Single-particle laser-induced-fluorescence spectra of biological and other organic-carbon aerosols in the atmosphere: Measurements at New Haven, Connecticut, and Las Cruces, New Mexico

Authors


Abstract

[1] This study focuses on organic carbon (OC) aerosols, including natural biological aerosols, in the Earth's troposphere, and on laser-induced fluorescence (LIF) spectral methods for studying these aerosols. LIF spectra of atmospheric OC and biological aerosols (having diameters greater than approximately 3 μm) measured at New Haven, Connecticut, and Las Cruces, New Mexico, are reported. A hierarchical clustering method was used to cluster approximately 90% of the single-particle LIF spectra into 8–10 groups. Some of these groups have spectra that are similar to spectra of some important classes of atmospheric aerosol, such as humic/fulvic acids and humic-like substances, bacteria, cellulose, marine aerosol, and polycyclic aromatic hydrocarbons (PAH). A comparison with previous measurements made at Adelphi, Maryland, reveals that the most highly populated clusters found at Adelphi, and some of the less populated ones, also appear in the LIF spectra at New Haven and Las Cruces, even though the regional climates at these locations is different (New England/Atlantic Coastal, for New Haven and Adelphi, and Chihuahuan Desert for Las Cruces), and the measurements were made in different seasons. The results are consistent with some (perhaps most) of the fluorors in OC and biological atmospheric aerosols being common to these three sites. On average, spectra characteristic of humic/fulvic acids and humic-like substances (HULIS) comprise 28–43% of fluorescent particles at all three sites; whereas cellulose-like spectra contribute only 1–3%.

1. Introduction

1.1. Organic Carbon in Atmospheric Aerosol

[2] Atmospheric aerosols are important to fields of science ranging from Earth climate to human health. This study focuses on the organic carbon (OC) aerosols, including biological aerosols, in the atmospheric boundary layer, and on laser-induced fluorescence (LIF) spectral techniques [Pinnick et al., 2004] for studying this aerosol component. Extensive research over many years has aimed to characterize the OC fraction of atmospheric aerosol using a variety of techniques ranging from culturing of bacteria, to single-particle mass spectrometry (MS), to chromatographic studies of extracts of collected particles. Typically the various methods do not provide the same information, but each tends to provide a different view of the extremely complex mixture of atmospheric aerosol.

1.2. Brief Summary of Some Previous Studies of Atmospheric OC Aerosol

[3] OC in atmospheric aerosols ranges from small molecules such as oxalic acid or phenols, to PAH, to viable bacteria, fungi and spores (each of which is composed of many thousands of different kinds of organic molecules), to highly complex mixtures of decomposition products of biological materials (e.g., humic substances), to secondary OC aerosols formed initially by oxidation of volatile organics (studied extensively since the pioneering studies of Haagen-Smit [1952] on smog, and Went [1960] emphasizing biogenic terpenes). OC aerosols can include alkanes, alkenes, carboxylic acids, ketones, phenols, furans, terpenoids, PAH, aromatic polyacids, lignans, cellulose, humic and fulvic acids, humic-like substances (HULIS), pollens, bacteria, bacterial spores, viruses and fungal spores [Shah et al., 1986; Larson et al., 1989; White and Macias, 1989; Novakov and Penner, 1993; Rogge et al., 1993a, 1993b; Saxena and Hildeman, 1996; Penner and Novakov, 1996; Lighthart, 1997; Shaffer and Lighthart, 1997; Seinfeld and Pandis, 2006; Havers et al., 1998; Zappoli et al., 1999; Krivacsy et al., 2000; Decesari et al., 2002; Bauer et al., 2002a, 2002b, 2002c; Lighthart and Mohr, 1994; Kiss et al., 2002; Huang et al., 2006; Qin and Prather, 2006; Dzepina et al., 2007]. The relative contributions of primary organic aerosol (particles directly injected into the atmosphere) and secondary organic aerosol (particles formed by gas-to-particle conversion) is highly variable [Alfarra et al., 2004; Kanakidou et al., 2005; Qin and Prather, 2006; Robinson et al., 2007].

[4] Concentrations of OC aerosols are significant but highly variable in the lower troposphere [White and Macias, 1989, Saxena and Hildeman, 1996; Larson et al., 1989; Hitzenberger et al., 1999; Zappoli et al., 1999; Krivacsy et al., 2001; Fraser et al., 2002], both in urban environments [Shah et al., 1986; Huntzicker et al., 1986; Dasch and Cadle, 1989; Holler et al., 2002; Qin and Prather, 2006], where typical concentrations are 1–15 μg m−3, and in rural/remote/North Atlantic regions [Shah et al., 1986; Cavalli et al., 2004], where concentrations are 0.5–4 μg m−3. Meteorological factors can affect concentrations of biological aerosols [Jones and Harrison, 2004]. OC aerosols can show both diurnal cycles [Zhang et al., 2004; Qin and Prather, 2006; Dzepina et al., 2007] and seasonal dependence [Rogge et al., 1993b; Bae et al., 2006]. HULIS accounts for a significant fraction of urban OC aerosol mass [Samburova et al., 2005].

[5] Processes that generate OC aerosols include biomass combustion [Zappoli et al., 1999; Cao et al., 2006], residential wood and coal burning, cooking [Herckes et al., 2002], the nucleation and growth of new particles in urban environments [Zhang et al., 2004; Qin and Prather, 2006], microbial decay of plant matter [Rogge et al., 1993a; Havers et al.,1998; Robinson et al., 2007], emission of volatile OC by vegetation and bacteria followed by oxidation, and then condensation into secondary OC aerosols [Kanakidou et al., 2005; Went, 1960], microbial and biochemical degradation of organic debris in the soil [Gelencsér et al., 2002], wind action around dairy and cattle feedlots [Rogge et al., 2006], farming operations, and sewage wastewater treatment plants [Bauer et al., 2002c]. Humic-like substances in surfactant matter on the ocean surface are aerosolized by wave action [Oppo et al., 1999]. Bubble bursting processes are believed to be an important source of water soluble OC (WSOC) particles over the north Atlantic [Cavalli et al., 2004]. Most airborne bacteria are derived from plants, soil, and water bodies. Their injection into the atmosphere is thought to derive from wind action (on plants and soil) and from bubble-bursting [Lighthart, 1997]. Both primary OC and secondary OC aerosols may be aged in the atmosphere. Bioaerosols and humic materials tend to be extremely complex even before they undergo aging in the atmosphere. Even with far simpler types of primary OC aerosol, the chemical reactions that age OC aerosols and the resulting aged aerosols can be extremely complex. The chemical makeup of a significant mass fraction of OC aerosol is not well known [Samburova et al., 2005]. Collectively the above references suggest that much more needs to be learned about OC aerosol in the Earth's atmosphere.

