In August 1999 an intensive field campaign was conducted in Atlanta, Georgia, focusing on the characterization of urban particulate matter. During the study an aerosol time-of-flight mass spectrometer (ATOFMS) was used to measure continuously the aerodynamic size and chemical composition of individual particles in the fine fraction (0.2–2.5 μm) of the atmospheric aerosol. The inorganic and organic components of the particles were analyzed using laser desorption ionization time-of-flight mass spectrometry, generating positive and/or negative ion mass spectra. Here an overview of the ATOFMS results is presented with respect to the major species detected, including sodium, carbon (EC, OC, EC/OC), dust (Li, Na, Al, K, Ca, Fe), sulfate, nitrate, and ammonium. As described, many of the dust particle types have similar composition to those observed in ATOFMS coal source characterization studies. The ion signals, size distributions, and temporal evolution (30–60 min resolution) of the different particle types are described. The complexity of the Atlanta aerosol is shown here on an individual particle basis, demonstrating how single particle data obtained with ATOFMS can be used to gain unique insights into the mixing state of urban aerosols. The second paper in this two part series focuses on the scaling procedures used for converting the unscaled ATOFMS data presented in this paper into atmospherically representative number concentrations [Wenzel et al., 2002].
 Recently, due to the growing concern over the inhalation and potential health hazards of airborne particles, a number of field studies have been conducted in different locations in the United States. A major goal of these campaigns involves characterization of the sources and types of fine particles, while providing more complete data for toxicological studies. With similar goals in mind, the nation's first Supersite was established in Atlanta, Georgia. The interest in the metropolitan Atlanta area comes mainly from its high pollution levels. In 1999, Atlanta failed to attain the air quality index (for ozone and particulate matter) on sixty-nine days in the spring and summer seasons. Also, particulate matter in eastern regions of the United States is far less characterized than that in the western United States. The main distinguishable attribute of the eastern United States is the presence of higher aerosol sulfate concentrations, primarily attributed to elevated SO2 emissions by coal fired power plants [Matyniak, 1989].
 While previous studies have been performed to monitor ozone and its effects in Atlanta, relatively little research had been performed concerning particulate matter (PM). The characterization of PM represents a growing concern particularly in regards to reaching compliance with the EPA's new PM2.5 standard [Lindsay et al., 1989]. Atlanta was chosen as an ideal site for the EPA study because relatively little data had previously been collected pertaining to aerosols in eastern regions of the United States. Two major goals of the Atlanta Supersite experiment were: (1) to respond to the need for a better understanding of particulate matter (PM) health effects and aerosol sources; (2) to test state-of-the-art particle measurement instruments, and compare their results to those obtained with more traditional techniques. As part of this comparison, a number of real-time single particle mass spectrometers were operated simultaneously [Middlebroook et al., 2003]. The instrument discussed herein is an aerosol time-of-flight mass spectrometer (ATOFMS) which was operated continuously in Atlanta for four weeks in August of 1999. ATOFMS couples time-of-flight aerodynamic sizing with laser desorption ionization (LDI) time-of-flight mass spectrometry, simultaneously measuring the aerodynamic size and chemical composition (as anions and cations) of individual aerosol particles in real time [Gard et al., 1997; Prather et al., 1994].
 Herein, an overview is given on the ATOFMS findings for PM in the Atlanta atmosphere over the entire Supersite field campaign. Specifically, the major particle types detected in the area are described, including sodium, dust (Li, Na, Al, K, Ca, Fe), carbon particles (OC, EC), as well as the associations with sulfate, nitrate, and ammonium. The ability of ATOFMS to collect single particle data allows for determination of the major chemical species, as well as the various, and often complex, chemical associations. Information can be obtained on internally and externally mixed particles and aerosol primary and secondary species, information that is essential for developing an understanding of the formation and evolution processes of ambient airborne matter.
 Using the ATOFMS mass spectral markers detected in Atlanta, particles were classified into seven major chemical classes: (1) sodium-containing, (2) sodium-containing with secondary species (ammonium, nitrate, sulfate), (3) carbonaceous, (4) carbonaceous with secondary species (ammonium, nitrate, sulfate), (5) dust, (6) dust with secondary species (ammonium, nitrate, sulfate), and (7) secondary (ammonium, sulfate, nitrate). ATOFMS information on aerodynamic particle size, chemical composition, and the temporal variations of different chemical species constitute three essential dimensions necessary for deciphering the convoluted aerosol chemistry in an urban atmosphere.
2.1. Single Particle Analysis: ATOFMS
 Ambient single-particle mass spectra were obtained using a transportable dual ion version of the ATOFMS [Gard et al., 1997]. Briefly, the transportable ATOFMS operates by first drawing polydisperse particles into the instrument through a converging nozzle. Next the particles pass through two skimmers in differentially pumped regions where much of the gas phase is removed, leaving a beam of particles passing directly through the center of the instrument. The particles then enter the light scattering region of the ATOFMS, where they cross two diode pumped Nd:YAG lasers operated in continuous mode (λ = 532 nm) and oriented at 90 degrees to one another. Upon passing through each laser beam, the particles scatter light. The scattered light signals are detected using photomultiplier tubes (PMT). From the time difference between the two detected scattering signals, one can determine the aerodynamic diameters of the particles using an external size calibration curve created using polystyrene latex spheres of known size. When the sized particle enters the mass spectrometer region of the ATOFMS, a pulse from a Nd:YAG laser (λ = 266 nm) is fired at the appropriate time based on the transit time of the particle measured in the light scattering region of the instrument. This laser pulse desorbs and ionizes species from the particle, producing both positive and negative ions, which are simultaneously measured using a dual ion time-of-flight mass spectrometer. For the duration of the study, the power of the Nd:YAG was approximately 108 W/cm2.
