Aerosol time-of-flight mass spectrometry (ATOFMS) was used for characterizing the aerodynamic size and chemical composition of individual particles during the Atlanta Supersite Experiment in 1999. During certain time periods, increased numbers of particles scattered light but did not produce mass spectra. Upon comparison of the size-resolved unscaled particle counts from an aerosol time-of-flight mass spectrometer with those from a laser particle counter, the presence of a chemical bias became apparent in the single particle mass spectral measurements. Upon further analysis, it was determined that these events occurred during time periods of elevated ammonium and sulfate mass concentrations measured with semicontinuous particulate analysis instruments. The missed particle type occurred mostly in the smallest size range (0.35–0.54 μm) and correlated well with optical scattering measurements. As described herein, a scaling procedure is developed that allows one to account for the ATOFMS chemical bias. This procedure is tested by comparing the scaled ATOFMS data with multiple measurements from other techniques made during the 1999 Atlanta Supersite study. This is the second paper in a two-part series focusing on ATOFMS data collected during the Atlanta Supersite experiment in 1999 [Prather et al., 2002].
 Aerosol particles play an important role in our atmosphere, in areas ranging from deleterious health effects to global climate and visibility. Developing a complete understanding of the overall effect of particles on human health and the environment requires accurate information on particle size and chemical composition in the ambient air. Recently, in order to obtain real time highly sensitive detection of the chemical combinations of species present in aerosol particles, mass spectrometry has become the tool of choice for a number of research groups [Suess and Prather, 1999; Noble and Prather, 2000; Johnston, 2000].
 In August of 1999, the first EPA Supersite experiment was conducted in Atlanta, Georgia with the main focus involving the characterization of urban particulate matter using a number of emerging particulate measurement techniques [Zhang et al., 2003]. As part of an effort to obtain chemical signatures for aerosols in the Southeastern region of the United States, several different prototype single particle mass spectrometers sampled alongside one another for the first time, along with various semicontinuous gas and particle phase monitors [Middlebrook et al., 2003]. A significant difference between single particle instruments lies in the method used for particle sizing. Precise sizing is necessary in order to scale data into atmospherically representative concentrations. A number of approaches are used for particle sizing, including measurement of the intensity of light scattered from particles [Thomson et al., 2000], determination of the time-of-flight between two points [Gard et al., 1997; Jayne et al., 2000], and dynamic focusing for selecting particles of a given size [Mallina et al., 2000]. Another significant difference between particle mass spectrometers is the method used for particle ionization. One instrument utilizes thermal desorption followed by electron impact ionization [Jayne et al., 2000], while the others operate using laser desorption/ionization (LDI) [Gard et al., 1997; Mallina et al., 2000; Thomson et al., 2000]. For the instruments utilizing LDI, research has shown that the choice of wavelength can have significant effects on the ionization and detection of the aerosol [Thomson et al., 1997].
 The aerosol time-of-flight mass spectrometer (ATOFMS) obtains precise (±1%) information on aerodynamic size using particle time-of-flight measurements. As described herein, precise size measurements allow one to use traditional particle sizing instruments to scale the raw ATOFMS particle counts for the different particle types detected into atmospherically relevant number concentrations. In previous publications, the transmission and optical detection biases of the ATOFMS have been described [Allen et al., 2000]. In this paper, the data from a laser particle counter (LPC) are compared to the ATOFMS raw particle counts, using one hour temporal resolution. The typical procedure involves dividing the ATOFMS particle counts occurring in each size and time bin by the LPC counts into the corresponding size bins to produce the necessary scaling factors (SF1) to match the raw ATOFMS counts to the LPC data. One assumption that is made when using this scaling procedure is that all particle compositions are hit with equal efficiency, and that the only instrumental biases are transmission losses and optical detection biases based on particle size. This assumption has been valid for data analyzed in the past for studies performed in California and on the west coast; however, as described herein, upon comparison with other measurements of the Atlanta aerosol, it is apparent that this assumption leads to errors in the number concentrations for the ATOFMS particle types detected in Atlanta. A discussion is presented on a new procedure that is developed to account for the observed chemical bias in Atlanta. This paper is the second of a two-part series detailing results obtained using the ATOFMS instrument during the 1999 Atlanta Supersite study [Liu et al., 2003]. As detailed in the first paper using the ATOFMS mass spectral markers detected in Atlanta, the particles were classified into seven major chemical classes: (1) sodium containing, (2) sodium containing with secondary species (ammonium, nitrate, or sulfate), (3) carbonaceous, (4) carbonaceous with secondary species (ammonium, nitrate, or sulfate), (5) dust, (6) dust with secondary species (ammonium, nitrate, or sulfate), and (7) secondary (ammonium, sulfate, or nitrate). For a more thorough discussion of the major different particle types detected in Atlanta and rules used for classifying these particles, the reader is referred to the accompanying paper which provides an overview of the Atlanta aerosol from a single particle perspective. In this paper, a new scaling procedure is described to account for an additional (eighth) type of particle consisting of relatively pure ammonium and sulfate.
