The WIBS uses light scattering and fluorescence spectroscopy to detect, size, and characterize the properties of interrogated aerosols on a single particle basis (instrument model 4A utilized here). Air is drawn into the instrument at a flow rate of 0.3 L min-1 and surrounded by a filtered sheath flow of 2.2 L min-1. The aerosol sample flow is then directed through a 635 nm, continuous wave (cw) diode laser, which produces elastic scattering measured in both the forward and side directions. Particle sizing in the range of approximately 0.5 to 20 µm is detected by the magnitude of the electrical pulse detected by a photomultiplier tube (PMT) located at 90∘ from the laser beam. Particles whose measured cw laser-scattering intensity (particle size) exceed user-determined trigger thresholds will trigger two xenon flash lamps (Xe1 and Xe2) to fire in sequence, approximately 10 µs apart. The two pulses are optically filtered to emit at 280 and 370 nm, respectively. Fluorescence emitted by a given particle after each excitation pulse is detected simultaneously using two PMT detectors. The first PMT is optically filtered to detect the total intensity of fluorescence in the range 310–400 nm and the second PMT in the range 420–650 nm. So for every particle that triggers xenon lamp flashes, Xe1 produces a signal in the FL1 (310–400 nm) and FL2 (420–650 nm) channels, whereas the Xe2 produces only a signal in the FL3 (420–650 nm) channel because elastic scatter from the Xe2 flash saturates the first PMT. The WIBS-4A has two user-defined trigger thresholds, T1 and T2, that define which data will be recorded. Particles producing a scattering pulse from the cw laser that is below the T1 threshold will not be recorded. This enables the user to reduce data collection during experiments with high concentrations of small particles. Particles whose scattering pulse exceeds the T2 threshold will trigger xenon flash lamp pulses for interrogation of fluorescence. Note that the triggering thresholds mentioned here are fundamentally different from the analysis thresholds that will be discussed in detail later.

Forward-scattered light is detected using a quadrant PMT. The detected light intensity in each quadrant are combined using Eq. (1) into an asymmetry factor (AF), where k is an instrument-defined constant, E is the mean intensity measured over the entire PMT, and Ei is the intensity measured at the ith quadrant (Gabey et al., 2010). AF=k(∑i=1n(E-Ei)2)1/2E This parameter relates to a rough estimate of the sphericity of an individual particle by measuring the difference of light intensity scattered into each of the four quadrants. A perfectly spherical particle would theoretically exhibit an AF value of 0, whereas larger AF values greater than 0 and less than 100 indicate rod-like particles (Kaye et al., 1991, 2005; Gabey et al., 2010). In practice, spherical PSL (polystyrene latex sphere) particles exhibit a median AF value of approximately 5 (Table 1). It is important to note that the AF parameter is not rigorously a shape factor like that used in other aerosol calculations (DeCarlo et al., 2004; Zelenyuk et al., 2006) and only very roughly relates a measure of particle sphericity.

The particle size reported by the internal WIBS calibration introduces significant sizing errors and critically needs to be calibrated before analyzing or reporting particle size. Size calibration was achieved here by using a one-time 27-point calibration curve generated using non-fluorescent PSLs ranging in size from 0.36 to 15 µm. This calibration involved several steps. For each physical sample, approximately 1000 to 10 000 individual particles were analyzed using the WIBS (several minutes of collection). Data collected for each sample were analyzed by plotting a histogram of the side-scatter response reported in the raw data files (FL2_sctpk). A Gaussian curve was fitted to the most prominent mode in the distribution. The median value of the fitted peak for observed side scatter was then plotted against the physical diameter (as reported on the bottle) for each PSL sample. A second-degree polynomial function was fitted to this curve to create the calibration equation that was used on all laboratory data presented here. The calibration between observed particle size and physical diameter may be affected by wiggles in the optical scattering relationship suggested by Mie theory. These theoretical considerations were not used for the calibrations reported here, and so uncertainties in reported size are expected to increase marginally at larger diameters.

Following the one-time 27-point calibration, the particle sizing response was checked periodically using a five-point calibration. The responses of these calibration checks were within 1 standard deviation unit of each other and so the more comprehensive calibration equation was used in all cases. These quicker checks were performed using non-fluorescent PSLs (Polysciences, Inc., Pennsylvania), including 0.51 µm (part number 07307), 0.99 µm (07310), 1.93 µm (19814), 3.0 µm (17134), and 4.52 µm (17135).