1.3. Aerosol Diagnostic Techniques

[6] A variety of aerosol characterization methods are used to provide complementary, and sometimes overlapping, insights into compositions of particles. Analyses of OC aerosols are often performed on sample collections because a significant mass of aerosol is required for analysis, although many single-particle studies using mass spectroscopy (MS), and some studies using laser-induced-breakdown spectroscopy (LIBS) and LIF have been reported [e.g., Hettinger et al., 2005; Lithgow et al., 2004; Pinnick et al., 2004].

1.3.1. Analysis Techniques Suitable for Aerosol Collections

[7] Optical and electron microscopy can indicate particle shape, which can be definitive for some types of particles, for example, some types of pollens. The combination of x-ray fluorescence and electron microscopy can also provide information about single-particle elemental composition. Culturing of bacteria, other microorganisms, and viruses collected in air samples can in some cases provide definitive information relating to viability, and identification to the genus or species level [Hensel and Petzoldt, 1995; Shaffer and Lighthart, 1997; Lighthart and Tong, 1998]. Culturing may be a first step in obtaining sufficient material for more detailed biochemical analyses. Biochemical assays for specific proteins and DNA or RNA sequences [Hindson et al., 2005] can indicate the presence of specific bacteria or viruses, or specific allergens. These assays typically indicate very little about molecules for which probes were not used. Mass spectrometers have been used extensively to characterize organic and nonorganic aerosols, apportion sources, etc. In Aerodyne's Aerosol MS (AMS), a small fraction of the molecules within a particle is vaporized and ionized, and the mass-to-charge spectra of the resulting ions is detected and accumulated for a collection of particles [Jayne et al., 2000]. Nuclear magnetic resonance (NMR) has been used to estimate the fractions of hydrogen atoms bound to C=O, and to C-O. Collections of aerosol can be fractionated on the basis of physiochemical properties such as solubility, tendency to elute into specific solvents, retention times in chromatographic columns, electrophoretic mobility, etc., and the resulting fractions can be analyzed using a variety of techniques, such as MS, NMR, or infrared spectroscopy. Gas chromatography-MS (GC-MS) has been developed and used extensively [Mazurek and Simoneit, 1997; Simoneit et al., 1999; Rogge et al., 2006; Williams et al., 2006].

1.3.2. Non-LIF Single-Airborne-Particle Diagnostic Techniques

[8] Techniques for rapidly analyzing single aerosol particles, even if they are not as definitive as those used on bulk samples, can be useful for better understanding the sources, chemistry, and fate of OC aerosol. In collections of particles: (1) definitive inferences cannot be made about the compositions of individual particles; (2) particles that occur in low concentrations may be difficult to detect using some analytical techniques; and (3) rapid time variations of specific, low concentration aerosol species are often not detected, because aerosol must be collected long enough to obtain material for analysis.

[9] Three primary non-LIF single-airborne-particle analysis techniques (beyond particle size), being used and/or developed, are as follows.

[10] 1. The laser-ablation aerosol time-of-flight MS (ATOFMS), particle analysis by laser mass spectrometry (PALMS), and the rapid single-particle MS (RSMS) provide elemental composition and masses of laser-ablated molecular fragments of single aerosol particles [Prather et al., 1994; Noble and Prather, 2000; Kane and Johnston, 2000; Lee et al., 2002; Murphy et al., 2003; Middlebrook et al., 2003; Sullivan and Prather, 2005]. Aerosol MS technology is the most widely used of the techniques listed here, and a commercially instrument is available (e.g., the ATOFMS from TSI Inc.).

[11] 2. LIBS yields atomic emission and plasma emission spectra that can help classify OC aerosols [Radziemski et al., 1983; Hahn and Lunden, 2000; Carranza and Hahn, 2002; Hybl et al., 2006]. LIBS has been applied to atmospheric aerosols [Hettinger et al., 2005; Lithgow et al., 2004; Hybl et al., 2006].

[12] 3. Angularly resolved elastic light scattering is a function of the shape and optical properties of a particle. Scattering patterns of single atmospheric particles have been measured in an attempt to classify particle morphologies [Kaye et al., 2000; Aptowicz et al., 2006].

1.3.3. Single-Particle Laser-Induced-Fluorescence Methods

[13] Instruments have been developed to measure LIF in one or two broadband wavelength channels [Pinnick et al., 1995; Hairston et al., 1997; Reyes et al., 1999; Seaver et al., 1999; Eversole et al., 1999; Kaye et al., 2000; Eversole et al., 2001; Ho, 2002], and such an instrument is available commercially, the UV-APS from TSI, Inc. Instruments have been developed to detect the dispersed LIF spectrum over an appreciable wavelength range [Hill et al., 1999; Pan et al., 2003a]. Pinnick et al. [2004] assembled a Particle Fluorescence Spectrometer (PFS) that measures, on the fly, the fluorescence spectra of individual airborne particles in the atmosphere, and used it to measure moderate-resolution single-particle spectra of ambient atmospheric aerosol. The spectra were clustered using an unstructured hierarchical technique. Some of the clusters found had fluorescent spectra appearing similar to some materials that are known to exist as aerosols. For example, (1) one was similar to that of bacteria and proteins; (2) one was similar to spectra of collections of marine aerosol [Oppo et al., 1999]; (3) one was similar to some spectra of cellulose (pure cellulose should not fluoresce, but compounds in plant cell walls such as ferulic acid, have spectra similar to that of typical cellulose); and (4) one was similar to some spectra of fulvic or humic acids [De Souza Sierra et al., 2000; Krivacsy et al., 2000; Klapper et al., 2002; Goldberg and Weiner, 1994] or humic-like substances (HULIS) [Ouatmane et al., 2002].

