Spatial distribution of carbonaceous aerosol in the southeastern United States using molecular markers and carbon isotope data

Authors


Abstract

[1] Spatial variations of source contributions to fine organic carbon (OC) and fine particles in the southeastern United States were investigated using molecular marker-based chemical mass balance modeling (CMB-MM) and carbon isotope analysis. Nine primary emission sources were resolved with wood combustion (average 1.73 μg m−3, 23 ± 14% of measured OC) being the most dominant contributor to OC, followed by gasoline engine exhaust (0.45 μg m−3, 6.1 ± 6.2% of OC), diesel engine exhaust (0.43 μg m−3, 4.8 ± 4.1% of OC), and meat cooking (0.30 μg m−3, 4.1 ± 2.6% of OC). Measurable contributions from vegetative detritus, cigarette smoke, road dust, and natural gas exhaust were found. The impact of coke facilities was estimated for the first time in Birmingham, Alabama, and contributed 0.52 μg m−3 on average to fine OC. The unexplained OC accounted for 54 ± 26% of measured OC, possibly because of contributions from secondary OC, other unidentified primary sources and the possible positive artifact of OC. The urban excess of OC from diesel exhaust, gasoline exhaust and meat cooking can be seen from the results of the urban-rural pair in Alabama. Detailed chemical analysis revealed the wood burning episode at the rural site and an episode of secondary formation in the study region. The 14C analysis, a tool to study the relative contributions of contemporary and fossil carbon contents of fine particles, agreed well with the CMB-MM analysis. Both reflected a higher fossil fraction of carbon at urban sites especially in Birmingham, Alabama.

1. Introduction

[2] Studies have shown that fine-particulate air pollution (particulate matter 2.5 micrometers in diameter and smaller or PM2.5) exhibits the greatest impact on human health [Dockery et al., 1993]. In response, the U.S. Environmental Protection Agency (U.S. EPA) promulgated a National Ambient Air Quality Standard (NAAQS) for PM2.5. The new NAAQS led to a number of areas in the United States being classified as nonattainment areas, particularly in the southeast. Two of the eight sites (Birmingham and Jefferson Street, Atlanta) of the Southeastern Aerosol Research and Characterization Experiment (SEARCH) have exceeded PM2.5 annual standard. On 17 December 2004, EPA designated some counties in Georgia and Alabama as nonattainment areas for fine particles. A prerequisite for the implementation of an effective control measure for PM2.5 is to gain an understanding of PM2.5 composition and the associated major source contributors. With the development of analytical techniques, a reliable method for organic tracer analysis has evolved [Mazurek et al., 1987; Schauer and Cass, 2000; Zheng et al., 2002]. These organic tracers can be applied in a chemical mass balance model to apportion the sources contributing to fine organic carbon [Schauer et al., 1996]. CMB-MM has been successfully used for source apportionment study in a number of regions of the United States such as the Los Angeles Basin [Schauer et al., 1996], the San Joaquin Valley in California [Schauer and Cass, 2000], and the southeastern United States [Zheng et al., 2002], and elsewhere, e.g. China [Zheng et al., 2005].

[3] As part of the SEARCH study, the Carbonaceous Aerosol Characterization Experiment (CACHE) aims to understand variability of organic components and source contributions to fine organic carbon and PM2.5 on a daily scale in the southeastern United States. In the present study, advanced analyses, including radiocarbon (14C) analysis and CMB-MM, were employed for the source apportionment of carbonaceous aerosol.

[4] There are various methods for source apportionment of aerosols including the source-oriented community multiscale air quality model (CMAQ) and the receptor-oriented model such as positive matrix factorization (PMF) and CMB. 14C analysis has been used as an effective tool in aerosol studies to identify the contemporary or fossil fuel associated origin of carbon since radioactive carbon is depleted in fossil fuel through decay [Tanner et al., 2004]. A better understanding of particulate matter (PM) sources can be achieved by combining two different methods in principle. To the best of our knowledge, this study is the first to report and compare the results of CMB-MM and 14C analyses conducted on the same aerosol samples.

2. Experimental Methods

2.1. Sampling

[5] Twenty-four-hour integrated PM2.5 samples were collected at sites in the SEARCH air quality monitoring network: North Birmingham, Alabama (BHM), Centreville, Alabama (CTR), Pensacola, Florida (PNS), and Jefferson Street, Atlanta, Georgia (JST) (see Figure 1). BHM (urban) and CTR (rural) is an urban-rural pair in Alabama and the other two are urban sites. The Southeast is characterized by a relatively long, hot and humid summer with light winds and a short, cool and dry winter. The generally rolling or hilly terrain is covered by subtropical vegetation (primarily loblolly pine) [Hansen et al., 2003]. During summer, VOCs from anthropogenic (transportation) and biogenic (forest) serve as precursors to the secondary formation of fine particulate matter.

Figure 1.

Sites of the Southeastern Aerosol Research and Characterization Experiment (SEARCH). BHM, N. Birmingham, Alabama; CTR, Centreville, Alabama; JST, Jefferson St., Atlanta, Georgia; and PNS, Pensacola, Florida.

[6] The details of site descriptions are given by Hansen et al. [2003]. In brief, BHM is located in Jefferson County of Alabama, representing a commercial-industrial-residential area. This site is about 3.6 km northeast of the intersection of I-65 and I-20, about 400 m west of a large cast iron pipe foundry, and about 2 km southwest of several coking plants. As an urban-rural pair, BHM and CTR are about 85 km apart in Alabama. CTR is about 250 m north of a lightly traveled road and surrounded by the Talladega National Forest with loblolly pine and mixed deciduous. JST is about 4.2 km northwest of downtown Atlanta and the surrounding areas include warehouse, storage building, parking lots, city streets, bus maintenance facility (250 m south), and residential neighborhood. PNS, an urban site in Florida, is located about 4.5 km northwest of downtown Pensacola and surrounded primarily by residential neighborhood and light commercial-industrial activity west and southwest.

[7] Simultaneous high-volume sampling was conducted every third day at four sites. Samples were selected to represent a range of concentrations across the CACHE network, including typical, low, and high for fall and winter. In addition, dates were picked to investigate difference of PM level and PM sources between the urban and rural sites in Alabama and suspected wood smoke events. Therefore PM2.5 samples at four CACHE sites from nine days during September 2003 and January 2004 (a total of 36 samples) were analyzed. PM2.5 samples for OC, elemental carbon (EC), 14C, and organic tracer analyses were collected on the quartz filter (8 × 10 in) with the high-volume sampler. A punch was taken from a quarter quartz filter for OC and EC analyses using the National Institute of Occupational Safety and Health (NIOSH) or thermal-optical transmittance (TOT) method. The remaining quarter filter was sent to the National Ocean Sciences Accelerator Mass Spectrometry (NOSAMS) facility at Woods Hole Oceanographic Institute for the 14C measurement.

