Ice‐Nucleating Particle Concentrations and Sources in Rainwater Over the Third Pole, Tibetan Plateau

The ice‐nucleating particles (INPs) modulate the microphysics and radiative properties of clouds. However, less is known concerning their abundance and sources in the most pristine and climatic sensitive regions, such as the Tibetan Plateau (TP). Here, to our best knowledge, we conduct the first investigation on INPs in rainwater collected in the TP region under mixed‐phase cloud conditions. The value of INP concentrations varies from 0.002 to 0.675 L−1 air over the temperature range from −7.1 to −27.5°C. This is within the INP spectra derived from precipitation under worldwide geophysical conditions and is also comparable with INP concentrations in the Arctic regions. The heat‐sensitive INPs account for 57% ± 30% of the observed INPs at −20°C and become increasingly important at higher temperatures, indicating biological particles as major contributors to INPs at temperatures above −20°C over the TP, especially on the day with additional input of biogenic materials carried by dust particles. Chemical analysis demonstrates that the rainwater components are mixture of dust particles, marine aerosol, and anthropogenic pollutants. Combined with the backward trajectory analysis, we show that dust particles transported from the surrounding deserts and originated from ground surface of TP may contribute to the heat‐resistant INPs at temperatures below −20°C.

As an integral part of Earth's atmosphere, clouds affect the energy balance of the Earth by absorbing or reflecting the solar and terrestrial radiation (Lohmann et al., 2016). The importance of cloud properties (Duan & Wu, 2006;Hua et al., 2018;Yang et al., 2012) and aerosol-cloud feedback processes (Zhao et al., 2019) in affecting the climatic warming over the TP has been pointed out. Aircraft observations indicated the majority of the detected clouds in summer were in mixed-phase state and accompanied by active ice processes (Chang et al., 2019). Yan et al. (2020) also found a maximum occurrence of mixed-phase clouds during the summer over the TP using satellite remote sensing techniques. Although many efforts have been made to investigate the properties and climatic effects of clouds (Kurosaki & Kimura, 2002;Qiu et al., 2019;Zhao et al., , 2017. Little is known concerning the microphysics of mixed-phase clouds (Qiu et al., 2019), resulting in large discrepancies in cloud simulation over the TP (Gao et al., 2016(Gao et al., , 2018. Primary ice formation in clouds can be initiated by atmospheric ice-nucleating particles (INPs) through heterogeneous ice nucleation (Lohmann et al., 2016;Pruppacher and Klett, 1997). Four pathways have been proposed for the heterogeneous ice nucleation in mixed-phase clouds, and immersion freezing has been widely recognized as the most important ice nucleation mechanism (Murray et al., 2012). INPs play a key role in affecting the lifetime and radiative properties of clouds (DeMott et al., 2010). However, compared to the Arctic or Antarctic areas, INPs get much less attention over the TP. An earlier balloon-borne observation was conducted by Tobo et al. (2007) at Lhasa. They concluded particles larger than 3.6 μm to be potential INPs and responsible for the cirrus formation. To our best knowledge, no direct INP measurement under mixed-phase clouds conditions was so far carried out in this region.
Atmospheric aerosols over the TP originate from multi-sources (Huang et al., 2006;Li et al., 2016;Liu et al., 2015). The dust particles from the Taklamakan desert (Huang et al., 2007) and the anthropogenic pollutants, such as black carbon (BC) from South Asia and north-western China (Li et al., 2016), influence the aerosol concentrations and chemical compositions in this region. In addition, TP has diverse natural underlying surfaces including glaciers, lakes and grassland (Kang et al., 2010), which could serve as sources of bioaerosols. As been identified by many laboratory and field studies, dust, biological particles, and BC constitute important INPs that can nucleate ice under different conditions (Hoose & Möhler, 2012;Kanji et al., 2017;Murray et al., 2012).
Up to now, the concentrations, sources, and the ice nucleation efficiency of INPs in pristine areas, such as TP, remain an ongoing debate in view of their relevance for Earth climate and their enhanced susceptibility to climate warming. In this study, INPs in rainwater at temperatures relevant to mixed-phase clouds were detected for the first time in Nam Co, a representative central site over the TP. The possible sources of INPs were furtherly identified by combining the chemical composition and source apportionment analysis.