[14] The LIF technique should be useful for measuring one or a few primary OC tracer molecules for different types of particles. Essentially all the fluorescence of aerosols is probably from aromatic OC, but only a small fraction of aromatics are strongly fluorescent. A LIF spectrum may be dominated by one or a few fluorescent moieties, even when the predominant fluorescent moiety contains a very small fraction of the mass. Also, many molecules that are fluorescent may not occur in sufficiently high concentrations in aerosol to contribute significantly to LIF spectra. For example, pure bacteria and bacterial spores, when illuminated with light in the 260 to 280 nm range, typically have fluorescence spectra that are strongly dominated by the aromatic amino acid tryptophan. Therefore tryptophan can serve as a LIF tracer for bacteria and proteins, similar to the way in which levoglucosan can be a tracer for particles from biomass burning when measured with GC-MS [Simoneit et al., 1999], and calcium dipicolinate can be a tracer for bacterial spores when measured with MS [Srivastava et al., 2005]. NADH and flavins are other strongly fluorescent biomolecules that occur in all cells, and which can be dominant fluorors from bacteria when the excitation wavelength is longer than, for example, 305 nm; however, when excited at 263 nm as in work by Pinnick et al. [2004] their contribution is swamped by that of tryptophan. Many other aromatic moieties in cells are negligibly fluorescent, for example, the bases for DNA and RNA (adenine, guanine, thymine, cytosine, and uracil), which absorb well at 263 nm, have very low quantum efficiencies (adenine's efficiency is about 0.0001 as compared to tryptophan's 0.15). Little of the aromatic material in noncombustion aerosol that is biological or of recent biological origin (bacteria, proteins, pollens, lignins, and humic substances) is PAH, although OC in vegetation, for example, pine tar, can contain PAH from terpenes, and wind action may blow particles of these into the atmosphere. Emission rates of phosphorescent mineral aerosols are probably too small for single-shot measurements with the PFS, with its 1-μs measurement window during the 10-ns laser pulse. Previously we thought some nonaromatic OC compounds (e.g., chlorophyll) might contribute significantly to these fluorescence spectra, but were unable to find evidence to support this.

[15] A class of atmospheric OC material that may not be seen with the PFS includes secondary organic aerosol (SOA) generated from volatile terpenes (e.g., pinene), terpenoids and other volatiles emitted from vegetation. We are not aware of terpenoids formed by atmospheric reactions of volatile terpenes/terpenoids that would be detected by the PFS with excitation at 263 nm. Vegetation can emit volatile (e.g., benzaldehyde) and semivolatile (e.g., methyl salicylate, benzyl acetate, estragole [Dudavera et al., 2004]) aromatic OC, which may also react in air and/or condense with the SOA, and possibly these aromatic materials could be sufficiently fluorescent for the SOA formed only from vegetation to be measured with the PFS.

[16] The single-particle LIF spectral technique can provide additional and complementary information to that obtained by other techniques for characterizing OC, and is well-suited for volatile particles. Because a small fraction of molecules (or monomers that go into polymers such as proteins or lignans) are fluorescent, the fluorescent molecules can, in some cases, act as tracers for certain particle types. Because the LIF technique does not destroy the particles, it could be combined with other online instrumentation, for example, MS, or with an airborne particle sorter which collects selected particles for further analysis [Pan et al., 2004].

1.4. Overview

[17] Here improvements to the particle fluorescence spectrometer (PFS) are briefly summarized. Measurements of the fluorescence spectra of individual atmospheric aerosol particles at New Haven, CT and at Las Cruces, NM, and the results of clustering of these spectra are presented. A key result is that the main template spectra found earlier at Adelphi, MD [Pinnick et al., 2004] are found again in New Haven and in Las Cruces, although the fraction of fluorescent particles and the relative populations of particles in various clusters are different.

2. Experimental Methods, Sites, and Analysis

2.1. Experimental Setup

[18] A schematic and photograph of the particle fluorescence spectrometer (PFS) is shown in Figure 1. The current PFS is a miniaturized version of a previous instrument [Pinnick et al., 2004] with improved sensitivity for measurement of smaller particles and with increased efficiency of sampling. A virtual impactor concentrator (MSP model 4220; sample rate of 330 L/min) is used to concentrate particles in the 2- to 10-μm range. The minority outlet flow of the concentrator (nominally 1 L/min) is fed to the PFS inlet nozzle (a double nozzle equipped with sheath flow) which forms a highly focused, laminar, cylindrical (around 300-μm diameter) aerosol jet within the relatively small (5-cm cube) optical chamber. Particles in the jet are detected by elastic-scattering signals from two focused, intersecting, different wavelength (650- and 685-nm), diode laser beams, which are used to trigger a pulsed probe laser and detection system. Fluorescence in particles is excited by a Nd:YLF probe laser frequency-quadrupled to a 263.5-nm wavelength and having a 2-mm beam diameter, 0.05-mJ energy per pulse, and 10-ns pulse length. The detector, a (Hamamatsu model H7260) 32-anode PMT, replaces the ICCD used previously [Pinnick et al., 2004] and increases the data acquisition rate, and reduces storage requirements, but retains sufficient spectral resolution (15 nm) for fluorescence emission. The PFS can: (1) measure fluorescence spectra of bacterial particles as small as 1-μm diameter (as opposed to 3 μm previously), and so can measure particles with sizes of 1–10 μm diameter; (2) measure spectra at rates of many thousands of particles/s (compared to about 10 particles/s previously); (3) measure each particle's elastic scattering which can be used to estimate particle size; and (4) provide a time stamp for each particle's arrival.

Figure 1.

Schematic and photograph of the Particle Fluorescence Spectrometer (PFS). Aerosol is sampled by a virtual impactor concentrator and the minority exit flow fed to a focusing nozzle. Flow through the nozzle forms a laminar aerosol jet within an optical cell. Single particles in the jet are probed on the fly with a pulsed UV laser exciting fluorescence in fluorophors within particles. Fluorescence emission (also elastic scattering) is collected by the reflective objective, passed through a long-pass filter (to attenuate the elastic scattering), dispersed by a concave grating, and sensed by a multianode photomultiplier tube (PMT). Trigger laser diodes (LD1 and LD2) and associated (PMTs), for triggering the pulsed UV laser and fluorescence recording system, are shown in the photo. This technique permits rapid measurement of single-particle light-scattering size and fluorescence spectra with aerosol sample rates of 100–220 L/min for 2- to 9-μm particles.