Fluorescence intensity in each WIBS channel was calibrated using 2.0 µm green (G0200), 2.1 µm blue (B0200), and 2.0 µm red (R0200) fluorescent PSLs (Thermo-Scientific, Sunnyvale, California). For each particle type, a histogram of the fluorescence intensity signal in each channel was fitted with a Gaussian function, and the median intensity was recorded. Periodic checks were performed using the same stock bottles of the PSLs in order to verify that mean fluorescence intensity of each had not shifted more than 1 standard deviation between particle sample types (Table 1). The particle fluorescence standards used present limitations due to variations in fluorescence intensity between stocks of particles and due to fluorophore degradation over time. To improve reliability between instruments, stable fluorescence standards and calibration procedures (e.g., Robinson et al., 2017) will be important.

Voltage gain settings for the three PMTs that produce sizing, fluorescence, and AF values significantly impact measured intensity values and are recorded here for rough comparison of calibrations and analyses to other instruments. The voltage settings used for all data presented here were set according to manufacturer specifications and are as follows: PMT1 (AF) 400 V, PMT2 (particle sizing and FL1 emission) 450 mV, and PMT3 (FL2, FL3 emission) 732 mV.

An individual particle is considered to be fluorescent in any one of the three fluorescence channels (FL1, FL2, or FL3) when its fluorescence emission intensity exceeds a given baseline threshold. The baseline fluorescence can be determined by a number of strategies but commonly has been determined by measuring the observed fluorescence in each channel when the xenon lamps are fired into the optical chamber when devoid of particles. This is referred to as the “forced trigger” (FT) process because the xenon lamp firing is not triggered by the presence of a particle. The instrument background is also dependent on the intensity and orientation of Xe lamps, voltage gains of PMTs, quality of PMTs based on production batch, orientation of optical components (i.e., mirrors in the optical chamber), etc. As a result of these factors, the background or baseline of a given instrument is unique and cannot been used as a universal threshold. All threshold values used in this study are listed in the Supplement Table S1. Fluorescence intensity in each channel is recorded at an approximate FT rate of one value per second for a user-defined time period, typically 30–120 s. The baseline threshold in each channel has typically been determined as the average plus 3× the standard deviation (σ) of FT fluorescence intensity measurement (Gabey et al., 2010), but alternative applications of the fluorescence threshold will be discussed. Particles exhibiting fluorescence intensity lower than the threshold value in each of the three channels are considered to be non-fluorescent. The emission of fluorescence from any one channel is essentially independent of the emission in the other two channels. The pattern of fluorescence measured allows particles to be categorized into seven fluorescent particle types (A, B, C, AB, AC, BC, or ABC) as depicted in Fig. 1 or as completely non-fluorescent (Perring et al., 2015).

Other threshold strategies have also been proposed and will be discussed. For example, Wright et al. (2014) used set fluorescence intensity value boundaries rather than using the standard Gabey et al. (2010) definition that applies a threshold as a function of observed background fluorescence. The Wright et al. (2014) study proposed five separate categories of fluorescent particles (FP1 through FP5). Each definition was determined by selecting criteria for excitation–emission boundaries and observing the empirical distribution of particles in a three-dimensional space (FL1 vs. FL2 vs. FL3). For the study reported here, only the FP3 definition was used for comparison, because Wright et al. (2014) postulated the category as being enriched with fungal spores during their ambient study and because they observed that these particles scaled more tightly with observed ice nucleating particles. The authors classified a particle in the FP3 category if the fluorescence intensity in FL1 > 1900 arbitrary units (a.u.) and between 0 and 500 a.u. for each FL2 and FL3.

Pollen samples were aerosolized using the dry powder vial (P1, P2) and tapping (P3) methods detailed above. Samples were also collected by impaction onto a glass microscope slide for visual analysis using a home-built, single-stage impactor with D50 cut ∼ 0.5 µm at flow rate 1.2 L min-1. Pollen was analyzed using an optical microscope (VWR model 89404-886) with a 40× objective lens. Images were collected with an AmScope complementary metal-oxide semiconductor camera (model MU800, 8 megapixels).

The WIBS is routinely used as an optical particle counter applied to the detection and characterization of fluorescent biological aerosol particles. Each interrogated particle provides five discreet pieces of information: fluorescence emission intensity in each of the three detection channels (FL1, FL2, and FL3), particle size, and particle asymmetry. Thus, a thorough summary of data from aerosolized particles would require the ability to show statistical distributions in five dimensions. As a simple, first-order representation of the most basic summary of the 69 particle types analyzed, Fig. 2 and Table 2 show median values for each of the five data parameters plotted in three plot styles (columns of panels in Fig. 2).

