg/m³. This is also found for Case 3B, although the decrease (from 1.85 to 1.03 µg/m³) is much smaller. Including organic species, the motor vehicle exhaust SCE using the MXTUN profile is 17 times higher than that of the BRVEH profile. In both cases, the MPIN diagnostic indicates that hopanes, i.e., α,β-norhopane (C₂₉; HOP3) and α,β-hopane (C₃₀; HOP6) are the key markers for motor vehicle exhaust. Figure 3 compares all fitting species in both source profiles. The lower BRVEH contribution is due to differences in the abundances of these two hopane species in the source profiles. Table S-5 shows that the abundances of HOP3 and HOP6 are factors of 10 and 16 higher, respectively, in the BRVEH than the MXTUN profiles. On average, all of the hopanes, as well as 1-octadecene, are 11 ± 3 times higher in the BRVEH than the MXTUN profiles. Using BRVEH as the motor vehicle exhaust profile, SCEs from the EV-CMB (0.06 ± 0.05 µg/m³ for Case 1B in Table 3) are not consistent with the PMF-CMB motor vehicle exhaust SCE for JST (2.45 ± 1 µg/m³). Although the SCE with organic species using the MXTUN profile is lower (1.03 ± 0.46 µg/m³), the source mix for Case 3B appears to represent the best solution, given the available source profiles.

Figure 3. Comparson of the BRVEH and MXTUN motor vehicle exhaust source profiles. (See Table 1 for profile descriptions and Tables S-2 and S-3 for species descriptions.)

Overall, the biomass burning and cooking SCEs are relatively consistent, given their uncertainties. This is also the case for the SCE of ~0.55 µg/m³ for coal combustion. The MPIN diagnostic indicates that Se is the key marker species for primary PM from coal combustion, as expected (e.g., Thurston and Spengler, 1985). The MPIN marker species for BRBURN (biomass burning) are K⁺ and K with conventional species (Case 3A) and HTRICONT and DTRICONT (n-C31 and n-C32, respectively, in Table S-3) as well as K⁺ and K with conventional plus organic species (Case 3B). The MPIN marker species for cooking is OC with or without organic speciation. The largest contribution is from secondary ammonium sulfate (4.4 ± 0.4 µg/m³). Road dust, secondary ammonium nitrate, and salt contributions are each less than 0.2 µg/m³. The differences between the results for Cases 3A and 3B indicate the general effect of the organic species on EV-CMB source apportionment, which lowers the motor vehicle exhaust (MXTUN) and biomass burning (BRBURN) contributions.

The EV-CMB solution was applied to individual samples using the source profile mix from Case 3 (Table 3), and average SCEs for the JST and BHM sites are shown in Table 4. The addition of organic species reduced the percent mass explained by 13–23%. R² of 0.70–0.78 and chi-square of 12–17 also fell off the performance target criteria. The performance measures for CMB SCEs from average PM_2.5 concentrations are expected to be better than average SCEs from the individual samples. This may be due to the large uncertainties in TD-GC/MS concentrations on a sample-to-sample basis that are smoothed out in the average of all samples. Average SCEs from the individual samples (Table 4) agree qualitatively with SCEs derived from the average of all samples (Table 3). Contributions from the combustion sources (i.e., motor vehicle exhaust, biomass burning, and cooking) are systematically lower when the organic species are included. Had the fractional abundances of the organic species in the source profiles been half as large, the average motor vehicle exhaust (MXTUN) and vegetative burning (BRBURN) SCEs at JST would have increased from 0.81 to 1.6 µg/m³ and from 1.44 to 3.1 µg/m³, respectively. Average cooking (BRCOOK) SCEs would not have changed (1.6 vs. 1.59 µg/m³), confirming that OC, not any of the organic species, is the key marker for this source. Note that although corrections were made for the positive OC sampling artifact in the ambient samples, this was not the case for the source samples. This could also contribute to unrealistically high organic species concentrations in the combustion source profiles.

