PM2.5 samples were collected from urban and rural locations in and around Atlanta, GA, USA. The five sampling sites, South DeKalb, Fire Station 8, Jefferson Street, Yorkville, and Fort Yargo State Park, are associated with two ongoing studies, the Southeastern Aerosol Research and Characterization (SEARCH) study (Edgerton et al., 2005, 2006; Hansen et al., 2003) and the Assessment of Spatial Aerosol and Composition in Atlanta (Butler et al., 2003). Samples were collected on Zefluor filters over 24 h periods at a flow rate of 16.7 L min-1 using cyclone inlet samplers (URG, Chapel Hill, NC, USA). The multichannel particle samplers were mounted approximately 2.5 m off the ground. Three samples were collected from Yorkville (33∘55′42.3 N, 85∘2′43.68 W), a rural background site for the Atlanta area. Two samples were collected from Jefferson Street (33∘46′38.94 N, 84∘24′59.58 W), an urban collection site that is not immediately influenced by any individual source or roadside traffic. Five samples were collected from Fort Yargo State Park (33∘58′4.18 N, 83∘43′29.16 W), a rural setting that is frequently impacted by biomass burning plumes and power plant emissions. Four samples were collected in South DeKalb, a mixed commercial–residential area approximately 1 km from a major interstate (33∘41′16.48 N, 84∘17′25.26 W). Seven samples were collected from Fire Station 8, located in an industrial area in close proximity to a rail yard, a fire station, and an intersection with heavy diesel truck traffic (33∘48′6.01 N, 84∘26′8.75 W; Table S1 in the Supplement).

PM2.5 samples were collected from gasoline and diesel exhaust, biomass burning, and coal fly ash (Oakes et al., 2012a). Emissions from ultra-low sulfur diesel fuel running a 10.8 L engine and conventional gasoline fuel running a 3.3 L engine were collected according to US Environmental Protection Agency protocols under typical urban driving conditions (Liu et al., 2008; Oakes et al., 2012a). Polydisperse coal fly ash, provided by Southern Company, from an electrostatic precipitator of a midsized coal-fired power plant was aerosolized and collected with a cyclone inlet sampler to separate the PM2.5 fraction (Oakes et al., 2012a). Smoke produced from the burning of materials collected from coniferous and deciduous trees native to Georgia, USA, was sampled during a controlled biomass burning experiment using a cyclone inlet sampler placed 3.5 m above the burn area at a flow rate of 16.7 L min-1 for approximately 30 min. The ash produced from the biomass burning experiment was analyzed in the same manner. The abovementioned primary emission source samples were collected on polytetrafluoroethylene (Zefluor) filters (Longo et al., 2014). In addition, two commercially available bacteria samples, Azotobacter vinelandii (Sigma A2135) and Bacillus subtilis (Sigma B4006), were homogenized with agate mortar and pestle and mounted on a cellulose acetate filter for analysis.

All samples were immediately stored after preparation in clean Petri dishes at -20 ∘C until analysis. Preparation for synchrotron analyses consisted of mounting an approximate 0.5 cm × 0.5 cm section of ambient aerosol and primary emission source filters over a slot on an aluminum support.

In order to create a database of inorganic sulfate standards necessary for the data analysis described below, 10 compounds were analyzed using the same experimental setup and synchrotron system as the ambient aerosols. These standards included ammonium sulfate (CAS 7783-20-2), barite, copper(II) sulfate (CAS 7758-98-7), gypsum, iron ammonium sulfate (CAS 7783-85-9), iron(III) sulfate (CAS 15244-10-7), jarosite (KFe33+(OH)6(SO4)2, magnesium sulfate (CAS 7487-88-9), potassium sulfate (CAS 7778-80-5), and sodium sulfate (CAS 7757-82-6). The sulfur standards were ground using an agate mortar and pestle to the consistency of a fine talcum powder (approximately 10 µm). A cellulose acetate filter was then gently dredged through a small quantity (less than 1 mg) of powder placed on a microscope slide. This procedure produced a thin and almost imperceptible coating on the filter in order to limit the thickness and thus self-absorption. S-NEXFS spectra of sulfate standards were collected in bulk mode. Self-absorption must be carefully controlled when measuring fluorescent X-rays from thick specimens; however, the effects of self-absorption are limited to the region of the spectrum above the K-edge (Bajt et al., 1993; Iida and Noma, 1993). In our repeated measurements, the post-edge features were consistent and reproducible, which allows us to distinguish between sulfate standards. A database of sulfate standards is provided in the Supplement.