2. Methods and Instrumentation
 A transportable dual ion version of the aerosol time-of-flight mass spectrometer (ATOFMS) was used to collect ambient single-particle mass spectra [Gard et al., 1997]. A brief instrumental and experimental overview is included in the first paper in this series. During the field study, laser particle counters (LPC; PMS, Inc. Model LASAIR 1002) were employed to measure particle concentrations in ambient air (particles/cm3) by the University of Minnesota [Woo et al., 2001]. Data from this instrument were considered representative of the “true” ambient concentration of aerosol particles and used to scale the unscaled ATOFMS particle counts to obtain atmospherically representative number concentrations. The LPC operated in the particle size range between 161 nm and 3 μm with 8 min temporal resolution. The ATOFMS analyzed particles between 0.2–2.5 μm with 30–60 min resolution.
 The carbon mass concentration and optical data described herein were obtained by Atmospheric Research & Analysis, Inc. using an R&P 5400 Ambient Particulate Carbon Monitor (APCM; Edgerton, E. Atmospheric Research & Analysis, 2001) and by Georgia Institute of Technology with a nephelometer [Carrico et al., 2003].
3. Results and Discussion
 As described in previous papers on ATOFMS, there are known biases based on the instrument design [Allen et al., 2000]. These biases create a transmission efficiency that favors particles of larger sizes, therefore providing higher detection efficiencies for larger particles. Because smaller particles (<1 μm) are present in much higher abundance than larger particles (>1 μm), the ATOFMS can detect particles over the entire size range (0.2–2.5 μm) with a significant number of particles of all sizes [Finlayson-Pitts and Pitts, 2000]. In order to correct for the known size biases, ATOFMS data must be scaled using more quantitative particle measurement instruments operated alongside the ATOFMS. Previous work has involved scaling ATOFMS data upon comparison with mass concentration data acquired with Micro-Orifice Uniform Deposition Impactors (MOUDI) [Allen et al., 2000]. However, with this scaling procedure, the results are potentially limited by the longer sampling times required for the MOUDI (typically 4 hours or more depending on concentrations) and lower size resolution.
 For all data in this study, scaling factors were obtained using hourly LPC data. The LPC data were corrected assuming a refractive index of 1.46 for atmospheric particles, which is similar to that of oleic acid. However, given the extremely complex relationship between optical and aerodynamic diameters of ambient particles [Hinds, 1998], no further conversions were applied. This assumption may affect the absolute ambient concentrations, however the basic trends in particle concentrations should remain the same. The LPC measures the size distribution and number concentrations of particles in size bins of 0.35–0.54 μm, 0.54–0.65 μm, 0.65–1.04 μm, 1.04–1.08 μm, 1.08–2.00 μm, and 2.00–3.00 μm in optical diameter. The 1.04–1.08 μm diameter LPC bin was not used for scaling because it was impossible to obtain statistically significant ATOFMS number counts using such a narrow size bin in one hour. The particle concentrations were recorded by the LPC every 8 min for 40 min of each hour. The values used here represent an average of five 8-min particle concentrations, and are assumed to be the average concentration for each hour.