Table 4. Average source contribution estimates (SCEs) from EV-CMB solutions for individual samples with and without organic compounds

		Source Contribution Estimates with Conventional Species		Source Contribution Estimates with TD-GC/MS Organic Species
Source Name	Source Mnemonic	(µg/m^3)	Percent Mass^a	(µg/m^3)	Percent Mass
Atlanta, GA (JST site)
Motor vehicle exhaust (Mexico City tunnel)	MXTUN	1.03 ± 0.49	9.4	0.81 ± 0.18	8.6
Biomass burning (BRAVO)	BRBURN	2.6 ± 0.9	24	1.44 ± 0.19	15.4
BRAVO cooking	BRCOOK	2.0 ± 0.5	18.6	1.60 ± 0.23	17.2
Coal-fired power plant emissions^b	CFPP	0.48 ± 0.07	4.4	0.54 ± 0.06	5.7
Paved road dust	PVRD	0.24 ± 0.10	2.1	0.33 ± 0.10	3.6
Pure ammonium sulfate	AMSUL	4.2 ± 0.4	38	4.3 ± 0.4	46
Pure ammonium nitrate	AMNIT	0.20 ± 0.10	1.79	0.19 ± 0.07	1.99
Pure sea salt	MAR	0.13 ± 0.11	1.19	0.15 ± 0.09	1.61
Average measured PM2.5 mass	Meas.^c	12.9 ± 0.6		12.9 ± 0.6
Average estimated PM2.5 mass	Est.	11.0 ± 1.2		9.4 ± 0.6
Average R-square	R^2d	0.93		0.78
Average chi-square	Chi-square^e	3.5		12.0
	PCT^f	88		75
Birmingham, AL (BHM site)
Motor vehicle exhaust (Mexico City tunnel)	MXTUN	2.8 ± 0.8	24	1.22 ± 0.11	13.3
Biomass burning (BRAVO)	BRBURN	2.5 ± 1.0	21	1.35 ± 0.16	14.7
BRAVO cooking	BRCOOK	1.97 ± 0.61	16.5	1.11 ± 0.20	12.0
Coal-fired power plant emissions^b	CFPP	0.38 ± 0.10	3.1	0.54 ± 0.06	5.9
Paved road dust	PVRD	0.66 ± 0.28	5.5	1.09 ± 0.33	11.9
Pure ammonium sulfate	AMSUL	3.3 ± 0.4	27	3.5 ± 0.3	38
Pure ammonium nitrate	AMNIT	0.12 ± 0.09	1.01	0.15 ± 0.07	1.66
Pure sea salt	MAR	0.23 ± 0.18	1.91	0.26 ± 0.13	2.9
Average measured PM2.5 mass	Meas.	12.8 ± 0.6		12.8 ± 0.6
Average estimated PM2.5 mass	Est.	12.0 ± 1.5		9.2 ± 0.6
Average R-square	R^2	0.94		0.70
Average chi-square	Chi-square	3.0		16.9
	PCT	96		73

Notes: ^aAverage percent of model calculated PM_2.5 mass accounted for by estimated source contributions. ^bCoal-fired power plant emissions from boiler stacks, referred to as coal combustion. ^cAverage measured PM_2.5. ^dAverage R-square (R²). ^eAverage chi-square. ^fAverage ratio of model calculated to measured PM_2.5 mass expressed as a percent.

The PMF and EV solutions to the CMB equations are qualitatively similar. Both analyses indicate important contributions from combustion (vehicle exhaust and biomass burning) sources. The average primary coal combustion SCE was ~0.4–0.5 µg/m³, although this source wasn’t identified by the PMF solution at BHM. An industrial source SCE (0.51 ± 0.27 µg/m³) identified at BHM is reasonable, as many such sources exist there. Since industrial source profiles with organic species were not available, these were excluded from the EV solution. The PMF solution did not produce a factor that could be associated with cooking. In the EV-CMB, the cooking source profile (BRCOOK) was mainly comprised of OC. Without an organic marker such as cholesterol, the association of a source factor with cooking is not definitive. Figures 4 and 5 show that average monthly motor vehicle exhaust SCEs with the EV solution did not vary systematically. This is also the case for the coal combustion SCEs at both sites. Monthly averaged coal combustion SCEs by PMF-CMB at JST (Figure 1) also show no annual pattern. Biomass burning SCEs peaked in November at JST with less variation at BHM. Cooking contributions peaked during the summer months at JST, which could suggest that this factor is more associated with SOC. The temporal patterns in SCEs for dust, ammonium sulfate (or secondary sulfate), ammonium nitrate (or secondary nitrate), and salt by the PMF and EV solutions were similar at both sites.

Figure 4. Monthly average EV-CMB source contributions (µg/m³) for Atlanta, Georgia (JST site), using conventional plus organic species. (Complete data were available from January through April on only 26 days. Therefore, averages are presented only for May through December.)

Figure 5. Monthly average EV-CMB source contributions (µg/m³) for Birmingham, Alabama (BHM site), using conventional and organic species.

