Changes in China's Snow Droughts Characteristics From 1993 to 2019

Snowpacks are natural water reservoirs providing a considerable amount of water for humans and ecosystems. However, current global snow products (e.g., ESA GlobSnow v3.0), lack high spatial resolution and regional calibrations necessary to capture the high heterogeneity of snow water equivalents (SWEs) in complex Asian mountainous terrains. Therefore, our understanding of snow drought characteristics in China remains limited. Herein, we used an improved SWE product calibrated specifically for China to explore the characteristics of snow droughts, delineated by a standardized SWE index (SWEI) between 1993 and 2019. Our analysis was focused over three main snow‐covered regions of China: Qinghai–Tibet Plateau (QTP), northern Xinjiang, Northeast China. Especially during the period from 1993 to 2010, we found that the SWEI increased significantly at rates of 0.022/yr (Northeast China), 0.017/yr (northern Xinjiang), and 0.011/yr (QTP) (p < 0.01, Mann‐Kendall trend test). Increased SWEI contributed to decreasing snow drought events across China, with an obvious short‐term characteristic, whilst area proportion of the identified 1‐month snow droughts was above 46.5% across three regions. Furthermore, we found that the occurrence of snow droughts was likely mediated by large‐scale atmospheric circulation, since increased water vapor transport caused a significant vapor flux convergence in cold seasons over three regions, especially in northern Xinjiang and Northeast China.

Mankin, 2021).Hence, understanding the characteristics of snow droughts is an important adaptation and mitigation step for managing regional climate change.
Previous studies have extensively explored snow droughts features at a basin scale, such as extent, intensity, and severity.For example, based on point-in-time measurements of peak SWE, Dierauer et al. (2019) quantified the historical frequency, severity, and risk of snow droughts in the western United States and southwestern Canada between 1951 and 2013.Furthermore, using a Standardized Snow Melt and Rain Index, Staudinger et al. (2014) effectively accounted for the influence of snow droughts in seven Swiss catchments.However, current snowpack metrics still face important bottlenecks to accurately estimate snow droughts (Dierauer et al., 2019;Gottlieb & Mankin, 2021;Staudinger et al., 2014).For example, point-in-time measurements in snowpacks, such as 1 April SWE (considered closest to the maximum accumulation of snowpack) or peak SWE, have been widely used to analyze the entire process of snow droughts around the world, retrospectively.However, these measurements are insufficient for accurately describing the different temporal characteristics of drought processes, such as the early signs of snow droughts and their duration.In particular, considering that topographic heterogeneity leads to differences in the peak timing of SWE in different regions (Stein et al., 2014), such point-in-time snow metrics limit the ability to characterize the spatiotemporal changes in snow droughts at high altitudes, such as the Qinghai-Tibet Plateau (QTP) or Tianshan Mountains.
The current availability of data sets has allowed for the reevaluation of drought characteristics, however, there is still considerable disagreement within the high-altitude regions of China (Jiang et al., 2022).In situ measurements of SWE at sparse observation sites in high-altitude regions limit the tracking of long-term trends and variability in snow droughts.Satellite remote sensing and reanalysis products provide spatially continuous estimates for SWE at different spatial scales.However, the coarse resolution, saturation interference from vegetation, and physical snow properties all make it difficult to accurately measure snowpacks (Barnett et al., 2005;H. Zhang et al., 2022).For example, a major omission in the European Space Agency Global Snow Monitoring for Climate Research (ESA GlobSnow) product is the SWE estimate of mountainous areas, owing to inversion flaws in areas of complex terrain (Bormann et al., 2018) (Figure S1 in Supporting Information S1).Furthermore, comparisons between different SWE gridded products have reported an evident uncertainty regarding the magnitude of snowpacks on a hemispheric ensemble scale (Gottlieb & Mankin, 2021).
Due to the insufficiency of time-point SWE in terms of characterizing the entire process of snow drought events, the Standardized SWE Index (SWEI) (Huning & AghaKouchak, 2020;Staudinger et al., 2014), similar to the Standard Precipitation Index (SPI), was developed and used to address the above limitations.Based on the continuous rolling time of SWE, the SWEI can be used to identify different characteristics of snow drought monitoring, such as extent, intensity, duration, and severity.For example, Huning and AghaKouchak (2020) previously described the duration and intensity of global snow droughts between 1980 and 2018.However, owing to complex topographical heterogeneity, current research is primarily based on publicly available data sets, such as Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) (Reichle et al., 2017), ESA GlobSnow v3.0 (Pulliainen, 2006), and Japanese 55-year Reanalysis (Kobayashi et al., 2015).These are all globally calibrated and thus cannot account for important regional features, failing to capture accurate changes in Asian mountainous areas, especially in the QTP (Huning & AghaKouchak, 2020;Jiang et al., 2014;Y. Li et al., 2020).Consequently, these research gaps still limit our understanding of snow-drought characteristics in China.Further examination using high-resolution and reliable SWE data sets on a Chinese scale is necessary to provide a better understanding of the risks posed by current and future snow droughts.
This study aims to quantify the characteristics of snow droughts (SWEI ≤ −0.5) in China between 1993 and 2019; specifically, elucidating (a) the intensity, duration, and spatial extent of droughts, (b) the potential influences of temperature and precipitation on snow droughts, and (c) the effects of large-scale water vapor flux.First, we used the SWEI to delineate changes in the intensity, duration, and spatial extent of droughts.Then, based on sensitivity analysis, we distinguished the spatial distribution of snow droughts category in grid cells, and the influence of temperature and precipitation on snow droughts over three snow-cover regions.Finally, we explore the possible association of snow droughts topographical heterogeneity and large-scale water vapor flux.Snow droughts identification on the regional scale is critical for water risk management and adaptation to changing snow conditions in China.

