Journal of Geophysical Research: Atmospheres

Size-resolved ultrafine particle composition analysis 1. Atlanta

Authors


Abstract

[1] During August 1999 as part of the Southern Oxidants Study Supersite Experiment, our group collected size-resolved measurements of the chemical composition of single ambient aerosol particles with a unique real-time laser desorption/ionization mass spectrometry technique. The rapid single-particle mass spectrometry instrument is capable of analyzing “ultrafine” particles with aerodynamic diameters ranging from 0.01 to 1.5 μm. Under the heaviest loading observed in Atlanta, particles were analyzed at a rate of roughly one per second in sizes ranging from 0.1 to 0.2 μm. Nearly 16,000 individual spectra were recorded over the course of the month during both daytime and nighttime sampling periods. Evaluation of the data indicates that the composition of the ultrafine (less than 100 nm) particles is dominated by carbon-containing compounds. Larger particles show varied compositions but typically appeared to have organic carbon characteristics mixed with an inorganic component (e.g., crustal materials, metals, etc.). During the experiment, 70 composition classes were identified. In this paper we report the average spectra and correlations with various meteorological parameters for all major compound classes and a number of minor ones. The major composition classes are identified from the primary peaks in their spectra as organic carbon (about 74% of the particles), potassium (8%), iron (3%), calcium (2%), nitrate (2%), elemental carbon (1.5%), and sodium (1%). Many of these compound classes appeared in repeatable size ranges and quadrants of the wind rose, indicating emission from specific sources.

1. Introduction

[2] Evidence continues build supporting the link between fine particles in the atmosphere and adverse health effects (see Pope [2001] for a review). Nevertheless, the physiological mechanisms are not known and therefore the physical and chemical properties of particles that cause these effects have not been identified. One impediment to identifying the particle properties that cause these effects is limited observational techniques that are available to determine the size and composition of particles. This is exacerbated by the relatively small mass contained in the small particle sizes that have been implicated.

[3] A related problem is identifying the source of particles in the atmosphere. Particles are emitted by a wide range of sources, become mixed in the atmosphere and undergo photochemical processing that adds secondary components. To develop source-receptor relationships, these particles are typically collected on filters or impactor stages. This mixture of particles is analyzed and statistical techniques are employed to numerically “unmix” them, thereby identifying their source.

[4] Over the past decade a number of groups worldwide have developed instruments that overcome many of the difficulties associated with the more conventional particle sampling techniques. These techniques, often termed single particle analysis or aerosol mass spectrometry, size and chemically analyze individual particles in real time [Johnston and Wexler, 1995; Johnston, 2000; Wexler and Johnston, 2001; Suess and Prather, 1999]. Each technique has its strengths and weaknesses. During the Atlanta Supersite measurement campaign, four of these instruments were operated simultaneously.

[5] The measurements that we report here were taken with an instrument called RSMS-II, in that it is the second generation of Rapid Single-Particle Mass Spectrometers to be developed in our group. The design and laboratory testing of RSMS-II have been described elsewhere [Mallina et al., 2000; Phares et al., 2001, 2003]. RSMS-II is distinguished from the other single particle instruments deployed in this campaign by its ability to analyze particles as small as a few tens of nanometers in diameter.

[6] This paper begins by briefly describing the function of RSMS-II and the particle classification algorithm employed. Then the measurement results are discussed in terms of the prominent particle classes, their sources, and atmospheric relevance, and some classes that are less frequent but have interesting sources or atmospheric relevance. Finally, the significance of the results are discussed.

2. Measurement Technique

[7] The Atlanta Supersite was the first field experience for a new instrument developed by us called RSMS-II (Rapid Single-Particle Mass Spectrometry, version 2). Unlike the other three single-particle class instruments at Atlanta [Lee et al., 2003; Liu et al., 2002; J. L. Jimenez and J. T. Jayne, Ambient aerosol sampling using an aerosol mass spectrometer in Atlanta and Houston, submitted to Journal of Geophysical Research, 2002], RSMS-II was able to analyze particles too small to be detected by light scattering and was more sensitive than thermal desorption techniques so able to analyze very small particles. RSMS-II design and its testing in laboratory settings have been reported elsewhere [Ge et al., 1998; Mallina et al., 2000; Phares et al., 2002, 2003], so will only be reviewed here briefly.