EV-CMB results from this study are compared in Table 5 with those of Zheng et al. (2007) for the Atlanta (JST) site during summer (June–August) of 2001. There are not enough sample days during winter for a quantitative comparison. Zheng et al. (2007) applied the EV-CMB solution using mainly organic compounds, including cholesterol and levoglucosan, determined by solvent extraction followed by GC/MS. The source profiles in Zheng et al. (2007) were normalized to OC rather than PM_2.5 mass. Source contributions to PM_2.5 were estimated from SCEs to OC and the ratios of OC to PM_2.5 mass in the source samples. With a source referred to as “unidentified” OM, Zheng et al. (2007) found the sum of model-calculated SCEs accounted for 83% of measured PM_2.5 mass during summer. The motor vehicle exhaust SCE (6.8%) of Zheng et al. (2007) agrees with estimates in this study (5.5% and 6.5% with and without organic species, respectively). The cooking contribution in this study (21–22%) was about twice that estimated by Zheng et al. (2007). The largest difference is for the biomass burning contribution: 0.44% in Zheng et al. (2007) versus 7.8% and 17.8% (with and without organic species, respectively) in this study. Note that n-C31 and n-C32 alkanes along with K⁺ and K were marker species for biomass burning when organic species were included in EV solution. The negligible biomass burning contribution estimated by Zheng et al. (2007) is inconsistent with prescribed burns and wildfires occurring during summer in the region (Christopher et al., 2009; Lee et al., 2005). Soluble K⁺ is a recognized marker species for biomass burning (Chow et al., 2007a). Table S-5 indicates that the K⁺/K ratio is low (0.054) in the paved road dust (PVRD) profile compared with ~1 in the biomass burning (BRBURN) profile. The average K⁺/K ratio in JST PM_2.5 samples was 0.47 ± 0.2 during summer, exceeding the expected ratio from a geological source and suggesting an impact from biomass burning contribution. Zheng et al. (2007) reported 23% unaccounted-for OM, and attributed it to SOC. As noted above, OC was accounted for within 7% of its average measured value in this study. SOC could have been accounted for, at least in part, by the cooking contribution, which constituted 69 ± 6% of OC. The sum of the combustion source contributions (42%, including SOC for Zheng et al., 2007) agree reasonably well (35–45%) with this study.

Table 5. Comparison of EV-CMB source contribution estimates for combustion sources in Atlanta, Georgia (JST site), during summer 2008–2010 with those of Zheng et al. (2007) during summer 2001

	Source Contribution Estimates (Percent of Calculated Mass)
Source	Zheng et al.	EV-CMB^a	EV-CMB^b
Motor vehicle exhaust	6.8%	6.5%	5.5%
Cooking	11.3%	21%	22%
Biomass burning	0.44%	17.8%	7.8%
Unidentified OM^c	23%	—	—
Sum	42%	45%	35%

Notes: ^aEV-CMB results based on conventional species from this study. ^bEV-CMB results based on conventional plus organic species by the TD-GC/MS method from this study. ^cBased on unaccounted-for OC in Zheng et al. (2007).

Discussion

The application of two CMB solutions to commonly acquired and enhanced source and receptor data sets reveals powers and deficiencies of the receptor modeling approach. The PMF solution makes full use of the available ambient data, but there is a clear mixing of different source types in the derived source factors. The EV solution allows for several measured source profiles to be included, but it is apparent that some are missing, as evidenced by excesses and deficiencies in some elemental and organic species concentrations that cannot be explained. Few source profiles from other studies contain the regional and time period applicability needed for widespread applications, and this can result in systematic biases that may over- or underestimate PM_2.5 mass and chemical components.

The addition of TD-GC/MS organic compounds is a useful adjunct to the more commonly available element, ion, and carbon fraction concentrations. This approach can be routinely applied to a small punch from the commonly available quartz-fiber filter used for thermal/optical carbon analysis for OC and EC. It is feasible to couple more specific organic compound detection to these analyzers (Diab et al., 2014; Grabowsky et al., 2011), though this is not yet practical. TD-GC/MS makes many more source markers available for both the PMF and EV solutions. However, the individual compounds are present in small abundances, much less than 1% of the mass in source and receptor samples. This allows them to be highly variable within a given source type. Their production, especially for the PAHs, is highly dependent on fuels and combustion conditions. The lack of oxygenated compounds, such as sugars, sterols, terpenoids, and organic acids, limits the ability of either solution to distinguish cooking and SOC sources from each other or from the other carbon-containing sources. Many of these nonpolar organic compounds can now be quantified by TD-GC/MS on a separate quartz-fiber filter punch after derivitization (Orasche et al., 2011), and this would be another useful adjunct to analysis of new and archived filters. Multiwavelength optical information that can be acquired along with the thermal/optical carbon analysis could also enhance the resolution of different source contributions (Chen et al., 2015).

