Measurements were made at the Kansas State University Northwest Research-Extension Center in Colby, KS, and at the University of Wyoming Sustainable Agriculture Research and Extension Center (SAREC) in Lingle, WY. Harvests of soybean, sorghum, and wheat were sampled in Kansas, and a corn harvest was sampled in Wyoming. Photographs of the fields pre-, during, and post-harvest are shown in Figs. S1 and S2 in the Supplement. The harvests usually lasted 2–4 h. Details of the sampling locations and dates are listed in Table 1. A mobile laboratory and gasoline-powered generators were used to sample in the fields during the harvests. Generators were always positioned downwind of the mobile laboratory. Additionally, the mobile laboratory was positioned downwind from the field being harvested and was repositioned if the plume was no longer reaching the laboratory due to a shift in wind direction or position of the combine harvester. Figure S3 shows an example GPS track of a combine harvester during a corn harvest and the mobile laboratory position in relation to the combine. Aerosols were sampled through an inlet comprised of a stainless-steel rain hat located 3.5 m above the ground with a 1/2 in. OD stainless-steel tube attached. From this tube, y-splitters were used to split the aerosol flow among various instruments. A schematic of the various sampling configurations used during the study is provided in Fig. S4 in the Supplement.

Ice-nucleating particle concentrations were measured online with the Colorado State University (CSU) Continuous Flow Diffusion Chamber (CFDC) (Rogers et al., 2001; Eidhammer et al., 2010). Aerosols are introduced into the CFDC chamber, which has two cylindrical walls that are coated with ice and held at different temperatures. The temperature difference results in a supersaturation gradient that permits calculation of the supersaturation and temperature at the predicted position of the aerosol lamina ring within a particle-free sheath flow between the walls. For the measurements performed here, the CFDC was operated at 5 % supersaturation with respect to water over a temperature range of -15 to -32 ∘C. As discussed in DeMott et al. (2015), supersaturation uncertainty ranges from < 1.6 to > 2.4 % over this range of temperatures. While these settings potentially permit deposition, immersion, and condensation-freezing modes of ice nucleation to occur, conditions in the supersaturated (or growth) region of the CFDC emphasize aerosols growing into water droplets via condensation and then droplets that contain INPs freezing into ice crystals. Hence, data collected during operation in this manner are often compared to methods that explicitly examine immersion-freezing nucleation. Downstream of the growth region, droplet evaporation is stimulated (evaporation region) by holding the two iced walls at the same temperature to create water subsaturated conditions. This design feature amplifies the size difference between larger ice crystals and smaller aerosol particles. All particles are sized and counted by an optical particle counter (OPC, Climet CI-3100), and particles larger than 3 µm are counted as ice crystals. A 2.4 µm impactor (50 % aerodynamic cut-size diameter) was used upstream of the CFDC inlet to limit the size of particles entering the CFDC, as large particles (> 3 µm) would otherwise interfere with counting the ice crystals that form in the instrument. The uncertainty in INP concentration is calculated by adding in quadrature the Poisson counting-statistics-derived standard deviations of the sample and background periods, which are measured by sampling through a particle filter upstream of the CFDC (Schill et al., 2016). A statistical significance test is also performed on the data. If INP concentrations are greater than the error in INPs multiplied by 1.64 (INP > (INPError ⋅ 1.64)), which corresponds to the Z statistic at 95 % confidence, then the data are considered statistically significant (Schill et al., 2016).

For select sampling periods, particles that formed ice in the CFDC were collected for chemical analysis via impaction onto electron microscopy (EM) grids (SPI Supplies Coated Grids Formvar^®/Carbon, 200 mesh, nickel) with a 2.9 µm single-stage inertial impactor (Kreidenweis et al., 1998). Scanning electron microscopy (SEM; Quanta FEG MK2) was used to image the particles, and energy dispersive X-ray (EDX; Oxford Instruments X-Max EDS detector) analysis was performed to obtain elemental composition. Analysis was completed at the University of Wyoming Materials Characterization Laboratory. Analysis was done by analyzing individual particles on the filters (73 particles for the sorghum sample and 67 and 72 particles for the corn samples). Characteristic combinations of elements were identified and then used to group the individual particles into classes. Particles containing at least one mineral dust marker, such as silicon, aluminum, or iron, were labeled as dust particles. These particles typically contained oxygen and sometimes carbon as well. Particles with oxygen, carbon, and either sulfur or nitrogen were labeled as organic. If particles contained phosphorous, along with organic markers (O, C, N or S), they were labeled as biological (Pratt et al., 2009). Mixtures of these particle types were labeled as both types. For example, if a particle contained silicon, carbon, oxygen, and nitrogen, it was labeled as dust-organic.