[8] Measurements of other chemical species essential for CMB-MM source apportionment, such as PM2.5 mass, aluminum, silicon, sulfate, nitrate, and ammonium, were made on the 24-hour integrated filter samples as described elsewhere in detail [Hansen et al., 2003]. Briefly, the FRM sampler (Rupprecht & Patashnick model 2025 sequential sampler) was equipped at each site to measure PM2.5 mass and elemental concentrations including aluminum and silicon with the 47 mm Teflon filter by the gravimetric method and X-ray fluorescence (XRF), respectively. Measurements of sulfate, nitrate, and ammonium were from the filters installed on Channel 1 of a three-channel filter-based particulate composition monitor sampling system (PCM; Atmospheric Research and Analysis, Inc., Plano, Texas) [Hansen et al., 2003] with the same sampling frequency as the high-volume sampler (every third day).

2.2. Organic Speciation

[9] An internal standard (IS) mixture solution containing 16 isotopically labeled organic compounds was spiked in each sample prior to solvent extraction. The IS mixtures include benzaldehyde-d6, dodecane-d26, decanoic acid-d19, phthalic acid-3,4,5,6-d4, acenaphthene-d10, levoglucosan-13C6 (carbon-13 uniform-labeled compound), hexadecane-d34, eicosane-d42, heptadecanoic acid-d33, 4,4'-dimethoxybenzophenone-d8, chrysene-d12, octacosane-d58, ααα-20R-cholestane-d4, cholesterol-2,2,3,4,4,6-d6, dibenz(ah)anthracene-d14, and hexatriacontane-d74. The mass of IS mixture spiked was proportional to the OC content in each sample. In this study, 250 μL of IS mixture was spiked per milligram of OC. Each sample was extracted twice with hexane, and then followed by three consecutive extractions by 2:1 mixture of benzene and 2-propanol (benzene: EMD Chemicals, Inc., HPLC grade; 2-propanol and hexane: Fisher Scientific International, Inc., Optima grade). Benzene was distilled prior to use. The extracts were filtered, combined, and concentrated into approximately 5–10 mL by rotary evaporation and then further reduced to about 250 μL by ultra pure nitrogen gas blow down. The final extract was derivatized by freshly prepared diazomethane to convert organic acids to their methyl ester analogues and analyzed by gas chromatography/mass spectrometry (GC/MS). Silylation was conducted with N,O-bis(trimethylsilyl) trifluoroacetamide (BSTFA) plus 1% trimethylchlorosilane (TMCS) (Pierce Biotechnology, Inc.) followed by GC/MS analysis for the quantification of levoglucosan and cholesterol. Methylated and silylated extracts were analyzed by GC/MS and the operation conditions have been described elsewhere [Zheng et al., 2002].

[10] Authentic standards, which contain the IS mixture, were analyzed by GC/MS along with the aerosol samples to assist identification and quantification of organic compounds. The identification scheme of organic compounds in aerosol samples was first developed by Simoneit and Mazurek [1982], followed by Rogge et al. [1993a, 1993b], Schauer et al. [1996], and Schauer and Cass [2000]. Relative response factor (RRF) for each target compound to its isotopically labeled IS, which had similar retention time and chemical structure, was calculated using authentic standards and corrected in the quantification of target compounds. The concentrations of organic species quantified in the present study were corrected from four field blanks (one field blank for each site). One laboratory blank analysis was conducted with every batch of samples to monitor possible contamination.

[11] The QA/QC strategy for the GC/MS system includes solvent blank runs, instrumental sensitivity checks, and relative response factor checks. The instrumental sensitivity to organic compounds was evaluated before the analysis of filter extracts. Two PAHs (pyrene and coronene) and cholesterol were chosen as the foundation of our QA/QC. For instance, with the authentic standards of coronene (0.05 ng μL−1), pyrene (0.02 ng μL−1), and cholesterol (0.20 ng μL−1), the instrument would be ready for the analysis of filter extracts if the coronene to pyrene ion ratio (ion 300 to ion 202) was above 0.25 and the absolute abundance of coronene and cholesterol could reach a minimum of 10,000 and 1,000,000, respectively. From previous studies, it was found that there were basically three main areas of instrument degradation that led to poor GC/MS performance: (1) injector degradation, (2) column degradation, and (3) MS detector degradation. The performance of these compounds is not indicative of the performance of other compounds as the criteria compounds tend to degrade first. Therefore, if these criteria compounds show good performance, then it is reasonable to assume that other target analytes exhibit good performance. However, the performance of many compounds can be good when the performance of criteria compounds is poor. Two RRFs of each target compound to its isotopically labeled internal standard were obtained from the first run of authentic standards (premixed with the IS mixture) at the beginning of the sequence and the second run of the same authentic standards at the end of the sequence. The drift of two RRFs of the same compound in each sequence was normally less than 50%. A target compound is regarded as undetectable if its signal-to-noise ratio (measured by peak area) is below 3.

2.3. 14C Analysis

[12] Recently, radiocarbon has been used as a tool to study the relative contributions of contemporary and fossil carbon contents of PM2.5 [Bench and Herckes, 2004] and the atmosphere of three European background sites [Mandalakis et al., 2005]. 14C in recently formed carbonaceous material or modern carbon is in equilibrium with current concentrations of atmospheric 14CO2 which has formed from atmospheric nitrogen and cosmic rays [Currie, 1992]. However, fossil carbon does not contain any 14C because of the short half-life of 14C (about 5730 y) compared to the formation time of fossil carbon. Therefore the fraction of modern carbon (fm) can be determined by measuring the 14C content in fine particulate matter and the 13C/12C isotope ratio to correct for isotopic fractionation processes [Tanner et al., 2004].

[13] A portion of each 8 × 10 in quartz filter was submitted to NOSAMS at Woods Hole Oceanographic Institute for 14C analysis. The details of this technique for high-volume samples using quartz filters have been described by Tanner et al. [2004]. Briefly, the filter was first packed into a prefired quartz tube containing CuO and Ag powder. The tube, in which the pressure was less than 3 mTorr, was sealed with a torch and then combusted at 850 degrees in a muffle furnace for 5 hours. Carbon dioxide was collected, purified, and divided into two portions with one portion for 14C analysis and a smaller portion for determining the 13C/12C ratio. The modern fraction (fm) was determined as the ratio of 14C/12C in an aerosol sample to 14C/12C in the standard, which was obtained from either of two standards: NBS Oxalic Acids I (NIST-SRM-4990) and Oxalic Acid II (NIST-SRM-4990C).

