Atmospheric sampling was conducted as a part of the BEACHON-RoMBAS (Bio–hydro–atmosphere interactions of Energy, Aerosols, Carbon, H2O, Organics, and Nitrogen – Rocky Mountain Biogenic Aerosol Study) field campaign conducted at the Manitou Experimental Forest Observatory (MEFO) located 48 km northwest of Colorado Springs, Colorado (39∘06′0′′ N, 105∘5′03′′ W; 2370 m elevation) (Ortega et al., 2014). The site is located in the central Rocky Mountains and is representative of the semi-arid montane pine-forested regions of North America. During BEACHON-RoMBAS, a large, international team of researchers conducted an intensive set of measurements from 20 July to 23 August 2011. A summary of results from the campaign are published in the BEACHON campaign special issue of Atmospheric Chemistry and Physics (https://acp.copernicus.org/articles/special_issue247.html). All the data reported here were gathered from instruments and sensors located within a < 100 m radius (Fig. 1).

UV-APS and WIBS-3 (model 3; University of Hertfordshire) instruments were operated continuously as a part of the study, and particle data were integrated to 5 min averages before further analysis. The UV-APS was operated under procedures defined in previous studies (Huffman et al., 2013; Schumacher et al., 2013). A total suspended particle (TSP) inlet head ∼ 5.5 m above the ground, mounted above the roof of a climate-controlled, metal trailer, was used to sample aerosol directed towards the UV-APS. Bends and horizontal stretches in the 0.75 inch tubing were minimized to reduce losses of large particles (Huffman et al., 2013). The UV-APS detects particles between 0.5 and 20 µm and records aerodynamic particle diameter and integrated total fluorescence (420–575 nm) after pulsed excitation by a 355 nm laser (Hairston et al., 1997). Both UV-APS and WIBS instruments report information about particle number concentration, but it is instructive here to show results in particle mass for comparison between all techniques. Total particle number size distributions (irrespective of fluorescence properties) obtained from the UV-APS and WIBS were converted to mass distributions assuming spherical particles of unit particle mass density, unless otherwise stated, as a first approximation. Total particle concentration values (in µg m-3) were obtained for each 5 min period by integrating over the size range 0.5–15 µm, and these mass concentration values were averaged over the length of the filter sampling periods. Uncertainty in mass concentration values reported here is influenced by assuming a single value for particle mass density and because of slight dissimilarities between size bins of the UV-APS and WIBS instruments at particle sizes above 10 µm that dominate particle mass.

A WIBS-3 was used to continuously sample air at a site ∼ 50 m from the UV-APS trailer and 1.3 m above the ground. Briefly, the diameter of individual particles sampled by the WIBS is estimated by the intensity of the elastic side scatter from a continuous wave 635 nm diode laser and analyzed by a Mie scattering model (Foot et al., 2008; Kaye et al., 2005). Particles that pass through the diode laser activate two optically filtered Xenon flash lamps. The first lamp excites the particle at 280 nm and the second at 370 nm. Emission from the 280 nm excitation is filtered separately for two photomultiplier tubes (PMTs), one which detects in a band at 320–400 nm and the other in a band at 410–650 nm. These excitation and emission wavelengths result in a total of three channels of detection: λex 280 nm, λem 320–400 nm (FL1 or channel A); λex 280 nm, λem 410–650 nm (FL2 or channel B); and λex 370 nm, λem 410–650 nm (FL3 or channel C) (Crawford et al., 2014). Individual particles are considered fluorescent here if they exceed fluorescent thresholds for any channel, as defined as the average of a “forced trigger” baseline plus 3 standard deviations (SD) of the baseline measurement (Gabey et al., 2010).

WIBS particle-type analysis is utilized to define types of particles that have specific spectral patterns. As defined by Perring et al. (2015), the three different fluorescent channels (FL1, FL2, and FL3) can be combined to produce seven unique fluorescent categories. Observed fluorescence in channel FL1 alone, but without any detectable fluorescence in channel FL2 or FL3, categorizes a particle as type A. Similarly, observed fluorescence in channels FL2 or FL3, but in no other channels, places a particle in the B or C categories, respectively. Combinations of fluorescence in these channels, such as a particle that exhibits fluorescence in both FL1 and FL2, categorizes a particle as type AB and so on for a possible seven particle types as summarized in Fig. S1 in the Supplement.