The CFDC sample flow rate of 1.5 L min-1 sets a limit of detection that restricts its useful temperature range for assessing INP number concentrations. This detection limit is also dependent on the background concentration measured. For the 10 min integrated sampling periods typically used and when background counts are low, this detection limit is ∼ 0.2 L-1. To help overcome this limitation, an aerosol concentrator (MSP 4240) was used upstream of the CFDC to concentrate aerosol using virtual impaction (Fig. S4b in the Supplement) (Romay et al., 2002). The concentrator was located on a cart 1 m above the ground. This method has been used in previous studies (e.g., Tobo et al., 2013). Measurements were taken for 10 min with and 10 min without the concentrator. A concentration factor (CF) was calculated by taking the number concentration of INPs during concentrated periods divided by the INP concentration during temporally adjacent non-concentrated periods. During harvesting, there were large spikes in concentration due to the passing of a combine harvester. This made CF calculations difficult when the concentrated and non-concentrated INP values were not equally affected by the spikes. It was important to use a time period within the overall harvesting experimental period that included stable aerosol concentrations during periods on and off the concentrator. Therefore, the CF calculated during a pre-soybean harvest period in Colby, KS (CF = 90 ± 3), was used as the CF for all of the harvests. We may note that this value is within 15 % of the value found by Tobo et al. (2013) in prior studies with the same aerosol concentrator and CFDC instrument. This is physically expected if most INPs reside in the size range above 0.5 mm. CF uncertainty was calculated by propagating the uncertainties in the INP values used to calculate it. Then, the INP number concentrations during periods using the concentrator were corrected by dividing INP number concentrations by the CF.

The Ice Spectrometer (IS) immersion freezing method uses aerosols collected onto filters over periods of 2–4 h, achieving 800–3500 L sample volumes that can extend the range of INP measurements to warmer temperatures and a detection limit of ∼ 0.001 INPs L-1. Although creating difficulties for comparing methods when higher-frequency changes in INP concentrations are occurring, these two methods are complementary, offering colder temperatures and higher time resolution with the CFDC and warmer temperatures and lower INP detection limits with the IS. For IS analysis, aerosols were collected onto 47 mm diameter, 0.2 µm pore diameter (sometimes 0.05 µm) polycarbonate Nuclepore filters (Whatman, GE Healthcare Life Sciences) fitted within open-faced Nalgene sterile filter units (Thermo Fisher Scientific Inc.). During the wheat and corn harvests, a 2.5 µm cyclone (50 % aerodynamic cut-size diameter at 16.7 L min-1, URG Corporation) was also used, upstream of a 47 mm diameter inline aluminum filter holder (Pall Corporation) fitted with a 0.2 µm diameter pore Nuclepore membrane. This limited the size of the particles collected to the same size range as the CFDC. Filters and dissembled filter holders were cleaned before use by immersion in 10 % H2O2 for 30 min followed by three rinses in deionized water (18 MΩ and 0.2 µm pore diameter filtered) and then dried by removal of excess water and placement on foil in a clean-air laminar flow cabinet.

For processing in the laboratory, filters were transferred to sterile, 50 mL Falcon polypropylene tubes (Corning Life Sciences); 7–10 mL of 0.02 µm pore diameter filtered water (Anotop syringe filter, Whatman) with 2 mM KCl was added to maintain activity of K feldspar, if present. Prior tests on dilute suspensions of pure K feldspar found that the use of deionized water reduced IN activity, presumably due to desorption of K+; the use of a suspension containing ≥ 0.1 mM K+ prevented this and so was used for dilutions. Tubes were tumbled end-over-end at 1 cycle s-1 for 20 min (Roto-Torque, Cole-Palmer) to resuspend particles. Measurements of immersion freezing were made on this suspension and 20-, 400- and 8000-fold dilutions of it. Thirty-two 50 μL aliquots of each dilution and a negative control (2 mM KCl) were then dispensed into two 96-well polymerase chain reaction (PCR) trays (μCycler, Life Science Products), which were then transferred to the cold blocks in the IS. The trays were then slowly cooled by lowering the temperature at a rate of 0.3 ∘C min-1 from 0 to -27 ∘C, and the numbers of wells frozen were counted at 0.5 or 1 ∘C intervals. Cumulative numbers of INPs per volume of liquid as a function of temperature were estimated using the formula -lnfu(T)/V, where fu(T) is the proportion of droplets not frozen at a given temperature and V is an aliquot volume (Vali, 1971). Values were then converted to concentrations per liter air samples. Uncertainties are given as binomial sampling confidence intervals (95 %) (formula no. 2, Agresti and Coull, 1998). For a detailed description of the IS, see Hiranuma et al. (2015).

