Ice-nucleating particles impact the severity of precipitations in West Texas

Abstract. Ice-nucleating particles (INPs) influence the formation of ice crystals in clouds and many types of precipitation. However, our knowledge of the relationship between INPs and precipitation is still insufficient. This study was conducted to fill this gap by assessing precipitation properties and INP concentrations (nINP) from a total of 42 precipitation events observed in the Texas Panhandle region from June 2018 to July 2019. We used a cold-stage instrument called the West Texas Cryogenic Refrigerator Applied to Freezing Test system to estimate nINP through immersion freezing in our precipitation samples. A disdrometer was used to measure the precipitation intensity and size of precipitating particles during each precipitation event. The analysis of yearlong precipitation properties as well as INPs for the samples shed a light on the seasonal variation of the nINP values in West Texas. Furthermore, we characterized the bacteria speciation of the storm and ambient dust samples collected at a commercial feedlot in West Texas to identify potential biological sources of INPs in our precipitation samples. Overall, our results showed a positive correlation between nINP and intensity of precipitation with notably large hydrometeor sizes in storm precipitations. Amongst all observed precipitation types, the highest INPs were found in the snow samples, and hail/thunderstorm samples have the highest INPs at high temperature −5°C.


atmospheric ice particles follows the process of homogeneous nucleation, in which it requires the cloud droplet to be supercooled to the temperature (T) of -32°C and below (depending on the pure water droplet size) to form ice crystals (Koop et al., 2000;Koop and Murray, 2016). Though our knowledge regarding INP remains insufficient, there have been advances in understanding the different modes of heterogeneous ice nucleation (IN) in the atmosphere in the last few decades. For example, deposition nucleation is induced by the direct deposition of water vapor on to an INP's surface and ice embryo formation on the surface under ice supersaturation conditions (Kanji and Abbatt, 2006;Möhler et al., 2008). Recently, some studies have argued that the deposition nucleation could be interpreted as pore condensation and freezing (Marcolli, 2014). The presence of water in pores of mineral materials and the resulting inverse Kelvin effect cause an instantaneous water saturation condition in the confined space, allowing the water to freeze even at water sub-saturated ambient conditions (David et al., 2019;Marcolli, 2014). Amongst various IN paths, perhaps the most important mode is immersion freezing (De Boer et al., 2010). This process starts with the formation of cloud droplet followed by freezing due to an INP immersed in the supercooled droplet. In addition, the past studies have identified other modes of heterogeneous nucleation, such as condensation freezing (Belosi and Santachiara, 2019), contact freezing (Hoffmann et al., 2013) and inside-out evaporation freezing (Durant and Shaw, 2005).

60
These modes are relatively less relevant in the mixed-phase clouds (MPCs) as discussed in the next section.

Importance of Immersion Freezing
INPs greatly influence cloud properties, especially in MPCs, which are typically observed in the altitude range of 2 km to 9 km above ground level (Hartmann et al., 1992). Out of all heterogeneous ice-nucleation modes, the immersion freezing is the most dominant mode of ice formation in MPCs (Ansmann et al., 2008;De Boer et al., 2010;Hande and Hoose, 2017;Vergara-Temprado et al., 2018). In Hande and Hoose (2017), different cloud types such as orographic, stratiform and deep-convective systems were simulated and analyzed for different freezing modes under various polluted conditions. The authors demonstrate that immersion freezing is the 70 predominant IN mode under various simulated circumstances, accounting for 85 to 99%, while other IN paths play a less significant role. Cui et al. (2006) also showed that immersion freezing is the primary mode of ice formation with little significance of the deposition mode in the early stages of the cloud development. Moreover, whereas contact freezing may be a highly efficient ice formation path, a previous simulation study showed that it is a negligible mode in the given MPC conditions (Phillips et al., 2007). Field et al. (2012) andDe Boer et al. (2011) showed that the formation of cloud droplets is a precondition for ice formation in MPCs, thus highlighting the importance of immersion nucleation. Furthermore, using multiple ground-based instruments, including Lidar, AERONET Sun Photometer, and Vaisala Radiosonde, Ansmann et al. (2008) found that a high INP concentration (n INP ) (i.e., ~ 1 -20 cm -3 ) in the Saharan dust. This high dust-including n INP episode coincided with the presence of liquid droplets at cloud tops at Ts of -22°C to -25°C. Similarly, Ansmann et al. (2009) shows the observation of 80 tropical altocumulus clouds having the liquid cloud tops. Due to the importance and dominance of immersion freezing, the current study focuses on measuring the immersion freezing efficiency of the precipitation samples collected in the Texas Panhandle region.