For the sake of WIBS analysis, each pollen type was broken into two size categories, because it was observed that most pollen species exhibited two distinct size modes. The largest size mode peaked above 10 µm in all cases and often saturated the sizing detector (see also fraction of particles that saturated particle detector for each fluorescence channel in Table 2). This was interpreted to be intact pollen. A broad mode also usually appeared at smaller particle diameters for some pollen species, suggesting that pollen grains had ruptured during dry storage or through the mechanical agitation process. This hypothesis was supported by optical microscopy through which a mixture of intact pollen grains and ruptured fragments was observed (Fig. S2). For the purposes of this investigation, the two modes were separated at the minimum point in the distribution between modes in order to observe optical properties of the intact pollen and pollen fragments separately. The list number for each pollen (Tables 2, S1 in the Supplement) is consistent for the intact and fragmented species, though not all pollen exhibited obvious pollen fragments.

The WIBS was developed primarily to discriminate biological from non-biological particles, and the three fluorescence channels broadly facilitate this separation. Biological particles, i.e., pollen, fungal spores, and bacteria (top row of Fig. 2), each show strong median fluorescence signal in at least one of the three channels. In general, all fungal spores sampled (blue dots) show fluorescence in the FL1 channel with lower median emission in FL2 and FL3 channels. Both the fragmented (pink dots) and intact (orange dots) size fractions of pollen particles showed high median fluorescence emission intensity in all channels, varying by species and strongly as a function of particle size. The three bacterial species sampled (green dots) showed intermediate median fluorescence emission in the FL1 channel and very low median intensity in either of the other two channels. To support the understanding of whole biological particles, pure molecular components common to biological material were aerosolized separately and are shown as the second row of Fig. 2. Each of the biofluorophores chosen shows relatively high median fluorescence intensity, again varying as a function of size. Key biofluorophores such as NAD, riboflavin, tryptophan, and tyrosine are individually labeled in Fig. 2d. Supermicron particles of these pure materials would not be expected in a real-world environment but are present as dilute components of complex biological material and are useful here for comparison. In general, the spectral properties summarized here match well with fluorescence excitation emission matrices presented by Pöhlker et al. (2012, 2013)

In contrast to the particles of biological origin, a variety of non-biological particles were aerosolized in order to elucidate important trends and possible interferences. The majority of non-biological particles shown in the bottom row of Fig. 2 show little to no median fluorescence in each channel and are therefore difficult to differentiate from one another in the figure. For example, Fig. 2g (lower left) shows the median fluorescence intensity of six different groups of particle types (33 total dots) but almost all overlap at the same point at the graph origin. The exceptions to this trend include the PAHs (blue dots), common household fibers (green), and several types of combustion soot (black dots). The fluorescent properties of PAHs are well known both in basic chemical literature and as observed in the atmosphere (Niessner and Krupp, 1991; Finlayson-Pitts and Pitts, 1999; Panne et al., 2000; Slowik et al., 2007). PAHs can be produced by a number of anthropogenic sources and are emitted in the exhaust from vehicles and other combustion sources as well as from biomass burning (Aizawa and Kosaka, 2010, 2008; Abdel-Shafy and Mansour, 2016; Lv et al., 2016). PAHs alone exhibit high fluorescence quantum yields (Pöhlker et al., 2012; Mercier et al., 2013) but as pure materials are not usually present in high concentrations at sizes large enough (> 0.8 µm) to be detected by the WIBS. Highly fluorescent PAH molecules are also common constituents of other complex particles, including soot particle agglomerates. It has been observed that the fluorescent emission of PAH constituents on soot particles can be weak due to quenching from the bulk material (Panne et al., 2000). Several examples of soot particles shown in Fig. 2g are fluorescent in FL1 and indeed should be considered as interfering particle types, as will be discussed. Three common household fiber particles (laboratory wipes and two colors of cotton t-shirts) were also interrogated by rubbing samples over the WIBS inlet because of their relevance to indoor aerosol investigation (e.g., Bhangar et al., 2014, 2016; Handorean et al., 2015). These particles (dark blue dots, Fig. 2 bottom row) show varying median intensity in FL1, suggesting that sources such as tissues, cleaning wipes, and cotton clothing could be sources of fluorescent particles within certain built environments.