Sampling Site and Rainwater Collection
The Nam Co Station for Multisphere Observation and Research (30.7°N, 90.0°E, 4,730 m above sea level) ( Figure S1) is located in the central part of the TP. The average aerosol optical depth (AOD) in Nam Co was found to be 0.05 at the wavelength of 500 nm (Cong et al., 2009) and comparable to that of Arctic (with a monthly average of 0.08) (Pokharel et al., 2019;Stone et al., 2014). Therefore, Nam Co represents a clean continental background site. The geophysical characteristics in the Nam Co region include mountains, glaciers, lakes, rivers, and grassland, which are representative of the main geophysical features over the TP.
A total of 34 rainwater samples were collected at Nam Co Station from January to October in 2018, covering 58.7% of the rainfall events during the sampling period. A rainfall event was defined when the amount of precipitation over 0.2 mm. The detailed sampling information was given in Table S1. Samples were collected by disposable Whirl-Pak bags (5L, Nasco, Ft., Wilkinson, Wis.) during each rain event. After the collection, samples were immediately transferred into polycarbonate bottles (Thermo Fisher Scientific, MA, USA) and kept frozen at −20°C until analysis. The meteorological parameters including air pressure, relative humidity (RH), temperature, wind speed, and direction were recorded by an automatic weather station, and discussed in Section 3.1.

Ice Nucleation Experiments
A cold-stage based instrument, the Peking University Ice Nucleation Array (PKU-INA) was applied for measuring the ice nucleation activity of rainwater samples following the protocol described in . Briefly, for each experiment, 90 droplets from each sample were pipetted onto a hydrophobic glass slide located on the cold stage and separated by a spacer with 90 compartments. The top of the spacer was sealed by a cover glass to avoid the Wegener-Bergeron-Findeisen process. The cold stage was cooled down to −35°C at a cooling rate of 1°C min −1 and monitored by a CCD (charge-coupled device) camera every 6 s. The recorded images were analyzed by a MATLAB program to identify the freezing of droplets based on the change of the gray values upon phase transition from liquid water to ice. Together with the recorded temperatures, the number of frozen droplets at a temperature can be determined.
The frozen fraction (ƒ ice ) is defined as Equation 1: N frozen is the number of frozen droplets at a certain temperature and N t is the total number of droplets (90 in this study). In the following data reduction and analysis, a time independent freezing of the supercooled droplets was assumed.
where V droplet is the volume of one droplet (1 μL in this study).
Conversion of INP per water volume to INP per air volume (N INP_air ) is achieved by assuming a cloud water content (CWC) of 0.15 g m −3 averaged over the CWC (ranged from 0.03 to 0.25 g m −3 ) detected in an aircraft measurement over the TP (Chang et al., 2019). This value was consistent with that estimated in precipitating convective clouds using multiyear satellite observation (<0.2 g m −3 ) (Chen et al., 2020). CWC is defined as cloud droplets of 1pL disperse in 1 m 3 of air weigh about 0.15 g, and the corresponding volume of cloud water per volume of air (F cloud _ air ) was 1.5 × 10 −7 m 3 water m −3 air (Gong et al., 2020 Confidence interval for N INP_water was estimated following the methods of Barker (2002) and O'Sullivan et al. (2018). Other uncertainties such as the impacts of dissolved solutes and chemical aging on the INP concentrations of rainwater were pointed out by , which were likely to contribute less than one order of magnitude in INP concentration. , and Cl − ) and water-soluble organic carbon (WSOC) in rainwater were measured by an ion chromatograph (DIONEX, ICS-2500/2000) ) and a TOC analyzer (SHIMADZU, TOC-L CPH CN200), respectively. The inductively coupled plasma-mass spectrometry (Bruker, aurora M90) was used to determine the metal element concentrations (Na, Mg, Al, K, Ca, Cr, Mn, Fe, Ni, Cu, Zn, Pb, and Ba). The mass concentrations of BC in rainwater were measured by a Single Particle Soot Photometer (SP2, Droplet Measurement Technologies, Inc., Boulder, Colorado) (Kaspari et al., 2011;Stephens et al., 2003). The BC experiment was conducted at Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences. The schematic of experimental set-up to measure the BC in rainwater sample by SP2 was depicted in Figure S2. Briefly, each rainwater sample was nebulized by a nebulizer to form jet stream (CETAC, U-5000AT, WA, USA). The jet stream was then introduced into an interaction region in SP2, which was created by the jet stream containing BC particles and the intracavity beam emitted from Nd:YAG laser. SP2 detected incandescence and scattering signals of BC-containing particles induced by laser. The mass concentration of BC is proportional to the peak intensity of incandesce. The concentrations of chemical components in rainwater, including WSOC, metal elements, and BC, are shown in Figure 1.