[14] Considering an anthropogenic increment in 14C by the nuclear bomb tests during 1950s and 1960s, a correction factor of 1.25, as suggested by Lewis et al. [2004], was employed to convert the modern fraction of carbon (fm) measured by NOSAMS to contemporary carbon (fc). Lewis et al. [2004] recommended that the factor should be at least 1.08 (tree age = zero), but no more than about 1.25 (tree age = about 50 y). Then the fossil fraction of carbon (ff) can be estimated by this equation: ff = 1 − fc. There was no prefractionation of particulate matter carbon before packing filter samples into a combustion tube. 14C analysis oxidizes everything that is combustible at 850 degrees in the presence of oxygen. Thus 14C analysis apportions total carbon into fossil and contemporary fractions. In the present study, CMB source apportionment results of primary carbon (organic carbon plus elemental carbon) are compared to the results of 14C analysis.

2.4. CMB-MM Analysis

[15] The principle of CMB modeling can be expressed by the following linear equation:

display math

where cik is the concentration of chemical species i at ambient receptor site k; m is the total number of emission sources; aij is the abundance of chemical species i (normalized to fine OC) from source j; and sjk is source contribution to fine OC from source j at site k. Therefore sjk can be solved with aij obtained from source tests and cik measured from ambient samples. An effective variance weighted least squares solution was used to solve the set of linear equations generated [Watson et al., 1984].

[16] Meeting the statistical performance measures (e.g., R-square and Chi-square) alone does not necessarily mean that the CMB output is significant. The knowledge of molecular markers provides a consistency check for selecting source profiles to be included in the model. The strategy for selecting source profiles is based on knowledge of existing sources and knowledge of tracers. As an example, the BHM site was located just about 2 km away from coke plants, thus such a source was considered. Nine source profiles were applied in CMB8.0 software for estimating source contributions to fine OC [Watson et al., 1998]. These nine sources are important sources of organic carbon. These sources included emissions from diesel-powered vehicles [Hildemann et al., 1991; Schauer et al., 1999a], combined catalyst and noncatalyst-equipped gasoline powered vehicles [Schauer et al., 2002], wood combustion [Fine et al., 2002a, 2002b], paved road dust [Schauer, 1998], meat cooking [Schauer et al., 1999b], vegetative detritus [Rogge et al., 1993a], cigarette smoke [Rogge et al., 1994], natural gas combustion [Rogge et al., 1993b], and emissions from coke facilities [Weitkamp et al., 2005; A. L. Robinson et al., Source apportionment of molecular markers and organic aerosol—1. Methodology for visually comparing source profiles and ambient data, submitted to Environmental Science and Technology, 2006]. Source profiles of wood combustion and paved road dust were modified to approximate emission characteristics of wood combustion and soil composition in the southeastern United States [Zheng et al., 2002]. The emission inventory of PM2.5 shows that wood burning emissions from fireplace and woodstove were far more important than emissions from prescribed burning and wild fires at three urban sites (BHM, JST, and PNS) during fall and winter (D. Tian and A. G. Russell, manuscript in preparation, 2006).

3. Results and Discussion

3.1. PM2.5 Composition

[17] PM2.5 mass concentrations range from 6.72 to 37.10 μg m−3 at BHM (urban), 3.89 to 39.14 μg m−3 at the rural CTR site, 4.85 to 25.64 μg m−3 at JST (urban), and 5.68 to 28.01 μg m−3 at PNS (urban). At all sites, the lowest PM2.5 mass concentration occurred on 17 December 2003, and the highest on 9 September 2003 (Figure 2). The sum of sulfate, nitrate, and ammonium accounted for about 49% of PM2.5 mass concentration on 9 September, indicating that secondary aerosols dominated. On 9 September, sulfur constituted 13 ± 2% of PM2.5 mass compared to about 7 ± 3% during other days. PM2.5 mass concentration on this sulfur-dominated day was also higher than the average of other days (32.5 μg m−3 versus 14.8 μg m−3). In addition, the spatial variation on this day was not as significant as that observed on other days, such as the 13 January sample in which PM2.5 mass concentration between CTR and JST differed by a factor of 3. On average, the sum of major chemical components including sulfate, nitrate, ammonium, OC, and EC explains 92% of PM2.5 mass at BHM, 86% at JST, 76% at PNS, and 77% at CTR. BHM exhibited the highest averaged OC and EC during the study period (9.87 μg m−3 OC and 3.36 μg m−3 EC), followed by JST (6.94 μg m−3 OC and 0.83 μg m−3 EC), PNS (6.37 μg m−3 OC and 0.52 μg m−3 EC), and CTR (6.26 μg m−3 OC and 0.27 μg m−3 EC). Carbonaceous aerosol (the sum of OC and EC) is obviously more important at BHM (64% of PM2.5 mass) and at JST (53% of PM2.5 mass).

Figure 2.

Major components of PM2.5 at four sites in the southeastern United States. OM stands for organic matter defined as 1.4 times of organic carbon concentration since organic matter includes other atomic constituents besides carbon [Gray et al., 1986].

[18] Table 1 presents the arithmetic mean and median concentration values of the major organic tracers as well as the concentration range for each tracer. Extreme day-to-day variability at the sites is evidenced by the wide tracer concentration ranges observed. The total ambient concentrations of these GC resolvable organic tracers in Table 1 were 362 ng m−3, 243 ng m−3, 239 ng m−3, and 293 ng m−3 at BHM, CTR, JST, and PNS, respectively. The mean value of organic tracers in Table 1 showed that PAH had the highest spatial variability with extremely high level at BHM, followed by JST, PNS, and CTR. The mean of levoglucosan exhibited the least spatial variation; however, it varied significantly at CTR (in the range of 26–1080 ng m−3). The organic tracers at the urban BHM site showed higher concentrations than the rural CTR site except for samples on 13 and 22 January, when PM2.5 mass concentrations were also higher at CTR. Levoglucosan was the dominating tracer species, accounting for 28%, 38%, 38%, and 47% of the total concentrations of all identified organic compounds at BHM, CTR, JST, and PNS, respectively. Exceptions were found in two samples with the most abundant species as benz(de)anthracen-7-one (397 ng m−3) at BHM on 2 November 2003 and as dehydroabietic acid (1095 ng m−3) at CTR on 22 January 2004.