[19] In this paper, only particles having diameter greater than about 3 μm are analyzed, in order to compare to previous results [Pinnick et al., 2004]. The effective flow rate of the PFS is particle-size dependent (mainly because of the virtual impactor sample inlet but also because of the focusing nozzle), and varies from about 100 L/min for nominal 3 μm particles to about 220 L/min for nominal 6-μm particles to about 190 L/min for nominal 9-μm particles.

2.2. Measurement Sites

[20] The two measurement sites in the present study, New Haven, CT, USA, and Las Cruces, NM, USA, are in regions with very different regional climate. The measurement site of our previous study [Pinnick et al., 2004] at Adelphi, MD, USA, (39°N latitude, elevation 75 m), in the Baltimore-Washington metroplex, is climatically more similar to CT than to NM. The three sites are abbreviated: MD for Adelphi, MD; CT for New Haven, CT; and NM for Las Cruces, NM.

[21] CT, at 41.2°N latitude, is in the Atlantic Coastal region of the USA. The elevation at the site is about 25 m. CT is a moderately populated city (New Haven-Meridan metropolitan area population was 542,000 in 2000) located in the heavily industrialized northeastern US about 10 km from Long Island Sound, 130 km from New York City, and 240 km from Boston, MA. The annual average temperature is 11°C. The New Haven area has moderately high precipitation (134 cm per year) and large deciduous forests with tall trees and other vegetation.

[22] NM (32.2°N latitude; elevation 1200 m) is in the Chihuahuan Desert of the southwestern US and northern Mexico. NM is a moderately populous urban area (metropolitan-area population about 193,000) with relatively low precipitation (about 25 cm per year). The average annual temperature is 18°C. The metroplex of El Paso, Texas–Ciudad Juarez, Mexico is located 80 km to the south. NM has relatively little vegetation, except in the nearby irrigated Mesilla Valley along the Rio Grande River (8 km west), at other riparian areas that are typically far apart, and at high mountain elevations which are mostly tens of kilometers distant. Outside of a small region along the Rio Grande, and near isolated riparian areas, there are essentially no trees near NM. Dust storms are common.

[23] Measurements were made during fall (24–25 October 2006) in CT, and during winter (22–23 January 2007) in NM. The CT and NM sites are located on university campuses (Yale University and New Mexico State University), with moderately traveled city streets located less than 50 m away. At both sites the virtual-impactor inlet for the PFS intake passes through a window or wall (through a 3-m-long, 5-cm-diameter, conductive tube) positioned about 2 m above ground level into a laboratory where the PFS is mounted on an optical table. Significant losses for aerosols less than 10-μm diameter are not expected in this tube because of the relatively high flow rate (330 L/min) and large tube diameter.

2.3. Analysis

[24] In general, between 5 and 50% of the particles drawn into the PFS have fluorescence intensity sufficiently above noise to permit analysis. Two approaches to analyzing the data are employed.

2.3.1. Hierarchical Cluster Analysis

[25] Initially a hierarchical cluster analysis approach was used so as not to bias the results by preconceived notions of possible classes of spectra. The cluster analysis scheme described by Murphy et al. [2003] for analyzing single-particle MS data was followed. Briefly, in this approach, the two spectra that are most similar are combined into a cluster with a spectrum that is the average of the spectra. This combining process is repeated until no two remaining spectra are more similar than some threshold measure of similarity.

[26] Each spectrum is treated as follows.

[27] 1. The elastic scattering at 263.5 nm is used to estimate particle size. Particles with estimated sizes smaller than 3 μm are not considered further (these data will be analyzed in the near future).

[28] 2. The wavelengths are determined from the positions of the first- and second-order diffraction of the 263.5 nm laser line by the grating (263.5 nm and 527 nm). The wavelength coverage is from 295 nm to 608 nm. Two pixels near 527 nm are not used in the analysis of the fluorescence spectra, because once the PFS is aligned such that the elastic scattering peak at 263.5 nm is located at the center of anode 4 of the PMT, the 527 nm is mainly on anode 20 but also partially on anode 21, and the wavelength calibration varies slightly depending on system alignment and on the position of the particle relative to the collection opties.

[29] 3. The averaged background from 100 trigger events was obtained by running the system without particle, but with all other conditions the same as when the fluorescence spectra of aerosol particles is acquired. This background, the average of the signals from anodes 6 through 19, and 22 through 25, was 0.313 counts per anode. This background was subtracted for all fluorescence spectra recorded.

[30] 4. The fluorescence spectra are corrected for spectral variations in detection efficiency caused by the grating, the filter and anode-to-anode variations in the multianode PMT.

[31] 5. Each spectrum was treated as a n-dimensional vector, u = Vv; where v is a unit vector (dot product of v with itself, v · v = 1), and the normalization constant (V), i.e., the unnormalized length of the vector (also the average fluorescence for the spectrum), is

equation image

[32] 6. Spectra that have V below the threshold, Vmin, which is set as twice the sum of averaged background from anode 6 to 19, and 22 to 25, are considered nonfluorescent particles.

[33] Hierarchical clustering of the spectra is done by combining spectra that have the largest dot products. First, the dot product between each vector and every other vector is calculated. Then the two spectra having the largest dot product are combined into one cluster, and the average spectrum for this new cluster is calculated and normalized. Then new dot products are calculated and the process is repeated until the largest dot product is smaller than the chosen threshold (typically between 0.9 and 0.93).

2.3.2. Nonhierarchical Cluster Analysis

[34] We also performed some unstructured nonhierarchical clustering where the two closest spectra found were used as the starting point for a cluster. Once this starting point is found, all other spectra that have a dot product above the threshold are added to that cluster.