Data
We collected high-quality daily SWE estimates at a spatial resolution of 25 km as the basis for snow droughts assessments (Table 1) (Jiang et al., 2022).Specifically, SWE was calculated by multiplication of the estimated snow depth using the linear unmixing method (LUT algorithm) and snow density (Yang et al., 2020).A series of calibrations were also performed to improve the quality of the snow depth data set.For bias correction, a regional semi-empirical model (Jiang et al., 2014;Yang et al., 2018) with a LUT algorithm (Derksen et al., 2005) was used to estimate the snow depth on a grid-cell scale weighted by the land cover fraction.Considering that the mixed-pixel problem interferes with the snow depth retrieval of the passive satellite microwave data, sensor cross-correction in brightness temperature was carried out to maintain the consistency between Scanning Multichannel Microwave Radiometer, Special Sensor Microwave/Imager (SSM/I), and Special Sensor Microwave Imager/Sounder (SSMI/S).Furthermore, observed weather station measurements and snow course data collected from local meteorological stations were used to correct snow depth estimates (Yang et al., 2020), with an overall unbiased root mean square error (RMSE) and bias value of 5.09 and −0.65 cm, respectively (Jiang et al., 2022).Thus, this data set was more reliable than others, such as MERRA-2 and GlobSnow-2.Furthermore, the RMSE of the SWE product in China was approximately 10-15 mm, meeting the requirements for snowpack change research in China (Jiang et al., 2022).Additionally, owing to the seasonal characteristics of snowpacks and the availability of daily SWE data, we only considered the changes in snow from November to April (winter and spring) and from 1992 to 2019.
The daily temperature and precipitation data used in our study from 1993 to 2019 were collected from the Climate Change Research Center of the Chinese Academy of Sciences.Additionally, the CN05.1 data set was generated by the "anomaly approach" using interpolation based on observational data from approximately 2,400 stations across China (Wu & Gao, 2013).For the "anomaly approach," a gridded climatology was first computed, before a corresponding gridded daily anomaly was added to obtain the final time series.Based on a considerable number of observation stations (∼2,416 stations), the validation overall showed significant agreement, despite uncertainties between CN05.1 and CN05 and between EA05 and APHRO.Moreover, this is a frequently employed climate data set in China due to its high spatial resolution (0.25°), and has been experienced extensive applications in the study of climate change (Ma & Zhang, 2022;Nie & Sun, 2022) and validation of models (Su et al., 2022;Yang et al., 2020).However, the uncertainty of meteorological data, especially the limitation of precipitation data, cannot be ignored.In general, China's meteorological observation stations are mainly located in the eastern economically developed regions, while relatively few in the west mountainous regions.Among them, from the northern QTP to Kunlun Mountains, there are basically no distribution of observation stations, which also determines that the data obtained in these areas have relatively large uncertainties (Wu & Gao, 2013).
Additionally, to better explore the large-scale atmospheric circulation patterns associated with snow droughts across China, hourly wind, surface pressure, and specific humidity fields (1° × 1°) provided by European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) were used to calculate the linear trend for water vapor flux and divergence between 1993 and 2019.In addition, we collected the monthly Northern Hemisphere Annular Mode (NAM) Index or Arctic Oscillation (AO) Index to explore the possible association with the occurrence of snow droughts.This index is defined as the normalized difference in the monthly zonal-mean sea level  pressure between 35° and 65°N, as previously proposed by J. Li and Wang (2003).This serves as a measure of the large-scale fluctuations in surface air pressure across the entire hemisphere, manifesting between the midand high-latitude Annular Belt of Actions.Moreover, multiple topographical variables, derived from Amatulli et al. (2018), were collected to explore the influence of topographical heterogeneity on the occurrence of snow droughts, including the elevation, slope, terrain ruggedness index (TRI), aspect sine, and aspect cosine, with a resolution of 10 km.Among them, the sine and cosine of aspect can be used to emphasize the difference between east/west and north/south exposure, respectively.