Another interesting point from the observations of median fluorescence intensity is that the three viable bacteria aerosolized in this study each show moderately fluorescent characteristics in FL1 and low fluorescent characteristics in FL2 and FL3 (Fig. 2a–c). A study by Hernandez et al. (2016) also focused on analysis strategies using the WIBS and shows similar results regarding bacteria. Of the 14 bacteria samples observed in the Hernandez et al. (2016) study, 13 were categorized as predominantly A-type particles, meaning they exhibited fluorescent properties in FL1 and only a very small fraction of particles showed fluorescence above the applied threshold (FT + 3σ) in either FL2 or FL3. The FL3 channel in the WIBS-4A has an excitation of 370 nm and emission band of 420–650 nm, similar to that of the UV-APS with an excitation of 355 nm and emission band of 420–575 nm. Previous studies have suggested that viable microorganisms (i.e., bacteria) show fluorescence characteristics in the UV-APS due to the excitation source of 355 nm that was originally designed to excite NAD(P)H and riboflavin molecules present in actively metabolizing organisms (Agranovski et al., 2004; Hairston et al., 1997; Ho et al., 1999; Pöhlker et al., 2012). Previous studies with the UV-APS and other UV-LIF instruments using approximately similar excitation wavelengths have shown a strong sensitivity to the detection of “viable” bacteria (Hill et al., 1999; Pan et al., 1999; Hairston et al., 1997; Brosseau et al., 2000). Because the bacteria here were aerosolized and detected immediately after washing from growth media, we expect that a high fraction of the bacterial signal was a result of living vegetative bacterial cells. The results presented here and from other studies using WIBS instruments, in contrast to reports using other UV-LIF instruments, suggest that the WIBS-4A is highly sensitive to the detection of bacteria using 280 nm excitation (only FL1 emission), but less so using the 370 nm excitation (FL3 emission) (e.g., Perring et al., 2015; Hernandez et al., 2016). A study by Agranovski et al. (2003) also demonstrated that the UV-APS was limited in its ability to detect endospores (reproductive bacterial cells from spore-forming species with little or no metabolic activity and thus low NAD(P)H concentration). The lack of FL3 emission observed from bacteria in the WIBS may also suggest a weaker excitation intensity in Xe2 with respect to Xe1, manifesting in lower overall FL3 emission intensity (Könemann et al., 2017). Gain voltages applied differently to PMT2 and PMT3 could also impact differences in relative intensity observed. Lastly, it has been proposed that the rapid sequence of Xe1 and Xe2 excitation could lead to quenching of fluorescence from the first excitation flash, leading to overall reduced fluorescence in the FL3 channel (Sivaprakasam et al., 2011). These factors may similarly affect all WIBS instruments and should be kept in mind when comparing results here with other UV-LIF instrument types.

The purpose of Fig. 2 is to distill complex distributions of the five data parameters into a single value for each in order to show broad trends that differentiate biological and non-biological particles. By representing the complex data in such a simple way, however, many relationships are averaged away and lost. For example, the histogram of FL1 intensity for fungal spore Aspergillus niger (Fig. S3) shows a broad distribution with long tail at high fluorescence intensity, including ca. ∼ 6 % of particles that saturate the FL1 detector (Table S2). If a given distribution were perfectly Gaussian and symmetric, the mean and standard deviation values would be sufficient to fully describe the distribution. However, given that asymmetric distributions often include detector-saturating particles, no single statistical fit characterizes data for all particle types well. Median values were chosen for Fig. 2 knowing that the resultant values can reduce the physical meaning in some cases. For example, the same Aspergillus niger particles show a broad FL1 peak at ∼ 150 a.u. and another peak at 2047 a.u. (detector saturated), whereas the median FL1 intensity is 543 a.u., at which point there is no specific peak. In this way, the median value only broadly represents the data by weighting both the broad distribution and saturating peak. To complement the median values, however, Table 2 also shows the fraction of particles that were observed to saturate the fluorescence detector in each channel.

The representation of median values for each of the five parameters (Fig. 2) shows broad separation between particle classes, but discriminating more finely between particle types with similar properties by this analysis method can be practically challenging. Rather than investigating the intensity of fluorescence emission in each channel, however, a common method of analyzing field data is to apply binary categorization for each particle in each fluorescence channel. For example, by this process, a particle is either fluorescent in a given FL channel (above emission intensity threshold) or non-fluorescent (below threshold). In this way, many of the challenges of separation introduced above are significantly reduced, though others are introduced. Perring et al. (2015) introduced a WIBS classification strategy by organizing particles sampled by the WIBS as either non-fluorescent or into one of seven fluorescence types (e.g., Fig. 1).