Chemical Analysis and Source Apportionment
Positive Matrix Factorization Model developed by Environmental Protection Agency (EPA-PMF)  was performed to identify the sources of chemical components in rainwater on the basis of the measured chemical components. The working principle and uncertainties of this model were detailed in Text S1. PMF has been widely used to identify the potential source categories and source contributions of particulate matter, volatile organic compounds, and other pollutants in water or atmosphere (Amato & Hopke, 2012;Daellenbach et al., 2020;Liu et al., 2019;Zhang et al., 2012).
In this model, information on chemical components (10-20 species) in each rainwater sample is required in the input files and the contributions of different aerosol types would be output (Pancras et al., 2013;Wang et al., 2020). In the present study, the concentrations of 13 elements, 4 water-soluble ions, WSOC, and BC (shown in Figure 1) were used as input parameters. In addition, information on source tracers is required for manual source identification. For example, the crustal elements including Al, Si, Ca, and Fe are markers for dust particles (Amato et al., 2009;Huang et al., 2014); Na + , K + , and Cl − are tracers to identify marine aerosol (Oduber et al., 2021). The present source profiles resolved from the PMF model are shown in Figure S3 and detailed in Text S2.
Overall, four aerosol types were identified by PMF model and on average contributed the following percentage of chemical compounds in rainwater: dust particles (14%), marine and salt-lake aerosols (27%), long-range transport anthropogenic pollutants (24%), and biomass burning aerosols (34%), indicating the impacts from multi-aerosol sources on chemical compounds in rainwater over the TP. The contributions of the four identified aerosol types are displayed in Figure S4.

Backward Trajectories and Geographic Information System Analysis
Ten-day backward trajectories were calculated using the NOAA HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) model (Rolph et al., 2017;Stein et al., 2016), with one trajectory related to each rainwater sample. Trajectories with 1-h time resolution ended at an altitude of 500 and 1,000 m above the ground level (AGL) were calculated. The 1,000-m altitude was chosen because precipitations usually start to form at the high altitude, as shown in   Figure S5, which fit well with the 1,000-m trajectories. As the spatial uncertainties of the backward trajectories grow rapidly over time, the backward trajectories calaculated at 1,000-m in the first 5, 7, and 10 days were shown in different colors in Figure 3. The trajectories were then coupled by land cover dataset obtained from Geographic Information System (GIS) to examine the land cover over which air mass had passed before reaching the sampling site. The details of the land cover analysis methodology are given in van Pinxteren et al. (2010). During the sampling period, the land cover over which air masses had been passed was categorized into five types: water/ice, natural vegetation, agricultural area, urban area, and bare areas (desert areas and rocky bare areas) (van Pinxteren et al., 2010).