Ambient aerosols were sized at aerodynamic sizes larger than 0.542 µm using an aerodynamic particle sizer (APS; TSI 3321) and counted using a condensation particle counter (CPC; TSI 3010). The wideband integrated bioaerosol sensor (Droplet Measurement Technologies WIBS-4A) was used to collect information on fluorescent and biological aerosols. The WIBS-4A, from here onward referred to as the WIBS, gives fluorescence information in three channels: FL1 (fluorescence at 310–400 nm, excited at 280 nm), FL2 (fluorescence at 420–650 nm, excited at 280 nm), and FL3 (fluorescence at 420–650 nm, excited at 370 nm). Particles from 0.8 to 20 µm are sized by light scattering. Based on fluorescence signatures, the data were classified into four particle classes: any particles measured with the WIBS (total particles), particles that fluoresced in at least one channel (FP), particles that fluoresced in two channels (termed fluorescent biological aerosol particles, FBAPs), and particles that fluoresced strongly in channel FL1 and weakly or not at all in channels FL2 and FL3 (FP3), as described in Wright et al. (2014). The WIBS instrument was non-functional during the corn harvest on 9 November 2015; thus, fluorescence data supplied for corn are from a corn harvest on a different day (4 November 2015) but at the same field in Wyoming.

This work utilized two different types of treatments to tease out the various biological and chemical compositional influences on INPs measured in the ambient environment. Upstream of the CFDC, a tube heated to 300 ∘C was used to deactivate organic components before they entered the CFDC. The heating tube setup, shown in Fig. S4c in the Supplement, consists of two tube furnaces (Thermolyne 21100) placed next to each other in series with a 1 in. diameter quartz tube running through the center of the tube furnaces. By measuring INP concentrations with and without passage through the heating tube, the fraction of organic INPs, which are deactivated by heating, can be measured in situ. Previously, heat and peroxide have been used to degrade organic INP components in bulk soil samples and aerosol generated from it post treatment (Tobo et al., 2014; Hill et al., 2016). While useful, these methods do not provide precise time resolution on single particles, which is important for episodic events. Thus, to enable higher time resolution of single ambient particles, an online heating technique was developed and used in this work. For initial optimization, to ensure that the heating tube method was comparable to the previous bulk heating results, the same soil sample (soil from a sugar beet crop collected in Wyoming) used by Tobo et al. (2014) in the bulk heating analysis was aerosolized and run with the heating tube setup. The previous study measured aerosolized pre-treated soil particles size-selected at 0.6 µm using a differential mobility analyzer (DMA; TSI 3080) before sampling with the CFDC. In this study, particles are size-selected at 0.5 µm. However, this difference in size did not greatly change the results. Figure S5 in the Supplement shows the previous results using the bulk heating method (Tobo et al., 2014) plotted with the results using the heating tube. Data from the heating tube at 300 ∘C agree well with results from the bulk heating experiment. This comparison demonstrates that, even though particles only pass through the heating tube for 98 s with a flow rate of 1.5 L min-1, the heating tube technique is as effective at degrading organic components as the bulk heating method, which entailed heating to 300 ∘C in an oven for 2 h.

Post-treatments were also applied to the IS filter wash water of the wheat harvest sample to selectively deactivate different INP components. To denature labile organic components (e.g., proteins), an aliquot was heated to 95 ∘C for 20 min, while to decompose and remove all organic INPs, an aliquot of the wash water was digested with hydrogen peroxide. The latter used the same method as detailed in McCluskey et al. (2018), except we used a more powerful UV source for hydroxyl radical generation. Briefly, this entailed adding 30 % H2O2 (Sigma Aldrich) to the aliquot to achieve a final concentration of 10 % and then immersing the suspension in water heated to 95 ∘C for 20 min while being illuminated with two 26 W UVB fluorescent bulbs (Exo Terra) to generate hydroxyl radicals. To remove residual H2O2, catalase (cat. number 100429, MP Biomedicals) was added in 20 µL aliquots to the cooled solution, allowing several minutes between each addition, until no further effervescence occurred. To lyse all bacteria (including known IN species) another aliquot was incubated with lysozyme to digest their cell walls (lysozyme also hydrolyzes fungal chitin oligosaccharides but not the chitin polymer itself). An aliquot of the aerosol suspension was amended to contain 4 mg mL-1 lysozyme, 10 mM Tris buffer, and 5 mM EDTA (both at pH 8) and incubated at 24 ∘C for 3 h. For a detailed description of this method, see Hill et al. (2016).

Measurements were made during a soybean harvest on 14 October 2014, a sorghum harvest on 15 October 2014, and during a wheat harvest (30 June and 1 July 2015, referred to as Wheat 1 and Wheat 2, respectively) in Colby, KS, and during a corn harvest in Lingle, WY, on 9 November 2015. Figure 1 shows CFDC and IS INP number concentrations measured during the harvests. The corn and wheat IS data were sampled through a 2.5 µm cyclone, in addition to open-faced Nalgene sterile filter units, to limit the size range of particles that were collected. While the use of the cyclone will not capture the IN activity of larger particles, these larger particles will sediment out faster than smaller particles and likely do not make it to cloud level. Therefore, the use of the cyclone and impactor offer a better representation of particles that could impact clouds. IS data with and without the cyclone are provided in Table 2 for a comparison of INP concentrations with and without larger particles.