Complementing the perspective from Fig. 2, stacked particle type plots (Fig. 3) show qualitative differences in fluorescence emission by representing different fluorescence types as different colors. The most important observation here is that almost all individual biological particles aerosolized (top two rows of Fig. 3) are fluorescent, meaning that they exhibit fluorescence emission intensity above the standard threshold (FT baseline + 3σ) in at least one fluorescence channel and are depicted with a non-gray color. Figure S4 shows the stacked particle type plots for all 69 materials analyzed in this study as a comprehensive library. In contrast to the biological particles, most particles from non-biological origin were observed not to show fluorescence emission above the threshold in any of the fluorescence channels and are thus colored gray. For example, 11 of the 15 samples of dust aerosolized show < 15 % of particles to be fluorescent at particle sizes < 4 µm. Similarly, four of five samples of HULIS aerosolized show < 7 % of particles to be fluorescent at particle sizes < 4 µm. The size cut point here was chosen arbitrarily to summarize the distributions. Two examples shown in Fig. 3 (Dust 10 and HULIS 3) are representative of average dust and HULIS types analyzed, respectively, and are relatively non-fluorescent. Of the four dust types that exhibit a higher fraction of fluorescence, two (Dust 3 and Dust 4) are relatively similar and show ∼ 75 % fluorescent particles < 4 µm, with particle type divided nearly equally across the A, B, and AB types (Fig. S4i). The two others (Dust 2 and Dust 6) show very few similarities between one another, where Dust 2 shows size-dependent fluorescence and Dust 6 shows particle type A and B at all particle sizes (Fig. S4i). As seen by the median fluorescence intensity representation (Fig. 2, Table 2), however, the relative intensity in each channel for all dusts is either below or only marginally above the fluorescence threshold. Thus, the threshold value becomes critically important and can dramatically impact the classification process, as will be discussed in a following section. Similarly, HULIS 5 (Fig. S4k) is the one HULIS type that shows an anomalously high fraction of fluorescence and is represented by B, C, and BC particle types, but at intensity only marginally above the threshold value and at 0 % detector saturation in each channel. HULIS 5 is a fulvic acid collected from a eutrophic saline coastal pond in Antarctica (Brown et al., 2004; McKnight et al., 1994). The collection site lacks the presence of terrestrial vegetation, and therefore all dissolved organic material present originates from microbes. HULIS 5, therefore, is not expected to be representative of soil-derived HULIS present in atmospheric samples in most areas of the world. We present the properties of this material as an example of relatively highly fluorescing, non-biological aerosol types that could theoretically occur, but without comment about its relative importance or abundance.

Several types of non-biological particles, specifically brown carbon and combustion soot and smoke, exhibited higher relative fractions of fluorescent particles compared to other non-biological particles. Two of the three types of brown carbon sampled show > 50 % of particles to be fluorescent at sizes > 4 µm (Fig. 3i, l), though their median fluorescence is relatively low and neither shows saturation in any of the three fluorescent channels. Out of six soot samples analyzed, four showed > 69 % of particles to be fluorescent at sizes > 4 µm, most of which are dominated by B particle types. Two samples of combustion soot are notably more highly fluorescent in both fraction and intensity. Soot 3 (fullerene soot) and Soot 4 (diesel soot) show FL1 intensity of 318 and 751 a.u., respectively, and are almost completely represented as A particle type. The fullerene soot is not likely a good representative of most atmospherically relevant soot types, but diesel soot is ubiquitous in anthropogenically influenced areas around the world. The fact that it exhibits high median fluorescence intensity implies that increasing the baseline threshold slightly will not appreciably reduce the fraction of particles categorized as fluorescent, and these particles will thus be counted as fluorescent in most instances. The one type of wood smoke analyzed (Soot 6) shows ca. 70 % fluorescent at > 4 µm, mostly in the B category, with moderate to low FL2 signal, which also presents similarly as cigarette smoke. Additionally, the two smoke samples in this study (Soot 5, cigarette smoke, and Soot 6, wood smoke) share similar fluorescent particle type features with two of the brown carbon samples, BrC 1 and BrC2. The smoke samples are categorized predominantly as B-type particles, whereas samples more purely comprised of soot exhibit predominantly A-type fluorescence. This distinction between smoke and soot may arise partially because the smoke particles are complex mixtures of amorphous soot with condensed organic liquids, indicating that compounds similar to the brown carbon analyzed here could heavily influence the smoke particle signal.

Biological particle samples were chosen for Fig. 3 to show the most important trends among all particle types analyzed. Two pollen are shown here to highlight two common types of fluorescence properties observed. Pollen 9 (Fig. 3a) shows particle type transitioning between A, AB, and ABC as particle size gets larger. Pollen 9 (Phleum pratense) has a physical diameter of ∼ 35 µm, so the mode seen in Fig. 3a is likely a result of fragmented pollen. Due to the upper particle size limit of WIBS detection, intact pollen of this species cannot be detected (Pöhlker et al., 2013). Pollen 8 (Fig. 3d) shows a mode peaking at ∼ 10 µm in diameter and comprised of a mixture of B, AB, BC, and ABC particles as well as a larger particle mode comprised of ABC particles. The large particle mode appears almost monodisperse, but this is due to the WIBS ability to sample only the tail of the distribution due to the upper size limit of particle collection (∼ 20 µm as operated). Particles larger than this limit saturate the sizing detector and are binned together into the ∼ 20 µm bin. It is important to note that excitation pulses from the Xe flash lamps are not likely to penetrate the entirety of large pollen particles, and so emission information is likely limited to outer layers of each pollen grain. Excitation pulses can penetrate a relatively larger fraction of the smaller pollen fragments, however, meaning that the differences in observed fluorescence may arise from differences the layers of material interrogated. Fungi 1 (Fig. 3b) was chosen because it depicts the most commonly observed fluorescence pattern among the fungal spore types analyzed (∼ 3 µm mode mixed with A and AB particles). Fungi 4 (Fig. 3e) represents a second common pattern (particle size peaking at larger diameter, minimal A-type, and dominated by AB and ABC particle types). All three bacteria types analyzed were dominated by A-type fluorescence. One gram-positive (Bacteria 1) and one gram-negative bacteria (Bacteria 3) types are shown in Fig. 3c and f, respectively.