4. Scaling Factor 1 (SF1)
 Initially, in order to obtain atmospherically representative ATOFMS number concentrations for the different particle types, the scaling factors for each of the 60-min periods were calculated using the assumption that the only sampling biases were due to transmission losses and optical detection. This proves to be an important, and in this case, erroneous assumption because of the neglect of any particle composition bias. The total number of particles for which a mass spectrum was obtained by the ATOFMS system was counted and subdivided into LPC size bins in 60-min intervals. Next, each 60-min time bin was analyzed for missing time when the instrument was not acquiring mass spectral data. This was accomplished by looking at each minute during the study to see if there was a particle that was sized during that time period. If no particle was sized during the entire minute, the period was noted. At the end of the 60-min interval, all noted temporal differences were summed, and the fraction of missing time (FMT) for each 60-min interval was calculated. If the FMT for a 60-min interval was greater than 25% or if the total counts of ATOFMS particles were below 5 (for that particular size bin), the point was omitted from the scaling factor calculations. If the FMT was equal to or less than 25% for the 60-min interval (i.e., the instrument was running for 45 out of the 60 min), the total number of hit particles was adjusted to the number of particles that would have been analyzed if the instrument were running the entire 60 min period. The number of particle hits in each size bin during the 60-min interval was divided by the total operating time. This number was then converted to particles/cm3 (using the measured ATOFMS flow rate of 0.91 L/min) and compared directly with the corresponding LPC value of particles/cm3. The result is a scaling factor one must multiply the ATOFMS number concentrations by in order to obtain the true ambient particle concentrations.
 The resulting scaling factors (SF1) obtained by comparing the ATOFMS to the LPC are shown in Figure 1 by LPC size bins. The most noticeable difference between Figures 1a–1e are the higher scaling factors that are required for smaller particles (0.35–0.54 μm) compared to those for larger sizes (2.00–3.00 μm). Also of note are the different temporal variations of the SF's observed for each of the size bins. This is easiest to see by comparing the scaling factors necessary for particles in the smallest (0.35–0.54 μm, Figure 1a) to those of the largest size bin (2.00–3.00 μm, Figure 1e). As shown, the smallest particle size bin SF vary over time from approximately 500,000 to 7,000,000 compared to the largest bin which vary from approximately 10 to 150, with a smooth continuum between (Figures 1b–1d). Although the ratio of scaling factors needed for each of the size bins is similar (i.e., approximately a factor of 15), the normalized standard deviation is greater for the smaller size bins where the largest standard deviations are more than twice those observed in the larger size bins. Another way of saying this is that the deviations from the mean (shown as dotted lines in Figure 1) are much larger in the smaller size ranges. These deviations are indicated graphically by the rapid, intense spikes occurring in the smallest size bins. This was the first time when sampling with the ATOFMS the hit rate of the ATOFMS varied from 1% to approximately 35% over a period of only 1–3 hours, with no instrumental modifications occurring. In previous ATOFMS laboratory and field studies, an average hit rate for all particle sizes ∼25% is typical. The hit rate for the ATOFMS is defined as the number of particles for which mass spectra are obtained divided by number of particles which are sized over a given time period. The variations that were first apparent in the hit rate observed in the field are reflected in the scaling factors. It is important to note that these particles were detected and sized by the ATOFMS; however, no mass spectra were obtained for these particles. The fact that the ATOFMS still detects and sizes the “missed” particles by light scattering is critical to knowing when the chemical bias is present and being able to accurately account for it.