2.3.3. Selection of Template Spectra

[35] Selection of the template spectra to be used as matched filters for analyzing all of the spectra in a consistent manner is described in section 3.2. These templates were chosen so as to have a relatively small number of clusters (8 to 10). When the normalized average spectra for these clusters were used to analyze all the spectra in the data sets, for a dot product threshold of 0.9, about 90% of the spectra fit into 10 clusters.

3. Results

3.1. Time Dependence of Total Particles and Subfraction of Fluorescent Particles

[36] The PFS was employed to continuously sample, over 48-hour periods, the concentration, size, and fluorescence spectra of individual atmospheric particles at CT and NM. Particle count rates varied from a few tens to a few hundreds per minute. Rates were typically higher during the day than the night (Figure 2). Particle sample rates are size dependent as explained in section 2.1. A count rate of 100 per minute corresponds to a concentration of roughly 1 particle per liter. The top (grey) line gives total particle rates and the lower (black) line indicates the fluorescent-particle rates. On average, about 49% of particles at CT, and 17% of particles at NM, have fluorescence spectra above noise (over the 295- to 608-nm wavelength range). Fluorescent particles are not always proportional to the total particle concentration. During the 2 day measurement campaign at CT, the atmospheric air temperature varied from 5°C to 13°C, and relative humidity from 38 to 93%; at NM, temperatures during the 2 day sampling period were −3°C to 10°C and relative humidity 38–100%. The effect of raising the temperature of aerosol particles from ambient temperature to room temperature (nominally 25°C) over the (approximately) 2 second transit time of particles through the 5-cm-diameter inlet tubing, virtual impactor, and inlet nozzle assembly is unknown but believed to be small. No obvious correlation between the temperature, humidity and the particle concentration was evident.

Figure 2.

Count rate of total particles and fluorescent particles (having aerodynamic diameter greater than approximately 3 μm) measured at New Haven, CT, and at Las Cruces, New Mexico. A count rate of 100 per minute corresponds to a concentration of about 1 particle per liter. Local time is shown.

3.2. Measurement of Particle Fluorescence Spectra

[37] A sample of PFS fluorescence-spectra data is presented in Figure 3. Shown are spectra for a series of 1000 consecutively arriving particles measured during 23 October 2006 in CT, over a 16 min period. Spectra of the small fraction of particles that have the largest fluorescence emission dominate this compressed view of the data. The attenuated elastic scattering (sizing) peak at 263.5 nm has considerably less variability than the fluorescence peaks because fluorescence is roughly proportional to fluorescence quantum efficiency and to particle volume [Hill et al., 2001], while elastic scattering by particles having a ratio of size to wavelength in this range is roughly proportional to cross-sectional area [Mitch, 1995].

Figure 3.

Raw fluorescence spectra of 1000 atmospheric particles measured on 23 October, 1000–1016 LT at New Haven, Connecticut. Most particles have weak fluorescence. The fluorescent particles emit mainly in the 300- 500-nm wavelength range. The peak at 263.5 nm is the elastic scattering of the pulsed 263.5-nm laser beam from the particles leaking through the long-pass filter (dimethyl formamide diluted with water in a 1-cm-thick cell). This scattering is used to estimate particle size.

3.3. Cluster Analysis of the Spectra

[38] An unstructured hierarchical cluster analysis was performed on two ensembles of spectra: one measured at CT over a 48 hr period from 1000 local time (LT) 23 October to 1000 LT 25 October 2006; the other measured at NM over a 48 hour period from 1600 LT on 22 January to 1600 LT on 24 January 2007. The threshold for fluorescence was chosen to be as small as possible and still retain spectra that do not appear too noisy. In CT, 124,418 total spectra were measured. Of these, 21,007 particles were larger than 3 μm according to the elastic scattering measurement. Of these particles larger than 3 μm, 10,367 (49%) had fluorescence above threshold, and were subjected to cluster analysis. In NM the data set had 363,846 total spectra, and 58,260 of these were from particles larger than 3 μm. Of these, 10,157 (17%) had fluorescence above threshold and were subjected to cluster analyses.

[39] Figure 4 illustrates the primary cluster spectra found at NM and CT along with the template spectra found at Adelphi, Maryland, MD [Pinnick et al., 2004] for comparison. The notation of the MD template spectra, which were labeled in order of increasing peak wavelength, was followed for the spectra that were similar to the MD templates. Most of the MD templates occur in both CT and NM, and are in most cases remarkably similar. These results are discussed in section 4.

Figure 4.

Fluorescence spectra templates derived from hierarchial cluster analyses for Adelphi, Maryland (MD), New Haven, Connecticut (CT), and Las Cruces, New Mexico (NM) data. The percentage of (fluorescent) particles that cluster into each template is also shown. The most highly populated cluster is that characteristic of humic acid/HULIS (cluster 8); the least populated is cluster 6 (no suggestion of possible fluorors).

[40] Here the approach to choosing the template spectra is somewhat less arbitrary than the method used for the spectra at MD. In the analysis of the MD data, an unstructured hierarchical cluster analysis was performed separately on each of 16 data sets. The cluster spectra from all data sets were examined to select representative clusters. A relatively small set of templates (8–10) was chosen that could: (1) represent the largest fraction of all the spectra, but also (2) represent particles that occur as small minorities, but have distinctive spectra that do not appear to be noise (e.g., cluster 6 has only 0.5% of the fluorescent particles at MD). In the analysis of MD data, the spectra were not combined into one ensemble for a cluster analysis of the ensemble, because the data were taken over a period of about two months, and there was concern that the spectra may vary as the season changed from winter to spring. Also, the computational power to cluster such a large ensemble was not available.

[41] Here the two large data sets analyzed (at NM and at CT) were each acquired over a period of about 48 hours, and so concerns of seasonal variability were not important. Also, for the present data, adequate computational resources were available. Thus the complete NM and at CT data sets were analyzed, where for each site there were over 10,000 fluorescence spectra for particles that exceeded both our size (roughly 3-μm diameter) and fluorescence thresholds. The most common spectra from each of these analyses were used as the templates, but expanded to include any weakly populated spectra templates that also occurred at MD (cluster 6).