Standardized Snow Water Equivalent Index
The daily SWEI used to standardize SWE was similar to the typical SPI (Mckee et al., 1993) or a Standardized Streamflow Index (SSI) (Staudinger et al., 2014).Therefore, SWEI could be interpreted as the unit standard normal deviation associated with the cumulative SWE percentile over a specific duration.In this study, we estimated the probabilities associated with the data using a nonparametric approach as previously described by Huning and AghaKouchak (2020), to standardize the daily SWE, before monitoring the snow droughts.Since the SWE data set contained both zero and non-zero values, we randomly perturbed the zero values by adding positive numbers that were smaller than the lowest non-zero value in the data set.To determine the probabilities associated with the data, we calculate the empirical Gringorten (1963) plotting position, instead of fitting a specific distribution function to the SWE data: where i represents the rank of the nonzero variable and N denotes the sample size.The ranks were determined by integrating the SWE data over 3 months for month m, given by where SWE m represents the integrated value of SWE for month m, as obtained from Jiang et al. (2022).The process involves integrating daily SWE values to calculate each of the components present on the right-hand side of Equation 2. Consequently, A m,i provides an integrated measure of both the SWE amount and its persistence over each 3-month interval, which was then standardized as described below.We calculate the nonparametric standardized SWEI by converting the empirical probability (p) into the standard normal distribution using the following Equation: where ф −1 represents the inverse standard normal distribution.
Although SWEI can be performed at different scales, such as 6, 12, 24, and 48 months, it is ineffective on longer time scales when compared to SPI or SSI (Huning & AghaKouchak, 2020).This is because snowpacks have strong seasonal characteristics and are commonly present for a few consecutive months, after which they disappear completely.Although a shorter time scale (i.e., 1 month) can also be used to conduct snow droughts analysis, this is more sensitive to infrequent extreme events.Therefore, we selected a 3-month time scale for this study.
Considering that some regions have zero SWE values, we specified that all grid cells meet the criterion that at least 75% of a given month for all years had a nonzero SWE value (Figure S2 in Supporting Information S1).Furthermore, we only considered those grid cells where actual values existed for three consecutive months, such as November-December-January or January-February-March.For spatial coverage, we used multi-year averaged SWEI grid cells with non-zero SWEI values to define the spatial cover of snow droughts in China.
Since the seasonal snowpacks (coverage and depth) decrease rapidly outside of the snow season (Lei et al., 2022), we only analyzed drought durations ranging from November to next April in this study, the longest of which did not exceed 6 months.Moreover, we only considered three main continuous seasonal snow cover regions in China: Northeast China (NE), northern Xinjiang (NXJ), and QTP.These areas of seasonal snow-covered area are relatively stable, generally accounting for 27% of the total area of China (Huang et al., 2016) and 58% of the terrestrial water storage.