 ATOFMS particle detection efficiencies vary for particles with different sizes, decreasing as a function of particle size. This paper shows the unscaled ATOFMS data in the format it was acquired in the field, demonstrating how instantaneous information can be acquired on PM trends and episodes. However, as described in the second paper in this series, ATOFMS particle counts can be scaled to atmospherically representative concentrations using the particle concentrations measured with a laser particle counter (LPC) operating at the same site during the field campaign. The ATOFMS operated almost continuously at the Atlanta Supersite from 8/6/99–9/1/99.
2.2. ATOFMS Data Analysis Program
 The mass spectra of 455,444 particles were acquired during the Atlanta Supersite Experiment. The aerodynamic size and the positive and/or negative ion mass spectra were recorded for each particle. All single particle mass spectra acquired were converted to a list of peaks at each mass/charge (m/z) using a minimum signal threshold of peak height 10 and peak area 20 with in-house software. The resulting peak lists were then imported into Microsoft Access 97 and Matlab version 126.96.36.199 R12 with YAADA (www.yaada.org) 1.0 (The Mathworks, Natick, MA). With the exception of differentiating between EC and OC and the last section on aerosol sources, all data presented in this paper resulted from searching through the established database for various m/z combinations to find the particle types of interest.
 The primary data analysis approach used in this paper is based on digital histograms (DIGHIS), used to summarize the mass spectra of a large number of particles [Liu, 2000]. With this technique, each single particle mass spectrum is first converted to a digital representation with unit mass resolution, where peaks above a given threshold are counted as “1's” and those below the threshold as “0's.” The “1's” and “0's” from multiple mass spectra are then summed to form a DIGHIS spectrum. As shown, this technique is useful for identifying m/z combinations and particle types.
3. Results and Discussion
 The particles were divided into general particle types based on rules used to search the database. The exact rules given for the particle types and secondary associations discussed in the following sections are provided in Table 1. The major chemical species constituting the aerosol in Atlanta can be determined by ATOFMS and tracked over time with 30–60 min resolution depending on PM concentrations. Moreover, individual particle analysis allows one to find associations between chemical species, and thus select search criteria to identify specific particle types. The detected particles are classified into exclusive classes and the ambient aerosol composition can be evaluated on the basis of percentage of particles belonging to each class for a certain size range, over a specified time period.
Table 1. Summary of the ATOFMS Markers Used to Search for the Major Species Observed in the Atlanta Data Set
Dust and sodium classes exclude particles containing m/z 36 (C3+).
 The major components detected in the Atlanta aerosol were: sodium and dust (primary species); nitrate, ammonium, and sulfate (secondary species); and carbonaceous matter. Carbon species occur in particles emitted directly from primary emission processes, as well as secondary processes (gas-to-particle conversion). Consideration of all possible combinations of these six major species to determine the particle categories results in an extremely large number of classes. Thus, in order to simplify the classification, three primary types (sodium, dust, carbon) were chosen and the label “+ secondary” was attached to these classes whenever secondary (ammonium, nitrate, sulfate) species were present in the same mass spectrum. If a mass spectrum does not contain any primary species, the particle is classified as pure secondary. Whenever a particle contains species that fall into multiple primary classes, a hierarchy of rules must be decided upon in order to assign each particle to only one exclusive class. The rules used herein are the following: (1) whenever a particle contains ion signals due to dust signatures, it is classified as dust, (2) of the remaining particles, any containing a signal due to sodium are classified as sodium, and (3) from the remaining particles, any particle with signal due to carbon is classified as carbon. In order to properly classify the particle types, a rule was added to the dust and sodium classes, excluding those particles with m/z 36 (C3+). This rule was necessary to insure organic particles with complex spectra were not erroneously classified as sodium or dust containing. An important thing to note is this rule results in organic-containing dust and sodium particles being classified as carbon. The implications of this will be further discussed. Using these rules, the final particle types are: (1) sodium-containing, (2) sodium-containing with secondary (ammonium, nitrate, or sulfate), (3) carbonaceous, (4) carbonaceous with secondary (ammonium, nitrate, or sulfate), (5) dust, (6) dust with secondary (ammonium, nitrate, or sulfate), and (7) secondary only (ammonium, nitrate, or sulfate). It must be noted that the classification is based only on the presence or absence of the marker m/z signals to qualitatively describe the aerosol composition and does not refer to the quantity of the chemical species. Efforts are underway to determine the quantities of various species in collections of particles of various types and will be the subject of future papers [Bhave et al., 2002].