[42] There was one adjustable parameter in the present analysis: the threshold for the dot-product, which was also an adjustable parameter in the MD analysis. The dot-product threshold was adjusted so that a relatively small (e.g., 8 to 10) number of templates could cluster with a large fraction (around 90%) of the spectra. The hierarchical cluster analysis was performed initially on each of the two (NM and CT) data sets using dot-product thresholds of 0.9, 0.91, 0.92 and 0.93. A threshold of 0.92 was then chosen because: (1) with a threshold of 0.9 or 0.91, spectra that seemed too different combined into one cluster; and (2) with a threshold of 0.93, the number of clusters populated with significant numbers of spectra appeared too large. A reason for 0.92 threshold seeming most appropriate for CT and NM, as compared to the 0.9 used at MD, may be that the smaller signal-to-noise of the MD data would, for an identical group of particles, result in a lower average dot product. A more detailed examination of the clusters that are obtained with different thresholds is left for later investigation.

[43] In addition to the hierarchical cluster analyses described above, a nonhierarchical cluster analysis using the approach mentioned in the experimental section was also performed. With a threshold of 0.9, most of the clusters found using either approach were similar, as would be expected for clusters that do not overlap to a significant extent.

[44] It may seem odd to use 0.92 for the unstructured cluster analysis, but then use 0.9 for analyzing the clusters using the known templates. In the unstructured analysis, clusters combine to make a new cluster with a new average spectrum. In doing so, many of the spectra in the new cluster may be farther from the average spectrum than the dot-product threshold used. Then, if this same average spectrum is used as a template with that dot-product threshold in a structured cluster analysis, a smaller number of spectra may combine with this cluster than they did in the unstructured analysis. By using 0.92 for the unstructured analysis, the generation of either too few or too many clusters is avoided.

3.4. Results of Analyzing the Atmospheric Aerosol Using the Chosen Template Spectra

[45] Figures 5 and 6show, for CT and NM: (1) template spectra found using the unstructured hierarchical analysis, and (2) a sample of some of the spectra that are in these clusters, i.e., those with dot products greater than 0.9. Not all the spectra combining with these templates are shown, because some clusters have thousands of spectra, and consequently there are too many overlapping lines. The percentages of the complete data sets that combine with these clusters are shown in parentheses. The 10 spectral clusters account for more than 91% of all fluorescent particles greater than about 3 μm. The most populated clusters are 2, 5, and 8; the least populated clusters 1, 6, and 7. Table 1 gives the percentages of particles in each of the clusters at the three sites.

Figure 5.

Normalized fluorescence spectra of single atmospheric aerosol particles (thin solid lines) measured at New Haven, Connecticut. The template spectrum for each cluster (derived from a hierarchical cluster analysis appearing in Figure 4) is shown by a thick dashed line. The sharp peaks at 263.5 nm are due to the (attenuated) elastically scattered light from each particle and give a measure of particle size. The peaks at 527 nm are due to the second-order diffraction of the 263.5-nm laser by the grating. For the clusters populated with many spectra, the measured spectra look like a very thick line with noise on the edges because so many spectra overlap.

Figure 6.

Same as Figure 5 except for Las Cruces, New Mexico, data.

Table 1. Percentage of Fluorescent Particles Found in Various Spectral Clusters (11 Total) in Maryland, New Mexico, and Connecticut Aerosol Dataa
 Cluster
1234567891011Not Fit>3 μm + flu>3 μmTotal
  • a

    The most populated cluster (cluster 8), which is characteristic of humic acids/HULIS, has approximately one third of fluorescent particles at all three sites. Also shown are the percentage of fluorescent particles “not fit” in any of the 11 clusters, the number of fluorescent particles in the aerosol ensembles at the different sites used in the cluster analysis, the total number of particles greater than 3 μm diameter, and the total number of particles used in the analysis.

MD(>3 μm), %3.423.711.88.310.80.53.228.2<0.5<0.5<0.510.211,050130,000130,000
NM(> 3μm), %1.213.8<0.51.510.70.51.339.417.44.30.99.010,15758,260363,846
CT (> 3μm), %0.528.28.81.314.00.053.233.5<0.51.4<0.59.110,36721,007124,418

4. Discussion

[46] Atmospheric aerosol is enormously complex. Each type of aerosol measurement technique (e.g., MS, NMR, collection of aerosol followed culturing or by extraction and GC-MS), provides a different, often partially overlapping view, of atmospheric aerosol.

4.1. What Do the Similarities in LIF Spectra Say About Atmospheric Aerosol?

[47] Most of the fluorescent particles can be clustered into a few spectral types that appear to mostly be quite distinct (by visual inspection; see Figure 4). These clusters are robust enough that most of the cluster types appear independently in measurements made in three different geographical locations having two very distinct climates, and in measurements made with two different PFS instruments.

[48] The seven clusters that are most similar (similarities determined by visual inspection; see Figure 4) at the three sites (clusters 1, 2, 4, 5, 6, 7, and 8) have the majority of all particles (77% in MD, 68% in NM, and 81% in CT). Clusters 1, 4, 5, 6, and 7 are remarkably similar at the three sites, and account for 10–20% of the fluorescent particles. (Cluster 1 is not quite as similar as the others in this group, mainly because of the difference at the shortest valid wavelength measured; the increased intensity near 295 nm in CT and NM is attributable to the different spectral characteristics of the filter used to block the 263.5-nm light.) The spectra in clusters 2 and 8 are also somewhat similar, and account for 52% in MD, 53% in NM, and 62% in CT. In cluster 2 the differences in the 380- to 510-nm range are more apparent. For cluster 8, the template spectrum found in MD seems to peak at somewhat longer wavelengths.

[49] Some factors contributing to the similarities between clusters at the different cites are as follows.

[50] 1. Long-range atmospheric transport and turbulent mixing tend to homogenize the ambient aerosol. For example, forest fire particulates can travel from Alaska to Nova Scotia, Canada [Duck et al., 2007]; dust from Asia has been measured across the USA [VanCuren and Cahill, 2002]; and Saharan dust can travel across the Atlantic Ocean to the USA.

[51] 2. Combustion (biomass, engines, etc.) emissions, both the primary and secondary organic aerosol, and the new compounds generated as these aerosols age, may be similar at different geographic locales. Humidity, temperature and atmospheric turbulent structure may affect the formation of aerosol particles, partitioning of combustion products between gas and particle phases, and processing of particles.