Samples were analyzed on the X-ray fluorescence microscope located at beamline 2-ID-B at the Advanced Photon Source, Argonne National Laboratory. The beamline is optimized to examine samples over a 1–4 keV energy range using a focused X-ray beam with a spot size of approximately 60 nm2 (McNulty et al., 2003). The energy was calibrated using an elemental sulfur standard (S0). The whiteline energy of the elemental sulfur standard was aligned to 2472 eV (Cozzi et al., 2009). Sulfur near-edge X-ray fluorescence spectroscopy (S-NEXFS) data were collected in two modes that differ based on spatial resolution. In the first mode, individual sulfur-rich particles with a diameter greater than 1 µm were identified in X-ray fluorescence maps; these particles were then interrogated with micro-S-NEXFS. The number of individual particles examined that provided usable spectra are provided in Table S1. The individual sulfur-rich particles seen in X-ray fluorescence maps are obvious contributors to the total sample sulfur. However, much of the total sulfur on an aerosol filter can also be contained in particles that are smaller than 1 µm or less sulfur-rich and therefore less apparent in X-ray fluorescence maps. Consequently, in the second mode, large areas of the filters were also examined with an unfocused beam (spot size = 0.25 mm).

In order to maximize the number of samples analyzed in the allotted time, X-ray fluorescence (XRF) maps were created for a subset of samples by rastering the sample through the focused beam in 0.5 µm steps with an incident energy of 2535 eV. At this resolution, individual sulfur-rich particles were clearly discernible. S-NEXFS spectra were scanned over a 50 eV range centered at 2485 eV in 0.33 eV steps, using a 1 s dwell time at each step. Here, dwell time is used to refer to the time spent at each step in the S-NEXFS spectrum or each pixel in an XRF map. Each S-NEXFS measurement for both bulk and individual sulfur-rich particles were repeated at least three times in a single location, creating a minimum effective dwell time of 3 s. For these settings, the total scan time would be approximately 7.5 min. X-ray spectromicroscopy data were collected using an energy dispersive silicon drift detector (Vortex with a 50 mm2 sensitive area). A flow of helium was introduced between the X-ray optical hardware and the sample to reduce elastically scattered X-ray background. An in-line monitor stick coated with an aluminum sulfate standard was measured in parallel with each sample in order to identify and correct for any potential drift in monochromator energy calibration that can occur during analyses (de Jonge et al., 2010), meaning that for every spectral measurement two spectra are collected: the specimen and an aluminum sulfate standard on the monitor stick. Because the energy was calibrated using elemental sulfur S0, all the data use 2472 eV as the reference energy for S0 during the data alignment to the monitor stick. Oxidation tests were also done on a few samples where the same region or particle was measured repeatedly, creating effective dwell times of more than 10 s. Even with this longer dwell time, there was no noticeable shift in oxidation state or the relative abundance of the oxidation states in the tested samples. Clean areas of filter were examined as blanks and showed negligible background signal.

S-NEXFS provides essentially the same information as another commonly cited technique, S-XANES (sulfur X-ray absorption near-edge structure) spectroscopy. The two techniques differ primarily in the method of signal detection. S-NEXFS uses the X-ray fluorescence signal, which is inversely proportional to the absorption signal used in a XANES measurement.