The CFDC INP number concentrations are averaged over 3 to 5 min periods, and the IS INP number concentrations represent the average over the whole harvest sampling period (typically 2–4 h). For a given CFDC operating temperature, there was a broad range of INP number concentrations for a given harvest due to the nature of harvest sampling: the concentrations vary rapidly in time due to the movement of the combine harvester up and down the field, laterally across, and closer and further away from the mobile laboratory and due to the stopping and starting of the harvesting. Thus, the difference in time resolution between the CFDC and IS techniques can explain some of the greater spread in CFDC data. Even so, the IS and CFDC data generally agree well in the overlap region between -15 and -25 ∘C. INP concentrations ranged from 0.5 to 147 L-1 at -30 ∘C as measured with the CFDC, and the IS data showed a maximum INP concentration of 922 L-1 at -25.5 ∘C during the wheat harvest. In general, these concentrations are very high compared to background INP concentrations and global averages (e.g., DeMott et al., 2010). This result was consistent with limited previous harvest sampling (Garcia et al., 2012) but not consistent when comparing crops harvested or even between harvests of the same crop (i.e., wheat). Further, measurements from a corn harvest in Nebraska described by Garcia et al. (2012) showed INP concentrations from drop freezing analysis between 30 and 80 L-1 at -20 ∘C and an average CFDC INP concentration of 5.9 L-1. In this study, average number concentrations of 8.1 and 3.6 L-1 were measured with the IS and CFDC, respectively, at this temperature. While the CFDC results agree quite well between the two studies, there was up to an order of magnitude difference between the immersion freezing measurements. The average distance from the combine harvester during sampling or differences in plant and soil properties at the time of harvesting could contribute to this. While the soil was dry during the Nebraska harvest, it was wetter during this study due to recent heavy rain, thus limiting the amount of soil dust kicked up during sampling. These results illustrate the complexity of harvest emissions due in part to varying concentrations in time, distance from the source, and soil moisture.

The shapes of all of the IS harvest spectra are similar, with a “hump” at the warm end (-5 to -22 ∘C), which is accentuated in wheat sample 1. This warm temperature hump, which is a frequent feature in terrestrial INP spectra and is commonly observed in precipitation samples, is suggested to be from biological sources (Petters and Wright, 2015). Interestingly, the crop dust emissions from the wheat field were considered particularly strong due its infestation with rust (Marv Farmer, personal communication, 2015), a parasitic fungal infection. Rust breaks down plant cell walls, which can result in more and finer plant dust particles being produced during harvesting. Furthermore, rust damage to leaf tissues would have allowed many adventitious phylloplane bacteria and fungi to have flourished. IN bacteria have been measured on wheat at populations of 108 g-1 of fresh green leaf in Wyoming (Hill et al., 2014) and at 3.5 × 106 g-1 of fresh dry leaf at harvest in Colorado (Garcia et al., 2012), and rust has been shown to be IN active at warm temperatures (Morris et al., 2013). Thus, these various biological particles could have contributed to the INP concentrations seen in the pronounced hump in the IS spectrum for this case. Also, total aerosol numbers, the concentration of fluorescent particles (see below), and INP concentrations on this day were the highest observed in all of the harvest measurements. This suggests that the direct and indirect consequences of the fungal infection of the wheat crop could be contributing to the large number of particles and could be altering the characteristics of the emitted particles.

It should be noted that the CFDC and IS INP spectra have different slopes and the concentrations can be quite different at colder temperatures. The CFDC INP concentrations are generally higher than or similar to the IS at warmer temperatures but lower at temperatures below -25 ∘C. The reasons for this are not fully understood; however, there are some possible explanations, as discussed in DeMott et al. (2017) and revisited here. Particle conglomerates could break up while in the IS wash water, which could provide more INPs in the bulk solution than are present as single particles measured in the CFDC. Alternately, small ice-nucleating entities (INEs), such as protein complexes (Hartmann et al., 2013) or macromolecules on pollen (Augustin et al., 2013; Pummer et al., 2012), could be present on and then released from the particle surfaces, and these INEs might be especially active at lower temperatures. Size could also play a role, as INP sizes are generally larger during harvests (Mason et al., 2016) and are not as effectively sampled into the CFDC. During the soybean and sorghum harvests, no size restriction was placed on the IS filters; thus larger particles (> 2.5 µm) are underrepresented in the CFDC data as compared to the IS. This is less of a concern (and indeed may be reflected in the data) for the corn and wheat harvests because a 2.5 µm cyclone was used to restrict particle sizes on the IS filters. Finally, there may be a time dependence of ice nucleation that accentuates differences between the CFDC and IS measurements at lower temperatures. Particles are in the CFDC growth region for approximately 5 s, while they are at a particular temperature for several minutes in the IS. However, previous results suggest there is little temperature dependence to stochasticity (Wright and Petters, 2013), and thus it is unlikely that the time difference is the cause of the discrepancy that occurred at the lowest temperatures.