INPs and Atmospheric Precipitation
It is known that INPs in MPCs have a notable impact on the properties of precipitation. Previously, Yang whereas an excessive increase of INPs may impede the updrafts in the storm. The reason for this complex effect of INPs may be explained by the variation in the latent heat release in the convective system at different stages of its development. This latent heat is further influenced by INP episode, thus affecting the dynamics of the 90 precipitation system. Furthermore, the increase in INP number might reduce the mean hail diameter (hail particles with smaller diameters melt more easily), which leads to decreased hail precipitation and an increased rain formation in contrast to the previous studies (Fan et al., 2017;Van den Heever et al., 2006). Similar results have been found by Chen et al. (2019). The authors show that an increased n INP in the simulated hailstorm can reduce the graupel size and reduce the concentration of hail stones. Likewise, the aircraft observations along 95 with the model simulations of convective storms in West Texas and U.S. High Plains have shown that the addition of INPs at the base of warm clouds would result in an increase of the precipitation amount by strong updrafts in the system (Rosenfeld et al., 2008), ultimately affecting the local hydrological cycle (Mülmenstädt et al., 2015). It has also been observed that INPs can be removed from the atmosphere through precipitation resulting in a net decrease in n INP , affecting the precipitation development (Stopelli et al., 2015). The estimation of n INP in this study 00 from the precipitation samples gives a quantitative approximation of INPs in the locally observed weather systems, potentially allowing us to parameterize the INP-precipitation relationship.
Several previous studies have characterized the nINP in the precipitation samples from various locations (Creamean et al., 2019;Petters and Wright, 2015;Levin et al., 2019). Petters and Wright (2015) reported a wide 05 range of n INP values in their local precipitation samples collected approximately 3 km west of Raleigh, NC, USA for July 2012 and October 2013. Their study shows a variation of 10 orders of magnitude in the concentrations of INPs with a high variability in the T range of -5°C to -12°C, suggesting inclusion of biological INPs, which are generally known to be active at relatively high freezing Ts (Després et al., 2012). The lower limit for the INP spectrum as a function of T derived from the cloud water and precipitation samples in Petters and Wright (2015) may highlight the extreme rarity of INPs at Ts warmer than -10°C. Particularly, the authors showed that the highest ever observed n INP at -8°C were three orders of magnitude lower than observed ice crystal concentrations in tropical cumuli at the same temperature. More precipitation studies may provide a constraint on minimum enhancement factors for secondary ice formation processes. In Levin et al. (2019)

Study Objectives
In this study, we calculated the n INP in precipitation samples collected in the Texas Panhandle region. All of our samples were analyzed at our laboratory using a cold stage instrument. The estimated n INP in the precipitation samples were compared with precipitation properties, such as the intensity of precipitation (mm hr -1 ) and hydrometeor particle size (mm). In addition, the seasonal variation of n INP in the Texas Panhandle 25 region was studied and compared with the particulate matter (PM) concentrations measured by our Internet of Things (IoT) sensors. A subset of the collected precipitation samples was analyzed for their bio-speciation to characterize potential biological INP sources in the West Texas region and also to investigate if the biological composition matches with any previously known high T biological INPs.

Precipitation Sampling
In this study, the precipitation samples were collected from different seasons throughout the year during June 2018 -July 2019. Sterilized Polypropylene tubes of 50 ml (VWR Centrifuge Tube) were used as sampling gauges. The gauges were placed at ~50 ft above the ground on the rooftop of Natural Science Building at West Texas A&M University, Canyon, TX. This particular location was chosen to avoid any obstruction for 35 our sampling activities. The sampling tubes were well exposed to the ambient air without any canopies throughout the sampling process. The sampling gauges were replaced every 24 hours to minimize the effect of dry deposition prior to the precipitation sample collection. A blank dry deposition sample (Sample# 34) was specifically collected for 24 hours from January 2-3, 2019 in order to examine and quantify the effect of dry deposition on n INP . The freezing spectrum of this dry deposition sample (suspended in HPLC grade pure water) 40 was later compared with the IN spectra of precipitation samples (see Sect. 3.3.1). For the duration of a given precipitation episode, some amount of sample was accumulated in the tube. The sampling tubes were then capped and stored at T of 4°C in the refrigerator, following the method described in Petters and Wright (2015), until the droplet-freezing assay experiments were commenced. The effect of storage conditions on the IN activity was not considered in this study. We note that Beall et al. (2020) recently found a decrease in precipitation n INP by 42% when stored at 4°C and suggested correction factors for the T range of -7°C to -17°C. After the freezing experiment, a subset of our samples was kept under deep-freezing conditions (-80°C) for further biological analysis (see Sect. 2.6). In total, 42 precipitation samples were collected from different weather systems. In this study we estimated the n INP values from (1) snows, (2) hails/thunderstorms, (3) long-lasted rains, and (4) weak rains. More information about the samples used in this study, precipitation types and the amount of the precipitation collected for each sample are provided in the Supplemental Information (SI) Table S1-1.