[52] 3. The predominant fluorophors in biological materials (tryptophan, NADH or similar compounds, and flavins) are preserved across species. Some of the main fluorors in plant cell walls (e.g., ferulates), and wood (lignans, sinapyl alcohols, etc.) are common in diverse environments.

4.2. Possible Fluorescent Materials or Particle Types in the Clusters

[53] Typically, the larger the distance over which electrons are conjugated the further the peak of the emission shifts toward the red.

4.2.1. Cluster 1

[54] Compounds having one aromatic ring or an aromatic ring with a small degree of additional conjugation could fall into cluster 1. The spectrum is cut at the short-wavelength end by the filter. The aromatic amino acid tyrosine has a similar spectrum [Hill et al., 1999], as would pure proteins that contain tyrosine but no tryptophan. In proteins that contain both tyrosine and tryptophan, the tyrosine fluorescence is typically very weak because it transfers absorbed energy to tryptophan, which then fluoresces [Lakowicz, 1983]. Compounds sometimes found in OC aerosol that may contribute are benzoic acids and phenols.

4.2.2. Cluster 2

[55] Compounds contributing to cluster 2 may include single-ring aromatic hydrocarbons having additional conjugated bonds. Double-ring aromatics may contribute.

4.2.3. Cluster 3

[56] The most similar spectrum in CT peaks near the same wavelength, but then decreases somewhat linearly with increasing wavelength. Cluster 3 is similar to spectra of some bacterial samples [Hill et al., 1999], especially ones that have not been washed well. The spectrum appears to be a mixture of compounds. The hump near 320 nm is similar to that in cluster 2. The broad hump is similar to cluster 8 which is characteristic of some mixtures of humic acids or HULIS compounds.

[57] Cluster 4's spectrum is similar to that of pure tryptophan and is characteristic of bacteria and tryptophan-containing proteins. Pinnick et al. [2004] measured fluorescence of tryptophan test particles with the previous fluorescence system, and found 96% of these test particles combined with this cluster [see also Hill et al., 1999, Figures 2 and 8]. Cigarette side-smoke had a similar spectrum [Hill et al., 1999], although it is not clear that particles as large as 3 μm would be in side-smoke. Other double ring aromatic compounds may have similar spectra. Bacteria or bacterial spores grown in a fluorescent medium and not washed well may not combine with this cluster [see Hill et al., 1999, Figure 9]. The concentration of particles in this cluster is about 90/m3 in CT and NH, about 2 orders of magnitude less than that of total bacterial cells measured by Tong and Lighthart [1999] at a rural site near Corvallis, Oregon. The fact that bacterial cells smaller than 3 μm likely dominate the concentration in Corvallis likely account for most of this difference. Bacterial particles were measured using single-particle MS at the San Francisco airport, and found to occur in small concentrations.

[58] Cluster 5's spectrum has similarity to spectra of collections of marine aerosol [Oppo et al., 1999]. It may include bacterial or other biological particles, but the peak emission is further to the red than for laboratory-grown bacteria. It is populated more uniformly (11% NM; 14% CT; and 11% MD) across the three sites than any other cluster.

[59] Cluster 6's spectrum (peaking near 380 nm) is strikingly narrow. This feature is rare. However, the feature is quite distinct and not believed to be noise.

[60] Cluster 7's spectrum (peaking near 430 nm) is similar to spectra reported for cellulose [Olmstead and Gray, 1997], except that cellulose has more of a shoulder near 380 nm. Spectra of humic substances are typically broader. Merola et al. [2001] present 266-nm-excited LIF spectrum of a condensed samples of diesel exhaust, rapeseed oil, and light oil. All of these spectra are broader than that of cluster 7.

[61] Cluster 8's spectrum is similar to some recorded for fulvic or humic acids [De Souza Sierra et al., 2000; Krivacsy et al., 2000; Klapper et al., 2002; Goldberg and Weiner, 1994] or humic-like substances (HULIS) [Ouatmane et al., 2002]. A distribution of PAH or other combustion aromatics (maybe partially oxidized PAH), could have such a spectrum. A 266-nm-excited LIF spectrum of condensed diesel exhaust [Merola et al., 2001] has similarities to cluster 8's spectrum in the 350 to 470 nm range, but has much more fluorescence than cluster 8 in the 310- to 340-nm range.

[62] Cluster 9's spectrum is somewhat similar to cluster 3, which may be thought of a mixture of the spectra of (and maybe the compounds contributing to) clusters 2 and 8. However, cluster 9 appears more humic-like, as the spectrum is relatively flat between 340 nm and 500 nm. Cluster 10's spectrum is similar to that of cluster 8 (humic-like), but having a prominent 570 nm peak. Cluster 11's spectrum is the cluster in NM that is most similar to Cluster 3. However, its large spike near 575 nm does not occur in cluster 3. We do not have suggestions for OC materials that would explain the sharp peaks near 380 nm (cluster 6) and 575 nm (cluster 10 and NM cluster 11), and conjecture that some inorganic materials may be responsible.

[63] Many combinations of fluorophors in atmospheric aerosol could have spectra similar to most of the clusters. Determining which of these possible combinations occur in single atmospheric particles will require more research. For example, selected distributions of PAHs and of aged PAHs may be assembled which could have spectra of any clusters other than clusters 1 and 6. Also, no single spectral template looks exclusively bacterial. Bacteria can occur in complex mixtures/agglomerates in airborne particles. Bacteria may be strong contributors to the spectra of clusters 2, 3 and 4, and may be less strong contributors to cluster 5. Bacteria in atmospheric particles may change chemically and/or photo chemically.

4.3. Possible Comparison of Fluorescence-Based Aerosol Classification With Aerosol MS

[64] There has been much research to develop and apply aerosol MS techniques to on-line aerosol measurement. Qin and Prather [2006], in an air quality study in Fresno, CA, classify broadly (using ATOFMS) “biomass burning particles” that have a mass-to-charge ratio characteristic of wood smoke and “high mass organic carbon particles” having mass-to-charge peaks characteristic of a number of organic molecules with molecular weights ranging from 110 to 202. Middlebrook et al. [2003], in an Atlanta experiment, employing side-by-side several laser ablation aerosol MS, classify particles into four categories: organic/sulfate, sodium/potassium sulfate, soot/hydrocarbon, and mineral. Zhang et al. [2004] used the Aerodyne AMS (not a single particle device) to classify particles into four broad categories: nonrefractory sulfate, nitrate, ammonium, and organics. The classification of OC aerosols according to fluorescence cluster templates (bacteria-like, cellulose-like, marine aerosol-like, humic-like, and HULIS-like, etc) does not appear to be directly comparable to the more detailed laser ablation aerosol MS data.