Fluorescent and biological particle concentrations and types measured by the WIBS were grouped into particle classes as described in the “Methods” section. These data are shown in Fig. 2 and Table 2. Interestingly, FBAPs were observed before the start of the soybean harvest period, which has been termed the “pre-soybean harvest” period. During the soybean harvest, the FBAP percentage increased from 6.8 % during the pre-soybean harvest period to 17.8 %, indicating that additional biological particles were emitted during the harvest. This pre-harvest period was likely strongly influenced by harvesting in the region even though we were not directly in a fresh harvest plume. The corn harvest produced the largest percentage and concentration of FBAPs out of all of the harvests sampled (33.5 %, 107.7 L-1). These results suggest an abundance of biological particles are emitted during corn harvests, which is supported by a previous study that showed corn harvests emit bacteria (Garcia et al., 2012), and could also indicate that more plant fragments are emitted during corn harvests than for other plant harvests. While wheat harvests emitted the highest number of fluorescent particles, they had the lowest percentages of FBAPs and Wheat 2 had the lowest concentration of FBAPs (2 L-1). This indicates that the wheat emissions only fluoresced in one channel and could point to the greater presence of plant material or soil dust, as opposed to other biological particles. Lignin is present in wheat and absorbs at 280 nm and emits at ∼ 360 nm, which would give a signal at FL1 (Albinsson et al., 1999). Wheat lignin also autofluoresces with excitation at 330–385 nm and detection at 420 nm, so it will give a signal at FL3. Therefore, if wheat lignin made up a bulk of the emitted particles, they might show up as FBAPs; however, FBAPs made up a low percentage of particles. This suggests that the bulk of the fluorescent wheat particles lack lignin and could be non-lignified plant cells or dead microbes.

The FP3 particles did not make up a significant percentage of particles, except during the first wheat harvest and the corn harvest, and average concentrations ranged from 5 L-1 in the pre-soybean harvest period to 193 L-1 during the Wheat 1 harvest. These particles have not been biologically or chemically identified but have been shown to correlate with INP concentrations (Wright et al., 2014). FP3 is indicative of the presence of tryptophan and the absence of nicotinamide adenine dinucleotide + hydrogen (NADH) and could indicate dead plant and microbial material containing protein (e.g., dead phloem cells, dead bacteria, and fungi). Figure 2 and Table 2 also include IS INP concentrations at three temperatures for comparison to the fluorescent particle concentrations. In general, the INP concentrations measured without a cyclone at -20 ∘C are on the same order of magnitude as the FP3 concentrations, and thus FP3 concentrations show potential as an indicator of INP concentration at -20 ∘C. The Wheat 1 case is somewhat of an outlier with the highest INP concentration at -15 ∘C, which was on the same order of magnitude as the FP3 concentration. Overall, the best agreement was between the FP concentrations and the -25 ∘C INP concentrations, which agreed very well for the three samples that had IS data at -25 ∘C.

To chemically characterize the harvest INPs, particles were collected via impaction onto SEM grids downstream of the CFDC. During the sorghum harvest, mineral dust made up 41 % of INPs at -17 ∘C, as shown in Fig. 3a. Organic material also made up a large fraction (29 %) of the INPs along with mixtures of dust and organics (13 %) and a small percentage of purely biological particles (2 %). Images of the particles, shown in Fig. 3b, include interesting structures that could be indicative of plant material or biological particles. There were plate-like structures with potassium, thread-like filaments containing silicon, which could be plant material (Lux et al., 2002), and flaky structures with the elemental composition of mineral dust. These results indicate that organic components make up a large percentage of sorghum harvest INPs at -17 ∘C.

SEM-EDX data from the corn harvest at -27 ∘C, shown in Fig. 4a, indicate mineral dust again comprised a significant portion of INPs (32 %), along with dust mixtures with organics (Dust-org, 19 %), biological particles (Dust-bio, 13 %), and sulfate (Dust-S, 10 %). Additionally, there was a significant number of biological particles (18 %), which were identified by the presence of carbon, nitrogen, and phosphorous (Pratt et al., 2009). Many of the measured INPs also had structured forms similar to the sorghum harvest emissions. These particles had oblong, granular shapes and some appeared to have tiny hairs on the surface, suggesting they were of biological origin. The large percentage of biological INPs agrees well with previous measurements during corn harvests, which also showed the presence of several genera of bacteria among CFDC residuals (Garcia et al., 2012); this included 19 IN bacteria L-1 air, quantified directly using quantitative PCR of the Ina gene. While SEM-EDX elemental compositions indicated the presence of biological particles, a full characterization of the biological components cannot be achieved with this method. Future work will utilize post-treatments of the filtered particles to further identify the types of biological particles that served as INPs. The WIBS data also revealed that the corn harvest produced the largest fraction and concentrations of FBAPs out of all the sampling locations. Taken together, these results indicate that organics and biological particles, along with mineral dust, make up a large percentage of harvest INPs between -17 and -27 ∘C. SEM samples were not collected during the soybean and wheat harvests; therefore, a comparison of organic and mineral components cannot be directly assessed from these crops.