As a separate tool for particle categorization, the University of Manchester has recently developed and applied a hierarchical agglomerative cluster analysis tool for WIBS data, which they have previously applied to the BEACHON-RoMBAS campaign (Crawford et al., 2014, 2015; Robinson et al., 2013). Here we utilize clusters derived from WIBS-3 data as described by Crawford et al. (2015). Cluster data presented here were analyzed with the open-source Python package FastCluster (Müllner, 2013). Briefly, hierarchical agglomerative cluster analysis was applied to the entire data set and each fluorescent particle was uniquely clustered into one of four groups. Cluster 1, assigned by Crawford et al. (2015) as fungal spores, displayed a 1.5–2 µm mode and a daily peak in the early morning that paralleled relative humidity (Schumacher et al., 2013). Clusters 2, 3, and 4 have strong, positive correlations with rainfall and exhibit size modes that peak at < 1.2 µm and were initially described by Crawford et al. (2014) as bacterial particles. Here we have summed clusters 2–4 to a single group referred to as ClBact, for simplicity when comparing with molecular tracers. It should be noted that assignment of name and origin (e.g., fungal spores or bacteria) to clusters is approximate and does not imply naming accuracy or particle homogeneity. Each cluster likely contains an unknown fraction of contaminating particles, but the clusters are beneficial to group particles more selectively than using fluorescent intensity alone. For more details see Robinson et al. (2013) and Crawford et al. (2015).

The WIBS-3 utilized here has since been superseded by the WIBS-4 (Univ. Hertfordshire, UK) and WIBS-4A (Droplet Measurement Technologies, Boulder, CO, USA). One important difference between the models is that the optical chamber design and filters of the WIBS-4 models were updated to enhance the overall sensitivity of the instrument (Crawford et al., 2014). Additionally, slight differences in detector gain between models and individual units can impact the relative sensitivity of the fluorescence channels. This may result in differences in fluorescent channel intensity between instrument models, as will be discussed later.

Total suspended particle samples were collected for molecular tracer and molecular genetic analyses using a high-volume sampler (Digitel DHA-80) drawing 1000 L min-1 through 15 cm glass fiber filters (Macherey-Nagel GmbH, Type MN 85/90, 406015, Düren, Germany) over a variety of sampling times ranging from 4 to 48 h (Supplement Table S1). The sampler was located < 50 m from each of the UV-LIF instruments described here, approximately between the WIBS-3 and UV-APS. Prior to sampling, all filters were baked at 500 ∘C for 12 h to remove DNA and organic contaminants. Samples were stored in pre-baked aluminum bags after sampling at -20 ∘C for 1–30 days and then at -80 ∘C after overnight, international transport cooled on dry ice. Due to the low vapor pressure of the molecular tracers analyzed, loss due to volatilization is considered unlikely (Zhang et al., 2010). A total of 36 samples were collected during the study, in addition to handling field blanks and operational field blanks. Handling blanks were acquired by placing a filter into the sampler and immediately removing it, without turning on the airflow control. Operational blanks were placed into the sampler and exposed to 10 s of airflow.

A direct-to-agar slit sampler (Microbiological Air Sampler STA-203, New Brunswick Scientific Co, Inc., Edison, NJ) was used to collect culturable airborne fungal spores. The sampler was placed ∼ 2 m above the ground on a wooden support surface with 5 cm × 5 cm holes to allow airflow both up and down through the support structure. Sampled air was drawn over the 15 cm diameter sampling plate filled with growth media at a flow rate of 28 L min-1 for sampling periods of 20 to 40 min. Growth media (malt extract medium) was mixed with antibacterial agents (40 units streptomycin, Sigma Aldrich; 20 units ampicillin, Fisher Scientific) to suppress bacterial colony growth. Plates were prepared several weeks in advance and stored in a refrigerator at ca. 4 ∘C until used for sampling. Before each sampling period, all surfaces of the samplers were sterilized by wiping with isopropyl alcohol. Handling and operational blanks were collected to verify that no fungal colonies were being introduced by handling procedures. A total of 14 air samples were collected over 20 days and immediately moved to an incubator (Amerex Instruments, Incumax IC150R) set at 25 ∘C for 3 days prior to counting fungal colonies formed. Each colony, present as a growing dot on the agar surface, was assumed to have originated as 1 colony-forming unit (CFU; i.e., fungal spore) deposited onto the agar by impaction during sampling. The atmospheric concentration of CFU per air volume was calculated using the sampler airflow. Further discussion of methods and initial results from the slit sampler were published by Huffman et al. (2013).