An extension of observation from the many particle classes analyzed is that particle type (A, AB, ABC, etc.) varies strongly as a function of particle size. This is not surprising, given that it has been frequently observed and reported that particle size significantly impacts fluorescence emission intensity (e.g., Hill et al., 2001; Sivaprakasam et al., 2011). The higher the fluorescent quantum yield of a given fluorophore, the more likely it is to fluoresce. For example, pure biofluorophores (middle row of Fig. 2) and PAHs (bottom row of Fig. 2) have high quantum yields and thus exhibit relatively intense fluorescence emission, even for particles < 1 µm. In contrast, more complex particles comprised of a wide mixture of molecular components are typically less fluorescent per volume of material. At small sizes the relative fraction of these particles that fluoresce is small, but as particles increase in size they are more likely to contain enough fluorophores to emit a sufficient number of photons to record an integrated light intensity signal above a given fluorescence threshold. Thus, the observed fluorescence intensity scales approximately between the second and third power of the particle diameter (Sivaprakasam et al., 2011; Taketani et al., 2013; Hill et al., 2015).

The general trend of fluorescence dependence on size is less pronounced for FL1 than for FL2 and FL3. This can be seen by the fact that the scatter of points along the FL1 axis in Fig. 2b is not clearly size dependent and is strongly influenced by particle type (i.e., composition dependent). In Fig. 2c, however, the median points cluster near the vertical (size) axis and both FL2 and FL3 values increase as particle size increases. It is important to note, however, that the method chosen for particle generation in the laboratory strongly impacts the size distribution of aerosolized particles. For example, higher concentrations of an aqueous suspension of particle material generally produce larger particles, and the mechanical force used to agitate powders or aerosolize bacteria can have strong influences on particle viability and physical agglomeration or fragmentation of the aerosol (Mainelis et al., 2005). So, while the absolute size of particles shown here is not a key message, the relative fluorescence at a given size can be informative.

As discussed, each individual particle shows increased probability of exhibiting fluorescence emission above a given fluorescence threshold as size increases. Using Pollen 9 (Phleum pratense, Fig. 3a) as an example, most particles < 3 µm show fluorescence in only the FL1 channel and are thus classified as A-type particles. For the same pollen, however, particles ca. 2–6 µm in diameter are more likely to be recorded as AB-type particles, indicating that they have retained sufficient FL1 intensity but have exceeded the FL2 threshold to add B-type fluorescence character. Particles larger still (> 4 µm) are increasingly likely to exhibit ABC character, meaning that the emission intensity in the FL3 channel has increased to cross the fluorescence threshold. Thus, for a given particle type and a constant threshold as a function of particle size, the relative breakdown of fluorescence type changes significantly as particle size increases. The same general trend can be seen in many other particle types, for example Pollen 8 (Alnus glutinosa, Fig. 3d), Fungi 1 (Aspergillus brasiliensis, Fig. 3b), and to a lesser degree HULIS 3 (Suwannee fulvic acid, Fig. 3j) and Brown Carbon 2 (Fig. 3i). The “pathway” of change, for Pollen 9, starts as A-type at small particle size and adds B and eventually ABC (A→AB→ABC), whereas Pollen 8 starts primarily with B-type at small particle size and separately adds either A or C en route to ABC (B→AB or BC→ABC). In this way, not only is the breakdown of fluorescence type useful in discriminating particle distributions, but the pathway of fluorescence change with particle size can also be instructive.

To further highlight the relationship between particle size and fluorescence, four kinds of particles (Dust 2, HULIS 5, Fungi 4, and Pollen 9) were each binned into four different size ranges, and the relative number fraction was plotted versus fluorescence intensity signal for each channel (Fig. 4). In each case, the fluorescence intensity distribution shifts to the right (increases) as the particle size bin increases. This trend is strongest in the FL2 and FL3 (middle and right columns of Fig. 4) for most particle types, as discussed above.