Source apportionment by receptor modeling is only one part of a weight-of-evidence approach (EPA, 2007) to justifying air quality management plans. Reconciliation of source contribution estimates with emission inventories and source-oriented modeling is the next step toward improving the application of emission reduction strategies and evaluating their long-term effectiveness.

Summary and Conclusion

Daily PM_2.5 concentrations were compiled for the Southeastern Research and Characterization (SEARCH) sites in Atlanta, Georgia (JST), and Birmingham, Alabama (BHM), from 2008–2010. Samples were analyzed for conventional species (i.e., mass, elements, ions, organic and elemental carbon and their thermal fractions) and 115 organic compounds (i.e., polycyclic aromatic hydrocarbons [PAHs], alkanes, alkenes, hopanes, and steranes) analyzed by thermal desorption-gas chromatography/mass spectrometry (TD-GC/MS).

The PMF-CMB solution found eight sources contributing to PM_2.5 at the JST site, with secondary sulfate being the largest (average of 45% of PM_2.5). Motor vehicle exhaust and biomass burning each contributed 19.4%. Coal combustion, road dust, secondary nitrate, and salt each contributed 4% or less. At the BHM site, the PMF-CMB solution attributed 43% to secondary sulfate, 21% to motor vehicle exhaust, 17.8% to biomass burning (including 11.3% to open burning and 6.5% to residential wood combustion [RWC]). Vegetative detritus contribution at BHM was 3.8%, higher than the 1.5% at JST. Road dust, industrial emissions, secondary nitrate, and salt contributed 6.2, 4.1, 2.2 and 0.8%, respectively. Coal combustion was not identified by PMF at the BHM site.

Based on the EV-CMB solution for individual samples, average motor vehicle exhaust contributions at JST were 0.81 ± 0.18 and 1.03 ± 0.49 µg/m³ with and without organic species, respectively. Corresponding contributions were 1.44 ± 0.19 and 2.6 ± 0.9 µg/m³ for biomass burning and 1.60 ± 0.23 and 2.0 ± 0.5 µg/m³ for cooking, respectively. For BHM, estimated motor vehicle exhaust, biomass burning, and cooking contributions were nearly a factor of 2 lower when organic species were included in the model.

The EV-CMB results indicate that the abundances of organic species were elevated in the source profiles compared with actual source emissions affecting the JST and BHM samples. Nonetheless, comparison with a previous source apportionment study in Atlanta showed similar motor vehicle exhaust contributions to PM_2.5 mass, with a factor of 2 higher cooking contribution, and a higher biomass burning contribution found in the current study.

Acknowledgment

The authors wish to thank Iris Saltus of the Desert Research Institute for manuscript preparation and editing.

Funding

This research was sponsored by the Electric Power Research Institute, Palo Alto, CA, Contract Number EP-P41625/C18286.

Supplemental Material

Supplemental data for this article can be accessed on the publisher’s website.

Additional information

Funding

This research was sponsored by the Electric Power Research Institute, Palo Alto, CA, Contract Number EP-P41625/C18286.

Notes on contributors

John G. Watson

John G. Watson, Judith C. Chow, and Douglas H. Lowenthal are Research Professors at the Desert Research Institute (DRI), part of the Nevada System of Higher Education.

Judith C. Chow

John G. Watson, Judith C. Chow, and Douglas H. Lowenthal are Research Professors at the Desert Research Institute (DRI), part of the Nevada System of Higher Education.

Douglas H. Lowenthal

John G. Watson, Judith C. Chow, and Douglas H. Lowenthal are Research Professors at the Desert Research Institute (DRI), part of the Nevada System of Higher Education.

L.-W. Antony Chen

L.-W. Antony Chen is an Assistant Professor at the University of Nevada, Las Vegas, and a DRI Affiliate Professor.

Stephanie Shaw

Stephanie Shaw is a Senior Technical Leader at the Electric Power Research Institute.

Eric S. Edgerton

Eric. S. Edgerton is President of Atmospheric Research & Analysis, Inc.

Charles L. Blanchard

Charles L. Blanchard is a Principal of Envair Corporation.