Meteorological data were recorded by a variety of sensors located at the site. Precipitation was recorded by a laser optical disdrometer (PARticle SIze and VELocity sensor – “PARSIVEL”; OTT Hydromet GmbH, Kempton, Germany) and separately by a tipping-bucket rain gauge. The disdrometer provides precipitation occurrence, rate, and physical state (rain or hail) by measuring the magnitude and duration of disruption to a continuous 780 nm laser that was located in a tree clearing (Fig. 1), while the tipping-bucket rain gauge measures a set amount of precipitation before tipping and triggering an electrical pulse. A leaf wetness sensor (LWS; Decagon Devices, Inc., Pullman, WA, USA) provided a measurement of condensed moisture by measuring the voltage drop across a leaf surface to determine a proportional amount of water on or near the sensor. Additional details of these measurements can be found in Huffman et al. (2013) and Ortega et al. (2014).

Increases in PBAP concentration have been frequently associated with rainfall (e.g., Bigg et al., 2015; Faulwetter, 1917; Hirst and Stedman, 1963; Jones and Harrison, 2004; Madden, 1997). Fungal polyols have also been reported to increase after rain and have been used as indicators of increased fungal spore release (Liang et al., 2013; Lin and Li, 2000; Zhu et al., 2015). Recently, it was shown that the concentration of fluorescent aerosol particles (FAPs) measured during BEACHON-RoMBAS increased dramatically during and after periods of rain (Crawford et al., 2014; Huffman et al., 2013; Schumacher et al., 2013) and that these particles were associated with high concentrations of ice nucleating particles that could influence the formation and evolution of mixed-phase clouds (Huffman et al., 2013; Prenni et al., 2013; Tobo et al., 2013). It was observed that a mode of smaller fluorescent particles (2–3 µm) appeared during rain episodes, and several hours after rain ceased a second mode of slightly larger fluorescent particles (4–6 µm) emerged, persisting for up to 12 h (Huffman et al., 2013). The first mode was hypothesized to result from mechanical ejection of particles due to rain splash on soil and vegetated surfaces, and the second mode was suggested as actively emitted fungal spores (Huffman et al., 2013). While the UV-APS and WIBS each provide data at a high enough time resolution to see subtle changes in aerosol concentration, the temporal resolution of the chemical tracer analysis was limited to 4–48 h periods defined by the collection time of the high-volume sampler. To compare the measurement results across the sampling platforms, UV-LIF measurements were averaged to the lower time resolution of the filter sampler periods, and the periods were grouped into three broad categories: Rainy, Dry, and Other, as will be defined below.

Time periods were wetness-categorized in two steps: first at 15 min resolution and then averaged for each individual filter sample. During the first stage of categorization each 15 min period was categorized into one of four groups: rain, post-rain, dry, or other. To categorize each filter period, an algorithm was established utilizing UV-APS fluorescent particle fraction and accumulated rainfall. The ratio of the integrated number of fluorescent particles to total particles was used as a proxy for the increased emission of biological particles. Figure 2a presents a time series of the size-resolved fluorescent particle concentration, showing increases during rain periods in dark red. A relatively consistent diurnal cycle of increased FAP concentration in the 2–4 µm range is apparent almost every afternoon, which corresponds to near-daily afternoon rainfall during approximately the first half of the measurement period. Disdrometer and tipping-bucket rainfall measurements were each normalized to unity and summed to produce a more robust, unitless measure of rainfall rate because it was observed that often only one of the two systems would record a given light rain event. If a point was described by total rainfall accumulation greater than 0.201 it was flagged as rain. A point was flagged as post-rain if it immediately followed a rain period and also exhibited a fluorescent particle fraction greater than 0.08. The purpose of this category was to reflect the observation that sustained elevated concentrations of FAPs persisted for many hours even after the rain rate, RH, and leaf wetness returned to pre-rain values. The only measurement that adequately reflected this scenario was of the fluorescent particles measured by UV-APS and WIBS instruments. The post-rain flag was continued until the fluorescent particle fraction fell below 0.08 or if it started to rain again (with calculated rain values greater than 0.201). Points were flagged as dry periods if they exhibited rainfall accumulation and fluorescent particle fraction below the thresholds stated above. Several periods were not easily categorized by this system and were considered in a fourth category as other. This occurred when fluorescent particle fraction above the threshold value was observed with no discernable rainfall.