Table 1. Concentrations of Major Organic Tracers in PM2.5 at Four Sites in the Southeastern United States (Arithmetic Mean, Median, and Range)a
CompoundBHMCTRJSTPNS
MeanMedianRangeMeanMedianRangeMeanMedianRangeMeanMedianRange
  • a

    Concentrations are in ng m−3; nd, not detected.

Pentacosane12.86.441.20–53.61.390.300.17–5.707.032.230.38–20.43.072.170.42–8.18
Hexacosane12.27.590.85–64.30.790.260.12–2.614.841.790.39–13.42.332.400.50–5.05
Heptacosane6.092.90.77–30.50.760.370.16–3.352.271.660.42–6.142.152.310.79–4.49
Octacosane4.441.580.35–23.80.410.190.07–1.731.250.670.22–3.701.531.690.45–3.21
Nonacosane9.842.831.57–59.12.270.740.40–12.43.292.100.58–14.56.615.631.55–20.4
Triacontane3.101.50.36–15.40.400.150.05–1.881.060.650.15–3.911.641.540.45–3.58
Hentriacontane8.123.241.43–45.03.110.870.36–18.23.592.570.71–14.97.845.811.93–18.5
Dotriacontane2.101.280.24–9.830.330.130.04–1.500.850.670.11–3.201.781.920.47–3.71
Tritriacontane2.721.410.40–12.70.700.270.12–3.641.331.100.23–4.962.332.490.73–5.09
17α(H)-21β(H)-29-norhopane1.361.420.48–2.470.050.030.02–0.100.620.540.13–1.390.300.280.11–0.59
17α(H)-21β(H)-hopane1.651.50.52–3.390.040.03nd–0.090.630.520.13–1.300.310.270.10–0.60
22,29,30-trisnorneohopane0.350.330.13–0.540.050.05nd–0.110.200.150.04–0.510.120.110.03–0.30
20S,R-5α(H),14β(H),17β(H)-cholestanes0.510.370.15–1.180.01ndnd–0.040.180.14nd–0.410.130.120.05–0.27
20R-5α(H),14α(H),17α(H)-cholestane0.450.460.19–0.85ndndnd–0.010.140.10nd–0.340.170.150.04–0.48
20S,R-5α(H),14β(H),17β(H)-ergostanes0.470.370.12–1.18ndnd 0.140.110.03–0.340.100.090.03–0.19
20S,R-5α(H),14β(H),17β(H)-sitostanes0.740.680.25–1.510.01ndnd–0.050.290.210.06–0.640.190.170.06–0.34
Benzo(b)fluoranthene10.92.620.08–56.70.150.030.01–0.640.430.240.07–1.570.380.210.05–1.29
Benzo(k)fluoranthene7.411.820.06–43.70.120.020.01–0.540.350.180.04–1.450.240.140.02–0.74
Benzo(e)pyrene8.992.240.09–50.50.120.020.01–0.460.430.260.05–1.740.360.240.07–1.04
Indeno(cd)fluoranthene1.960.480.03–9.360.050.010.004–0.170.160.060.02–0.680.110.110.03–0.30
Indeno(cd)pyrene9.772.060.10–48.30.110.020.01–0.420.460.180.05–2.010.320.270.06–0.88
Benzo(ghi)perylene8.892.590.15–45.20.110.020.01–0.401.000.330.06–5.050.560.500.16–1.47
iso-Nonacosane0.03ndnd–0.280.02ndnd–0.130.120.00nd–0.750.140.08nd–0.28
anteiso-Triacontane0.900.610.04–2.140.02ndnd–0.120.450.240.04–1.530.380.330.13–0.75
iso-Hentriacontane0.840.560.07–2.47ndnd 0.380.35nd–1.050.500.440.12–0.98
Levoglucosan236183117–39222968.026–108020196.042.0–69925426787.0–499
Cholesterol0.780.360.10–1.960.090.08nd–0.240.270.20nd–0.760.300.250.09–0.65
Nonanal8.634.320.67–42.82.621.36nd–9.486.322.73nd–20.25.303.401.27–13.5

[19] Spatial distributions of total organic compounds and some key groups of organic compounds are shown in Figures 3 and 4. Concentrations of the total identified organic compounds were the highest at BHM for all fall samples (September to December), while for the January samples, it was at CTR or JST. Levoglucosan exhibited a very similar pattern as the total organic compounds with the correlation coefficient as 0.9181 at CTR, 0.9756 at JST, and 0.8829 at PNS. Levoglucosan concentrations at BHM showed a slight decrease in January, while a significant increase of levoglucosan concentrations was observed at CTR in January. Both levoglucosan and one of the resin acids, pimaric acid, had similar distribution patterns at CTR, JST, and PNS (the average correlation coefficient as 0.7226). At BHM, levoglucosan was less correlated with both total organic compounds and pimaric acid. The distributions of hopanes and steranes (H+S) showed significant difference between BHM (urban) and CTR (rural) (Figure 4). Concentrations of H+S were always the highest at BHM, and the lowest at CTR, clearly demonstrating the strong impact of fossil fuel associated sources at BHM. PAHs also showed significant higher level at BHM for most period of time, except on 22 January 2004, when PAHs were the highest at CTR. The high level of PAHs at CTR might be from wood combustion. In some samples at BHM (e.g., 2 November) and CTR (e.g., 13 and 22 January), total PAH concentrations were significantly higher than the levels in other samples, suggesting additional sources of PAHs might exist. This is further supported by the extremely high PAH to H + S concentration ratios on 2 November at BHM (76), on 13 January (45) and 22 January (118) at CTR. Excluding these samples, the average ratio of PAH to H + S was only 7 at BHM and 4 at CTR. Aromatic diacids and aliphatic diacids in Figure 4 refer to the sum of 1,2-benzenedicarboxylic acid, 1,4-benzendicarboxylic acid, and 1,3-benzenedicarboxylic acid, and the sum of C3–C9 aliphatic diacids, respectively. Aromatic diacids were not found in diesel engine exhaust source tests by Rogge et al. [1993c] and Schauer et al. [1999a]. No aliphatic diacids were detected in the diesel engine exhaust source test by Rogge et al. [1993c], while some octanedioic acid and nonanedioic acid were observed in the diesel engine source test by Schauer et al. [1999a]. Aliphatic diacids in Figure 4 are the sum of C3–C9 aliphatic diacids and the most abundant aliphatic diacids are propanedioic acid and butanedioic acid. Some octanedioic acid and nonanedioic acids were measured in the ambient PM2.5 samples, which accounted for 17 ± 8% of the sum of C3–C9 aliphatic diacids. Secondary formation was most likely the major source of aromatic and most aliphatic diacids.