Disdrometer Measurements of Precipitation Properties
For our precipitation measurements, we used the OTT Parsivel 2 (Particle Size Velocity 2) sensor. This device is a modern laser-optical disdrometer (λ = 780 nm) which measures the size and fall velocity of precipitating particles. The OTT Parsivel 2 was deployed in side-by-side position with the precipitation gauge collector for the duration of our study period. A detailed technical description of OTT Parsivel 2 is given in a previous study (Tokay et al., 2014), so only a brief description is provided here. A combination of the laser transmitter and receiver component was integrated as a single cluster in a weatherproof housing and detect precipitation particles passing through a horizontal strip of light. A nominal cross section area of a laser beam detection was 54 cm 2 , and the system recorded the number of hydrometeors in a 32 x 32 matrix (i.e., fall velocity x diameter) in the ≥ 30 seconds time resolution. The measurable size range of hydrometeor particles was 0.062 -24.5 mm in diameter (D p ) with bin size intervals (∆D p ) varying from 0.125 to 3.0 mm. Our disdrometer was coupled with an OTT netDL Hydrosystem logger (40 channels). The OTT Parsivel 2 also measured the intensity of precipitation (mm hr -1 ) and the number of precipitation particles passing through the horizontal strip of light in the event of precipitation. The OTT Parsivel 2 automatically categorized the precipitation type according to the National Weather Service (NWS) weather code based on the measured precipitation properties. Due to the intermittent nature of the precipitation, the OTT Parsivel 2 assigned multiple NWS precipitation codes during a single precipitation event (Table S1-1 column 'NWS Code'). We compared our manual observations with the NWS precipitation code assigned by the disdrometer, and we categorized all 70 observed precipitations into four different types. These four major precipitation types defined in this study included snow, hail/thunderstorm, long-lasted rain, and weak rain. More detailed methodology of precipitation categorization is discussed in SI S1.1.

IoT Air Quality Sensor Measurements
A cluster of Arduino-based IoT air quality sensors was developed to measure ambient air conditions at 75 our precipitation sampling location. This IoT cluster was deployed alongside the disdrometer and sampling gauge to complement this study. A DFRobot PM laser dust sensor measured PM with size ranges of < 1 µm (PM1.0), < 2.5 µm (PM2.5), and < 10 µm (PM10). Other ambient conditions, including T, barometric pressure, and humidity, were measured with a precision Bosch BME280 environmental sensor. We calibrated our sensors against a commercially available sensor (GlobalSat Inc., LS-113). Our sensors utilized Long Range and Wide Area Network (LoRaWAN) technology for data transmission. A LoRaWAN transceiver is connected to our sensors for wireless data transmission. This small IoT device operated with 915 MHz signal frequency, transmitting encrypted and signed packets of captured air quality data through a hosted LoRa network server to a Kibana visualization server. This data interface enabled in situ monitoring and processing of the data. The PM concentrations were later time-averaged for comparison with the precipitation properties and n INP in the precipitation samples.

Immersion Freezing Experiment
All immersion freezing experiments in this study were conducted using an offline instrument called West Texas -Cryogenic Refrigerator Applied to Freezing Test (WT-CRAFT) system (Hiranuma et al., 2019;Cory et al., 2019;Rodriguez et al., 2020). The WT-CRAFT system is a cold-stage technique in which the droplets are placed on an aluminum plate and cooled until they are frozen. A commercially available digital camera was used to record the droplet freezing events, and we visually evaluated the freezing Ts based on the shift in droplet brightness while freezing. If there was an uncertainty in determining the T at which a droplet was completely frozen, we used the ImageJ software for further image analysis of those droplets (see Table S4 in Hiranuma et al., 2019). This system was used to obtain T-resolved n INP in -25°C < T < 0°C. The lower T limit was -25°C to ensure measuring INPs with negligible artefacts (Hiranuma et al., 2019). Our system is susceptible to low INP detection, and the minimum INP detection limit of the WT-CRAFT system for this study was 0.002 L -1 air. To minimize any contamination during the IN measurement, the WT-CRAFT system was placed in a ventilated fume hood. For each experiment an aluminum plate surface was freshly coated with a thin layer of thermally conductive and IN-inert Vaseline to physically isolate individual droplets from the aluminum surface (otherwise, aluminum can act as a heterogeneous IN surface). A total of 70 suspension droplets of 3L volume 00 each were prepared for each run. The aluminum plate with the droplets on it was then placed inside a portable cryogenic refrigerator (Cryo-Porter). Freezing Ts were measured by the sensor taped on the aluminum surface with a resolution of 0.1°C, and the external keypad controller was used to control cooling rate (°C min -1 ). In this study, the freezing experiments were carried out at a cooling rate of 1°C min -1 . The validity of using this cooling rate and another test regarding time trial aspect are demonstrated in SI S2 (Figs. S2-1 and S2-2). The droplets were cooled until all 70 droplets were frozen before warming up the system to 5°C to be prepared for a subsequent experiment.
If all the droplets were frozen at T > -25°C, a HPLC-grade ultrapure water was used to prepare different serial dilutions for the precipitation samples. The diluted suspensions were made to compute the n INP down to -25°C. Some of our precipitation samples were diluted until the frozen fraction (the ratio of number of droplets 10 frozen to the total number of droplets) curve was conformed to the background curve (i.e., frozen fraction curve for the HPLC ultrapure water). At the end each WT-CRAFT experiment, the frozen fraction and ambient n INP were estimated as a function of T with an interval of 0.5°C. The IN measurements from the undiluted and diluted runs were merged by taking the lower n INP values for the overlapped T region.
The total systematic T uncertainty in our experiments for this study was ± 0.5°C (Hiranuma et al., 2019).