Overview of Meteorology
The time series of the meteorological parameters during the entire sampling period are shown in Figure 2. Over the TP, there are two distinguished seasons, that is, dry season (October-April) with lower temperature and RH and a humid monsoon season (June-September) with higher temperature and RH (Figure 2a). On average, the wind speed was 3.4 ± 1.2 m s −1 (mean ± 1 standard deviation), with southernly prevailing wind (see Figures 2b and S6). Most of precipitations fell in monsoon season under the strong influence of Indian summer monsoon after June (Figure 2c). In the dry season, westerlies dominated the large-scale air circulation with less precipitation (Li et al., 2007;Xu et al., 2008) (Figure 2c).
The backward trajectories (5, 7, and 10 days back) corresponding to the 34 rainwater samples are depicted in Figure 3. Approximately 82% air masses after June 14 originated from Indian Ocean, passed over India and Bangladesh, and reached the inland of TP (marked by blue). The remaining six trajectories (marked by red) came from the west or the north. The resulting backward trajectories corresponded well to the atmospheric circulation in TP, that is, air masses were impacted by Indian monsoon and westerlies in humid and dry season, respectively. Therefore, the rainwater samples were categorized into two groups, that is, monsoon (from June 15-September 22) and dry season samples (January 3, April 1, June 14, and October 3, 4, and 5), based on the meteorological conditions and backward trajectory analysis. In dry season, the rainfall was limited; thus, only six samples were presented here. Notably, the sample collected on April 1 was impacted by the air masses from the north, passing over the Taklimakan desert.

INP Concentrations
The    (Figure 4). Exceptions were few measured N INP_air below the lower limitation of the spectra ( Figure S8), but deviations were within one order of magnitude. This is due to the lower CWC value reported in TP resulted from high elevation than those obtained in thicker clouds (0.2 and 0.8 g m −3 , Rangno & Hobbs, 2005). If assuming a CWC value of 0.4 g m −3 , the same as that used in , the resulting N INP_air ( Figure S9) were within the range of the spectra summarized in .
The N INP_air in Nam Co spanned three orders of magnitudes over the determined temperature ranges, indicating a large variation of the INP concentrations observed in different days. Such wide freezing temperature of INPs can be attributed to INPs originated from complex aerosol sources, as verified by source apportionment of rainwater components (Figure S4). No significant differences of N INP were observed in dry and monsoon seasons (Figure 4), exceptions were one dry season sample (April 1) and three monsoon samples (August 17, 30, and 31) with evidently higher N INP concentrations (>0.2 L −1 air) and onset temperatures (>−10°C) (Figure S8). Already at a first glance, the shape of ice nucleation curves, that is, bumps appeared in the curves (see Appendix in Welti et al. (2018) for more details) in temperatures above −20°C and the higher onset temperatures of these four samples hinted the INPs from biological sources ( Figure S9). Details about the origins of INPs for these samples will be furtherly explained in the next section.
The comparison of the observed N INP_air at −20°C (N INP_air-20 ) with those reported in Arctic region is given in Table 1. The maximum discrepancy between N INP_air-20 over the TP and those in Arctic was only one order of magnitude (Creamean et al., 2018;Hartmann et al., 2020;Irish et al., 2019;Mason et al., 2016;Prenni et al., 2009;Si et al., 2019;Tobo et al., 2019), indicating comparable N INP_air-20 in these two areas. As indicated by chemical analysis, the rainwater components came from both natural (mainly marine and dust, Figure S4) and anthropogenic sources CHEN ET AL.   (biomass burning and anthropogenic pollutants, Figure S4). Resemble to TP, the previous field studies carried out in Arctic region (Table 1)