3.1. Aerosol Chemical Composition: Primary Particle Types
Figure 1 shows the DIGHIS for the 3 major classes (dust, sodium/no dust, carbon-containing) detected in Atlanta along with the corresponding temporal profiles in 30 min temporal resolution (1c, 1f, 1i) for each of these types over the duration of the study. A discussion of each of the primary types is given in this section. Table 1 provides the rules used to search for each of these particle types. Table 2 shows a summary of the combinations of major particle types with secondary species
Table 2. Summary of Percentage of Secondary Species Detected on Each Primary Particle Typea
Only particles producing both positive and negative spectra are included in the analysis.
 As shown in Table 1, to search for all dust particles, the following set of rules was used: the presence of signal with absolute area ≥1500 in the m/z range 25.5–28 (due to Al), or in the m/z range 54–57 (due to Fe or CaO), or in the m/z range 1.2–2.2 (H+), or m/z 40 ± 0.5 (Ca) with absolute area ≥2000, or signal with absolute area ≥200 in the m/z range 5–9 (Li), or m/z 96 ± 0.5 (Ca2O) with absolute area 100 and the absence of m/z 72 ± 0.5 and in all cases the absence of m/z 36 ± 0.5 signal with relative area larger than 1% (Table 1). The rule excluding m/z 72 ± 0.5 is necessary to find all Ca-rich particles without interference from elemental carbon (i.e., C6+).
Figures 1a and 1b show the positive and negative ion DIGHIS of all dust containing particles. In the positive ion spectra, peaks are present for Li (m/z 7), Na (m/z 23), Mg (m/z 24), Al (m/z 27), K (m/z 39), Ca (m/z 40), Ti (m/z 48), and Fe (m/z 56). In negative polarity for dust particles, one detects carbon (m/z −12, −24), nitrate (m/z −46 and −62), sulfate (m/z −32, −80, −96, and −97), chloride (m/z −35 and −37), silicates (m/z −60 and −76), fluoride (m/z −19), and phosphate (m/z −63 and −79) ions. The common occurrence of many of these peaks (i.e., carbon, nitrates, sulfates) demonstrates how transformed the dust particles are in the Atlanta atmosphere. This finding contrasts a number of other previous ATOFMS field studies where typical dust particle mass spectra show “clean” signatures and contain only metal ions. Figure 1c shows the temporal variations of the dust particle types. The temporal variations of dust particles in Atlanta show rapid spikes with durations of 1–2 hours, usually at night or during the early morning hours, superimposed on a constant background. The rapid nature of these spikes makes their detection difficult using standard PM mass measurement techniques and demonstrates one of the unique insights single particle mass spectrometers can provide in such a study. It is important to note that the temporal profiles of phosphate (m/z −79 [PO3]−), silicate (m/z −76 [SiO3]−), and fluoride (m/z −19 [F]−) typically track these dust particles, an observation that is consistent with other ATOFMS studies [Silva and Prather, 2000; K. R. Coffee et al., manuscript in preparation, 2002].
 The wind direction when the dust particles were observed was highly variable. This is not surprising due to the very general (inclusive) nature of the search criteria used which results in a mixing of multiple dust types from many sources. A more refined data analysis procedure, yielding more distinct dust particle types, allows the wind dependence and possible sources for the temporal spikes to be determined as discussed in the section below on Aerosol Sources.
Figures 2a and 2b show the positive and negative ion mass spectra from one of the more common dust particle types, showing large positive ion peaks at m/z 7, 23, 27, 39, and 56, representing lithium, sodium, aluminum, potassium, and iron. Typical negative ion signals show the presence of m/z −16 [O]− and −17 [OH]−, silicates (m/z −60 [SiO2]−, −76 [SiO3]−, m/z −103 [AlSiO3]−, and m/z −119 [AlSiO4]), and nitrates (m/z −46 [NO2]− and −62 [NO3]−). A distinct calcium-rich particle type was also detected in Atlanta (Figure 2c). ATOFMS mass spectra for this dust type show intense signals due to Ca (40 & 44), m/z 56 [CaO]+, m/z 57 [CaOH]+, m/z 75 [Ca(OH)H2O]+, m/z 93 [Ca(OH) 2H2O]+, m/z 96 [Ca2O]+, m/z 113 [Ca2O2H]+, m/z 131 [Ca2(OH)3]+, m/z 169 [Ca3O3H]+, m/z 187 [Ca3O3H3O]+. Less frequently, Ca-related clusters are observed at higher masses, including m/z 225 and 261, [Ca4O3H]+ and [Ca4O6H5]+, respectively. This Ca-particle type is most likely from calcium carbonate [Bruynseels and Van Grieken, 1983; Silva and Prather, 2000].
 Further classification of the dust particle type can be performed, based on the presence of other metal species in the form of positive ions. For instance, two other relatively common dust types were often detected during the study, one containing titanium with peaks at m/z 48 [Ti]+ and m/z 64 [TiO]+, and the second with intense barium signals at m/z 138 [Ba]+ and m/z 154 [BaO]+ (Figure 2d). While these elements have been detected in single particle spectra from soil dust [Silva and Prather, 2000; K. R. Coffee et al., manuscript in preparation], Li can also be present in coal fly ash [Pougnet et al., 1990], and barium has been detected in brake dust [Silva, 2000]. Moreover, in one short period of time (∼9:30 am on 26 August), the dust particle mass spectra showed signals also due to zinc at m/z 64, 66, 68 that could be from a local industrial or machining processes [Lannefors and Akselsson, 1977; K. A. Coffee et al., manuscript in preparation]. Almost all dust particle mass spectra contain carbonaceous ion peaks (98%) due to organic species naturally present in soils, secondary reactions, or possibly the original source (i.e., coal combustion). Dust particle mass spectra also contain sulfates (61%), nitrates (90%), and ammonium (63%) ion markers, indicating aged dust particles in the Atlanta area (Table 2).