15
The uncertainty in our estimated n INP were estimated using the 95% confidence interval method described in Schiebel (2017). Background contamination tests for WT-CRAFT were carried out weekly to make sure negligible background freezing at -25°C. In this study, we consider the frozen fraction ≤ 0.05, accounting for less than 3% of pure water activation, as negligible background (Hiranuma et al., 2019). For these background tests, only HPLC grade ultrapure water was used for preparing the droplets.

IN Parameterization
Here we describe the parameterization used to estimate ambient n INP . Initially, we computed the C INP (T) https://doi.org/10.5194/acp-2020-863 Preprint. Discussion started: 14 September 2020 c Author(s) 2020. CC BY 4.0 License. value, which is the nucleus concentration in precipitation suspension (L −1 water) at a given T as described in Vali (1971). This C INP (T) value was calculated as a function of unfrozen fraction, f unfrozen (T) (i.e., the ratio of number of droplets unfrozen to the total number of droplets) as: in which, V d is the volume of the droplet (3 µL).
Next, we used the cloud water content (CWC) parameter in order to convert C INP (T) to n INP (T), INP in the unit volume of atmospheric air at standard T and pressure (STP) conditions, which is 273.15 K and 1013 35 mbar. We presumed CWC to be a constant of 0.4 g m −3 , covering the continental clouds in our study. Our assumption would be reasonable since Petters and Wright (2015) showed that the variation of n INP with CWC values for different cloud types in the atmosphere would typically be limited within a factor of two, and our n INP uncertainties could be larger than that. Thus, the effect of CWC on the n INP would be negligible.
The sample air volume (V air ) at the cloud level was calculated by converting the volume of the 40 precipitation sample collected (V l ) using the Eqn. (2) from Petters and Wright, 2015: where ρ w is a unit density of water (1 g ml -1 ). V air is in liters (L), whereas V l is given in ml. The multiplication 45 factor '1000' is used to convert the volume from cubic meter (m 3 ) of air to liter of air. The cumulative n INP per unit volume of sample air, described in the previous study DeMott et al. (2017), was then estimated as: where DF is a serial dilution factor (e.g., DF = 1 or 10 or 100 and so on). The overall goal of our metagenomics analysis was to identify known ice-nucleation-active bacterial and fungal species in feedlot dust and precipitation samples collected in the West Texas region. In this study, we have examined a heterogenous set of samples including feedlot samples, hail, long-lasted rain and a 24-hour dry 55 deposition sample. Next, we describe our microbiome analysis procedure in four different steps, including (1) DNA Extraction, (2) 16S rRNA Amplicon Diversity Sequencing, (3) Bioinformatics, and (4) Data Analysis. For DNA extraction, Genomic DNA was first extracted from all samples using PowerSoil DNA Isolation Kits (MoBio Laboratories, Inc., Carlsbad, CA, USA). Extraction proceeded following the manufacturer's protocol, with the following minor changes: solutions C1 and C6 were heated to 65°C and solution C6 was allowed to remain on the 60 filter membrane for at least one minute before centrifugation. Additionally, the C6 step was repeated. Library preparation for bacterial 16S DNA amplicon sequencing utilized primers for the V1-V3 hypervariable region of the 16S gene. These primers were constructed for the 16S amplicon using a combination of the 28F and Illumina i5 sequencing primer and the Illumina i7 sequencing primer with the 519R primer. Amplifications were performed in 25 μl reactions with Qiagen HotStar Taq master mix (Qiagen Inc, Valencia, CA, USA). Reactions were performed with 1 μl of each 5μM primer and the template DNA. Amplification was performed on an ABI Veriti thermocycler (Applied Biosytems, Carlsbad, CA, USA) under the following thermal profile: 95°C for 5 min, then 25 cycles of 94°C for 30 sec, 54°C for 40 sec, 72°C for 1 min, followed by one cycle of 72°C for 10 min and 4°C hold. An ethidium bromide stained gel was used to qualitatively determine the amount of the amplification product to add to the second amplification stage. Primers for the second PCR were designed based on the Illumina Nextera PCR 70 primers. The second stage amplification proceeded using the same cycling protocol as the first round, except it was amplified for only 10 cycles. SPRIselect beads (BeckmanCoulter, Indianapolis, IN, USA) were used at a 0.7 ratio to size-select the DNA amplicons from an equimolar pooled sample. Pooled samples were then quantified using a Quibit 2.0 fluorometer (Life Technologies) and loaded on an Illumina MiSeq (Illumina, Inc. San Diego, CA, USA) 2x300 flow cell at 10pM.
For bioinformatics, raw data were initially processed using a standard microbial diversity analysis pipeline (QIIME2-2020). Raw data was first checked for sequencing quality and chimeric sequences, before being parsed through a microbial diversity pipeline. During the cleanup stage; denoising of the raw data was performed using various techniques to remove short sequences, singleton sequences, and reads with poor quality scores. Next, 80 chimera detection software was used to filter out any potentially chimeric sequences. Finally, remaining highquality sequences were corrected base by base to check for sequencer miscalls. The diversity analysis pipeline clustered all sequences based on 97% similarity to yield operational taxonomic units (OTUs), before running a seed sequence from each OTU through a taxonomic database curated in-house by RTLGenomics. Finally, the taxonomy was assigned to each sequence using a classifier that was pretrained on GreenGenes database with 85 99% OTUs. The relative abundance of bacterial taxa within each sediment sample was determined by dividing each OTU by the total number of reads. Alpha diversity was carried out by taking phylogenetic distances into account and by looking at how diverse the phylogentic tree is for each sample. Next, beta diversities were analyzed using weighted (by bacterial abundance) or unweighted Unifrac distances calculated from a mid-point rooted tree. Multivariate differences in beta diversity were analyzed using Permutational Multivariate Analysis of Variance Using Distance Matrices function (ADONIS), which uses an ANOVA-like simulation to test for sampling location differences (McMurdie and Holmes 2013).