Sensitivity Analysis
To evaluate the sensitivity of snow droughts to precipitation and temperature, we used snow drought classification to identify different drought types on a grid scale between 1993 and 2019.Dierauer et al. ( 2019) previously proposed combining anomalies in temperature and precipitation to classify winters with a below-normal peak SWE as warm, dry, and dry and warm snow droughts.Although this classification is useful for easily distinguishing the influence of temperature and precipitation on the occurrence of snow droughts, the use of peak SWE causes an apparent uncertainty in China due to strong topographical heterogeneity (Gottlieb & Mankin, 2021).Therefore, to remain consistent with the snow droughts classifications proposed by Dierauer et al. (2019), we replaced the peak SWE with negative SWEI (<−0.5) to classify snow droughts in grid cells as follows: where SWEIi, Pi, and Ti are the daily negative SWEIs, precipitations, and temperatures between 1993 and 2019, respectively; and are the monthly averages of temperature and precipitation in specific months between 1993 and 2019; and Dry, Warm, and Dry and Warm represent the different drought types.
According to the above formula, snow droughts type was determined by the mean daily negative SWEI, temperature, and precipitation at each grid box.Based on the above calculations, we further estimated the distributional characteristics of each drought type at the grid scale from 1993 to 2019.Due to snow droughts at any grid point being matched with a pair of temperature and precipitation values, we further distinguished the distribution characteristics of different types of snow droughts by fitting the temperature and precipitation (Rahbek et al., 2019).

Atmospheric Water Vapor Transport
The occurrence of snow droughts during cold seasons has previously been linked to the effect of planetary-scale internal dynamical processes and atmospheric-ocean interactions, such as water vapor changes in the winter Northern Hemisphere.To evaluate the impacts of the large-scale atmospheric circulation on snow droughts, we used the vertically integrated water vapor flux and the divergence of water vapor flux, given by where u represents the zonal wind, v represents the meridional wind, g represents the acceleration due to gravity, and q represents the specific humidity.The pressure of the uppermost layer, Pt, was assumed as 200 hPa in the computation, since water vapor was considered to be negligible beyond that altitude, while Ps denotes the pressure at the Earth's surface.The divergence of water vapor flux may also be evaluated using the following expression: where  ⃖⃖ ⃗  represents the horizontal wind vector, and q represents the specific humidity, the initial term on the right side denotes moisture advection, contributing to the overall convergence or divergence of water vapor.The second term on the right also reflects the impact of wind divergence, which corresponds to the water vapor transport field.

Statistical Analysis
In this study, the nonparametric Mann-Kendall test (Kendall, 1975;Mann, 1945) was used to assess the significance of trends in time series.The null hypothesis in the Mann-Kendall test indicated that the data were independent and randomly ordered.The trend test results elucidated the magnitude and direction of the correlation.
The coefficient values ranged from +1 to −1, representing positive and negative correlations, respectively.The statistical significance of this trend was assessed, where p values below 0.05 and 0.01 represented trends significant at the 95% and 99% confidence levels, respectively.Pearson's correlation coefficient (r) was used to analyze the statistical relationships between the climatic variables and SWEI.The significance of the correlation was assessed using the two-side test approach, where p values below 0.05 represented trends significant at the 95% confidence levels.Furthermore, Z-scores as a standardized statistical measure were used to explore the potential responses of snow drought to Northern Hemisphere Annular Mode Index or AO Index (NAMI/AO).