3.1.2. Sodium (No Dust)
 Sodium-containing particles are produced primarily from mechanical generation and thus occur mostly in the coarse size mode. A major source of atmospheric sodium is ocean-spray, which can occur in continental air masses due to long-range transport. In addition to being of marine origin, sodium may occur in re-suspension of sodium-rich dust. The search for all Na-containing particles is performed based on the presence of m/z 23 ± 0.5 signal with relative area (peak area divided by the total area of the mass spectrum) larger or equal to 4% or absolute area of 100 and the absence of m/z 36 ± 0.5 signal with relative area larger than 1%. To distinguish between sodium/dust and sodium/salt particles, it is possible to search for sodium without the ion markers for dust described in the previous section. As shown in the DIGHIS in Figures 1d and 1e, typical sodium/no dust-containing single particle ATOFMS mass spectra present an intense signal at m/z 23, together with other sodium-associated species that vary depending on the age and evolution history of the particles [Gard et al., 1998; Song and Carmichael, 1999; Vogt et al., 1996], including m/z 62 [Na2O]+, m/z 63 [Na2OH]+, m/z 81 [Na2Cl]+, m/z 108 [Na2NO3]+, m/z 165 [Na3SO4]+, m/z −119 [NaSO4]−, and m/z −147 [Na(NO3)2]. The temporal profile for sodium/no dust-containing particles is shown in Figure 1f, with major sodium events occurring on the 13th, 20th, and 31st August. These represent times when sodium type particles correlated with sulfate containing particles, and thus represent transformed salt particles formed through reactions with gaseous SOx species. This particle type shows a high correlation with relative humidity (R2 = 0.74) and periods when the wind was blowing from the south, providing evidence they could be of marine origin. The last episode of Na-containing particles during the study showed sodium sulfate and sodium nitrate rich particles. The mass spectra of sodium-containing particles analyzed in Atlanta typically contained signals from secondary species (sulfate (93%), nitrate (84%), ammonium (26%), and carbon (92%)), indicating secondary reactions with NOx and SOx, as well as adsorbed low volatility and semivolatile organic compounds (Table 2).
 A unique type of sodium-particle mass spectra was observed in Atlanta (Figure 2e) where Na (m/z 23) was detected with water cluster ion signals at m/z 41 [Na H2O]+, m/z 59 [Na (H2O)2]+, m/z 77 [Na (H2O)3]+, m/z 95 [Na (H2O)4]+, and m/z 99 [Na2OH (H2O)2]+. This particle type, in contrast to other salt particles observed during the study showed a very strong wind dependence, originating exclusively from the south. These particles were sampled at night. Most likely the particles were directly transported to the site at night under high relative humidity conditions without the necessary time for full evaporation to occur.
 Combustion processes, including vehicular emissions and coal combustion, produce carbon particles containing carbon in the form of a core of elemental carbon (EC) often coated with semivolatile organic carbon (OC) species which condense from the exhaust gases [Amann and Siegla, 1982]. Organic carbon can be incorporated into the particles as a result of atmospheric photochemical processes producing low and semivolatile carbon compounds [Bowman et al., 1997; Odum et al., 1996]. Summer aerosol samples, relative to winter samples, usually contain a larger fraction of organic matter [Turpin et al., 1991], particularly in the southeastern United States [Tanner and Parkhurst, 2000; Zheng et al., 2002].
 The criteria used to search for all carbon-containing particles within the Atlanta data set was the presence of at least one of the following signals (±0.5): m/z 12 [C]+, m/z 36 [C3]+, m/z 37 [C3H]+, m/z 60 [C5]+, m/z 72 [C6]+, m/z −12 [C]−, m/z −24 [C2]−, m/z −36 [C3]−, m/z −48 [C4]−, m/z −72 [C6]−, with an absolute area of 50 or relative area of 0.5 (Table 1). This choice of markers allows one to select all particles containing organic compounds, elemental carbon, or any combination. The resulting DIGHIS is shown in Figures 1g and 1h. The digital histograms show the presence of carbon clusters [Cn]+, (e.g., m/z 12, 24, 36, 48, 60) both in positive and negative polarity, mostly due to elemental carbon, as well as organic fragments (m/z 15, 27, 43). As Table 2 shows, most of the carbon-containing particles contain markers indicative of secondary processing including sulfates (83%), nitrates (84%), and ammonium (78%).