In situ heating during real-time CFDC measurements was utilized to assess the contribution of minerals and organics to INPs emitted from harvests. Heating at 300 ∘C has a similar impact on organics as peroxide digestion and will degrade heat-labile organics and biological particles (Tobo et al., 2014), and thus a comparison of heated and non-heated INP concentrations reveals the percentage of organic versus inorganic INPs. SEM-EDX results presented in Fig. 4b show the chemical changes in INPs that occurred with heating at 300 ∘C during the corn harvest. The percentage of mineral dust increased, which is expected because, as organics are degraded with heating, minerals will remain IN active and make up a larger percentage of INPs. The percentage of biological INPs was reduced from 18 % to 7 %, but they were not totally deactivated. The heat treatment dramatically reduced the percentage of Dust-bio INPs from 13 % to 1 %, which suggests the IN-active biological components were degraded with heat, suggesting that the biological components played a larger role in the IN activity than the minerals within this class. The latter scenario is consistent with previous studies that show organic and protein residues on mineral surfaces can enhance the ice nucleation ability of the minerals (O'Sullivan et al., 2014, 2016; Conen et al., 2011). Interestingly, the percentage of Dust-org particles was not reduced with heating. This could indicate that the organics that were internally mixed with minerals were not susceptible to heat at 300 ∘C or that, after the organics were degraded, the INP activity was unchanged because the minerals in these mixed particles were serving as the active sites for ice nucleation. Alternatively, heating the Dust-org particles could have evaporated off some volatile organics uncovering active sites on the dust. This study cannot differentiate between those scenarios, but future studies should investigate the physical changes caused by heating, including how heating might change mixing state and surface morphology.

The quantitative changes to ice-nucleating ability with heating are shown in Fig. 5. The fraction of INPs with respect to the concentration of total particles larger than 0.5 µm, as measured with the CFDC OPC (n0.5µm), is plotted on the y axis. INP fraction is shown instead of INP concentration to allow for direct comparison between heated and non-heated sampling periods. This is because there were large changes in particle concentrations due to sampling in and out of the harvesting plume and changing wind directions, which complicate a direct concentration comparison. This figure displays only statistically significant data points, as determined with the significance test described in the “Methods” section, unless otherwise noted. Results indicate that heating had a large impact on INP number concentration for soybean harvest emissions as cold as -25 ∘C. Similarly, for the sorghum harvest, INPs were reduced by heating to below detection levels at -18 ∘C, but a smaller impact was noted at temperatures ≤ -22 ∘C. During the corn harvest, heating reduced the fraction of INPs at warm temperatures (-19 ∘C) to below the instrumental detection limits, but at colder temperatures (-28 ∘C) there was only a slight change in the fraction. A similar situation appears for the wheat harvest, although data were not collected for heating trials below -22 ∘C. These results suggest a general, albeit variable, impact in which organic (including biological) particles from harvesting exert more influence at warmer temperatures, while at colder temperatures mineral dust components likely dominated the ice nucleation activity. Further characterization of the emissions is necessary to identify the nature of the organic particles, but these results suggest harvest emissions are distinct for different crops.

The observed decrease in INPs with heating is presented in a different way in Fig. 6. Fractional change in INPs is shown for each temperature and crop in cases where heating measurements were made. At temperatures between -17 and -19 ∘C (the warmest temperature accessible for comparison via CFDC data), all of the harvest samples had large decreases in INP activity with heating. The fractional changes at these relatively warm temperatures were between -0.7 and -0.98, which suggests that a large percentage of these warm temperature INPs are of organic or biological origin. At -32 ∘C, the INPs fractionally decreased by ∼ 0.5 for all crops, indicating that minerals and possibly 300 ∘C heat-stable organics are contributing up to 50 % of the INPs. However, the fact that there was still a reduction in INPs at these cold temperatures agrees with previous results that showed that organics contributed significantly to the INP population of soil dust even at cold temperatures where, traditionally, minerals are expected to dominate the activity (Tobo et al., 2014).

During the wheat harvest, heat treatment resulted in a 98 % reduction in INP number concentrations at -18 ∘C. This suggests that biogenic particles make up almost all of the INPs at temperatures ≥ -18 ∘C. Additional focus in the wheat sampling was placed on the evaluation of the contributions to this degradation observed in situ and in real time with the CFDC. Post-treatments on the wheat harvest sample via IS immersion freezing measurements, shown in Fig. 7, revealed a variety of biological/organic INP compositions contributing to the IN activity > -20 ∘C, along with an underlying mineral or non-organic contribution to the IN activity, as suggested by the dashed grey line. Lysozyme digestion indicated that bacteria likely contributed foremost to the INP population. By digesting bacterial cell walls, lysozyme will cause rupturing of all bacteria. For the known species of ice nucleation active bacteria (e.g., Pseudomonas syringae, Pantoea agglomerans, and Xanthomonas campestris), clusters of the protein anchored in the outer membrane will, as the outer membrane disintegrates, disaggregate into smaller clusters active at ∼ -7 to -10 ∘C or into single proteins active at -12 to -13 ∘C (Govindarajan and Lindow, 1988). However, in the wheat harvest sample the effect was observed as cold as -21 ∘C, suggesting that other, as yet unidentified, IN bacteria were not only present but abundant in the wheat dust. While WIBS data suggested that < 1 % of particles were FBAPs, the lysozyme digestion shows that a large amount of bacteria was generated from the harvest.