[65] Single-particle aerosol MS techniques can measure far smaller particles than can LIF spectral techniques. Single-particle LIF is weak, especially for particles as small as 1-μm diameter. The laser intensity required to measure good spectra of 1-μm particles (about 30 MW/cm2) is approaching the plasma breakdown threshold, which is of the order of 100 MW/cm2 for organic particles [Pinnick et al., 1988]. For lower spectral resolution LIF, the particles could be somewhat smaller. TSI's ATOFMS can measure particles in smaller size ranges: 0.03 to 0.3 μm for the model 3800-030 and 0.3 to 3 μm for the model 3800-100.

[66] Single-particle MS spectra are far more detailed than single-particle LIF spectra, for example, the single-particle MS measurements of bacteria and spores, where dipicolinate and several amino acids are observed [Srivastava et al., 2005; McJimpsey et al., 2006], and may be expected to provide much more ability to discriminate between samples than should LIF spectra. We note, however, that statements regarding possible species identification using single-particle MS alone may be premature, because: (1) there are hundreds of species of bacterial spores, and some may only be aerosolized in specific locations under certain conditions; (2) bacteria in the environment may occur mixed with many different materials, and (3) differences in washing the prepared bacteria or spores may make identification of a specific bacterium much more difficult, as was found in the LIBS work of Hybl et al. [2006]. Also, for extremely complex samples, for example, humic material that may contain some bacteria and fungal parts, it might be that for certain applications, the MS spectra are exceedingly complex, and the LIF spectra may provide information related to tracer fluorophors that may be more easily interpreted.

[67] Instrumentation for LIF is simpler than for MS. Even the probe laser in LIF has been replaced by light emitting diodes in some efforts [Pan et al., 2003b]. For some studies where monitoring at multiple sites is needed, the lower cost of LIF could make it more suitable than MS, even though MS provides more detailed spectra.

4.4. Potential Future Research and Extensions of the LIF Technique

[68] To assess the potential uses of LIF-spectral techniques for extensive monitoring of atmospheric aerosol, either by itself or in combination with other single-particle techniques such as MS or LIBS, much more needs to be understood about the fluorescence spectral properties of atmospheric aerosols and potential tracer fluorophors in these aerosols.

4.4.1. Further Analyses of Aerosol Fluorescence Spectra

[69] Measurements could be made to study the diurnal and seasonal variability of fluorescence cluster types at a particular site or to investigate OC particles associated with sources (combustion engines, sewage treatment plants, animal feedlots, agricultural activity, industrial pollution, etc.). The clusters shown here only include about 90% of the particles. The other 10% should be examined. The dot product thresholds chosen, 0.9 and 0.92, likely forced some clusters to combine; these subclusters could be worth examining on their own. The tree structures should be examined, as should statistics relating to robustness of the clusters. For the clusters shown here, the separations are sufficiently distinct that differences are clearly apparent upon visual inspection. However, for analyses of subclusters more sophisticated statistics are desirable.

4.4.2. PFS With Different-Wavelength or Multiple-Wavelength Excitation Lasers

[70] Single-particle measurements show that changing the laser excitation wavelength can excite different fluorophors [Pinnick et al., 1998, Pan et al., 1999, Hill et al., 1999; Sivaprakasam et al., 2004] and provide additional spectral information about aerosols. For example, bacteria excited with 355-nm light do not exhibit fluorescence from amino acids, but exhibit spectra of NADH and flavins. Sivaprakasam et al. [2004] assembled a laboratory instrument that excites the same particle with both 266- and 355-nm light (separated by 400 ns), and measures the fluorescence in three spectral bands. These measurements provided for an increased ability to discriminate among particles. A PFS that measures more detailed spectral information for each of several excitation sources would provide an increased ability to discriminate and to allow more detailed comparisons with excitation-emission (EEM) fluorescence spectra that have been measured for different materials.

4.4.3. PFS Integrated With Particle Sorter (LIF-Puffer)

[71] To determine the composition of a collection of particles within a spectral cluster, an LIF-spectrum-triggered particle sorter [Pan et al., 2001; Davitt et al., 2006] has been demonstrated to collect only selected particles, which can subsequently be analyzed using other techniques, for example, GC-MS [Williams et al., 2006] or culturing [Lighthart and Tong, 1998].

4.4.4. PFS Integrated With MS or LIBS

[72] The LIF spectral technique is nondestructive and could potentially be used in tandem with other techniques. Hybl et al. [2006] have built a combined LIF/LIBS detector, that uses nonspectrally resolved fluorescence in the 300- to 400-nm band to cue the LIBS system, and measured both test particles and outdoor air. In the aerosol MS system described by McJimpsey et al. [2006], LIF in two spectral bands was used to determine whether to measure a particles' MS, and thereby eliminate some background clutter. The combination of a PFS with LIBS or MS should provide an improved overall discrimination capability, for, for example, the bioaerosols emphasized by Hybl et al. [2006] and McJimpsey et al. [2006]. It could also be used to clarify what the PFS measures, so that the PFS can be more useful when used by itself.

5. Concluding Comments

[73] A most interesting finding of this paper is that the majority of the spectral clusters found with different instruments, during different seasons, and at sites as diverse as New Haven, CT, Las Cruces, NM, and Adelphi, MD are so similar. Although the proportions of particles in the various clusters are different, the similarities in most of the clusters are remarkable. The reasons for these similarities will be interesting to explore further.

Acknowledgments

[74] This research was supported by the Defense Threat Reduction Agency under the Physical Science and Technology Basic Research Program and by ARL mission funds. John Bowersett (ARL) machined the double-nozzle (clean-air sheath) assembly for the Particle Fluorescence Spectrometer.

Ancillary