Figure 1i shows the temporal variations for all supermicron carbon-containing particles observed for the study. The temporal trends for submicron carbon particles are discussed later in the section on secondary species. Evident upon comparison of Figure 1i with 1c and 1f are the peaks that occur at the same times for all three primary particle types (i.e., 8/9/99 7:30, 8/12/99 10:30, 8/13/99 6:30, 8/21/99 7:00, 8/30/99 9:00). During these time periods, based on these coinciding temporal peaks as well as the ion associations within the mass spectra, some of the particle mass spectra classified as carbon are actually OC coatings on pre-existing dust and sodium particles. These particles were classified as carbon due to the rule excluding particles with C3+ (m/z 36) from being classified as the primary particle types, sodium or dust. As mentioned, this rule was necessary to insure carbonaceous soot particles with complex mass spectra were not classified as dust and sodium types. If necessary for a certain application, an additional rule could be added to further separate the dust/carbon and sodium/carbon particles from the carbon particles. Although this rule resulted in a larger fraction of coarse (>1 μm) carbon particles, it is important to note that in Atlanta a significant number of supermicron carbonaceous primary particle types were detected. This indicates there are sources in the region that produce larger carbon-containing particles.
 When analyzing single particle mass spectra, carbon cluster ions, [Cn]+, are often detected simultaneously with organic fragment ions. In some cases, this may be due to organic compounds that fragment to carbon clusters, but more often this pattern is related to particles containing internally mixed elemental carbon (EC) and organic carbon (OC). Figure 3 shows representative single particle mass spectra for “pure” EC positive (3a) and negative ions (3b), as well as a positive ion spectra representative of an EC/OC mixture (3c). It is possible to track the relative variations of EC, EC/OC, and OC particle types over time. However, the analysis method chosen to identify these separate carbonaceous types goes beyond simple rules-based searching. The method uses an adaptive resonance theory-based neural network algorithm, ART-2a, for automatically sorting and classifying particles with similar spectral characteristics into specific particle types referred to as clusters [Song et al., 1999]. This algorithm separates particles into distinct classes of chemically similar particles within large ATOFMS data sets and generates new classes whenever a data point (mass spectrum) falls outside a certain proximity to all existing classes. The assignments for the specific patterns of the different particle types produced in the ART-2a analysis of the Atlanta data set to EC, EC/OC, and OC are based on laboratory studies performed recently in our laboratory and described in further detail elsewhere [Silva and Prather, 2000].
 From the ART-2a analysis, many distinct types of EC, OC, and EC/OC particle types were observed with distinct temporal variations and meteorological dependences, indicating different sources. The multiple types matching the patterns for the different carbonaceous forms were added together. Figure 4 shows the temporal variations of the sum of EC, EC/OC, and OC particles. Evident in these figures are strong diurnal variations in both the OC and EC/OC particles. The trends track relative humidity and show anti-correlations with temperature, suggesting the partitioning of organic compounds from the vapor phase into the particle phase and/or the formation of a stable nocturnal inversion layer. Peaks also occur in the early morning hours possibly due to vehicular traffic. In contrast, the EC particles do not show such diurnal trends, and their temporal profiles are most likely correlated with particular sources and events.
 The typical wind directions for the EC and OC particles types suggest different sources on many occasions. Specifically, there were two distinct wind directions observed for the EC particles. One type of EC particle originated from the northeast (30°) and north (350°) while another originated mostly from the southwest (250°). The EC particles originating from the southwest (250°) showed an enrichment in sulfate species, and a relatively pure EC signature indicative of fresh emissions. The sources of the EC particles were most likely vehicular exhaust (i.e., diesel) and coal combustion processes. The temporal variations (and wind dependence) for some of the EC particle types correlated with those for dust particles, suggesting these particles were most likely produced by local coal combustion facilities. Other forms of EC correlated with some of the OC particle types that were most likely vehicular exhaust. These correlations will be explored further in a future ATOFMS source apportionment paper.
 In contrast to the EC particles, in general, the OC particle temporal trends did not show as strong a wind dependence. The most abundant OC types showed a slight enhancement when the wind was coming from the northeast (50°) and southwest (200–250°). Based on the mass spectral patterns, temporal profiles, and wind directions observed, vehicular exhaust is a likely source for many of the OC particle types. When examining the temporal profile of the OC/EC particle type in detail over the course of the study, two diurnal peaks are apparent with an initial smaller peak occurring between 23:00 and 2:00 and the second occurring between 5:00 and 8:00. One OC/EC type shows major peaks in Figure 4b on 8/25/99 at 1:00 and 8:00 am. The composition of this particle type shows enrichment in nitrate (m/z −46 and −62) and sulfate species (m/z −80 and −97), as well as water (m/z 19). The addition of water possibly explains the size distribution of this particle type, which shows a bimodal character with peaks above and below 1 μm. This particle type showed a strong dependence on wind direction, peaking when the wind was coming from the south. A bus station was located directly south of the sampling site, so it is possible this particle type originated from this location.