[8] The instrument was housed in a trailer at the site. Air was sampled through a 7 m long duct that extended 2 m above the roof of the trailer and about 5 m above the ground. The duct was a 10 cm diameter stainless steel tube and air was drawn through it at a volume flow rate of 110 L/m yielding laminar flow (Re = 1500, residence time = 32 s) to minimize losses. The particles were sampled from the center of the duct and the majority of the flow was discarded. The air then passed through 2 m of 1cm O.D. conductive tubing and entered a Nafion™ drier (Perma Pure, Inc., model PD-100T-12 SS) where the humidity was reduced to below 20%.

[9] The sample air now enters the instrument itself, which is illustrated in Figure 1. The inlet sections select a narrow particle size range for analysis, transmit these particles to the source region of the mass spectrometer, and remove the air. The air first entered a computer-controlled 10-position rotary valve (Valco Intruments Co., Valve Type MW), which sends the airflow to a bank of 10 critical orifices (O'Keefe Controls Co., Type N, see Table 1 for sizes). Each critical orifice limits the air flow into the instrument thereby controlling the inlet pressure. Critical orifices provide a stable airflow control while keeping particles losses to a minimum. The inlet pressure is controlled because the pressure is used to select the particle size that is sampled.

Figure 1.

Schematic of the RSMS-II instrument as configured for the Atlanta Supersite experiment.

Table 1. Orifice Size, Inlet Pressure, and Particle Diameter Relationship
Particle Diameter, nmPressure, TorrOrifice Diameter, μm
141.1102
302.3127
473.6178
594.5203
1058.0254
15311.7330
20415.7381
25419.7432
35428.1508
49140.8635
82381.8889
995115.41067
1285229.51600

[10] After the critical orifices, a section of pipe allows the flow to stabilize and become laminar. The pressure is sensed in this section of pipe. The flow then passes through a 3 mm orifice 1.5 mm thick that obstructs a 38 mm diameter pipe. This orifice focuses particles of a given size depending on the upstream pressure that was selected by the bank of critical orifices. An approximate relationship between pressure and diameter is

equation image

where Dp is the particle diameter that is focused and λ is the mean free path evaluated at the temperature and pressure of the choked orifice, roughly 0.53 times the upstream pressure and 0.83 times the upstream temperature. The maximum particles size that can be focused in this way is Dp,max = (18μDn StkfpUsonic)1/2, where ρp is the particle density, the velocity through the orifice, Usonic, is sonic since the flow is choked, μ is the viscosity of air, Stkf = 2 is the Stokes number that is focused [Phares et al., 2001], and Dn is the orifice diameter; μ and Usonic must also be evaluated at the orifice choke conditions. Table 1 shows the orifice sizes employed during the Atlanta Supersite experiment along with the corresponding inlet pressures and nominally focused particle aerodynamic diameter. Note that the particle sizes are their dry aerodynamic diameters because the air stream and particles are dried before they reach the focusing orifice.

[11] The 3 mm critical orifice focuses particles that have Stokes number near 2 to a beam that then passed through three subsequent 1 mm orifices separating skimmer stages. Two vacuum pumps (Varian, SD-451, 450 L/m) draw away the gas and unfocussed particles after the first stage, and single pumps draw on the two subsequent stages. A turbo pump (Varian, V550, 550 L/s) evacuates the source region of the mass spectrometer.

[12] The particles are analyzed by laser desorption/ionization (LDI). An Ar/F excimer laser beam (MPB PSX-100 operating at 193 nm) is oriented parallel and counter propagating with the particle beam. If a particle is in the source region of the mass spectrometer when the laser fires, the particle is ablated and ionized, and the ions are analyzed in the mass spectrometer. The laser was typically fired at a rate of 30 Hz, emitting 2.0 mJ/pulse that was focused to a 0.6 mm spot yielding a laser irradiance of 7.0 × 103 J/m2 at the focus.