Contribution of Biological INPs
INPs being sensitive to heat are inferred as protein-based biological INPs (Christner, et al., 2008a). Thus, to quantify the concentrations of biological INPs in rainwater, the experiments with samples being heated to 95°C for 10 min were conducted (refer to Joly et al., 2014). Figure 5a depicts the reduction ratio of N INP_air at different temperatures (−15, −18, −20, and −22) after heating treatment. Here, the reduction ratio means the difference in N INP_air before and after heating treatment divided by the original N INP_air . On average, the reduction of N INP_-15 due to heat was 84 ± 17% (mean ± standard deviation), indicating a large fraction of the observed INPs at −15°C were in biogenic origin. The reduction ratio down to 57 ± 30% (mean ± standard deviation) at −20°C, and as expected, the lower contribution from biological INPs with decreasing freezing temperatures. These results indicate the prevalence of biological INPs in the rainwater and these INPs would become more important in higher temperatures (Christner, et al., 2008a;Christner, et al., 2008b;Gong et al., 2020;Hill et al., 2014;Joly et al., 2014;Stopelli et al., 2014).
Moreover, the reduction ratio kept constant at temperatures below −20°C, for example, a value of 57% ± 30% (mean ± standard deviation) and 58 ± 22% (mean ± standard deviation) was observed at −20 and −22°C, respectively. This can also be seen in Figure S10 (-18) ). We note that heating treatment leads to a significant reduction in INPs (∼86%) at −18°C for the above-mentioned four samples (April 1 and August 17, 30, and 31) with high INP concentrations above average (Figure 5b), implying the bioaerosols elevated N INP_air in these samples to extremely high level.
One interesting thing is that the rainwater collected on April 1 showed obvious signs from dust particles. The backward trajectory of this day passed over the northern Taklimakan desert (Figure 3)   times higher than average, were detected accordingly (Figure 1b). The strong influences of dust can also be seen from source apportionment ( Figure S4). Dust particles can act as INPs at temperatures below −20°C (Hill et al., 2016;Kanji et al., 2017;Steinke et al., 2016), or become ice active in higher temperature by carrying ice-active biogenic macromolecules (Conen et al., 2011;O'Sullivan et al., 2016). Feldspar can act as INPs at higher temperatures (>−10°C) as well (Atkinson et al., 2013). While the heat-resistant feldspar cannot be the heat-sensitive INPs species observed in the present study in temperature above −20°C. Therefore, biogenic materials carried by dust particles (Maki et al., 2019;Puspitasari et al., 2016) bring additional heat-sensitive biological INPs, as a result, lead to the improvement of ice nucleating activities of dust particles and a higher INP concentration on April 1.
Backward trajectories and GIS analysis showed that the air masses averagely spent considerable time over the natural vegetation (54.3%) and agricultural areas (32.2%), but less time over water/ice (10.8%), bare (2.6%), and urban (0.1%) areas during the sampling period ( Figure S11). The agricultural soil (Conen et al., 2011;O'Sullivan et al., 2014) was found to contribute to the atmospheric biological INPs. So, the long residence time of air masses over natural vegetation and agricultural areas, indicating such underlying surfaces can priorly serve as sources of biological INPs. While the water/ice surface may not contribute to the INPs significantly, due to limited residence time and low ice nucleating efficiency of INPs when air mass passed over the open water and ice (Gong et al., 2020;Irish et al., 2019;Si et al., 2019). The source apportionment based on chemical compositions of rainwater on July 29 and August 26 showed strong impacts from marine and salt-lake sources ( Figure S5), but these days were not associated with high biological INPs ( Figure 5). Thus, the observed biological INPs were likely from continental source other than the open water sources (marine and salt-lake).