Once wetness categories were assigned by the algorithm at 15 min resolution, each high-volume filter sample was categorized by a similar nomenclature, but using only three categories. These were defined as Dry, Rainy (combination of rain and post-rain categories), or Other based on the relative time fraction in each of the four original 15 min categories. For each sample, if a given category represented more than 50 % of the 15 min periods, the sample was assigned to that category. Despite the effort to categorize samples systematically, several sample periods (5 of 35) appeared mis-categorized by looking at FAP concentration, rainfall, RH, and leaf wetness in more detail. In some circumstances, this was because light rainfall produced observable increases in FAPs, but without exceeding the rainfall threshold. Or in other circumstances a period of rainfall occurred at the very end or just before the beginning of a sample, and so the many-hour period was heavily influenced by aerosol triggered by a period of rain just outside of the sample time window. As a result, several samples were manually re-categorized as described here. Samples 20 and 21 (Table S1) were 4 h samples that displayed high relative humidity and rainfall; thus, samples were originally characterized as Rainy. This period was described by an extremely heavy rain downpour (7.5 mm in 15 min); however, that seemingly placed the samples in a different regime of rain–aerosol dynamics than the other Rainy samples and so these two samples were moved to the Other category. Sample 23, originally Rainy, presented a FAP fraction marginally above the 0.08 threshold, but visually displayed a trend dissimilar to other post-rain periods and so was re-categorized as Dry. Sample 28 showed no obvious rainfall, but the measurement team observed persistent fog in three consecutive mornings (samples 25, 27, 28), and the concentration of fluorescent particles (2–6 µm) suggested a source of particles not influenced by rain, and so this Rainy sample was re-categorized as Other. Sample 38 displayed a fluorescent number ratio just below the threshold value, and was first categorized as Dry; however, the measurement team observed post-rain periods at the beginning and end of the sample, and the sample was re-categorized as Other. For all samples other than these five, the categorization was determined using the majority (> 0.50) of the 15 min periods. In no cases other than the five that were re-categorized was the highest category fraction less than 0.50 of the sample time. Note that we have chosen to capitalize Rainy, Dry, and Other to highlight that we have rigorously defined the period using the characterization scheme described above and to separate the nomenclature from the general, colloquial usage of the terms. Wetness category assignment for each high-volume filter sample period is shown in Fig. 2 as a background color (brown for Dry samples, green for Rainy samples, and pink for Other samples) and Table S1.

To validate the qualitative differences between wetness categories described in the last section, we present observations about each of these groupings. First, we organized the WIBS data according to the particle categories introduced by Perring et al. (2015). By this method, every fluorescent particle detected by the WIBS can be defined uniquely into one of seven categories (i.e., A, AB, ABC). By plotting the relative fraction of fluorescent particles described by each particle type, temporal differences between measurement periods can be observed, as shown in Fig. 2e. To a first approximation, this analysis style allows for coarse discrimination of particle types. For example, a given population of particles would ideally exhibit a consistent fraction of particles present in the different particle categories as a function of time. By this reasoning, sample periods categorized as Dry (most of the latter half of the study; brown bars in Fig. 2) would be expected to have a self-consistent particle-type trend, whereas sample periods categorized as Rainy (most of the first half of the study; green bars in Fig. 2) would have a self-consistent particle-type trend, but different from the Dry samples. This is broadly true. During Rainy periods, as seen in Fig. 3a, there is a relatively high fraction (> 65 %) of ABC type particles (light blue) and a relatively low fraction (< 15 %) in BC (purple) and C (yellow) type particles, suggesting heavy influence from the FL1 channel. In contrast, during Dry periods the fraction of ABC particles (light blue) is reduced (< 25 %), whereas BC (purple) and C (yellow) type particles increase in relative fraction (> 30 and > 40 %, respectively), which suggested a diminished influence of FL1 channel.