[13] The linear mass spectrometer uses a dual gradient to accelerate the ions, which were detected by a dual microchannel plate. Table 2 lists the voltages and plate spacings employed in the mass spectrometer. The dual gradient configuration was employed to ease the conversion of the time-based ion current to a mass/charge-based ion current.

Table 2. Mass Spectrometer Specifications
PlatePosition, cmVoltage, kV
Backing0+5.0
First accelerator1.67+2.7
Second accelerator2.67Gnd
Front microchannel117−2.2
Back microchannel118.5−1.2

[14] The ion current was digitized by a custom 500 MHz A/D converter (Precision Instruments, model 9847). The laser sync out triggers the A/D converter, but most of the laser shots miss a particle. The A/D converter is programmed to scan acquired spectra for peaks that exceed a desired level. If this level is exceeded, a particle was hit and the spectra is acquired into memory and stored on disk by LabView™ software (National Instruments).

3. Measurement Results

[15] Figure 2 shows a timeline of the Atlanta Supersite project period illustrating the time periods where the instrument was acquiring data. The instrument was run before August 10, but a single voltage gradient was employed in the mass spectrometer making data inversion highly uncertain. The instrument was run during the daytime, except when evening runs were coordinated with other measurement equipment. The instrument could not be run continuously due to overheating problems associated with the mechanical pumps. Overall, particles were sampled for about 120 hours out of the 20-day period, and spectra from 15,989 particles were obtained. The particle aerodynamic diameter, defined by equation (1), ranged from about 15 nm to over 1 μm.

Figure 2.

Timeline of the measurements. Each dot represents the time and size of an acquired spectrum.

[16] During the Atlanta Supersite experiment, as in other locations where single particle analysis instruments have been deployed, the particle spectra fall in composition classes. That is, some of the particles have similar spectra, indicating that they derive from the same or similar sources, and have undergone similar atmospheric transformations before they were detected. Thus it is instructive when interpreting the often massive amount of information obtained from single particle instruments to classify the spectra first.

[17] The Atlanta Supersite spectra were classified with the ART-2a algorithm [Carpenter et al., 1991; Hopke and Song, 1997; Song et al., 1999], which was recently validated with spectra generated LDI/MS (laser desorption/ionization mass spectrometry) from laboratory generated composition standards [Phares et al., 2001]. In brief, the algorithm starts by taking the first spectrum as a seed for the first composition class. The dot product is taken between this particle and the seed, and if the dot product is greater than an arbitrary value, called the vigilance factor, the particle is added to that composition class. If the dot product is less than the vigilance factor, a new class is formed. A second arbitrary factor, the learning rate, is used to adjust the class spectrum to reflect new membership. As determined by laboratory experiments, the vigilance factor was 0.4, the learning rate was 0.05, and 20 iterations were employed to confirm that convergence was obtained [Phares et al., 2001].

[18] During the Atlanta Supersite measurements, over 70 composition classes were categorized, but only 40 composition classes were identified that contained over 5 members, lending them some statistical significance. Of these, 9 major composition classes were found, arbitrarily defined as having more than 1% of the total number of particles observed. Table 3 shows the number of particles measured in each compound class, for classes with 5 or more spectra. The vast majority of the particles were classified as organic carbon, which will be discussed in more depth later. Note that even if a particular compound class comprises only a small percentage of the particles observed, it may comprise a substantial fraction or even a majority of the particles for a given time and particle size. In this work, the intermittent nature of the instrument operation precluded this plume or puff identification, but in subsequent work this identification was more successful [Phares et al., 2001].

Table 3. Number of Spectra Obtained in Each Compound Class
Compound ClassGroupingNumber of Spectraa
  • a

    Total number of spectra collected was 15,989.