Changes in SWEI
We observed an increasing trend of SWEI over the three regions between 1993 and 2019 (Figure 1a).Particularly in northern Xinjiang and Northeast China, the high values of linear trends were concentrated within mountainous areas such as the Tianshan (∼+0.04/yr),Altai (∼+0.03/yr),Greater Khingan (∼+0.04/yr), and Changbai Mountains (∼+0.05/yr).In the inner section of these areas, such as Junggar Basin of the northern Xinjiang and Northeast Plain, the changes were lower, since water vapor flux was likely blocked by terrain and descends.Meanwhile, we also found that the spatial pattern of SWEI variation was heterogeneous across the QTP between 1993 and 2019.Among them, the SWEI in the Himalayas (∼+0.03/yr) and Tanggula Mountains (∼+0.03/yr) in the eastern QTP showed an upward trend, while the inner part of the QTP showed a relatively downward trend at a rate of ∼0.03/yr.Subsequently, we found similar temporal trends over three regions between 1993 and 2019 (Figures 1b-1d).Specifically, the SWEI was rapidly increasing before 2010, after which these changes tended to slow.In Northeast China, northern Xinjiang, and the QTP, the SWEI increased at rates of 0.022/yr, 0.017/yr, and 0.011/yr, respectively, all of which passed the 0.01 significance test (MK).Since 2010, the upward trend of SWEI has been flattened, which is likely to have been affected by both synergistic changes in temperature and precipitation (Figure S3 in Supporting Information S1).
Our analysis further showed a monthly distribution in the three regions over the full period, with different magnitudes (Figure 2).For example, in Northeast China, the number of grid cells in April accounted for the largest proportion of negative SWEI, accompanied by the most pronounced decreasing trend of −16.4/yr in grid cells (p < 0.01, MK) (Figures 2a and 2d).Meanwhile in northern Xinjiang, the monthly distribution was similar to that of Northeast China, where the area proportion of negative SWEI was highest in April at a rate of −9.4/yr in grid cells (p < 0.01, MK) (Figures 2b and 2e).In the QTP, the number of grid cells in April also accounted for the greatest proportion of negative SWEI at a rate of −16.2/yr in grid cells (p < 0.01, MK) (Figures 2c and 2f).In general, the occurrence of snow droughts across the three regions was likely concentrated during April and decreased significantly, implying that the potential risks of snow droughts to agriculture in spring have been decreasing over the past 30 years.

Changes in Drought Duration
For the identified snow drought events with different durations between 1 and 6 months, our analysis showed an overall similar change in area, depicted by count of grided cells, and duration between Northeast China, northern Xinjiang, and the QTP from 1993 to 2019 (Figures 3a, 3c, and 3e).From 1993 to 2019, the number of grid cells decreased significantly by −34.5/yr in Northeast China, −18.2/yr in northern Xinjiang, and −51.6/yr in the QTP (p < 0.01, MK).We also found that the area proportion of snow droughts constantly decreased as the duration increased (Figures 3b, 3d, and 3f).The thicker the transect, the greater the area proportion of snow drought.In particular, the short-term snow droughts within 1-month had the largest area proportion of 46.5% (NE), 47.3% (NXJ), and 55.7% (QTP) in grid cells, while area proportion at 2-and 3-months was about 20%, rather small at 5and 6-months.In general, the identified events of snow droughts over these three regions have decreased significantly over the past three decades, with most of them being concentrated during a short period within 1-month, which was closely related to the substantial increase in SWEI intensity (Figures 1b-1d).

Influences of Temperature and Precipitation on Snow Droughts
To identify the possible impacts of temperature and precipitation on the occurrence of snow droughts, we carried out classification and driving analysis on negative SWEI in grid cells.In the warm type (Figure 4a), the high values (>300) in grid cells were mainly distributed in the eastern region of the QTP.In the dry and warm type (Figure 4b), the largest count in grid cells was mainly distributed in the western region of QTP (>500) and Changbai Mountains of Northeast China (>300).Similarly, to the dry and warm type, the high values of dry types in grid cells were also distributed in the western region of QTP, Greater Khingan and Changbai Mountains.Figures 4d-4f further shows that the patterns of fitting points composed of temperature and precipitation in the dry and dry and warm types were mainly limited by corresponding precipitation (<1 mm/day), compared to that of the warm type.Also, the dry type is constrained by colder temperature than that of other types, in addition to precipitation.In contrast, the distribution density of warm types depicts much greater precipitation than other types.Overall, the change in temperature and precipitation synergistically influence the occurrence of snow drought in the QTP, while changes in precipitation are likely to affect the occurrence of snow drought in northern Xinjiang and Northeast China.