Ambient and Precipitation Properties
The time series summary of ambient and precipitation properties measured by our disdrometer as well as IoT cluster, respectively, is shown in Fig. 1. All the individual T and relative humidity data points shown in Fig. 1a and 1b correspond to the sampling start date of each precipitation event. A notable seasonal variation of atmospheric T was observed. Each data point in Fig. 1a show the average temperature measured over the sampling period of a given precipitation event. The highest average temperature measured during a 00 precipitation event was 34.9°C ± 12.2, which was in the summer of 2018 (i.e., 16 July; ID# 7; a long-lasted rain sample), while the lowest T was -6.5°C ± 6.7, measured during the winter of 2018 (i.e., 28 Dec; ID# 23; a snow sample). The annual mean T for Canyon, TX region measured at our sampling site was 17.7°C. The details of each precipitation event and its properties are shown in the SI Table S1-1 and S1-2. The diurnal cycles of ambient properties are not shown in Fig. 1a. Nevertheless, we typically observed suppression of T before precipitation events in our study. It has been understood that the T gradient plays a major role in the development and growth of the precipitation systems (Vaid and Liang 2015). The relative humidity shown in Fig. 1b was the averaged value for each precipitation sampling period. The highest and lowest relative humidity values measured were 70.7% ± 2.3 (on 12 March 2019; ID# 26; a weak rain sample) and 30.8% ± 0.7 (on 16 July 2018; ID# 7; a long-lasted rain sample). Figure 1c displays the time series of the cumulative number of detected precipitation particles in individual precipitation events and the overall mean number of detected particles (dashed line). In our study period, the precipitation events during September 2018 -January 2019 exhibited a substantial number of precipitation particles with a cumulative number of 2E+05 to 6.6E+05 per event. This high number of precipitation particles, greater than the overall mean cumulative number (i.e., 7.9E+04) was observed in conjunction with snow/hail-involving precipitation events during this period. Out of all the 42 15 samples, the highest number of precipitation particles detected were 6.6E+05 (on 5 Nov, 2018; ID# 19; a snow sample), while the lowest was 1.0E+04 (2 Sep, 2018; ID# 13; weak rain). Additional information is detailed in SI Table S1-2 and S1-3. There were other occasional snow/hail precipitation events in our study period, but the frequency of their occurrence was indeed high in Fall -Winter. Overall, a high average number of precipitation particles were detected for all the snow samples combined, 2E+05 ± 2E+02, followed by hail/thunderstorm 20 samples, 7.1E+04 ± 1.9E+04. On the other hand, the weak rain episode had lowest average number of precipitation particles, 1.8E+04 ± 5.4E+03 (SI Table S1-3). Figure 1d shows the average, maximum and minimum precipitation intensity (mm hr -1 ) measured during each precipitation event. Due to the intermittent nature of the precipitation, the intensity widely ranged from 0 to 150 mm hr -1 per event. The measured lower values of the average intensity were due to the influence of low intensities observed over a prolonged period of a given 25 precipitation event. The highest maximum intensity of 129.3 mm hr -1 was measured during a hail/thunderstorm event (ID# 40), while the lowest was 1.1 mm hr -1 during a snow event (ID# 23). The average intensity ± standard error for each precipitation category is shown in the SI Table S1-3. Hail/thunderstorm events have recorded the highest average precipitation intensity of 5.3 ± 7E-01 mm hr -1 , which was greater than the average intensity measured for the weak rain episodes, 1.5 ± 3.8E-01 mm hr -1 by a factor of 3 (SI Table  S1-3). https://doi.org/10.5194/acp-2020-863 Preprint. Discussion started: 14 September 2020 c Author(s) 2020. CC BY 4.0 License.
The variation of precipitation properties was further investigated by analyzing the size distribution of precipitation particles. Figure 2 shows the log-normal particle diameter distribution for each category of precipitation system. These size distributions were computed from the size-resolved precipitation particle measurements by the OTT Parsivel 2 disdrometer during each precipitation event. The size of precipitation particles was represented at the mid-value of the corresponding size bin. As shown in the Fig. 2a and 2b, both the snow and hail/thunderstorm samples had particles of diameter greater than 10 mm, and the maximum particle diameter was 17 mm. Although there are three episodes of long-lasted rain with a particle diameter greater than 14 mm (Fig. 2c), a clear trend of overall decrease in the hydrometeor size was seen for this category as well as the weak rain samples (Fig. 2d). Even though the number of samples in each precipitation category was different, 66.7% (4 samples) of the total snow samples (n=6) had precipitation particles of diameter ≥ 8.5 mm. Less compared to snow, but 55.6% (10 samples) and 46% (6 samples) of the hail/thunderstorms (n=18) and long-lasted rain (n=13) samples, respectively, had recorded precipitation particles of diameter ≥ 8.5 mm. In contrast, none of the weak rain samples (n=5) had hydrometeors of diameter ≥ 8.5 mm, and all weak rain 45 samples contained particles only smaller than 6.5 mm. Moreover, the mode precipitation particle diameter for the snow, hail/thunderstorm, and long-lasted rain samples was 0.44 mm, whereas it was 0.31 mm for the weak rain samples (see SI Table S1-3). This variation in mode diameter along with the results shown in Fig. 2 exhibited the shift in hydrometeor particle size distribution towards a higher diameter with an increased intensity of precipitation. Further discussion regarding the variation of n INP with the severity of precipitation was analyzed 50 and is followed in Section 3.3.