Figure 3.

Spatial distributions of total organic compounds, levoglucosan, and pimaric acid in PM2.5.

Figure 4.

Spatial distributions of hopanes and steranes, polycyclic aromatic hydrocarbons (PAHs), aromatic and aliphatic diacids in PM2.5.

3.2. Source Contributions to Fine OC

[20] Impacts of up to nine sources to fine organic carbon concentrations were estimated in the southeastern United States (Figure 5 and Table 2). The CMB results are statistically significant with R2, χ2, and DF as 0.86 (target 0.8–1.0), 2.98 (target 0–4.0), and 18 (target >5), respectively. The “Other OC” represents the difference between the measured organic carbon concentrations and the sum of the apportioned contributions from the resolved nine primary sources. On average, “Other OC” accounted for 54 ± 26% of “measured OC” for the four sites, which may include secondary OC and other unidentified primary sources. In addition, since OC was measured on the 8 × 10 in quartz filter installed on a high-volume sampler without any denuder, positive artifact was most likely to occur, which might also contribute to the “Other OC”. Among the eight emission sources (except for emissions from coke facilities), wood combustion (1.73 μg m−3, 23 ± 14% of measured OC) dominated identified sources impacting fine OC, followed by gasoline exhaust (0.45 μg m−3, 6.1 ± 6.2% of OC), diesel exhaust (0.43 μg m−3, 4.8 ± 4.1% of OC), and meat cooking (0.31 μg m−3, 4.1 ± 2.6% of OC). Contributions from vegetative detritus (0.17 μg m−3), cigarette smoke (0.14 μg m−3), road dust (0.02 μg m−3), and natural gas exhaust (0.01 μg m−3) were smaller. At BHM, the estimated contributions from coke facilities ranged from an undetectable level to 2.90 μg m−3 (0.52 μg m−3 on average). The results indicate that this source might play an important role in contributing to fine OC in Birmingham, as is suggested by the National emissions inventory [U.S. Environmental Protection Agency, 2004].

Figure 5.

Source contributions to fine organic carbon (OC) identified by molecular marker-based chemical mass balance (CMB-MM) modeling.

Table 2. Source Apportionment of Fine Organic Carbon in PM2.5 at Four Sites in the Southeastern United Statesa
Sampling Site and DateDiesel ExhaustGasoline ExhaustWood CombustionMeat CookingRoad DustVegetative DetritusCigarette SmokeNatural Gas CombustionCoke FacilitiesOther OCMeasured OCR2χ2DF
  • a

    Unit is μg m−3. The uncertainty associated with the source contribution estimate is from CMB8 and represents one standard deviation, which reflects the uncertainty of the ambient data and the source profile.