It is important to note a few caveats here. First, the ability of the WIBS to discriminate distinctly between PBAP types is relatively poor and it is still unclear exactly how different particle types would appear by this analysis method. Particles of different kinds and from different sources are likely convolved into a single WIBS particle type, which could either soften or enhance the relationships with rain discussed here. Second, the assignment of particle types is heavily size dependent and sensitive to subtle instrument parameters, and so it is unclear how different instruments would present similar particle types. For example, Hernandez et al. (2016) used two WIBS instruments and found differences in relative fraction of particle categories for samples aerosolized in the lab. They reported fungal spores to be predominately A, AB, and ABC type particles, whereas Rainy sample periods, suggested to have a heavy fungal spore influence by Huffman et al. (2013), show predominantly C, BC, and ABC type particle fractions. These discrepancies may be due to the comparison of ambient particles to laboratory-grown cultures. The highly controlled environment of a laboratory may not always accurately represent the humidity conditions in which fungal spore release occurs in this forest setting (Saari et al., 2015). This could impact the fluorescence properties of fungal spore particles that have different amounts of adsorbed or associated water (Hill et al., 2009, 2013, 2015). More likely, however, is that the WIBS-3 used here exhibits differences in fluorescence sensitivity from the WIBS-4A used by Hernandez et al. (2016). Even a slight increase in sensitivity in the FL3 channel with respect to the FL1 or FL2 channels could explain the shift here towards particles with C-type fluorescence. One piece of evidence for this is the quantitative comparison of particle measurements presented by the UV-APS and WIBS-3 instruments co-deployed here (Fig. 4). The number concentration of particles exhibiting fluorescence above the FL2 baseline of the WIBS-3 is approximately consistent with the number of fluorescent particles measured by the UV-APS, and significantly below the concentration of FL3 particles. The UV-APS number concentration shows the highest correlation with the WIBS-3 FL2 channel: during Rainy periods, R2=0.70; Dry, R2=0.82; and Other, R2=0.92. These observations are in stark contrast to the trends reported by Healy et al. (2014) that the UV-APS fluorescent particle concentration correlated most strongly with the WIBS-4 FL3 and that the number concentration of FL3 was the lowest out of all three channels. Given that the FL3 channel of the WIBS and the UV-APS cover similar excitation and emission wavelengths, it is expected that these two channels should correlate well. Based on these data, we suggest that the WIBS-3 utilized here may present a very different particle-type breakdown than if a WIBS-4 had been used. So, while caution is recommended when comparing the relative breakdown of WIBS particle categories shown here (Fig. 3) with other studies, the data are internally self-consistent, and comparing qualitative differences between, e.g., Rainy and Dry periods, is expected to be robust. The main point to be highlighted here is that there is indeed a qualitative difference in particles present in the three wetness categories, as averaged and shown in Fig. 3a, which generally supports the effort to segregate these samples.

Further evidence that there is a qualitative difference in the three wetness categories is shown using molecular genetic analysis (Fig. 3b, c). The analysis of fungal DNA sequences from 21 of the high-volume samples found 406 OTUs belonging to different fungal classes and phyla. When organized by wetness type it was observed that 106 of these occurred only on Rainy samples, 148 of these occurred on Dry samples, and 37 on Other samples, with some fraction occurring in overlaps of each (Fig. 3c). This shows that the number of OTUs observed uniquely in either the Rainy or Dry periods is greater than the number of OTUs present in both wetness types, suggesting that the fungal communities in each grouping are relatively distinct. Further, Fig. 3b shows a breakdown of fungal taxonomic groupings for each wetness group. This analysis shows that there is a qualitative difference in taxonomic breakdown between periods of Rainy and Dry. Specifically, during Dry periods there is an increased fraction of Pucciniomycetes (green bar, Fig. 3c), Chytridiomycota (yellow), Sordariomycetes (orange), and Eurotiomycetes (pink) when compared to the Rainy periods.

To estimate fungal spore emission to the atmosphere, the concentration of arabitol and mannitol (Fig. 5a, b, Table 1) in each aerosol sample was averaged for all samples in each of the three wetness categories. The average concentration of arabitol collected on Rainy TSP samples (35.2 ± 10.5 ng m-3) increased by a factor of 3.3 with respect to Dry samples, and the average mannitol concentration on Rainy samples was higher by a factor of 3.7 (44.9 ± 13.8 ng m-3). Figure 5a, b show the concentration variability for each wetness category, observed as the standard deviation from the distribution of individual samples. For each polyol, there is no overlap in the ranges shown, including the outliers of the Rainy and Dry category, suggesting a definitive and conceptually distinct separation between dry periods and those influenced by rain. The concentrations observed during Other periods is between those of the Dry and Rainy averages, as expected, given the difficulty in confidently assigning these uniquely to one of these categories. The observations here are roughly consistent with previous reports of polyol concentration, despite differences in local fungal communities and concentrations. For example, Rathnayake et al. (2016a) observed 30.2 ng m-3 arabitol and 41.3 ng m-3 mannitol in PM10 samples collected in rural Iowa, USA. In addition, Zhang et al. (2015) reported arabitol and mannitol concentrations in PM10 samples of 44.0 and 71.0 ng m-3, respectively, from a study in the mountains on Hainan Island off the coast of southern China. More recently, Yue et al. (2016) studied a rain event in Beijing and observed increased polyol concentrations at the onset of the rain. The observed mannitol concentration (45 ng m-3) was approximately consistent with observations reported here and with previous reports, while the arabitol concentration values observed were approximately an order of magnitude lower (0.3 ng m-3).