Organic carboncarbonaceous11828
Potassiummineral1240
Ironmetal380
Potassium/calciummineral315
Nitratenitrate274
Elemental carboncarbonaceous231
Sodium/potassiummineral209
Aromatic hydrocarboncarbonaceous208
Calciummineral160
Organic carbon 2carbonaceous130
Iron 2metal100
Calcium 2mineral96
Aluminummetal80
Iron 3metal63
Nickelmetal62
Sodiummineral57
Leadmetal57
Tin/antimony/leadmetal47
Zincmetal39
Silicon oxidesilicon oxide35
Coppermetal34
Chromiummetal30
Unidentifiedunidentified29
Unidentifiedunidentified28
Unidentifiedunidentified26
Nitrate 2nitrate24
Oxygenated hydrocarboncarbonaceous20
Nitrate 3nitrate20
Aluminum oxidemetal19
Silicon oxide 2silicon oxide13
Unidentifiedunidentified12
Silicon oxide/nitratesilicon oxide11
Elemental carbon 2carbonaceous10
Unidentifiedunidentified8
Unidentifiedunidentified7
Tinmetal7
Leadmetal7
Sodium 2mineral6
Chromiummetal6
Mercurymetal5

[19] The 40 compound classes with over 5 spectra were classified into 7 groups corresponding to hydrocarbons, minerals, nitrates, iron, silicon dioxide, other metals, and unidentified. Spectra in the carbonaceous group are dominated by CxHy peaks characteristic of particulate organic and elemental carbon. Spectra in the mineral group have predominant Na, K, and Ca peaks. Nitrate is present in a range of compound classes and indicates both organic and inorganic nitrate. Silicon and its oxides are present in a few compound classes and is primarily associated with crustal material. Finally, a number of compound classes could not be identified. Table 3 summarizes the compound classes and groupings.

[20] Figure 3 summarizes the wind direction and relative size distribution of these compound class groupings. Notice that carbonaceous PM dominates the observations. The majority of spectra were hydrocarbon almost regardless of the particle size or wind direction. Metal and mineral particles derived from the southeasterly direction in substantial numbers, especially those smaller than about 100 nm. Similarly, a substantial fraction of the particles deriving from the northeasterly direction were identified as iron and mineral, especially in the 100 to 300 nm size range. Particles in larger size ranges, those over about 300 nm, were much less likely to be hydrocarbon-derived. Nitrate particles were quite prevalent, along with mineral particles presumably of crustal origin.

Figure 3.

Relative prevalence of the compound class groupings as a function of wind direction and particle size.

[21] In what follows are discussions about the sources and significance of each of the composition classes. Figure 4 illustrates the overall distribution of the spectra obtained; this is not necessarily indicative of the overall particle burden in Atlanta because the instrument was run infrequently at night and not uniformly throughout the day, and the instrument transmission efficiency is a function of particle size. Notice in the upper left panel that the particle hit rate correlated with elevated CO concentrations indicating primary source, combustion-oriented emissions dominate the mix of particles detected at the site. The upper right panel shows that the instrument sampled particles most efficiently in the 100 to 300 nm size range, although substantial numbers of particles were obtained at other sizes. The hit rate is lower for smaller particles because of the smaller amount of analyte available for analysis. Kane and Johnston [2000] have shown that analysis of particles is less efficient as the particles become smaller, for those smaller than about 100 nm, with the hit rate increasingly dependent on the particle composition as the particle size decreases.

Figure 4.

Overall distribution of spectra obtained during the experiment. Upper left panel: correlation of spectra with CO concentration. Upper right panel: overall size distribution. Lower left panel: diurnal distribution of measurements. Lower right panel: wind rose of the measurements. Note that since the instrument was not operated evenly throughout the day and has a transmission efficiency that is a function of particle size, the fractions shown here do not represent the fractional amounts in the atmosphere, but rather the fractional amounts that were measured.