 While the identification of EC particles by ATOFMS is possible, fragmentation occurring during the LDI process typically complicates the interpretation of ATOFMS OC-containing particle mass spectra. In addition, the complexity of the particle composition, with virtually hundreds of compounds simultaneously present, makes the interpretation challenging. In spite of these difficulties, some organic compound classes were identified in the Atlanta particle mass spectra, such as aromatics, alkylamines, and polycyclic aromatic hydrocarbons (PAHs). Aromatic signatures were mostly present in the carbon-containing particle mass spectra obtained during two time periods, 4:00 am on the 15th, and 8:30 am on the 25th. A typical mass spectrum of an aromatic-containing particle is displayed in Figure 5a, which is characterized by the presence of organic fragments at m/z 15 [CH3]+, m/z 27 [C2H3]+, as well the typical MS fragmentation pattern for aromatics [McLafferty and Turecek, 1993; Silva and Prather, 2000], with markers m/z 91 [C7H7]+, m/z 77 [C6H5]+, and m/z 51 [C4H3]+. A significant number of spectra collected during the same time period also show higher mass fragments, with or without the lower mass peaks and/or presence of potassium. Two major patterns were recognized, one characterized by signals m/z 118, 132, 146, and 160, and a second one by signals m/z 98, 112, 126, 140, and 154 (Figures 5b and 5c). The former fragment sequence may be attributed to substituted phenyls of the general structure CnH2n−8, while the latter could be due to unsaturated esters or diketones [McLafferty and Turecek, 1993].
 Background levels of PAH-containing particles, with mass spectra similar to the one shown in Figure 6a, were often detected and spiked during two episodes together with lead. In particular, during two short time windows, with peaks on the 24th at 7:30 pm and the 31st at 8:30 pm, the ATOFMS detected a high number of Pb-containing particles, mostly associated with other heavy metals (Zn, Mo, Cd) (Figures 6b and 6c). These spikes lasted only a few minutes, and the sudden increase in ATOFMS counts indicates a very local and concentrated source [Preston et al., 1992; Wang et al., 2000].
3.2. Secondary Species: Sulfur, Nitrate, and Ammonium
 As shown in Table 2 and described previously, in Atlanta most of the major particle types showed the presence of peaks in their mass spectra due to secondary processing. The following section provides further details on the associations between the major particle types and secondary species including sulfur, nitrate, and ammonium. Carbon associations have already been described so will not be discussed in any further detail.
 Particle-phase sulfate results predominantly from sulfur dioxide oxidation in the cloud free atmosphere [Tyndall and Ravishankara, 1991] and more commonly SO2 conversion in clouds and fogs [Gurciullo and Pandis, 1997; Hegg and Larson, 1990; Kreidenweis et al., 1997; Zhang et al., 1999]. These processes often lead to sulfate representing a major component of the bulk chemical composition of urban aerosol samples. The PM2.5 fraction in the eastern United States is known to have a higher sulfate concentration (exceeding nitrate) than the western United States [Karamchandani and Seigneur, 1999; West et al., 1999]. This is mostly due to the relative importance of electricity generation from fossil fuels and road traffic as pollution sources (the former tends to emit SO2, the latter predominantly NOx) and the greater abundance of ammonia, which is able to fix nitrate in the particle-phase in the western United States.
 Common ATOFMS signals assignable to sulfur species occur at: m/z −32 [S]−, m/z −80 [SO3]−, m/z −96 [SO4]−, m/z −97 [HSO4]− in the negative ion spectra and m/z 165 [Na3SO4]+, and m/z 213 [K3SO4]+ in the positive ion spectra. Some of these signals can result from fragmentation of higher m/z peaks (for instance m/z −80 from m/z −96 and −97), as well as different oxidative forms of S species (e.g., sulfate and sulfite). The presence or absence of these markers can also indicate different particle types. For example, m/z 165 is present in sea salt, whereas m/z −96 and −97 are mostly found coupled with dust and organic particle types, respectively.
 Sulfate was detected in most particle types in Atlanta, combined with the major positive ion species, including metals (K, Na, Ca, Pb), organic fragments, and [Cn]+ clusters (Table 2). Sulfates are not the only anions present in the particles, but are typically coupled with nitrate (92%), chloride, silicate, and phosphate, depending on the particle type. A representative example of an OC/EC sulfate-containing particle mass spectra is shown in Figure 7. As was typical, also present in the spectrum are peaks due to nitrates (−46 and −62). The most abundant form of sulfate in inorganic particles was the sodium salt. In dust particles, the presence of sulfates was typically indicated by a peak at m/z −96 [SO4]− and sometimes −80 [SO3]−. As shown in Table 2, the majority of the dual ion single particle ATOFMS mass spectra collected in Atlanta shows some signal due to sulfate, coupled with other components. Further details on the sulfate particulate matter observed in Atlanta are provided in the accompanying publication.
 Nitrogen oxides, mostly emitted by fossil fuel combustion, are the main precursors of tropospheric particulate nitrate, a major component of the urban aerosol [Malm and Sisler, 2000]. Atmospheric nitric acid is mainly formed from the oxidation of nitrogen [Finlayson-Pitts and Pitts, 2000] and ammonium nitrate, one significant nitrate species, is in equilibrium in the troposphere with the precursor gases ammonia and nitric acid [Harrison and Msibi, 1994; Sievering et al., 1994; Watson et al., 1994]. Because nitrogen dioxide oxidation is more rapid than sulfur dioxide oxidation, and due to the dependence on ambient ammonia concentrations that influence the ammonium nitrate dissociation process, the spatial patterns of nitrate are expected to be less uniform than those of sulfate. As already described, the eastern portion of the United States is known to have lower nitrate aerosol concentrations relative to the western regions.