The fact that particle fluorescence type can change so dramatically with increasing particle size becomes critically important when the Perring-style particle type classification is utilized for laboratory or field investigation. For example, Hernandez et al. (2016) aerosolized a variety of species of pollen, fungal spores, and bacteria in the laboratory and presented the breakdown of particle types for each aerosolized species. This first comprehensive overview summarized how different types of biological material (i.e., pollen and bacteria) might be separated based on their fluorescence properties when presented with a population of relatively monodisperse particles. This was an important first step, however, because differentiation becomes more challenging when broad size distributions of particles are mixed in an unknown environment. In such a case, understanding how the particle type may change as a function of particle size may become an important aspect of analysis.

Particle type analysis is critically affected not only by size but also by the threshold definition chosen. Figure 5 represents the same matrix of particle types as in Fig. 3 but shows the fluorescence intensity distribution in each channel (at a given narrow range of sizes in order to minimize the sizing effect on fluorescence). Figure 5 can help explain the breakdown of particle type (and associated colors) shown in Fig. 3. For example, in Fig. 5a, the median fluorescence intensity in FL1 for Pollen 9 (2046 a.u., detector saturated) in the size range 3.5–4.0 µm far exceeds the 3σ threshold (51 a.u.), and so essentially all particles exhibit FL1 character. Approximately 90 % of particles of Pollen 9 are above the 3σ FL2 threshold (25 a.u.), and approximately 63 % of particles are above the 3σ FL3 threshold (49 a.u.). These three channels of information together describe the distribution of particle type at the same range of sizes: 9 % A, 26 % AB, 63 % ABC, and 2 % other categories. Since essentially all particles are above the threshold for FL1, particles are thus assigned as A type particles (if < FL2 and FL3 thresholds), AB (if > FL2 threshold and < FL3 threshold), or ABC (if > FL2 and FL3 thresholds). Thus, the distribution of particles at each fluorescence intensity and in relation to a given thresholding strategy defines the fluorescence type breakdown and the pathway of fluorescence change with particle size. It is important to note differences in this pathway for biofluorophores (Fig. S4g and h). For example, Biofluorophore 1 (riboflavin) follows the pathway B or C→BC, while Biofluorophore 11 (tryptophan) follows the pathway A→AB→ABC.

By extension, the choice of threshold bears heavily on how a given particle breakdown appears and thus how a given instrument may be used to discriminate between biological and non-biological particles. A commonly made assumption is that particles exhibiting fluorescence by the WIBS (or UV-APS) can be used as a lower limit proxy to the concentration of biological particles, though it is known that interfering particle types confound this simple assumption (Huffman et al., 2010). Increasing the fluorescence threshold can reduce categorizing weakly fluorescent particles as biological but can also remove weakly fluorescing biological particles of interest (Huffman et al., 2012). Figure 6 provides an analysis of eight representative particle types (three biological, five non-biological) in order to estimate the tradeoffs of increasing fluorescence threshold separately in each channel. Once again, the examples chosen here represent general trends and outliers, as discussed previously for Fig. 3. Four threshold strategies are presented: three as the instrument fluorescence baseline plus increasing uncertainty on that signal (FT + 3σ, FT + 6σ, and FT + 9σ), as well as the FP3 strategy suggested by Wright et al. (2014). Using Dust 4 as an example (Fig. 6d), by increasing the threshold from 3σ (red traces) to 6σ (orange traces), the fraction of dust particles fluorescent in FL1 decreases from approximately 50 to 10 %. Increasing the fluorescence threshold even higher to 9σ reduces the fraction of fluorescence to approximately 1 %, thus eliminating nearly all interfering particles of Dust 3. In contrast, for biological particles such as Pollen 9 (Fig. 6b), increasing the threshold from 3σ to 9σ does very little to impact the relative breakdown of fluorescence category or the fraction of particles considered fluorescent in at least one channel. Changing threshold from 3σ to 9σ decreases the FL1 fraction minimally (98.3 to 97.9 %), and for FL2 and FL3 the fluorescence fraction decreases from 90 to 50 % and from 60 to 42 %, respectively. Figure 6 also underscores how increasing particle size affects fluorescence fraction, as several particle types (e.g., Pollen 9 and HULIS 5) show sigmoidal curves that proceed toward the right (lower fraction at a given size) as the threshold applied increases and thus removes more weakly fluorescent particles.