[22] The overheating problems forced the instrument to be attended continuously, which limited measurements in the nighttime hours. The lower left panel shows the prevalence of measurements over the time of day. Finally, the lower right panel shows the correlation of the particles acquired to the wind speed over the course of the measurements. The wind derived primarily from westerly and northwesterly directions although a large number of particles also came from the east.

[23] Figures 514 summarize the findings for each composition class. Each figure contains 6 graphs that elucidate a different feature of the composition class. The title is our identifier for the compound class and the percentage is the fraction of spectra identified with this class. The large graph showing Normalized Peak Area as a function of Mass/Charge is the average spectrum for the composition class, which are called the generalized weights by the ART-2a neural network community. The table located below the title within this graph contains the weights for each mass/charge ratio, and their mass and ion assignment, and are a quantitative representation of what the graph represents qualitatively. Their sum of squares of the area is defined as one and the magnitude indicates the relative peak area such that larger values correspond to larger ion current in the spectra. Note that this does not necessarily translate to a larger quantity of the compound in the particle because the ionization efficiencies for each compound may differ greatly.

Figure 5.

Summary statistics for the organic carbon compound class. See text for more details. Large graph: average mass spectrum for this class. Table: peak areas for each m/z. Size versus sample time: the periods of time when this compound class was observed. Polar: Wind rose showing directional preference for this class. Lower left histogram: fraction of particles in this class by time of day. Lower middle histogram: fraction of particles in this class by CO concentration. Left middle histogram: fraction of particles in this class by particle size.

Figure 6.

Summary statistics for the potassium compound class. See caption of Figure 5 and the text for a detailed description of each panel.

Figure 7.

Summary statistics for the iron compound class. See caption of Figure 5 and the text for a detailed description of each panel.

Figure 8.

Summary statistics for the lead compound class. See caption of Figure 5 and the text for a detailed description of each panel.

Figure 9.

Summary statistics for the tin/antimony compound class. See caption of Figure 5 and the text for a detailed description of each panel.

Figure 10.

Summary statistics for the iron 2 compound class. See caption of Figure 5 and the text for a detailed description of each panel.

Figure 11.

Summary statistics for the calcium compound class. See caption of Figure 5 and the text for a detailed description of each panel.

Figure 12.

Summary statistics for the nitrate compound class. See caption of Figure 5 and the text for a detailed description of each panel.

Figure 13.

Summary statistics for the elemental carbon compound class. See caption of Figure 5 and the text for a detailed description of each panel.

Figure 14.

Summary statistics for the sodium/potassium compound class. See caption of Figure 5 and the text for a detailed description of each panel.

[24] Within the axes of this graph are two others. The upper one shows when and at what sizes particles of this class were obtained; each dot corresponds to one spectrum. The lower one is a wind rose and illustrates the fraction of the total spectra that are in this composition and derive from a given petal of the wind rose. Note that the fraction in each wind rose is the fraction of total particles observed in that petal, so an even distribution implies that there was not a significant directional dependence of the particles observed, as in Figure 5. Spectra were only classified into a wind rose petal if the wind speed was over 1.0 m/s.

[25] Below the graph of the spectrum are three histograms. The left most shows the fraction of particles in this class by time of day, for all days during the measurements. Again, these are normalized to the total number of particles observed at that time of day so a relatively even distribution, as in Figure 5, means that these particles were as likely to observed any time of the day. The middle histogram shows the correlation with carbon monoxide concentration, a gas phase tracer for certain combustion sources, and again is normalized so that correlations with CO concentration are evident if they occur. The right most histogram shows the size distribution of the spectra, and it too is normalized by the number of particles observed at each size. In the discussion to follow, we will summarize the findings for each of the 10 major classes (those with more than 1% of the total particles sampled) and also discuss a selection of the more minor classes that were observed. The rest occur in either a few instances indicating the occasional passage of a plume from a unique source or in a narrow size range indicating a more prevalent source but at only one size.