Influences of Topographical Factors on Snow Droughts
As illustrated by the maps in Figure 5, we found that snow droughts primarily occurred in high-elevation and terrain complex regions (Figure S4 in Supporting Information S1), such as the Himalayas, Kunlun, and Hengduan Mountains in the QTP; Tianshan and the Altai Mountains in northern Xinjiang; and the Greater Khingan and Changbai Mountains in Northeast China.Moreover, the spatial extent of snow droughts within a 1-month duration was clearly larger than that with a longer duration, such as 3-month.As the drought duration increased, the frequency tended to decrease in the Northeast China (NE), QTP, and northern Xinjiang (NXJ) regions.We further found that there was a strong association between complex topographic features and high frequency as longitude changes, characterized by heatmaps (Figures 5b-5d).Specifically, the frequency of snow droughts tended to slow down alongside decreasing elevation and terrain complexity.There was an apparent difference in the frequency of snow droughts in the north/south aspect, whilst the cosine of aspects was wider than that of the sine of aspect between the duration of 2 and 4 months (Figures 5e and 5f).This was consistent with the direction of most mountain ranges in China, such as the Kunlun and Tianshan Mountains, and many mountains on the QTP.The occurrence of snow droughts exhibited significant disparities between the north and south aspects of 10.1029/2023JD039297 9 of 14 the mountains due to the persistent geographical features.The complex terrain of the high mountains acted as a barrier hindering the inland penetration of moist air, particularly when the atmospheric water vapor originating far from the ocean encountered these elevated landforms.The intricate topology of the mountains posed challenges for the advancement of moist air masses, impeding their ability to reach inland areas of China and contribute to substantial snowfall.This phenomenon is attributed to the interaction between the atmospheric circulation patterns and the obstructive effect of the mountainous terrain, creating a notable contrast in the frequency of snow droughts between the distinct aspects of the mountains.Meanwhile, the rapid rise in winter temperatures was likely causing the snow to melt rapidly (Figure S5 in Supporting Information S1).These factors together made high-altitude mountainous areas and inland basins such as the Himalayas, inland Tianshan Mountains, and Kunlun Mountains more prone to snow droughts.

Possible Association of Snow Droughts and Water Vapor Flux
The NAMI/AO index has been recognized as an influential factor in climate variability within the Northern Hemisphere, exerting a substantial influence on China's weather and climate conditions.This index plays a pivotal role in modulating atmospheric circulation and water vapor transport, thereby impacting weather phenomena and subsequently influencing regional climate characteristics.Given the potential as a driver of climate variability, it is imperative to investigate the impact of the NAM/AO index on specific events such as snow droughts.To diagnose and predict the occurrence of snow droughts and the relationship with potential atmospheric circulation modes, we analyzed the changes in the vertical water vapor field in the cold seasons, and further explored the potential response between snow drought and NAMI/AO index.
First, we examined the linear trends in the integrated layer water vapor flux patterns during the cold season between 1993 and 2019 (Figure 6).Specifically, we found that the southward (southwestward) water vapor transport toward northern Xinjiang (Northeast China) was enhanced over these years, leading to a significant convergence trends of water vapor flux over there.This suggested that due to the strengthening of atmospheric dynamic 10.1029/2023JD039297 10 of 14 transport, the water vapor in northern Xinjiang and Northeast China will increase, thus providing favorable climatic conditions for increasing snowfall.In contrast, enhanced divergence of water vapor flux in the northern Heilongjiang province was found, likely to aggravate the occurrence of snow drought events in this area.Meanwhile, we found that the inner and eastern regions of the QTP were also in a trend of water vapor convergence, while the water vapor in the southeastern and southwestern margins of the QTP showed a trend of divergence.This indicated that the increase in water vapor likely reduced snow drought events in the inland plateau and Hengduan Mountains, whilst aggravating drought in the Himalayas and Hindu Kush.
Furthermore, as shown in Figure 7, we found the occurrence cycle of snow drought in the cold season (November-April) was basically negative correlated with the periodic change of NAMI/AO.Especially since 2010, the changes of NAMI/AO are significantly negative correlated to the number of snow droughts (r = −0.32,p < 0.05).With the strengthening of NAMI/AO, more water vapor convergence has been likely caused by the large-scale circulation, which has contributed to the decline of snow drought events across China.This is attributed to the fact that, on one hand, a positive phase of NAMI/AO is frequently associated with a reinforced westerly wind belt, facilitating the transport of moisture from oceanic regions to inland areas, likely contributing to more snowfall, such as in the northern Xinjiang.On the other hand, a positive phase of NAMI/ AO may enhance the stability of atmospheric circulation, fostering favorable meteorological conditions.This may involve increased upward motion and convective activity, promoting the sublimation and condensation of water vapor.Furthermore, a positive NAMI/AO can influence terrain effects in regions like the QTP, impacting atmospheric circulation and potentially leading to enhanced water vapor convergence in specific areas, creating more favorable conditions for precipitation or snowfall.