The square of the correlation coefficient (R2) here between concentration values of arabitol and mannitol during Rainy samples is very high (0.839; Table 2) suggesting that arabitol and mannitol originated primarily from the same source, likely active-discharge fungal spores. The correlation is similar to the 0.87 R2 reported by Bauer et al. (2008a) and the 0.93 R2 reported by Graham et al. (2003). In contrast, the same correlation seen between mannitol and arabitol concentrations for Dry samples is relatively low (0.312). This is consistent with reports that arabitol can be used more specifically as a spore tracer, but that mannitol has additional atmospheric sources besides fungal spores. The same correlation was also performed between arabitol or mannitol and other molecular tracers (endotoxins and (1→3)-β-D-glucan), but all R2 value were less than 0.43, suggesting that the endotoxins and glucans analyzed were not emitted uniquely from the same sources as arabitol and mannitol.

Results from the two UV-LIF instruments were averaged over high-volume sample periods, and a correlation analysis was performed between tracer mass and fluorescent particle mass showing positive correlations in all cases. The FAP mass from the UV-APS shows high correlation with the fungal polyols during Rainy periods, with R2 of 0.732 and 0.877 for arabitol and mannitol, respectively (Table 3; Fig. 5c, d). The same tracers correlate poorly with the UV-APS during Dry conditions. This is expected, because Ascomycota and Basidiomycota spores emitted by wet-discharge methods are the only fungal spores reported to be associated with arabitol and mannitol (Elbert et al., 2007; Feofilova, 2001; Lewis and Smith, 1967). This high correlation suggests that the UV-APS does a good job of detecting these wet-discharge spores, and corroborates previous statements that particles detected in ambient air by the UV-APS are often predominately fungal spores (Healy et al., 2014; Huffman et al., 2012, 2013). In contrast, the low slope value and the poor correlation during Dry periods suggest that the UV-APS is also sensitive to other kinds of particles, as designed. The small positive x offset (FAP mass; Table S2, Fig. 5c, d) during Rainy periods is likely due to particles that are too weakly fluorescent to be detected and counted by the UV-APS, which is consistent with observations made in Brazil (Huffman et al., 2012).

Particle mass from WIBS Cl1, assigned to fungal spores (Crawford et al., 2015), also correlate strongly with the same two molecular tracers. Both Rainy periods (R2 0.824) and Dry periods (R2 0.764) correlate well with arabitol (Fig. 5e), while mannitol (Fig. 5f) only shows a strong correlation during the Rainy periods (R2 0.799). Mannitol is a common polyol in higher plants while arabitol is only found in fungal spores and lichen (Lewis and Smith, 1967). So the strong correlation of each polyol with UV-LIF mass during Rainy periods when actively discharged spores are expected to dominate and the similarly strong correlations associated with arabitol suggest that the Cl1 cluster does a reasonably good job of selecting fungal spore particles. The poor correlation between mannitol and Cl1 during dry periods illustrates that the background mannitol concentration is likely not due to fungal spores alone, but has contributions from other higher plants that contain mannitol. Particle concentrations detected by individual WIBS channels and in the other clusters were also compared with polyol concentrations, but each correlation is relatively poor compared to that with respect to Cl1. As seen in Table 3 and Figs. S2–S3, correlations in FL1, 2, and 3 with arabitol are poor (< 0.4) in the Dry category and good (0.4 < R2 < 0.7) in the Rainy category. For mannitol, all the UV-LIF instruments show high correlation (> 0.7) in all cases. This is likely due to mannitol being a non-specific tracer and suggests that the majority of UV-LIF particles observed during all periods was dominated by PBAPs.