Birmingham, Alabama (BHM)
   9 Sep. 20031.79 ± 0.200.79 ± 0.153.05 ± 0.870.28 ± 0.080.10 ± 0.040.03 ± 0.010.13 ± 0.03 0.19 ± 0.037.3513.71 ± 0.740.834.7618
   21 Oct. 20031.35 ± 0.172.53 ± 0.262.40 ± 0.711.09 ± 0.200.06 ± 0.030.10 ± 0.060.64 ± 0.13 0.74 ± 0.10 12.56 ± 0.680.912.3121
   2 Nov. 20033.47 ± 0.400.59 ± 0.142.98 ± 0.841.30 ± 0.38 0.29 ± 0.090.74 ± 0.14 2.90 ± 0.365.3917.65 ± 0.940.932.2814
   11 Nov. 20030.22 ± 0.040.36 ± 0.040.45 ± 0.110.27 ± 0.050.02 ± 0.010.04 ± 0.010.02 ± 0.004  3.384.75 ± 0.290.892.2021
   11 Dec. 20030.47 ± 0.070.96 ± 0.102.34 ± 0.600.09 ± 0.02 0.03 ± 0.010.06 ± 0.01 0.07 ± 0.010.484.50 ± 0.270.892.4721
   17 Dec. 20030.12 ± 0.040.69 ± 0.080.83 ± 0.250.23 ± 0.05 0.12 ± 0.020.14 ± 0.03 0.07 ± 0.01 2.70 ± 0.180.863.0617
   13 Jan. 20041.17 ± 0.131.19 ± 0.141.23 ± 0.350.27 ± 0.060.05 ± 0.020.06 ± 0.020.17 ± 0.04 0.11 ± 0.025.679.92 ± 0.550.931.8318
   16 Jan. 20041.33 ± 0.121.03 ± 0.131.53 ± 0.410.28 ± 0.06  0.38 ± 0.05 0.59 ± 0.079.7514.89 ± 0.790.805.5321
   22 Jan. 20040.51 ± 0.070.63 ± 0.071.10 ± 0.300.33 ± 0.060.02 ± 0.010.03 ± 0.010.11 ± 0.02 0.01 ± 0.0035.388.12 ± 0.460.921.9521
Centreville, Alabama (CTR)
   9 Sep. 20030.11 ± 0.01 0.69 ± 0.100.21 ± 0.040.04 ± 0.010.01 ± 0.002   5.566.63 ± 0.450.872.4114
   21 Oct. 20030.11 ± 0.02 0.20 ± 0.050.16 ± 0.030.03 ± 0.010.03 ± 0.005   5.636.17 ± 0.360.891.7714
   2 Nov. 20030.12 ± 0.02 0.40 ± 0.090.06 ± 0.020.01 ± 0.0030.04 ± 0.005   3.764.38 ± 0.270.656.3015
   11 Nov. 20030.11 ± 0.02 0.32 ± 0.070.20 ± 0.040.01 ± 0.0030.04 ± 0.005   3.884.56 ± 0.280.803.6514
   11 Dec. 20030.07 ± 0.020.03 ± 0.010.23 ± 0.050.15 ± 0.04 0.02 ± 0.0030.01 ± 0.005  1.582.10 ± 0.150.822.8016
   17 Dec. 20030.06 ± 0.020.02 ± 0.010.20 ± 0.050.01 ± 0.0030.02 ± 0.0060.02 ± 0.002   1.071.39 ± 0.120.812.8414
   13 Jan. 20040.05 ± 0.15 7.64 ± 2.28  0.86 ± 0.11   8.8617.42 ± 0.920.873.015
   16 Jan. 20040.07 ± 0.030.11 ± 0.021.28 ± 0.160.04 ± 0.03 0.07 ± 0.010.01 ± 0.008  1.803.38 ± 0.220.813.4118
   22 Jan. 2004 0.13 ± 0.055.25 ± 0.550.37 ± 0.13 0.08 ± 0.010.13 ± 0.03  4.3810.34 ± 0.570.842.4717
Jefferson Street, Atlanta, Georgia (JST)
   9 Sep. 20030.24 ± 0.040.09 ± 0.020.51 ± 0.070.36 ± 0.070.04 ± 0.010.01 ± 0.0050.05 ± 0.01  4.395.70 ± 0.330.931.4819
   21 Oct. 20030.62 ± 0.070.19 ± 0.030.67 ± 0.180.31 ± 0.060.08 ± 0.030.04 ± 0.010.11 ± 0.020.02 ± 0.006 6.268.30 ± 0.460.922.0019
   2 Nov. 20030.71 ± 0.090.60 ± 0.070.98 ± 0.270.45 ± 0.10 0.05 ± 0.020.34 ± 0.050.01 ± 0.005 8.0611.23 ± 0.610.892.7620
   11 Nov. 20030.48 ± 0.060.32 ± 0.040.81 ± 0.140.34 ± 0.090.02 ± 0.0070.06 ± 0.020.19 ± 0.03  5.727.93 ± 0.450.902.4420
   11 Dec. 20030.17 ± 0.030.14 ± 0.020.53 ± 0.090.04 ± 0.01 0.03 ± 0.0050.02 ± 0.004  1.802.73 ± 0.190.872.7417
   17 Dec. 20030.14 ± 0.030.09 ± 0.010.36 ± 0.100.08 ± 0.020.01 ± 0.0040.02 ± 0.0040.03 ± 0.007  1.432.16 ± 0.160.892.3817
   13 Jan. 20040.20 ± 0.061.14 ± 0.122.27 ± 0.690.18 ± 0.060.02 ± 0.0090.11 ± 0.020.10 ± 0.020.06 ± 0.02  6.48 ± 0.370.844.2615
   16 Jan. 20040.41 ± 0.111.16 ± 0.144.78 ± 1.420.91 ± 0.18 0.33 ± 0.050.42 ± 0.070.12 ± 0.04 4.2912.41 ± 0.670.834.0919
   22 Jan. 20040.17 ± 0.050.54 ± 0.061.80 ± 0.560.09 ± 0.040.02 ± 0.0070.09 ± 0.020.05 ± 0.010.02 ± 0.01 2.735.51 ± 0.330.853.8115
Pensacola, Florida (PNS)
   9 Sep. 20030.14 ± 0.040.24 ± 0.031.81 ± 0.250.46 ± 0.090.02 ± 0.0090.08 ± 0.020.14 ± 0.03  3.366.25 ± 0.360.892.1921
   21 Oct. 20030.14 ± 0.040.25 ± 0.031.48 ± 0.200.59 ± 0.120.04 ± 0.020.07 ± 0.010.12 ± 0.02  3.005.69 ± 0.330.852.8324
   2 Nov. 20030.08 ± 0.050.38 ± 0.052.55 ± 0.360.08 ± 0.05 0.52 ± 0.080.30 ± 0.05  4.528.43 ± 0.470.852.8818
   11 Nov. 20030.07 ± 0.030.13 ± 0.020.50 ± 0.090.15 ± 0.03 0.33 ± 0.050.07 ± 0.01  2.884.13 ± 0.260.862.4418
   11 Dec. 20030.08 ± 0.030.15 ± 0.021.39 ± 0.270.21 ± 0.040.01 ± 0.0040.06 ± 0.010.05 ± 0.01  1.693.64 ± 0.230.882.2220
   17 Dec. 20030.04 ± 0.030.12 ± 0.020.97 ± 0.250.22 ± 0.05 0.31 ± 0.040.03 ± 0.007  1.363.05 ± 0.20.842.9417
   13 Jan. 20040.28 ± 0.070.60 ± 0.072.52 ± 0.670.32 ± 0.07 0.16 ± 0.030.26 ± 0.040.02 ± 0.01 4.628.78 ± 0.490.853.6120
   16 Jan. 20040.22 ± 0.050.30 ± 0.041.69 ± 0.480.23 ± 0.05 1.59 ± 0.200.03 ± 0.020.02 ± 0.009 3.787.86 ± 0.440.823.9718
   22 Jan. 20040.26 ± 0.090.64 ± 0.084.39 ± 1.160.44 ± 0.14 0.54 ± 0.100.25 ± 0.050.05 ± 0.02  9.52 ± 0.530.863.1615

[21] CMB-MM analysis identified the highest average OC from wood burning at PNS (1.92 μg m−3). Wood burning was so important at PNS that it accounted for 29 ± 10% of measured fine OC. At all sites except BHM, enhanced wood combustion activities can be seen in January, the coldest period of the year. This may contribute to a higher OC to PM2.5 ratio in January (0.54 ± 0.10) compared to that before January (0.41 ± 0.11). D. Tian and A. G. Russell (manuscript in preparation, 2006) noted that prescribed burning, which was more important at the rural CTR site, was relatively low during the fall, and increased in the winter and early spring. The two highest levels of wood combustion OC were found at CTR (7.64 μg m−3 on 13 January and 5.25 μg m−3 on 22 January). At CTR, wood combustion accounted for only 9% of measured OC before January, however, it dominated OC in the 13 January sample (44%) and the 22 January sample (51%). Thus the unusual higher PM2.5 mass and OC concentrations at this rural site was attributed to wood burning. This corresponded well to the elevated levels of levoglucosan and resin acids (Figure 3) as well as PAHs (Figure 4) measured in these two samples at CTR.

[22] Spatial differences in diesel exhaust contribution showed that the average contribution from diesel exhaust was the highest at BHM (1.16 μg m−3 of fine OC) followed by JST, PNS, and CTR (0.35 μg m−3, 0.15 μg m−3, and 0.08 μg m−3). Diesel engine exhaust impacts between BHM and CTR differed by a factor of 15. The lowest level of diesel engine exhaust was consistently seen at CTR or at PNS. CTR is the only rural site among the four sites in the present study. It is surrounded by heavily wooded National Forest and located north of a lightly traveled road. PNS is an urban site in Florida, however, it is different from another two urban sites, BHM and JST. The surrounding area is primarily residential with light commercial-industrial activity west and southwest [Hansen et al., 2003]. The lowest averaged EC concentration was found at PNS (0.52 μg m−3) and CTR (0.27 μg m−3) as well.