INPs from Abiotic Sources
As shown in Figure 4, the HR-N INP_air became more important with the decreasing temperature. The contribution of HR-N INP_air to total N INP_air reached up to 40% at temperatures below −20°C. The chemical composition analysis showed that almost all the detected samples consisted of crustal elements (Ca, Mg, Al, and Fe) ( Figure S2), indicating the presence of dust materials in rainwater. The effects of dust particles on rain samples can also be confirmed by source apportionment ( Figure S4). The observed heat-resistant INPs in this study were found to dominate the ice nucleation in rainwater at temperatures below −20°C (Figures 4 and S8), being consistent with the typical ice active temperatures for dust particles (Hoose & Möhler, 2012;Kanji et al., 2017). The surrounding deserts and the ground surface over the TP can be the sources of dust particles Pokharel et al., 2019), contributing to the observed HR-N INP_air in rainwater.
BC, another heat-resistant material, was detected in the rainwater samples. Its mean concentration in the detected rainwater was 1.07 ± 1.05 ng mL −1 . BC may come from South Asia with intensive biomass burning and fossil fuel combustion via the long-range transport (Li et al., 2016). This was evidenced by the source apportionment ( Figure S9). Assuming a CWC of 0.15 g m −3 , the mean value of BC concentration in per volume of air was 0.16 ± 0.16 ng m −3 , which was remarkably lower than the mean values reported in two of our previous studies (3,200 and 7,700 ng m −3 ) in urban areas in China Gong et al., 2016). No obvious linear correlation between HR-N INP_air and BC mass concentration was found CHEN ET AL. (R 2 = 0.087), consistent with the INP measurement in Beijing showing no correlation between BC mass concentration and N INP_air at temperatures above −25°C .In addition, most of the previous studies showed inefficient ice nucleating activities of BC over the temperature ranges relevant for mixed-phase clouds (Chou et al., 2013;Kanji et al., 2020;Mahrt et al., 2018;Schill et al., 2020;Vergara-Temprado et al., 2018). Combined with these results, we suggest the BC generated from biomass burning and fossil fuel combustion may play a minor role in the INPs in rainwater over the TP due to low concentration and poor ice nucleating activities.
As shown in Text S2, the inorganic species (NO 3 − , SO 4 2− , and NH 4 + ) and metallic elements (Cu, Zn, Ni, and Pb) were important constitutes in factor 3, which was considered as the indicator of long-range transport anthropogenic pollutants. Inorganic species cannot serve as INPs in the determined conditions. Similar to BC, metallic elements were not likely to serve as INPs, as (1) all these elements were in an extremely low concentration (1-10 ppb); (2) no obvious linear correlation between these elements and HR-N INP_air were observed (Cu: R 2 = 0.01; Zn: R 2 = 0.03; Ni: R 2 = 0.02; Pb: R 2 = 0.06). Thus, we exclude metals generated from anthropogenic pollution as INP contributors.
Ash particles can be generated through coal combustion and biomass burning, few studies have identified their ice nucleating activities in immersion freezing (Garimella et al., 2016;Grawe et al., 2016Grawe et al., , 2018Umo et al., 2015). In the present study, we cannot quantify the concentration of ash particles through PMF model due to the similar chemical tracer of ash particles to those of dust particles (i.e., both include several common mineral elements, such as Si, Na, Ca, and Fe).

Conclusions
The abundance and the potential sources of atmospheric INPs at central TP (i.e., Nam Co) are quantified and identified by integrating detailed chemical composition analysis with INP measurements of rainwater samples. The observed INP number concentration varied from 0.002 to 0.675 L −1 air over the temperature range from −7.1 to −27.5°C, being within the range of INP spectra derived by  and were comparable to those measured in the Arctic region.
Heating experiments demonstrate that biological INPs on average accounted for 57% ± 30% (mean ± standard deviation) of the total INPs at −20°C and become even more important at higher temperature regime. The continental underlying surfaces over TP, such as natural vegetation and agricultural areas likely serve as sources of these biological INPs. Chemical analysis showed the rainwater components may be influenced by mixed sources including dust particles, marine aerosols, and anthropogenic pollutants. The components associating with dust particles either transported from surrounding deserts or originated from the ground surface over the TP can contribute to the heat-resistant INPs at temperatures below −20°C. BC in rainwater, with the mean concentration of 1.07 ± 1.05 ng mL −1 , may come from South Asia with intensive biomass burning and anthropogenic activities via long-range transport. Due to much lower content and inefficient ice nucleating activities in the determined temperature range, BC may not be responsible for the observed heat-resistant INPs.
Our work indicates rainwater components collected in TP are impacted by multi-aerosol sources. As a result, the variations and properties of INPs are modulated by both natural and anthropogenic aerosols. In this study, the INPs were quantified indirectly from rainwater, and most of samples were collected in monsoon seasons. Given lack of study over the TP, the limited works are unable to provide a full picture of INPs over the third pole regions. Thus, we suggest more field measurements are needed in the future by coupling the INP and high-resolution aerosol chemistry measurements. In addition, the long-term INP measurements over the year are also needed, for example, in dry seasons with relatively more dust events, to better elucidate the seasonal characteristics of INPs in the third pole region.

Data Availability Statement
The data that support the collusion of this work are at this site (