Bauer et al. (2008a) reported measurements of fungal spore number concentration in Vienna, Austria, using epifluorescence microscopy, and also measured fungal tracer mass collected onto filters in order to estimate the mass of arabitol (1.2 to 2.4 pg spore-1) and mannitol (0.8 to 1.8 pg spore-1) associated with each emitted spore. Bauer et al. (2008a) and Yttri et al. (2011b) reported ratios of mannitol to arabitol of ca. 1.5 (± standard deviation of 26 %) and 1.4 ± 0.3, respectively. Our measurements show slightly lower ratios of mannitol to arabitol, but that the ratio is dependent on wetness category: Rainy, 1.29 ± 0.17; Dry, 1.12 ± 0.23; and Other, 1.24 ± 0.54. The mannitol to arabitol ratio would be expected to vary as a function of fungal population present in the aerosol, whether between different wetness periods at a given location or between different physical localities.

Using the approximate mid-point of the Bauer et al. (2008a) reported ranges, 1.7 pg mannitol per spore and 1.2 pg arabitol per spore, atmospheric number concentrations of spores collected onto the high-volume filters were calculated from the polyol mass concentrations measured here. Based on these values, and assuming all polyol mass originated with spore release, the mass concentration averages (Fig. 5) were converted to fungal spore number concentrations (Fig. 6). The trends of spore concentration averages are the same as with the polyol mass, because the numbers were each multiplied by the same scalar value. After doing so, the analysis reveals an estimated spore concentration during Dry periods of 0.89×104 (± 0.21) spores m-3 using the arabitol concentration and 0.70×104 (± 0.19) spores m-3 using the mannitol concentration (Table 4). The estimated concentration of spores increased approximately 3-fold during Rainy periods to 2.9×104 (± 0.8) spores m-3 (arabitol estimate) and 2.6×104 (± 0.8) spores m-3 (mannitol estimate) (Fig. 6a, b). These estimates match reasonably well with estimates reported by Spracklen and Heald (2014), who modeled the concentration of airborne fungal spores across the globe as an average of 2.5×104 spores m-3, with ca. 0.5×104 spores m-3 over Colorado.

The UV-LIF instruments discussed here are number-counting techniques and in this instance have been applied as spore counters. As a first approximation, each particle detected by the UV-APS was assumed to be a fungal spore with the same properties used in the assumptions by Bauer et al. (2008a). Figure 6d, e, g, h show correlations of fungal spore number concentration estimated from polyol mass on the y axes and from UV-LIF measurements on the x axes. The first, and most important observation is that the estimated fungal spore concentration from each technique is on the same order of magnitude, 104 m-3. Looking at individual correlations reveals a finer layer of detail. These results show that the number concentration of fungal spores estimated by the UV-APS is greater than the number of fungal spores estimated by the tracers, as evidenced by slope values of ca. 0.2 and 0.35 for Rainy and Dry conditions, respectively (Table S3, Fig. 6d, e). Again, this suggests that the UV-APS detects fungal spores as well as other types of fluorescent particles. The R2 values (∼ 0.5) during Rainy periods indicate that the additional source of particles detected by the UV-APS is likely to have a similar source, such as PBAPs mechanically ejected from soil and vegetative surfaces with rain splash (Huffman et al., 2013). The magnitude of the over-estimation is higher during Dry periods, which would be expected because Rainy periods exhibited much higher particle number fractions associated with polyol-containing spores.

The Cl1 cluster from WIBS data shows correlations with estimated fungal spores from arabitol and mannitol that have slopes much closer to 1.0 than correlations with UV-APS number (Fig. 6g, h, Table S3). For example, the slope of the Cl1 correlations with each polyol during Rainy periods is ca. 0.87. This suggests only a 13 % difference between the spore concentration estimates from the two techniques during Rainy periods. The average number concentration of Cl1 during Rainy periods is 1.6×104 (± 0.8) spores m-3. In both cases the slopes with respect to Cl1 are greater than 1.0 during Dry periods, suggesting that the cluster method may be missing some fraction of weakly fluorescent particles. Huffman et al. (2012) similarly suggests that particles that are weakly fluorescent may be below the detection limit of the instrument and Healy et al. (2014) suggested that both UV-APS and WIBS-4 instruments significantly under-count the ubiquitous Cladosporium spores that are most common during dry weather and often peak in the afternoon when RH is low (De Groot, 1968; Oliveira et al., 2009). Fundamentally, however, the results from the UV-APS, and even more so the numbers reported by the clustering analysis by Crawford et al. (2015), reveal broadly similar trends with the numbers estimated from polyol-to-spore values reported by Bauer et al. (2008a).