3.1. Organic Carbon

[26] Figure 5 summarizes the data for spectra dominated by carbon peaks. These spectra are characteristic of particles dominated by organic carbon content and often occur in laboratory tests for a wide range of organic compounds [Kane and Johnston, 2000; Phares et al., 2001]. As with most carbonaceous particles, this class is enriched in trace components in fossil fuels such as Si and K. Although we observed these spectra at nearly all particle sizes, there is a tendency to observe a higher fraction of these particles at smaller particle sizes, consistent with the source being high temperature combustion. A somewhat lower fraction of particles are carbon when the wind is from the east, and a somewhat higher fraction when the wind is from the west. Note that downtown Atlanta is to the east of the site, the suburbs are to the west, and railroad lines surround it. About 74% of the particles that we observed fell into this composition class. That carbonaceous particulate matter dominates the mix of particle types in Atlanta as has been determined by others [Butler et al., 2003].

3.2. Potassium

[27] Biomass is burned as part of crop management and power production. Figure 6 summarizes the data for the spectra that most likely derive from biomass burning because of the dominance of their potassium peak. Potassium has a very high ionization efficiency compared to other compounds typically found in the atmosphere so the large peak does not necessary indicate a preponderance of potassium in the particles. About 8% of the particles were in this class. Unlike soot, the size distribution is confined primarily to sizes between about 100 and 500 nm, where nearly 10% of the particles are biomass. There is a strong tendency to observe these particles in the late evening hours and their sources primarily lie to the north and east of the site, toward downtown.

3.3. Iron

[28] The site was located within a few kilometers of a metal shredding/recycling plant to the east. Another few kilometers further east is downtown Atlanta. Most of the iron-containing particles were found between 100 and 400 nm in diameter in the wind roses indicating a source to the east, and were found throughout the day but mostly in early morning hours (Figure 7). As expected, most of these particles are from the east indicating the metal shredder, but the size distribution is probably not indicative of the shredder, which is expected to produce rather coarse particles not the size range observed. Similar particle size, wind-direction, and time-of-day patterns were observed for some of the other metal-dominated composition classes such as lead (Figure 8), and tin/antimony (Figure 9). The other iron-dominated spectra, iron 2 (Figure 10) and iron 3, demonstrated similar time-of-day and wind-direction patterns but their size distribution is coarser and more indicative of what we anticipated from the shredder. The differences in the spectra between the iron (Figure 7), iron 2 (Figure 10) and iron 3 are relatively minor and mostly based on the presence or absence of sodium peaks at m/z = 23. The correlation with time of day may be due to inversion trapping and concentrating of the emissions near the site. The anti-correlation with CO may simply indicate its lower emission level during the hours when the particles were observed.

3.4. Tin-Antimony Spectra

[29] Three instances, one major plume and two more smaller occurrences, were observed where particles contained a mixture primarily composed of tin and antimony, with some trace amounts of lead and iron (Figure 9). The size distribution was narrowly centered near 200 nm. The small size of these particles indicates a high temperature source, since the boiling point for antimony is 1587 C while that for tin is 2603 C. The source of this metal mixture is probably the alloy called Babbitt, which is used to coat bearing surfaces because it entrains particles and deforms to irregularities in the bearing surface [Avallone and Baumeister, 1978]. Most likely, these particles are derived from failing, overheating bearings in equipment near the site. The easterly source direction also indicates possible bearings in the metal shredder, or a source from downtown, which would most likely indicate machinery associated with transportation.

3.5. Calcium

[30] Figure 11 shows spectra that are likely derived from crustal particles. The spectra are dominated by calcium and the size distribution shows that this is probably the tail of a coarser mode in the size distribution. These particles are observed in all petals of the wind rose but predominantly come from the north and easterly directions, a very similar pattern to that of the potassium compound class.