 An ATOFMS marker for nitrate is m/z 30 [NO]+, which has previously been successfully employed to track nitrate-containing aerosol in southern California [Liu et al., 2000]. Other markers indicative of the presence of particulate nitrate often occur in the negative ion mass spectra at m/z −46 [NO2]−, m/z −62 [NO3]−, m/z −125 [HNO3 NO3]−, and occasionally m/z −188 [(HNO3)2 NO3]−. In this study, the search for nitrate-containing particles was based on the presence of one of the following signals (±0.5): m/z 30 with absolute area 30, or m/z 108 (due to [NaNO3]+) with absolute area 50, or m/z −46 with absolute area 1000, or m/z −62 with absolute area 1000 (Table 1).
 In the submicron size mode, the presence of nitrate-containing particles corresponds to the increased presence of carbonaceous particulate matter with a correlation (R2) of 0.93 (Figure 8). Also in the submicron mode, nitrate is typically coupled with ammonium in the same spectra, indicating the presence of ammonium nitrate. In the supermicron mode, nitrate was also associated with carbonaceous particles, but more commonly combined with sodium and dust particles in the forms of sodium and calcium nitrate. An extensive episode of high sodium nitrate supermicron particle counts occurred over the last two days (8/31/99–9/1/99) of the study. Particle mass spectra collected during this episode are characterized by negative ion signals at m/z −85 [NaNO3]−, m/z −101 [NaO NO3]−, m/z −115 [Na(NO2)2]−, m/z −131 [NaN2O5]−, m/z −147 [Na(NO3)2]−. When searching for different combinations of nitrate coupled with the other major species, namely sulfate, carbon, sodium, and ammonium, it was found that only a small number belong to “simple” classes of ammonium nitrate (0.35%) and sodium nitrate (0.26%). Most of the nitrate is present instead as complex mixtures (Table 2).
 Gas-phase ammonia is emitted by both anthropogenic and biogenic sources and constitutes the most abundant base present in the atmosphere [Asman et al., 1998; Bouwman et al., 1997; Misselbrook et al., 2000]. Ammonia undergoes acid/base reaction with the acids in ambient air, mostly nitric and sulfuric, forming the respective ammonium salts [Murano et al., 1998], and thus plays a fundamental role in determining aerosol pH. A reliable ATOFMS marker for ammonium occurs at m/z 18 ± 0.5 [NH4]+ (minimum absolute area 30, see Table 1). Although there is the possibility of interference from water at this peak, in lab studies using 266 nm radiation, the dominant peak for water occurs at m/z 19 from H3O+ (G. Khairallah, manuscript in preparation, 2002).
 As shown in Figure 8, the temporal trends for ammonium containing track the profiles for nitrate and carbonaceous matter submicron particles in Atlanta. The correlations between all of these species in the submicron particles are high (R2 ≥ 0.9). The percentage of particles containing ammonium for each primary particle type for all particle sizes over the detected size range is shown in Table 2. It should be noted the PM-ammonium data obtained at the Supersite using other techniques do not show the diurnal pattern that is especially evident over the last week of sampling. In contrast, the semicontinuous data show a more constant presence of ammonium, with attenuation of the diurnal pattern. As will be explained in more detail in the accompanying paper, this is most likely due to the presence of high amounts of ammonium sulfate in the particles during the day with less organic and nitrate species. Ammonium sulfate and bisulfate and sulfuric acid are characterized by a very high ionization threshold and thus are not easily detected at the LDI wavelength used in this study (266 nm).
3.3. Size-Composition Relationships
 In Figure 9, the percentage of particles as a function of their aerodynamic diameters are shown for the three major primary types identified in Atlanta, namely sodium, dust, and carbon. These are then further separated by the presence or absence of secondary species on the particles, specifically nitrate, sulfate or ammonium. Due to their low contributions, particles that did not fit into any one particle type, dust and sodium without secondary species, and particles with only secondary species were grouped into one class referred to as “other types.” This figure shows the percentage of particles as a function of size (in 0.1 μm bins) in order to demonstrate which compositions contribute most to each particle size. As expected, mechanically generated dust and sodium-containing particles with secondary species are found mostly in the coarse mode (>1 μm). The size distribution of the carbonaceous particles is bimodal with significant fine and coarse mode fractions. This distribution is noteworthy because typical urban distributions show carbonaceous particles predominantly in the fine mode. The presence of carbonaceous aerosol particles above 1 μm in Atlanta results from the partitioning of both primary and secondary low volatility organic compounds onto pre-existing sodium and dust particles [Odum et al., 1996], as well as unique source(s) of coarse organic particles.
4. Aerosol Sources
 One major strength of real-time single particle mass spectrometry methods lies in the ability to determine chemical associations between species within individual particles and in turn to use these associations to understand the chemistry and sources of the observed aerosol. The data analysis method of peak searching used for this paper was chosen to provide an overview of the major correlations between chemical species within individual particles. Using this method, one can readily see that during the study, the Atlanta aerosol was a complex mixture of many chemical species within the same particles. However, when using simple rules-based searching methods such as the one described herein, one often sacrifices valuable single particle information that shows unique combinations of certain species, which can lead to additional insight into particle sources and processing. Going into all of the particle types observed in Atlanta in further detail is beyond the scope of this paper; however, as an example of how single particle mass spectrometry can be used for source identification, a brief description of results from comparing ATOFMS source signatures acquired in coal testing with the Atlanta data set is given here.