5. Scaling Factor 2 (SF2)
 In order to correct for this newly observed chemical bias, the number of hit particles (i.e., those which produced positive, negative, or both mass spectra) and the number of missed particles (those which were sized but did not produce mass spectra) in a given time and size bin were totaled. An average threshold hit rate for each size bin was determined as the “typical” hit rate (i.e., rate expected when there is no chemical bias). This was determined by plotting the hit rate in each size bin and determining the average hit rate for that size range of particles over the entire study. The values used are as follows: 12%, 14%, 17%, 36% and 26% for 0.35–0.54 μm, 0.54–0.65 μm, 0.65–1.03 μm, 1.08–2.00 μm and 2.00–3.00 μm, respectively. The trend in these values is consistent with laboratory calibration experiments that show increasing hit rates for increasing sizes, as well as efficient focusing of supermicron particles with the current nozzle/skimmer setup. Any time the hit rate of the ATOFMS dropped below these average threshold values, a new “undetected” or missed particle type was added to the total hit particles, producing a second set of scaling factors (SF2). As mentioned, these missed particles were sized with the scattering lasers but no mass spectrum was obtained.
 In order to test the effect of the new scaling factors (SF2), ATOFMS data are compared with semicontinuous measurements made during the study. Shown in Figure 2a are the carbon mass concentrations obtained with the R&P APCM over the final 9 days of the study compared to ATOFMS data scaled using the original SF data. Figure 2b shows the same carbon data plotted against the ATOFMS data scaled using the new SF2 values. Comparing Figure 2a with Figure 2b, it is evident that once the ATOFMS data are scaled using the second set of scaling factors (SF2), trends in the number concentration of carbon-containing particles detected by the ATOFMS track the carbon mass concentration data. This comparison confirms that using the new SF2s represents a more accurate way of scaling the ATOFMS data for this study.
 Further evidence for the SF2 values being more appropriate for scaling than the SF1 scaling values is provided in Figure 3, which shows the temporal data from a nephelometer operated by the Georgia Institute of Technology [Carrico et al., 2003] compared to the total scaled ATOFMS concentrations and the missing particle class over the same 9 consecutive days. Figure 3a shows optical scattering data as well as the ATOFMS scaled carbon counts and “missed” particles. From this figure, one can see the trends of the missed particle type and the scattering data track well, suggesting this particle type is composed of a highly reflective material. This is consistent with the fact that the ATOFMS could detect these particles by light scattering but no mass spectra were observed. In Figure 3b, a correlation between optical absorption values and carbon-containing particles detected by the ATOFMS is also apparent, again suggesting the second set of scaling procedures produce more representative ATOFMS number concentrations for the different particle types.
 Once the new scaling factors (SF2) are calculated, they can be applied to data collected over the entire study for the different particle types. As described, the spectra were divided into broad classes of carbon, sodium containing, and dust based on the combinations of ions present. These classes were further subdivided into particles containing secondary inorganic components, including sulfate, nitrate, and ammonium. In this paper, the missing type category is added as described. Exact search criteria used for identification of these particle classes, along with an in-depth analysis of each particle class detected can be found in the accompanying paper in this issue.
Figure 4 shows the temporal profiles of the scaled number concentrations for the major particle types observed in Atlanta. In this figure, the smallest particles (Figure 4a) are composed of primarily carbon containing (EC, OC, EC/OC) particles, whereas particles greater than 1 μm (Figure 4e) are composed of both dust and sodium containing particles. A gradual shift can be seen between the two extreme size bins (Figures 4b–4d) with an overall composition change around 1 μm. This is similar to previous ATOFMS studies; however this analysis shows a more correct assessment of the ambient number concentration of particles taking into account all possible instrumental biases. Figure 4 shows a substantial fraction of particles possessing secondary components, leading to the conclusion that particles detected during this study represented a complex mixture of primary particles having undergone secondary processing. Also one should note that the number of missed type particles increases in decreasing particle size bins. This would suggest the smaller particles are more chemically simplistic. As they grow in size, the particles become more and more coated with organics and nitrates from the atmosphere, enhancing the absorption of the 266 nm light and thus increasing their detection efficiency by LDI.