IoT Air Quality Sensor Results
An IoT air quality sensor-measured PM concentrations were also analyzed for each precipitation sampling period to understand the effect of wet deposition of PMs on INPs. Figure 3 shows the time series of average PM concentrations observed during each precipitation episode, overall mean PM values, and the hourly PM data. The overall mean ± standard error PM concentrations calculated from over one year of data were 3.9E+00 ± 9.2E-02 (PM 1.0 ), 4E+00 ± 4.5E-02 (PM 2.5 ), and 1E+01 ± 2.2E-01 (PM 10 ) µg m -3 . Although, there was an inconsistent variation of PM concentrations with precipitation type, we observed a substantial increase in all PM values for the period July -Aug 2018 and May 2019. In contrast, a decrease in all PM concentrations was observed during Sep 2018 -Mar 2019. This increase in PM values during summer and decrease during winter suggested the seasonal variation at the sampling site. In addition, the influence of PM values on n INP from each precipitation event was analyzed at -10°C, -15°C, -20°C, and -25°C. The Pearson correlation coefficients (R-value) at -10°C, -15°C, -20°C, and -25°C were statistically insignificant and negative for all PM types (SI Fig. S3). These results suggested no strong positive correlation between the PM and n INP for our sampling period. Moreover, we did not observe a clear sign of wet deposition during a given precipitation event (Stopelli et al., 2015), as there was no decrease in the original hourly PM concentrations during or after the precipitation. Overall, our PM analysis had suggested a local seasonal variation in PM concentrations, and no significant relation between PM and n INP values from our precipitation samples.  . This figure also displays other reference IN spectra, including our 24-hour dry deposition blank sample (collected from January 2 -3, 2019 at our sampling site) and IN spectra measured for dust suspension samples collected from the downwind side of a local feedlot (identity purposely concealed), where substantial and consistent dust emission historically persists (Whiteside et al., 2018). For the measured T range, n INP values from dry deposition blank sample were at least an order of magnitude lower than that from our precipitation samples. This finding corroborated our assumption of negligible contribution of dry deposition in our WT-CRAFT estimated n INP from precipitation samples. Interestingly, the feedlot IN spectra and most of our precipitation samples shown in Fig. 4 80 were greater than the previously derived precipitation IN upper limit, implying abundant IN active feedlot dusts, which might be involved in the precipitation formation and thereby our samples. It is noteworthy that adjacent feedlots (> 45,000 head capacity) are located within 33 miles of our sampling site. We observed approximately a two orders of magnitude increase in the upper limit of feedlot n INP compared to previous precipitation study at -15°C. For T ≥ -15°C, there was at least one sample from each precipitation category falling in the IN spectra region 85 of feedlot dust. Some of the hail/thunderstorm type samples had n INP values in the range of feedlot samples even for the entire T range of 0°C to -25°C (Fig. 4b). For example, at -5°C, n INP from our precipitation samples were in the range of 0.01 -0.11 L -1 of air in the atmosphere. These findings suggest the influence of local feedlot dust on precipitations and on increase of n INP upper limit for precipitation samples. Furthermore, for 0°C ≤ T ≤ -25°C, we found no precipitation sample with n INP values below the lower limit from the Petters and Wright (2015) study.

INP Results
Compared to all other precipitation types, hail/thunderstorm type had the highest average n INP ± standard error of 0.1 ± 0.01 and 118 ± 68.1 L -1 at -5°C and -25°C. In addition, the snow type had the highest average n INP of 0.4 ± 0.3, 0.8 ± 0.5, and 5.7 ± 2.5 L -1 at -10°C, -15°C, and -20°C (SI Table S3-1). The lowest n INP values were observed in both the long-lasted and weak rain samples at most of the temperatures. Interestingly, we observed an order of magnitude increase in the maximum n INP calculated at -5°C and -25°C in hail/thunderstorm type compared to long-lasted and weak rains. Such high values of nINP in snow and hail/thunderstorm samples suggested that the INPs impact the severity of a precipitation at least in the West Texas region. These feedlot dusts could reach cloud height and be involve in local aerosol-cloud-precipitation interactions, influencing the local hydrological cycle. Further discussion regarding the feedlot contribution in INPs 00 in our precipitation samples are provided in Section 3.4. We observed a reduced uncertainty in n INP from precipitation samples at T > -10°C. For example, a two order magnitude difference was estimated at -8°C in this study, which is lower than previously reported n INP uncertainty at the same temperature . Nonetheless, the discrepancy in n INP still remains at high Ts. Furthermore, the lower n INP values from this study were greater than the lower limit presented in Petters and Wright (2015). The upper and lower n INP limit 05 derived from this precipitation study could help in comparison studies of n INP at the cloud level to the observed ice-crystal concentrations. Overall, our findings imply that the local feedlot dust contribute to the regional INPs, with an increase in the high T n INP in our precipitation samples.