Summary and Discussion
Using regional high-quality SWE estimates and their derived SWEI index, we assessed the changes of snow droughts in Northeast China, the QTP, and northern Xinjiang over the past three decades.Our results revealed that SWEI values increased significantly, at rates of 0.022/yr (Northeast China), 0.017/yr (northern Xinjiang), and 0.011/yr (QTP) (p < 0.01, MK).The increase in SWEI was accompanied by a decrease in drought extent and intensity.The occurrence of snow droughts with a 1-month duration (>46.5%) was predominant, whereas the frequency of snow droughts decreased as duration time increased.Furthermore, the occurrence of snow droughts in April had the largest proportion, although decreased significantly over the past three decades across the three regions.Classification analysis showed that the warm type events were more apparent in the eastern region of QTP than the other regions, while the dry and dry and warm type events mainly occurred in the western region of QTP, Northeast China and northern Xinjiang.Spatially, we found that the occurrence of snow droughts was significantly dependent on complex topography, with higher frequencies being found in high-altitude mountainous areas and inland basins.Additionally, we found that the number of snow droughts was negatively correlated with the NAMI/AO with a 0.05 significance level (t-test) since 2010.Alongside the strengthening of NAMI/AO, an apparent water vapor flux convergence was detected during the cold seasons, which likely contributed to the increase in snowfall, especially in northern Xinjiang and Northeast China.
Not only AO but also other internal dynamical processes such as El Niño-Southern Oscillation and Pacific Decadal oscillation are also directly or indirectly associated with the occurrence of snow droughts, particularly in the QTP (Deser et al., 2015;J. Zhang et al., 2019J. Zhang et al., , 2021)).Usually, in the positive stage of these dynamical processes, the deeper India-Burma trough and intensified cyclonic circulation near Lake Baikal commonly having a higher water vapor flux jointly brought more snowfall to the plateau.In contrast, during the negative stage, the East Asian through and the East Asian winter monsoon strengthens frequently, and dry cold air flows, resulting in a decrease in snowfall and inducing snow droughts (You et al., 2020).Furthermore, since the 1980s, the rapid warming in the cold season across the plateau has been roughly twice as fast as in the warm season.A trend that is expected to intensify in the future, and cause more snowpack loss.Due to snow-and ice-albedo feedback mechanisms (Che et al., 2008;Cook et al., 2020;Qiu et al., 2019;Zheng et al., 2019), the loss of highly reflective snow cover will continue to increase the amount of solar energy absorbed, resulting in more warming, which is not conducive to the maintenance of snow, likely prompting the exacerbation of future snow droughts in some parts of QTP (Deng et al., 2017).
Substantial uncertainties persist in remote sensing retrieval of SWE product used in our study (Bormann et al., 2018;Jiang et al., 2022).One particular  challenge arises in mountainous regions with complex terrain, such as the Tianshan Mountains, where the presence of thicker snow cover is common.However, the coarse spatial resolution of microwave pixels hinders the representativeness of snow depth measurements at a single site in mountainous areas (Jiang et al., 2014(Jiang et al., , 2022)).This limitation results in an unbiased RMSE of approximately 16 cm, leading to a notable underestimation of deep snowpack areas (>40 cm).Additionally, passive microwave data are influenced by underlying surface conditions.For instance, the SWE product we employed is notably affected by forest canopy effects on microwave observations within forested regions, with an unRMSE of 7.5 cm (Jiang et al., 2022).Besides, the influence of liquid water content on passive microwave retrieval is a well-documented challenge in snow-related studies (Dietz et al., 2012;Nolin, 2010).This factor becomes especially pertinent in spring conditions, where melting and refreezing of snow can occur.The presence of even a small amount of liquid water in the snowpack can evidently influence the emissivity of the snow, making it challenging to differentiate from snow-free land.Overall, these uncertainties impact the reliability of SWE products in mountainous regions.Nevertheless, our findings reveal that snow droughts primarily occur in the spring and in high-altitude mountainous regions, indicating the stability of these snow droughts to a certain extent, which is highly consistent with the identified stable snow-cover areas in China (Huang et al., 2016;Wang et al., 2018).Furthermore, the computation of SWE Index (SWEI) mirrors the principles of Standardized Precipitation Index (SPI) (Mckee et al., 1993).It relies on statistical distributions to uncover long-term dynamics in SWE, rather than solely relying on actual measurements at a single site or pixel.
In other words, SWE data is transformed into a standard normal distribution, and then the degree of deviation of the SWE relative to the long-term statistical data is calculated.Therefore, despite the inherent uncertainties associated with the SWE product in mountainous regions, we contend that SWEI can still effectively reflect the characteristics of snow droughts in these regions.
In addition, due to complex topographical heterogeneity, the current understanding and capacity to diagnose and predict the hydrological and ecological consequences of snow droughts are still poor (Gottlieb & Mankin, 2021;Huning & AghaKouchak, 2020;Y. Li et al., 2020).By using a SWE data set calibrated specifically for China, this study improves our understanding of snow drought occurrence in China.Due to strong seasonal characteristics, changes in snow droughts are likely to impact the downstream runoff and water usage.Especially in spring, the runoff of the Brahmaputra, and Lancang-Mekong Rivers around the QTP is strongly mediated by the meltwater of the snowpack upstream, although snowmelt runoff only represents a relatively lower fraction than that of rainfall-runoff over the total runoff (Irannezhad & Liu, 2022;Liu et al., 2021;Wang et al., 2021).Considering the continuous changes in drought characteristics, policymakers must develop different adaptation strategies and manage water resources in the spring to mitigate the drought and agricultural risks associated with the changing snowpacks (Huning & AghaKouchak, 2020).Nonetheless, the complex relationship between snow droughts and runoff needs to be investigated extensively to clarify the specific hydrological response.