6. Missed Particle Type Composition
 A question exists as to the composition of the particles during the time periods when the ATOFMS hit rate plummets. In order to determine the composition of these particles, it is necessary to examine results from other simultaneously operated instruments during the time periods when the missed particle type is obvious in the ATOFMS measurements. A strong correlation exists between semicontinuous measurements of ammonium and sulfate mass concentrations collected at the Atlanta Supersite and the presence of the missing particle type [Stolzenburg and Hering, 2000; Erisman et al., 2001; S. Hering, personal communication, 2002]. Weber et al.  describe events between 20 August 06:00 and 24 August 01:00 of pollution plumes corresponding to peaks in sulfate concentrations [Weber et al., 2001]. These time periods also represent times with high missed particle concentrations by the ATOFMS, even in the larger sizes, Figure 4. Semi-continuous OC and nitrate mass concentration measurements showed oscillations particularly evident during the last part of the study that were not present in the ammonium and sulfate data, suggesting the missing particle type was relatively pure ammonium sulfate. Particles composed of significant amounts of ammonium and sulfate would scatter light efficiently, which is consistent with the nephelometer results shown. Furthermore, relatively pure ammonium sulfate particles would absorb light less efficiently at the 266 nm wavelength used for this study. This has been demonstrated in a lab-based study by Thomson et al. , where ammonium sulfate particles were difficult to detect with LDI at various wavelengths. In fact, the authors had the foresight to propose this could result in an entire missing class of particles in ambient studies [Thomson et al., 1997].
 In order to gain insight into a possible source for these ammonium sulfate-rich particles, the ATOFMS mass spectra were further examined to determine if any other types followed similar temporal trends. In this analysis, it was determined that a number of forms of relatively pure EC particles were common during these time periods [Liu et al., 2003]. In fact, over the course of the study, these EC types showed strong temporal correlations with the missing particle type. During time periods where the missing type was present, the wind was often blowing from the east and southeast (90°–135°). Most likely, a unique local source was producing these small relatively pure particles that were being transported to the site relatively quickly so as not to pick up organic and nitrate species from secondary processing.
 In order to account for the ATOFMS sampling biases, it is necessary to scale the data using other measurements to obtain atmospherically representative number concentrations. The method for accomplishing this using an LPC has been described. For the ATOFMS instrument, it is possible to scale ambient data because the precise size of each particle analyzed is measured. Accurate size measurement of each particle is imperative if data are to be directly compared to those acquired with supporting instrumentation. Once the chemical and transmission biases are accounted for, number concentrations of the various single particle types detected by the ATOFMS data can be obtained. The scaled results show that smaller particles (less than 1 μm) were composed primarily of carbon species (OC, EC, OC/EC), whereas larger particles (particles greater than 1 μm) were composed of predominately dust and sulfate with the addition of secondary species. When sampling in the Atlanta environment, the presence of a missed particle type was evident based on observed variations in the hit rate of the instrument and ultimately the scaling factors necessary to scale the ATOFMS data. This missed particle type was most prevalent in submicron particles and most significant in the smallest size bins (0.35–0.54 μm). Upon comparing the presence of this missed particle type to mass concentration data obtained with semicontinuous instrumentation, strong correlations with time periods with high ammonium and sulfate mass concentrations as well as times of high light scattering were observed, indicating these particles were most likely composed of significant amounts of ammonium sulfate with relatively small amounts of absorbing organic species. Pure particles have been shown in laboratory studies to have a major effect on instrumental detection efficiencies [Kane and Johnston, 2000]. However, typically in ambient studies, particle compositions are observed to be much more complex, at least in the fine size mode (0.1–2.5 μm), providing higher, more uniform detection efficiencies (i.e., less chemical bias) than those observed for more chemically simplistic particles. In this study, it appears the particles were generated by a unique local source and transported in a distinct air mass containing pure EC particles relatively quickly to the sampling site without the necessary time to undergo secondary processing.
 This research was supported by Georgia Institute of Technology (contract G-35-W06-G1) and the U.S. Environmental Protection Agency (contract R826240-01-0). The authors thank Professor Deborah Gross, Alexander R. Barron, and Benjamin S. Warren from Carleton College for their help with data collection while in Atlanta. The authors would also like to thank Peter McMurry of University of Minnesota for providing the LPC data used for scaling ATOFMS counts to atmospherically representative number concentrations and Michael Bergin of Georgia Institute of Technology for providing nephelometer absorption and scattering data for comparison.