Seasonal variability and INP-precipitation relationship
The time series of cumulative n INP from precipitation samples at different Ts (i.e., -5°C, -10°C, -15°C, -20°C, and -25°C) is shown in Fig. 5. The overall average, cumulative n INP ± standard error is also presented in In order to elucidate this seasonal variation, we further subcategorized our sampling period into four different periods; i.e., May-August (May-Aug; which is a summer season at Canyon, TX), September-October (Sep-Oct), November-January (Nov-Jan; which is a winter season at Canyon, TX), and February-April (Feb-Apr), shown in Fig. 6a. Most of the high T (-5°C) n INP were observed during May -Aug, while there was a decrease in the following seasons, with no INPs at -5°C in the Feb -Apr period. The May -Aug season was dominated by hail/thunderstorms, whereas Feb -Apr had seen mostly long-lasted and weak rains. Likewise, significantly (p-value = 0.09; student's independent t-test) higher INPs were measured at -10°C during Nov -Jan than in Feb -Apr. These findings suggested a strong seasonal variation in INPs, specifically in the high T (≥ -15°C) INPs in Canyon, TX.
The variation of INPs among the precipitation types is shown in Fig. 6b. A statistically significant (p-30 value ≤ 0.01; student's independent t-test) increase in high T (-5°C) n INP was found in hail/thunderstorm samples compared to long-lasted rains. Additionally, we observed only one sample from weak rain type with n INP at -5°C, supporting the decrease of high T INPs in the less severe precipitation types. Similarly, the distribution of n INP at -25°C for weak rain type was shifted towards relatively lower values than compared with more severe precipitation types, such as hail/thunderstorm. For example, at -25°C, hail/thunderstorm type had a median nINP of 22.44 L -1 , which was greater than what was measured in weak rain (6.19 L -1 ) type. These results of increase in severity of precipitation with an increase in INPs were further corroborated by our findings from maximum intensity range based n INP analysis (shown in Fig. 6c). For this intensity -n INP analysis, we grouped all our precipitation samples into three different categories based on the observed maximum intensity (mm hr -1 ) in each precipitation event. A significant (p-value ≤ 0.01) increase in INPs at -5°C was found when the maximum 40 intensity was > 50 mm hr -1 compared to the range of 10 -50 mm hr -1 . The samples from this high intensity range (> 50 mm hr -1 ) were mostly coincided with the hail/thunderstorm precipitation types, supporting our previous findings of increase in severity of precipitation with INPs. It is also important to note that there was only one hail/thunderstorm sample which fell in the low intensity range (< 10 mm hr -1 ). Overall, we found a strong seasonal variation in INPs, especially in the high T (-5°C and -10°C) INPs from our yearlong precipitation study. Moreover, we observed an increase in the severity of precipitation with INPs, which highlights the importance of INPs in the development and growth of severe precipitation systems in the West Texas region.

Microbiome of Feedlot and Precipitation Samples
We successfully generated data on the bacterial microbiome of our precipitation and feedlot dust samples. Unfortunately, our attempt to extract fungal microbe was not successful due to the limitation in sample amount. Thus, we focus on bacterial discussions hereafter. In most cases, bacterial phyla were classified to the level of genus. The majority of bacteria in all samples belonged to phyla Proteobacteria and Bacteroidetes ( Fig. 7a and Table S4-1). In hailstorm samples, the main taxa of Proteobacteria were Massilia (a genus found in clinical samples, mammals, but also soil, rhizosphere and even aerosols), genera belonging to the order Sphingomonadales (bacteria with wide metabolic abilities), Caulobacterales (bacteria living in diverse 55 terrestrial and aquatic habitats, some are minor human pathogens), and Rhizobiales (nitrogen-fixing bacteria forming symbioses with the roots of legumes). Among the Bacteroidetes phylum, the genus Marinoscillum was relatively the most abundant. This genus is a recently described marine bacterium and it is interesting that it was found in hailstorm samples at percentages from 17.3% to 3.2% of the microbiome. Our results perhaps indicate some connection with storms or winds originating from the North Atlantic Ocean (back-trajectories 60 analyses done, but not shown). Other Bacteroidetes taxa with notable presence in hailstorm microbiome included Saprospirales and Chitinophagales orders with bacteria living on animals and in the gut of animals as expected.
The microbiome of one long-lasted rain sample shared members also found in hailstorms: the genus Massilia in significant numbers (11.3% of the microbiome), bacteria of the Proteobacterial orders Rhizobiales, Sphingomonadales and Burkholderiales; a significant percentage (8.5%) of the marine genus Marinoscillum and bacteria in order Sarospirales of phylum Bacteroidetes. Our results suggest that no known ice nucleation active species detected in precipitation microbiomes. The order Pseudomonadales, which includes most known ice nucleation active species, was found at the limit of detection.