The fungal culture samples show similar division during Rainy and Dry periods as arabitol and mannitol concentrations (Fig. 6c), with an increase of ca. 1.6 × during Rainy periods. The trend of a positive slope with respect to the UV-LIF measurements is also similar between the tracer and culturing methods. In general, however, the R2 value correlating CFU to fungal spore number calculated from the UV-LIF number is lower than between tracers and UV-LIF numbers (Table 3, Fig. S4). This is not unexpected for several reasons. First, the short sampling time of the culture samples (20 min) leads to poor-counting statistics and high number concentration variability, whereas each data point from the high-volume air samples represents a period of 4–48 h. Second, culture samplers, by their nature, only account for culturable fungal spores. It has been estimated that as low as 17 % of aerosolized fungal species are culturable, and so it is expected that the CFU concentration observed is significantly less than the total airborne concentration of spores (Bridge and Spooner, 2001; Després et al., 2012). Nonetheless, the culturing analysis here supports the tracer and UV-LIF analyses and the most important trends are consistent between all analysis methods. The concentration of fungal spores is higher during the Rainy periods, and there is a positive correlation between both tracer and CFU concentration and UV-LIF number.

In a pristine environment, such as the Amazon, supermicron particle mass has been found to consist of up to 85 % biological material (Pöschl et al., 2010). Total particulate matter mass was calculated here from the UV-APS number concentrations (m-3) and converted to mass for particles of aerodynamic diameter 0.5–10 µm. In only this case a density of 1.5 g cm-3 was utilized to calculate a first approximation of total particle mass to which all other mass measurements were compared. An average TSP mass density of 1.5 g cm-3 was utilized, because organic aerosol is typically estimated with density < 1.0 g cm-3, biological particles are often assumed to have ca. 1.0 g cm-3 density, and mineral dust particles have densities of up to ca. 3.5 g cm-3 (Dexter, 2004; Tegen and Fung, 1994). Fungal spore mass was estimated here using the fungal spore concentrations calculated from arabitol and mannitol mass (Fig. 6) and then using an estimated 33 pg reported by Bauer et al. (2008b) as an average mass per spore. Dividing the resultant fungal spore mass by total particulate mass provides a relative mass fraction for each high-volume sample period. These calculations suggest that fungal spores represent ca. 23 % ± 9 (using arabitol) or 21 % ± 8 (using mannitol) of total particulate mass during Rainy periods (Table 4, Fig. 7). This represents a nearly 6-fold increase in percentage compared to Dry periods (4.8 % ± 1.4 and 3.7 % ± 1.1, respectively). A similar increase during Rainy periods was also seen in the mass fraction of fungal cluster Cl1, which represented 17 % ± 10 of the particle mass during Rainy and 2 % ± 1 during Dry periods (Table S4).

Endotoxins measured in the atmosphere are uniquely associated with gram-negative bacteria (Andreae and Crutzen, 1997). Here, we show correlations between total endotoxin mass and WIBS ClBact, which was assigned by Crawford et al. (2015) to be bacteria due to the small particle size (< 1 µm) and high correlation with rain. This assignment of particle type to this set of clusters is quite uncertain, however, and should be treated loosely. The correlation between endotoxin mass and UV-APS and the WIBS clusters was very poor, in most cases R2 < 0.1 (Table 3, Fig. 8), suggesting no apparent relationship. Analysis of bacteria by both UV-LIF techniques is hampered by the fact that bacteria can be < 1 µm in size and because both instruments detect particles with decreased efficiency at sizes below 0.8 µm. So weak correlations may not have been apparent due to reduced overlap in particle size. Despite the lack of apparent correlation between the techniques, the relatively variable endotoxin concentrations were elevated during Rainy periods, consistent with Jones and Harrison (2004), who showed that bacteria concentrations were elevated after rainy periods.

Glucans, such as (1→3)-β-D-glucan, are components of the cell walls of pollen, fungal spores, plant detritus, and bacteria (Chow et al., 2015b; Lee et al., 2006; Stone and Clarke, 1992). In contrast to the observed difference in endotoxin concentration during the different wetness periods, (1→3)-β-D-glucan showed no correlations with UV-LIF concentrations (Table 3) and no differentiation during the different wetness periods.