Additionally, XRF microscopy was conducted in two modes. First, to map the distribution of sulfur regardless of oxidation state, an incident energy above the S K-edge (2535 eV) was used, and samples were rastered through the focused beam in 0.5 µm steps, as mentioned above. In the second mode, the spatial distribution of S0 and S+VI oxidation states were quickly mapped by selecting an incident X-ray energy tuned to their specific whiteline energies. Because of the energy drift at this beamline, it is important to identify the correct whiteline energies for the multi-energy mapping; therefore, a S-NEXFS spectrum was collected for the particle of interest immediately before mapping. The corresponding whiteline energies for S0 and S+VI were then taken directly off this spectrum. Individual maps were then collected at the whiteline energies determined for S0 and S+VI. Although it is possible that the energy drifted during mapping, the interval between the two measurements is short enough that this is not a problem. Due to the 6 eV difference between the whiteline energies of S0 and S+VI, we do not expect significant overlap between the two oxidation states in multi-energy mapping. Energy drift during mapping may reduce signal intensities of S0 and S+VI, but the overall distribution patterns of the sulfur oxidation states in the individual particle should remain relatively unaffected. These multi-energy XRF maps were completed in 0.3 µm spatial resolution and 0.5 s dwell time per pixel. X-ray focusing was adjusted to maintain a consistent beam spot size at both energies; however, some shift in the final image did occur. The images were aligned using the stack registration function of the Fiji software (Schindelin et al., 2012).

S-NEXFS data were normalized to create a relative intensity value of approximately 1 for post-edge area of the spectra. The data were also processed using a three-point smoothing algorithm built into the software package Athena to remove high-frequency noise (Ravel and Newville, 2005). The relative contribution of each oxidation state can be determined by assuming the entire area under the whiteline peaks is representative of total sulfur, and the area under each whiteline peak represents the relative contribution of each oxidation state (Huffman et al., 1991; Xia et al., 1998). The area under the whiteline peaks was determined using the Gaussian peak fitting function of Athena (Ravel and Newville, 2005) (Fig. S1 in the Supplement). Because of the relatively low contribution of other sulfur oxidation states, only sulfur species containing S+VI oxidation state could be further characterized via linear combination fitting. Linear combination fitting is an effective tool for the deconvolution of spectra of known mixtures (Longo et al., 2014; Long et al., 2014; Huffman et al., 1991; Solomon et al., 2003; Prietzel et al., 2011). Using Athena software, individual particle and bulk S-NEXFS spectra were fit with previously characterized sulfate standard materials using a linear combination approach to determine both speciation and relative abundance of sulfate phases (Ravel and Newville, 2005) (Fig. S2). Athena uses a nonlinear, least-squares minimization approach to fit spectra of unknown materials with spectra of standard materials and computes an error term, R factor, to quantify the goodness of fit produced by a particular linear combination of standard S-NEXFS spectra. The linear combination of standards that yielded the lowest R factor reflects the best fit (Ravel and Newville, 2005).

Azotobacter vinelandii, Bacillus subtilis, gasoline and diesel exhaust, biomass burning, and coal fly ash were only characterized at the bulk level (approximately 0.28 mm2 filter area). These common primary emission sources, with the exception of one bacteria sample, contained sulfur solely in the S+VI oxidation state. Azotobacter vinelandii was the only emission source to contain both oxidized and reduced sulfur, with approximately 44 % S+VI and 56 % S0 (Fig. 1).