Bulk heating of the IS sample to 95 ∘C resulted in a larger reduction of INPs that can be attributed to heat-labile INPs, such as proteins in bacteria and fungi, on the plants, and in soil dusts raised by the harvester. There was also a modest amount of organic material that was not susceptible to 95 ∘C heat, but was degraded with peroxide digestion, that was contributing to INP concentrations and is evident in the shaded regions between the red markers and the black markers in Fig. 7. If arable soil dust contributed largely to the INP concentrations, peroxide treatment would show a greater reduction in INPs than was observed here. The large reduction due to heating indicates that biogenic particles make up a large percentage of INPs at temperatures warmer than -18 ∘C. These biogenic particles come from a variety of sources, which highlights the complex nature of INPs emitted from agricultural and soil perturbation activities. No one particle type can accurately describe the nature of INPs for agricultural areas in general, but rather a mixture of biogenic particle types best represents these emissions. The findings in this study suggest that harvesting and plant litter emissions stimulated by wind at the surface provide the most viable explanation of the ubiquity of heat-labile INPs in the High Plains boundary layer even in the absence of harvesting, as found by Garcia et al. (2012).

Results presented herein, especially those shown in Figs. 6 and 7, emphasize the potential need to include harvesting INP emission impacts in regional cloud models to assess their subsequent impacts on clouds and precipitation in both agricultural and naturally vegetated regions. Harvesting emits mineral, organic, and biological particles into the atmosphere in large quantities. PM10 emission factors ranging from 10 to over 1000 kg km-2 have been reported for different crops harvested in California, and these emission factors vary based on crop, relative humidity, and soil moisture (Flocchini et al., 2001). A full characterization of the emitted organic matter is beyond the scope of this work and would involve intensive chemical, biological, and plant pathological investigations. Even from a single source such as harvesting, there are several distinct inputs including but not limited to pulverized plant tissues, dust, bacteria, fungi and other biological particles present on plant surfaces, various biological, organic and mineral INPs lofted from the soil, and even residual fertilizer on the soil surface. This complex combination of sources is difficult to untangle, especially because it can change with geographic location, crop type, plant and soil states, environmental conditions during harvesting, and year-to-year differences in the many parameters. Additionally, different pathogens can grow on the crops, as was shown with the rust-infected wheat crop sampled in Colby, KS. All of these factors can change the ratio of mineral to organic components in the INPs, which has implications for how these emissions should be represented in models.

To assess the ability of existing INP parameterizations to model harvesting INP concentrations, the measured CFDC and IS INP number concentrations were compared to predicted INPs using parameterizations for average global INP concentrations (D10) (DeMott et al., 2010), mineral dust (D15) (DeMott et al., 2015), and biological particles (T13) (Tobo et al., 2013) (Fig. 8). The D10 and D15 parameterizations predict INP concentrations at a given temperature based on particle number concentrations above 0.5 µm (n0.5µm). For comparison with the IS, particle number (n0.5µm from the CFDC) was averaged over the IS sampling times. The D15 parameterization results presented here do not include the factor of 3 increase suggested in DeMott et al. (2015) for use in predicting atmospheric concentrations of relevance to clouds, because comparisons here are made to uncorrected CFDC and IS INP concentrations. The T13 parameterization uses biological particle concentrations, derived from the WIBS FBAP concentrations in this study, instead of n0.5µm to predict INP concentrations. FBAP concentrations were averaged over the CFDC and IS sampling times to compare to CFDC- and IS-derived INP concentrations, respectively. In applying WIBS FBAPs within the Tobo et al. (2013) parameterization, we must note that FBAP concentrations used to develop the parameterization were based on an ultraviolet aerodynamic particle sizer (UV-APS) that senses FBAPs at sizes above 0.5 µm, while the WIBS FBAP signal is for > 0.8 µm particles. Hence, we expect that predicted values may be somewhat underestimated in this case. Note that all temperatures are integrated into such an analysis, so that biases may enter due to changes in the contributions of different compositions at different temperatures, as has been discussed. Also, the CFDC data presented here cover a narrower temperature range (-17 to -32 ∘C) than that used in developing these parameterizations (e.g., -9 to -34 ∘C for D10).

Comparisons shown in Fig. 8 indicate that different crops have different relationships with CFDC-derived n0.5µm as described by different parameterizations. For instance, the corn CFDC and IS with cyclone INP data are predicted most accurately by the D15 parameterization. D15 is used to model dust INP activity; thus the correlation suggests that dust was serving as a source of INPs during the corn harvest. The SEM-EDX results presented in Fig. 4a confirm this and show that dust and dust mixtures made up 74 % of INPs measured at -27 ∘C. However, the IS data for all sizes (no cyclone) are predicted best with the D10 parameterization. The WIBS instrument was inoperable during the corn harvest; therefore, the T13 parameterization could not be tested against the corn data.