[23] Average contributions from gasoline exhaust were 0.97 μg m−3, 0.47 μg m−3, 0.31 μg m−3, and 0.03 μg m−3 of fine OC at BHM, JST, PNS, and CTR, respectively. About 22% of measured fine OC were attributed to vehicular emissions (diesel and gasoline engine exhaust) at BHM, followed by 12% at JST, 7% at PNS, and 2% at CTR. Compared to the samples collected from September to December 2003, higher levels of OC from gasoline exhaust were found in January 2004 samples at JST (0.95 versus 0.24 μg m−3), PNS (0.51 versus 0.21 μg m−3), and CTR (0.08 versus <0.01 μg m−3) while this difference was minor at BHM.

[24] Average contributions from meat cooking were the highest at BHM (0.46 μg m−3), followed by JST (0.31 μg m−3), PNS (0.30 μg m−3), and then CTR (0.15 μg m−3). The difference for meat cooking between BHM and CTR was observed with the two highest impacts found at BHM on 21 October (1.09 μg m−3) and 2 November 2003 (1.30 μg m−3). The average impact of meat cooking at JST and PNS was comparable (Figure 5). It should be pointed out that although cholesterol was used as one of meat cooking tracers in CMB-MM, the accuracy of source apportionment results for meat cooking is largely determined by the uncertainties in measurements for these compounds and the accuracy of the source profiles used. In a study at Seney National Wildlife Refuge sampling site in Northern Michigan, which is located in a large wetland, Sheesley et al. [2004] suggested that emissions of sterols from aquatic organisms could be additional source of cholesterol.

[25] BHM was the only ambient monitoring station impacted by the emissions from coke facilities. Since the source profile applied in the present study described the emission characteristics from Pittsburgh coke facilities, the contribution from coke facilities at BHM here was an estimate. At BHM, this source alone may account up to 2.90 μg m−3 of fine OC. The highest level was found on 2 November 2003, almost as high as that from wood combustion. There are two large, active coke plants in Birmingham in 2003 and both are located northeast of the BHM site. Several coking ovens are just about 2 km away [Hansen et al., 2003]. The prevailing wind was from northwest on 2 November. The estimated emissions from coke facilities exhibited positive correlation with total PAH concentrations (correlation coefficient as 0.9912). The median level of total PAH concentration over the study period was 49 ng m−3, however, on 2 November, as much as 795 ng m−3 of total PAH concentration was measured.

[26] Other sources, such as vegetative detritus, cigarette smoke, paved road dust, and natural gas combustion contributed less but measurable amounts of fine OC. Some may be important at a specific site or on a specific day. For example, the highest concentration of fine OC from vegetative detritus was found at PNS on 16 January 2004 (1.59 μg m−3). Very small amount of natural gas combustion OC was resolved at JST and PNS. Similar to wood burning, the highest relative contributions of vegetative detritus to fine OC were also found at PNS.

3.3. Source Contributions to PM2.5 Mass

[27] Contributions to PM2.5 mass from nine primary emission sources can be calculated using the ratios of fine OC to total fine particulate matter from source tests (Figure 6 and Table 3). Quantification of contribution from a specific source to PM2.5 was based on its contribution to primary carbon identified by CMB. For example, CMB analysis showed that diesel exhaust contribution to fine OC was 1.785 μg m−3 on 9 September at BHM. Therefore diesel exhaust contribution to PM2.5 in that sample was estimated as 9.06 μg m−3 since the OC to PM2.5 ratio was reported as 0.197 in the diesel engine exhaust test by Schauer et al. [1999a]. Ratios used in this study for various sources are listed in Table 3. The secondary fraction of inorganic ions was estimated by subtracting the amount of ions estimated to be due to the primary source emissions from measured ambient concentrations. PM2.5 mass was thus reconstructed from the primary source contributions as well as the secondary ions. The “Other OM” was estimated by multiplying the “Other OC” by a factor of 1.4 to account for other atoms present in the organic compounds. The “Others” is an estimate of PM2.5 mass that cannot be explained by all of the identified primary sources and the secondary inorganic ions. Contributions from the primary sources identified plus secondary aerosol formation accounted for 126 ± 21%, 97 ± 11%, 114 ± 15%, and 102 ± 17% of the measured PM2.5 mass concentrations at BHM, CTR, JST, and PNS, respectively. Although in CMB, values ranging from 80 to 120% of explained mass are considered acceptable, there are still several cases when the sum of identified mass exceeded the measured mass by more than 20% (see Table 3). Since OC measurement was performed on the 8 × 10 in quartz filter using a high-volume sampler while PM2.5 mass concentration was measured with a low-volume sampler (PCM) equipped with a denuder, positive OC artifact from the high-volume sampler (no denuder equipped) may partially contribute to an overestimate of PM2.5 mass concentration. As mentioned earlier, PM2.5 source apportionment was primarily based on CMB apportionment results of organic carbon in combination with the organic carbon to mass ratios from source tests. Therefore underestimation of the ratio can lead to an overestimate of the mass. In addition, modeling error is based on measurement errors and the average uncertainty of organic compounds measured was estimated as 20%. The modeling error from CMB output should be regarded as a lower estimate of the uncertainty of source contribution estimates since the modeling error does not include other potential variability such as fuel difference and operating conditions of vehicles.

Figure 6.

Sources of PM2.5 in the southeastern United States.

Table 3. Reconstructed PM2.5 Source Contributions Calculated Using Apportionment Results of Organic Carbon (OC) and OC to Mass Ratios From Source Testsa
Sampling Site and DateDiesel ExhaustGasoline ExhaustWood CombustionMeat CookingRoad DustVegetative DetritusCigarette SmokeNatural Gas CombustionCoke FacilitiesOther OMSecondary SulfateSecondary NitrateSecondary AmmoniumIdentified MassPM2.5 Mass% Mass Explained
  • a

    Unit is μg m−3.