At the bulk level, the oxidation state of sulfur in ambient PM2.5 samples is dominated by S+VI; only two samples from South DeKalb contain S0 at quantities of less than 10 % total sulfur (Fig. 2). In contrast, individual particles with aerodynamic diameters of greater than 1 µm consistently contain both S+VI and S0 (Fig. 3). Only 1 out of 23 individual particles analyzed did not contain S0. At Fire Station 8 and Jefferson Street, the individual particles sampled contain on average 10 % S0. Individual particles from South DeKalb average only 4 % S0. The rural individual particles sampled from Fort Yargo and Yorkville contain on average 5 and 9 % S0, respectively. Multi-energy maps revealed that S+VI was present throughout the aerosol particle. Often the highest concentration of S+VI was in the center of the particle, where the most mass is present (Fig. S3). The spatial distribution of S0 was more varied. In some cases, the S0 was concentrated in the center of the particle with a less prominent ring on the outside of the particle. In other cases, the S0 was most concentrated in only one area of the particle (Fig. S3).

The relative contribution of different sulfur species can be determined by the deconvolution of S-NEXFS spectra with linear combination fitting (Prietzel et al., 2011; Ravel and Newville, 2005). In these samples, only S+VI could be further characterized through linear combination fitting because it was by far the most abundant species of sulfur present in the samples. The regions of the S-NEXFS spectra represented by the lower sulfur oxidation states did not possess sufficient detail to yield compositional information. Because of spectral similarities between various metal sulfates, linear combination fits using copper(II) sulfate or iron(III) sulfate often yielded fits with similar R factors. This makes the absolute determination of metal sulfate composition difficult; thus, iron(III) sulfate, copper(II) sulfate, and jarosite are referred to collectively as metal sulfates. The specific compositional information derived from the best linear combination fits, i.e., iron(III) sulfate, copper(II) sulfate, or jarosite, is presented in Tables S2 and S3, and suggests that a combination of iron and copper sulfates is likely present in these samples. In common primary emission sources and ambient PM2.5, S+VI corresponded to sulfate aerosol. In emission sources, only biomass burning, coal fly ash, and diesel exhaust had robust enough signals to characterize specific sulfate composition, and each source had a unique sulfate composition (Table S3). Biomass burning contained potassium sulfate (100 ± 0.005 %). Coal fly ash contained gypsum (100 ± 0 %), the mineral form of calcium sulfate. Diesel exhaust contained ammonium sulfate (70 ± 11 %) and metal sulfate (30 ± 12 %).

The composition of sulfate was generally consistent within a sampling location (Table S3 and Fig. 4). At the bulk level, rural sampling locations contain a mix of ammonium sulfate and metal sulfate. Yorkville samples are comprised of 61 ± 9 % ammonium sulfate and 39 ± 9 % metal sulfate, and Fort Yargo samples contain 84 ± 7 % ammonium sulfate and 16 ± 7 % metal sulfate. Bulk urban samples all contain a large fraction of ammonium sulfate. Samples from Fire Station 8 contain ammonium sulfate (84 ± 7 %) mixed with metal sulfate (14 ± 16 %), and Jefferson Street contains ammonium sulfate (51 ± 8 %) combined with metal sulfate (50 ± 10 %). South DeKalb has three different compositional groups at bulk level: (1) ammonium sulfate (63 ± 7 %), metal sulfate (29 ± 6 %), and potassium sulfate (8 ± 13 %); (2) ammonium sulfate (42 ± 14 %) and metal sulfate (58 ± 14 %); and (3) gypsum (43 ± 15 %) and metal sulfate (58 ± 15 %).

Only individual particle S-NEXFS spectra from Fire Station 8, Fort Yargo, and South DeKalb contained enough sulfur to determine the sulfate composition (Table S3 and Fig. 4). In total, 17 individual particle S-NEXFS spectra were used to gain insights on the sulfate composition of ambient PM2.5. Consistent with the bulk results, individual particles sampled from Fire Station 8 are largely ammonium sulfate and metal sulfate, with the remainder made of gypsum. Individual particles from South DeKalb contain either ammonium sulfate or metal sulfate. Individual particles from Fort Yargo contain a mixture of mostly ammonium sulfate and potassium sulfate with additional contributions from metal sulfate and gypsum. Spectral signals of samples from Jefferson Street and Yorkville were not strong enough to characterize the composition of sulfate in the ambient PM2.5 samples.