Figure 2 .
Figure 2. (a-c) Monthly distribution and (d-f) linear trends of negative snow water equivalent index in grid cells in Northeast China (NE), the Qinghai-Tibet Plateau, and northern Xinjiang (NXJ) regions between 1993 and 2019.Bar plots show the data distribution, with error plots showing the outlier.The symbols "*" and "**," respectively represent the significance levels of 0.05 and 0.01 (M-K test).

Figure 3 .
Figure 3. (a, c, and e) Count of snow droughts in grid cells in Northeast China (NE), the Qinghai-Tibet Plateau, and northern Xinjiang (NXJ) between 1993 and 2019; (b, d, and f) Kite diagrams represent the proportions of snow droughts at different durations from 1-to 6-months.The thicker the transect, the greater the proportion of snow droughts.

Figure 4 .
Figure 4. (a-c) Maps of warm, dry and warm, and dry types in grid cells (unit: count) between 1993 and 2019; (d-f) Fitting climate space of temperature and precipitation corresponding to warm, dry and warm, and dry types in grid cells between 1993 and 2019.

Figure 6 .
Figure 6.Linear trends of the integrated layer water vapor flux (vectors, unit: kg/m/s/yr) and water vapor flux divergence (shadings, unit mm/day/ yr) for the cold seasons (November-April) between 1993 and 2019.Negative values indicate the trends of water vapor convergence.The doted regions are statistically significant at the 95% confidence level according to the student's t test.

Figure 7 .
Figure 7. Z-scores of Northern Hemisphere Annular Mode Index/Arctic Oscillation (red bars) and identified snow droughts (blue bars) in the cold seasons between (a) 1993and 2000, (b) 2001 and 2010, and (c) 2011 and 2019.

Table 1
Remote Sensing and Meteorological Reanalysis Data Set Characteristics