OC to mass ratio0.1970.5380.7490.3380.13060.3240.4600.8490.4000.714      
Birmingham, Alabama (BHM)
   9 Sep. 20039.061.464.080.840.740.100.29 0.4710.310.8 3.1741.437.1112
   21 Oct. 20036.874.713.203.210.430.311.40 1.845.114.020.441.1932.724.2135
   2 Nov. 200317.61.093.983.84 0.881.60 7.247.565.50 1.9251.231.1165
   11 Nov. 20031.100.680.600.800.130.140.03  4.731.780.0320.5810.69.7109
   11 Dec. 20032.371.793.130.27 0.090.13 0.170.672.790.181.0112.610.4121
   17 Dec. 20030.581.291.110.68 0.370.30 0.170.711.490.140.487.326.72109
   13 Jan. 20045.942.211.650.780.400.190.36 0.297.943.530.141.0324.524.1102
   16 Jan. 20046.751.912.050.82  0.84 1.4813.62.371.110.6831.720.9152
   22 Jan. 20042.581.181.460.980.120.080.24 0.037.542.051.200.5818.114.2127
Centreville, Alabama (CTR)
   9 Sep. 20030.54 0.930.640.340.04   7.7816.60.0033.4930.439.178
   21 Oct. 20030.56 0.260.470.260.11   7.893.930.0051.4214.915.994
   2 Nov. 20030.63 0.530.160.060.12   5.262.980.0190.8810.711.891
   11 Nov. 20030.58 0.430.580.070.13   5.431.690.0490.529.489.26102
   11 Dec. 20030.370.060.310.430.030.050.02  2.222.140.0050.606.256.3998
   17 Dec. 20030.290.040.260.030.170.05   1.491.390.0440.504.263.89110
   13 Jan. 20040.25 10.2  2.66   12.43.110.271.2330.134.787
   16 Jan. 20040.350.211.710.100.030.200.03  2.522.150.130.798.227.26113
   22 Jan. 2004 0.237.011.10 0.250.29  6.131.860.480.7918.218.399
Jefferson Street, Atlanta, Georgia (JST)
   9 Sep. 20031.230.170.681.060.300.030.12  6.1510.70.0113.2923.825.693
   21 Oct. 20033.140.360.900.930.580.110.240.03 8.773.470.0231.2819.819.3103
   2 Nov. 20033.621.121.311.340.130.160.730.01 11.34.260.0291.5025.525.1102
   11 Nov. 20032.450.601.081.000.150.170.40  8.011.930.0500.6716.512.6131
   11 Dec. 20030.840.260.710.13 0.080.04  2.532.680.0891.028.377.84107
   17 Dec. 20030.710.170.490.230.110.060.05  2.001.490.0220.555.874.85121
   13 Jan. 20040.992.113.030.540.130.350.210.08 3.372.410.0010.8914.112.8110
   16 Jan. 20042.082.156.382.69 1.010.900.15 6.001.250.360.4923.516.6142
   22 Jan. 20040.861.012.410.260.170.280.100.02 3.821.741.561.0913.311.4117
Pensacola, Florida (PNS)
   9 Sep. 20030.730.442.421.370.130.250.30  4.7010.7 3.2824.428.087
   21 Oct. 20030.700.461.981.740.310.200.26  4.214.400.0021.5315.818.287
   2 Nov. 20030.400.713.400.24 1.600.65  6.342.01 0.6116.015.4104
   11 Nov. 20030.370.230.670.44 1.010.15  4.041.770.0850.519.2710.390
   11 Dec. 20030.410.271.850.620.060.190.11  2.371.820.0390.688.429.3190
   17 Dec. 20030.200.221.290.65 0.970.06  1.911.560.0360.597.495.68132
   13 Jan. 20041.401.113.370.950.060.500.570.02 6.464.480.281.7220.921.498
   16 Jan. 20041.120.552.250.68 4.910.070.02 5.302.900.0871.0719.015.2125
   22 Jan. 20041.331.195.861.29 1.680.530.06 4.142.060.200.7519.117.8107

[28] Secondary sulfate, wood combustion and diesel exhaust were the three major sources impacting PM2.5 at all sites over the study period. However, their contributions to PM2.5 differed across sites. At BHM, diesel engine exhaust had higher impact, while secondary sulfate contributed more to PM2.5 than diesel engine exhaust at JST, CTR, and PNS. Impacts from gasoline exhaust are also higher at both BHM (1.81 ± 1.18 μg m−3) and JST (0.88 ± 0.79 μg m−3). Emissions from coke facilities at BHM could explain about 7% of measured PM2.5 mass concentrations. Samples collected on 9 September 2003 showed high levels of secondary sulfate and secondary ammonium at all sites, suggesting that the high PM2.5 mass concentrations measured on that day were due to a secondary aerosol formation event.

3.4. Comparison Between 14C and CMB-MM Analyses

[29] The nine PM2.5 sources resolved by the CMB-MM method were further classified into two categories as having fossil and contemporary origins. Fossil sources (fossil fraction) consist of carbon (OC+EC) from diesel exhaust, gasoline exhaust, natural gas combustion, and coke production. The fraction with contemporary origin (contemporary fraction) includes carbon from wood combustion, meat cooking, vegetative detritus, cigarette smoke, and road dust. CMB-MM also gave the apportionment results of each fitting species such as EC, which was mainly from diesel exhaust (83%). 14C method apportions the fossil and contemporary fractions of carbon that can be oxidized at 850 degrees under oxygen. For each source type, the CMB-MM apportioned EC was added to the apportioned OC to get total carbon. The EC and OC from primary sources apportioned by CMB-MM are given in Figure 7. As can be seen in Figure 7, unexplained OC dominated OC in most samples at CTR. Figure 8 illustrates the estimate of fossil fractions by two independent methods (14C and CMB-MM). A strong correlation exists between the two approaches (correlation coefficient as 0.7936), with the highest average fossil fractions at BHM (0.70 by 14C versus 0.60 by CMB-MM), followed by JST (0.54 versus 0.49), PNS (0.40 versus 0.26), and CTR (0.28 versus 0.29).

Figure 7.

Source contributions to elemental carbon (EC) and OC concentrations using CMB-MM analysis.

Figure 8.

Comparison of fossil fractions resolved by 14C and CMB-MM analyses.

4. Conclusions

[30] Spatial variations of major sources of fine OC and PM2.5 were investigated in the southeastern United States during fall and winter. Carbonaceous aerosol and sulfate are the major constituents of PM2.5. Up to nine sources of fine OC were identified by CMB-MM, which exhibited different impacts at four sites. Vehicular emission dominated fine OC at BHM, while wood burning was the major contributor of fine OC at PNS. Compared to the paired rural site, BHM was more impacted by sources such as diesel exhaust, gasoline exhaust, and meat cooking. This study resolved two types of episode which resulted in high PM2.5 mass concentration, e.g., a wood burning episode on 13 January at CTR and an episode on 9 September due to secondary aerosol formation. Source apportionment analyses using the 14C and the CMB-MM methods showed similar results in the estimate of fossil and contemporary fractions. The radiocarbon analysis provided supportive evidence that fossil fuel combustion had the highest impact at BHM.

Acknowledgments

[31] The authors would like to thank Allen Robinson from Carnegie Mellon University for kindly providing emission profiles from a coke facility and Bo Wang for the assistance of organic tracer analysis. This project was supported by the Southern Company.

Ancillary