The first wheat harvest (Wheat 1) INP concentrations are predicted well with the D10 parameterization for CFDC and IS with cyclone data. This “global” INP parameterization represented a diverse range of INP sources (i.e., not distinct to one source), which may explain why it captures the diverse range of INPs that wheat harvests emit. Furthermore, this again supports the idea that elevated INP activity at temperatures higher than -20 ∘C observed in data compilations like DeMott et al. (2010) have their major sources from plant and microbial INPs. The cyclone-IS INP concentrations are predicted best by the D15 parameterization, although they are consistently overpredicted. The relationship between n0.5µm and CFDC INPs during the second wheat harvest, Wheat 2, is not captured by any of the tested parameterizations. This could indicate changing emissions throughout the harvest or a mixture of minerals, organics, and biological particles that does not have a consistent relationship between n0.5µm or FBAPs and INPs. It might also suggest a nonfluorescing population or an especially active biological population not being represented by the T13 parameterization, which was modeled on data collected in a region rich in fungal spores and which might not capture the behavior of other biological types. To explore this scenario, the markers in Fig. 8 were colored by CFDC and IS operating temperature and are displayed in Fig. S6 in the Supplement. The Wheat 2 data points, as well as some Wheat 1 and soybean points that are not well predicted by T13, are mostly at warm temperatures (around -20 ∘C). Thus, this could indicate a non-spore biological particle type was contributing to the INP population during these harvests.

The soybean harvest was modeled well by both the D10 and D15 parameterizations for both CFDC and non-cyclone IS data; however, the T13 parameterization also accurately predicted the CFDC data except for a few points. This might suggest that the soybean emissions had contributions to INPs from both dust and biological particles. The soybean emissions had a large percentage of FBAPs (17.8 %) according to the WIBS and a strong reduction in INP activity after heat treatment, which is indicative of biological INPs. However, the heat did not totally wipe out the IN activity, which suggests the presence of minerals in the INPs as well.

Sorghum CFDC INP concentrations are modeled well by both the D10 and T13 parameterizations, while the IS non-cyclone data were best represented by D15. However, there is a hump in the data between -15 and -18 ∘C, which is better predicted by T13. This again suggests there is a mix of particle types including mineral dust and biological INPs present. Indeed, ∼ 11.8 % of particles measured on the WIBS were FBAPs, and heating reduced the INP activity of the sorghum to below detection limit at -17 ∘C and by 63 % at -32 ∘C, indicating a strong organic contribution to INPs. SEM-EDX analysis of INPs active at -17 ∘C shows 41 % of INPs were mineral dust and, while a small fraction (2 %) of particles were of biological origin, a large percentage of INPs had organic components (42 %). Some of these organic particles could be from biological sources but based on their low (undetectable) levels of phosphorus, they were labeled as organic. Phosphorus is often a limiting nutrient in plants and is re-mobilized into living tissues or seeds when plants are senescing. The high organic but low biological signature could indicate that phosphorous was relocated from leaf and stem tissues to the sorghum grain before the harvest, which has been observed in sorghum when phosphorous is limited (Roy and Wright, 1974). Images of some of these particles (Fig. 3b) show structured shapes indicative of biological origin. This evidence points to the importance of organic and biological particles as well as mineral dust serving as INPs during the sorghum harvest.

The results presented here suggest that different crops have different relationships between aerosol number concentrations and INP concentrations. No single INP parameterization accurately predicts INPs released during harvest periods for all crops, but both D10 and D15 could be used in agricultural regions to predict ambient INP concentrations during harvest months, given measurements and/or forecasts of aerosol concentrations. FBAP concentration data are not readily available, and thus the comparison to the T13 parameterization is provisional at this point.

The large seasonal increases in harvest emissions could have effects on precipitation, especially in the Plains states where deep convection is frequently occurring. Several modeling studies have investigated the effects of increased aerosol concentrations on convection. One study showed that increases in aerosols modify storm structure but have minimal effects on warm-front precipitation (Igel et al., 2013), while another suggested deep convection in the Great Plains is modified by larger aerosol loading, by raising cloud-top height in mixed-phase clouds and by increasing precipitation rates in clouds with large amounts of liquid water (Li et al., 2011). Increases in biomass burning aerosols have been linked to increases in severe weather (Wang et al., 2009) and the likelihood of tornado formation (Saide et al., 2015), while mineral dust has been shown to have competing effects on squall lines with an overall weakening due to larger dust concentrations (Seigel et al., 2013). It is important to note that these studies have focused on the effects of aerosols serving as cloud condensation nuclei and have not included aerosols serving as INPs. The varying effects of aerosols on convection highlight the need to further investigate these scenarios and include INPs in these simulations, which could change the results and lead to a better representation of clouds and precipitation in agricultural regions in models.