3.6. Nitrate

[31] Ammonium nitrate and organic nitrate containing particles are known to comprise a substantial fraction of the particulate mass in many urban environments. The organic carbon compound class presented a significant NO+ peak indicative of nitrate. In addition about 1.7% of the spectra were dominated by this NO+ peak. Figure 12 shows spectra from particles that are most likely composed of ammonium nitrate. The NO peak is indicative of nitrate and ammonium peaks are not only present in the spectra but are the predominant cation in particles observed far from the marine sources of sodium in sea salt. These particles are predominantly from westerly directions. The particles were observed mostly in the 0.3 to 1 μm size range and contain a substantial fraction of carbon. Presumably, these particles were emitted as soot some distance from the site. During their transport to Atlanta, ammonia was emitted from the surface and nitric acid was formed by gas-phase photochemistry so that ammonium nitrate condensed onto the pre-existing soot particles causing them to grow into this size range [Wexler and Seinfeld, 1990; Kerminen and Wexler, 1995].

3.7. Elemental Carbon

[32] Two carbonaceous classes exhibit high m/z ions. These classes have nearly equal preponderance (1.4% and 1.3%) but quite different size distributions with the more common class (Figure 13) occurring in a size range form about 100 to 500 nm while the less common class is mostly observed in particle sizes near 1 μm and above. The class shown in Figure 13 contains many high m/z carbon cluster ions that are characteristic of elemental carbon. The other class is more characteristic of aromatic hydrocarbons and contains ions at m/z 92, 106, 118 and 128 in addition to low m/z carbon clusters. Both classes are observed in all petals of the wind rose with a higher likelihood of observance from the south. To the south of the site is a Greyhound Bus repair depot and buses were observed idling there during many occasions.

3.8. Sodium

[33] Two compound classes possessed spectra that were dominated by their sodium peaks. The first was present in 1.3% of the particles (Figure 14) and also has a significant potassium peak, while the second was much less prevalent (0.3%) and had little besides sodium in the spectrum. Both spectra derive from easterly petals of the wind rose and correlate with somewhat lower concentrations of CO, indicating sea salt as the likely source. The particle sizes in the more common class are relatively fine since coarser particles should have been removed during the transit to Atlanta, which is about 200 miles from the coast.

4. Conclusions

[34] Using RSMS-II, over 15,000 individual particles were analyzed for their composition and size as part of the Atlanta Supersite experiment. Particles were sampled from 14 nm to 1.3 μm and categorized by their spectra into 70 compound classes, comprising organic compounds, minerals, metals and inorganics. The strengths of single particle measurement is lack of blank and artifact contamination because there are no substrates and the particles are sampled in situ. RSMS-II is specifically designed to sample the fine, ultrafine, and nanoparticle size ranges where size differentiated composition determination is particularly challenging.

[35] As has been determined by others, Atlanta's PM is predominantly carbonaceous in nature [Lee et al., 2002; Butler et al., 2003]. We found primarily carbonaceous particles throughout the size spectrum, time of day, and wind rose, even for particles as small as 14 nm. Fossil fuel combustion in many sectors of the local economy is probably responsible. But a substantial fraction of the ambient particles were composed of metals, minerals and other inorganic compounds that are more conducive to identification with the LDI/MS technique employed by RSMS-II. These particles were classified by the similarity of their spectra and found to have characteristic source directions and particle size distributions. Source apportionment has been performed on bulk samples but is fraught with difficulties including colinearities that impede identification of the source-receptor relationship. By analyzing individual particles, these colinearities are eliminated because the particles from disparate sources are not mixed together before the analysis is performed.

[36] Finally, increasing evidence implicates the particles smaller than 1 μm as responsible for the epidemiologically observed health effects of PM. The particles that are most likely to pose a health effect are those that are transmitted to the sensitive pulmonary regions of the human airway. These are the particles with small deposition fractions in the upper airways and lie between about 100 and 1000 nm. These are the particles retained in the airways after a complete breathing cycle and thus it is these particles that may be transported to the pulmonary airways on subsequent breaths [Sarangapani and Wexler, 2000]. RSMS-II is especially suited to sizing and analyzing these particles, and therefore helping close the link between ambient PM and human health effects.

Acknowledgments

[37] This work was supported by funding from the U.S. Environmental Protection Agency via STAR grant R826234-01-0 and the Atlanta Supersite.

Ancillary