 For the purposes of direct comparison of the Atlanta data set and previous ATOFMS coal source characterization studies, ART-2a analysis was used. The coal tests were analyzed using ART-2a to create clusters or source signatures of the common particles formed by coal combustion at both low and high temperatures. These clusters were then compared with the clusters obtained by running ART-2a on the Atlanta data set. Figure 10 shows a comparison of one of the more common coal combustion particle ART-2a clusters observed plotted with an ART-2a particle cluster for a particle type commonly observed in Atlanta. The clusters shown represent the average area spectrum for the particle type from the coal combustion study (narrow (dark) lines), and the average spectrum for the particle type detected in Atlanta (wider (lighter) lines). Evident in both positive spectra are peaks at m/z 7, 23, 27, 39, 56 due to Li, Na, Al, K, Ca, and Fe, respectively, with m/z −46 (nitrates), −60 and −76 (silicates), and −97 (sulfates) in the negative ion spectra. As discussed previously, Li is a potential unique marker for coal combustion [Guazzotti et al., 2003]. The majority of the spectral intensities match quite well with the exception of the peak at m/z −46 (NO2−) observed in the Atlanta particle type. This peak results from the heterogeneous reaction of this particle type with NOx gas-phase species. The temporal profile (1 hour resolution) for this particle type between 8/09/99 and 9/1/99 is shown in Figure 11. One can see rapid temporal bursts of this particle type superimposed on a background. Most likely the bursts occur when the particles are coming directly from the source and the background counts are due to particles with longer lifetimes and/or soil particles that are difficult to differentiate from the coal signature. Many of the observed temporal bursts are evident in the dust temporal plot shown in Figure 1c as well, suggesting a significant fraction of the dust peaks are due to coal combustion. The wind direction was examined during the temporal bursts and was found to be blowing mostly from 230–250° and 290–320°. Three large coal fired power plants are located within 80 km of the sampling site in the northwest and southwest [Lee et al., 2002], so these could explain the dust peaks observed when the wind was blowing from these directions. The dust types occurring more consistently and showing less wind dependence could be of crustal origin. Many of the other inorganic dust particle types containing various combinations of Al, K, Ca, and Fe also matched the signatures from coal testing, indicating that coal production in the Atlanta area is mostly likely a significant contributor to ambient PM2.5 in the Atlanta region. The inorganic nature of these particles is consistent with particles produced by high temperature coal combustion (2000 F), which is used for producing coal power in the area (E. Edgerton, personal communication, 2002). This high temperature process converts most of the carbonaceous species into CO2. After further analysis, other particle types will be correlated with wind direction and known local sources to determine the exact source(s) of other specific particle types.
 The ATOFMS mass spectra collected at the Supersite show a very complex picture of the urban aerosol during the Atlanta summer. The major species observed in the particles were sodium, ammonium, sulfate, nitrate, dust, EC, and OC, giving rise to hundreds of complex mixtures. One of the primary types detected was dust, which was often K- and Ca-rich, in the coarse mode, and typically coupled with secondary species including nitrate, sulfate, and carbon. Other metals are often present in dust particles including lithium, aluminum, iron, and titanium. Zinc and lead are often associated with these particles and show very short temporal bursts with high number concentrations, indicating they are most likely emitted by a local and very concentrated source. The chemical signatures of many of the dust types observed in Atlanta are similar to those observed in previous ATOFMS source characterization studies. Correlations are observed between many of these common dust particle types for short time periods (i.e., 1–2 hours) and wind direction, suggesting many of the dust particle types detected in the Atlanta area are produced by local coal fired power plants.
 Sulfate was the most common anion observed and was associated with most particle types. Simple inorganic mixtures such as ammonium sulfate, sodium sulfate, sodium nitrate, and ammonium nitrate were infrequently detected with ATOFMS in Atlanta particulate matter. However, when using the current ionization scheme, ATOFMS is insensitive to the presence of certain pure species such as sulfuric acid and ammonium sulfate, so their presence is most likely underestimated as described in the associated paper. Complex mixtures of primary particle types mixed with secondary species dominated the aerosol, indicating the high reactivity of the summer urban atmosphere. The fact that the ATOFMS detects secondary markers on all particle types shows the aerosol is highly transformed. EC and OC were detected as “pure” EC and OC particles, as well as internal EC/OC mixtures that showed different wind dependences, indicating different sources. During certain episodes, several classes of organic compounds were identified, including alkylamines, PAHs, other aromatics, and possibly unsaturated esters. The high relative humidity was reflected in numerous water clusters found in sodium and calcium rich particles. Detailed information on individual particle combinations measured by single particle mass spectrometry techniques can significantly contribute to the understanding of aerosol sources and atmospheric transformation processes, as well as provide unique data on size resolved chemical associations for toxicological studies. The next step involves deriving atmospheric concentrations of the different primary particle types as described in the following paper.
 The authors would like to acknowledge Deborah Gross of Carleton College and her students Benjamin Warren and Alexander Barron for their help in data collection for the first two weeks in Atlanta. Funding for this study was provided by Georgia Institute of Technology contract (G-35-W06-G1) and the U. S. Environmental Protection Agency (contract R826240-01-0). Last the authors would like to thank Stefania Angelino for the many useful discussions about the Atlanta data set.