70
Massilia was also relatively dominant in all four feedlot samples with percentages from 6.5% to 65.4% of the microbiome. Marinoscillum was also found in feedlot samples from 3% to 8.5% of the microbiome, perhaps indicating some connection of these genera either with the formation of precipitation or with their presence in aerosols during precipitation events. It is noteworthy to mention that neither of these two genera (Massilia and Marinoscillum) was detected in the background deposition blank sample and it is not known 75 whether they have any ice nucleation activity. Other bacterial taxa with significant presence in feedlot samples included members of orders Caulobacterales and Burkholderiales.
Alpha diversity analysis (Shannon's Faith PD index of diversity) indicated that feedlot and hailstorm samples had a lower bacterial diversity than the long-lasting rain sample (Fig. 7b). We sought to identify a possible 80 connection between the feedlot microbiome and the microbiome of hail and rain. Beta diversity analysis compared the microbiome diversity distance of feedlot samples between themselves, as well as the microbiome diversity distance of the background deposition, hailstorm and long-lasted rain samples to feedlot samples (Fig.  7c). In all comparisons, the distance was at least 0.70, a high value not indicating a "cause and effect" connection between the feedlot and precipitation microbiomes. However, our detailed phylogenetic analysis showed evidence of such a connection by identifying common bacterial taxa in feedlot and precipitation microbiomes. Their absence from the background deposition blank sample may indicate local aerosol-cloud interactions leading to precipitation events, but it is not known if this is a result of any bacterial traits such as ice nucleation activity.

Conclusion
We have successfully estimated n INP (per liter of air) in the immersion freezing mode from different precipitation samples collected in Canyon, TX, USA. IN spectra were derived for MPC T range (0°C to -25°C) from four different precipitation types (snows, thunder/hail storms, long-lasted rains, and weak rains) using a coldstage instrument (WT-CRAFT). Our disdrometer measurements showed a clear variation in the precipitation properties among the four different categories of precipitation samples. Severe precipitations, such as hail/thunderstorms, had the highest rainfall intensity (mm hr -1 ) and the number of precipitation particles were highest in the snow samples. We also found an increased number of large hydrometeors (> 8.5 mm in diameter) in both the snow and hail/thunderstorm samples. In contrast, there were no precipitation particles > 6.5 mm in 00 diameter observed in the weak rain samples. Our PM concentration measurements showed no strong correlation with the measured INPs from precipitation samples. The IN spectra from each precipitation category in this study were compared with the nucleus spectra from previous precipitation based INP study . Previously derived IN spectra from local feedlot dust samples (Whiteside et al., 2018) was also used for comparing n INP from precipitation samples. We have found that n INP values from our precipitation samples were greater than 05 previously derived IN upper limits from precipitations. Especially, the high T (≥ -15°C) INPs in some of our precipitation samples were in the same order of magnitude as of local feedlot dust samples. These findings suggested the importance of local feedlot dusts as INP sources. Moreover, we have observed a strong seasonal variation in n INP in this precipitation based INP study. The May -Aug period had seen the most INPs at -5°C, while none during Feb -Apr season. It is important to note that, hail/thunderstorms were predominantly observed in the May -Aug season. A statistically significant increase in high T (-5°C) INPs was observed in hail/thunderstorms compared to long-lasted rains. Except in one case, we observed no weak rain samples with INPs at -5°C. These findings suggested an increase in high T (-5°C and -10°C) INPs in severe precipitation systems like hail/thunderstorm and snow in the West Texas region. These results were further supported by our findings of increased high T INPs when the rainfall intensity was > 50 mm hr -1 . Overall, our results showed that the INPs impact the severity of precipitation systems observed in Texas Panhandle, which represents the importance of more precipitation based INP studies in the future. We also identified the similarity in bacterial microbiome between our hailstorm and local feedlot dust samples, nevertheless, it is not known whether these microbiomes are IN active. Regardless, we did not find the previously known bacterial INPs, such as Pseudomonas and Xanthomonas (Morris et al., 2004) in both the precipitation and feedlot samples. Our preliminary analysis showed 20 that organic component was predominant in our precipitation residuals (>70%, not shown in this study), which is similar to the composition of local animal feeding dust (Hiranuma et al., 2011). This similarity might explain the observed increase in n INP from precipitation samples at high Ts (≥ -15°C). Therefore, it is worthwhile to characterize the local feedlot dust, as they can be the sources of INPs and can impact the local hydrological cycle.
The relationship between pollen and the PM concentrations has not been analyzed in this study. Collecting long-term observational pollen and other biogenic aerosol particles data for multiple years may add important knowledge regarding the role of local bioaerosols on the precipitation INPs. As for more future studies, INPs derived from precipitation samples collected over multiple years would give comprehensive insight into their impact on local precipitation systems. This work highlights this need for more precipitation