To better understand how the different thresholding strategies qualitatively change the distribution of particle fluorescence type, Fig. 7 shows stacked fluorescence type distributions for each of the four thresholds analyzed. Looking first at Dust 3 (Fig. 7d), the standard threshold definition of 3σ shows approximately 80 % of particles to be fluorescent in at least one channel, resulting in a distribution of predominantly A, B, and AB-type particles. As the threshold is increased, however, the total percentage of fluorescent particles decreases dramatically to 1 % at 9σ and the particle type of the few remaining particles shifts to A-type particles. A similar trend of fluorescent fraction can also be seen for Soot 6 (wood smoke) and Brown Carbon 2, where almost no particle (10 and 16 %, respectively) remain fluorescent using the 9σ threshold. Soot 4 (diesel soot), in contrast, exhibits the same fraction and breakdown of fluorescent particles whether using the 3σ or 9σ threshold. Using the FP3 threshold (which employs very high FL1 threshold), however, the fluorescent properties of the diesel soot change dramatically to non-fluorescent. As a “worst-case” scenario, HULIS 5 shows ca. 60 % of particles to be fluorescent using the 3σ threshold, but this material is unlikely to be representative of commonly observed soil HULIS, as discussed above. In this case, increasing the threshold from 6σ to 9σ only marginally decreases the fraction of fluorescent particles to ca. 35 and 22 %, respectively, and the breakdown remains relatively constant in B, C, and BC types. Changing the threshold definition to FP3 in this case also does not significantly change the particle type breakdown, since the high FP3 threshold applies only to FL1.

As stated, the WIBS is most often applied toward the detection and characterization of biological aerosol particles. For the biological particles analyzed (Fig. 7, top rows), increasing the threshold from 3σ to 9σ shows only a marginal decrease in the total fluorescent fraction for Pollen 9, Fungal Spore 1, and Bacteria 1 and only a slight shift in fluorescence type as a function of size. Using the FP3 threshold, however, for each of the three biological species the non-fluorescent fraction increases substantially. Wright et al. (2014) found that the FP3 threshold definition showed a strong correlation with ice nucleating particles and the authors suggested these particles with high FL1 intensity were likely to be fungal spores. This may have been the case, but given the analysis here, the FP3 threshold is also likely to significantly underestimate fungal spore number by missing weakly or marginally fluorescent spores.

Based on the threshold analysis results shown in Fig. 7, marginally increasing the threshold in each case may help eliminate non-biological, interfering particles without significantly impacting the number of biological particles considered fluorescent. Each threshold strategy brings tradeoffs, and individual users must understand these factors to make appropriate decisions for a given scenario. These data suggest that using a threshold definition of FT baseline + 9σ is likely to reduce interferences from most non-biological particles without significantly impacting most biological particles.

As a part of the comprehensive WIBS study, particle asymmetry (AF) was analyzed as a function of particle size for all particles. As described in Sect. 2.1, AF in the WIBS-4A is determined by comparing the symmetry of the forward elastic scattering response of each particle, measured at the quadrant PMT. Many factors are related to the accuracy of the asymmetry parameter, including the spatial alignment of the collection optics, signal-to-noise ratio and dynamic range of the detector, agglomeration of particles with different refractive indices, and the angle at which a non-symmetrical particle hits the laser (Kaye et al., 2007; Gabey et al., 2010). Figure 8 shows a summary of the relationship between AF and particle size for all material types analyzed in Table 2. Soot particles are known to frequently cluster into chains or rings depending on the number of carbon atoms (Von Helden et al., 1993) and, as a result, can have long aspect ratios that would be expected to manifest as large AF values. The bacteria species chosen have rod-like shape features and thus would also exhibit large AF values. These properties were observed by the WIBS, as two types of soot (diesel and fullerene) and all three bacteria showed higher AF values than other particles at approximately the same particle diameter. For an unknown reason, all three brown carbon samples also showed relatively high AF values given that the individual particles of liquid organic aerosol would be expected to be spherical with low AF. Similarly, the intact pollen showed anomalously low AF, because a substantial fraction of each was shown to saturate the WIBS sizing detector, even if the median particle size (shown) is lower than the saturating value. For this reason we postulate that the forward-scattering detector may not be able to reliably estimate AF when particles are near the sizing limits. Intact pollen, soot samples (diesel and fullerene soot), bacteria, and brown carbon samples were excluded from the linear regression fit because they appeared visually as outliers to the trend. All remaining particle groups of material types (seven in total) are represented by blue in Fig. 8. A linear regression R2 value of 0.87 indicates a high degree of correlation between particle AF and size across the remaining particles. The strong correlation between these two factors across a wide range of particle types, mixed with the confounding anomaly of brown carbon, raises a question about the degree to which the asymmetry factor parameter from the WIBS-4A can be useful or, conversely, to what degree the uncertainty in AF is dominated by instrumental factors, including those listed above.