Study on the Snowfall Amount Triggering Regional Avalanches in Southeastern Tibet

Abstract

Global climate warming has exacerbated extreme snowfall events. The Southeastern Tibet (ST) region has become a high-incidence area for avalanches due to its unique topographical and climatic conditions. However, current research has paid insufficient attention to the thresholds for avalanches triggered by extreme snowfall. Therefore, the aim of this study is to construct the I-D (intensity-duration) thresholds for avalanche events triggered by extreme snowfall in southeastern Tibet, providing a scientific basis for disaster prevention and mitigation work in this region. Based on the snowfall data from 1951 to 2020, this study calculated four extreme snowfall indices, namely SF1d, SF90p, SF95p, and SF99p, to determine extreme snowfall events. And 33 avalanche events during this period were verified through the confusion matrix. This study found that the intensity of extreme snowfall events in southeastern Tibet has increased while the frequency has decreased. The I-D threshold parameters α (from 5.79 to 14.88) and β (from −2.81 to −0.66) within the study area were determined, and the overall threshold is I = 9.29 × D^−2.27 (D represents the duration of snowfall, with the unit being days.). It was also found that extreme snowfall in the study area has a significant positive correlation in with the ST. The terrain has a greater impact on the snowfall intensity, but its regulation on the duration of events is limited. Overall, in southeastern Tibet, if the single-day snowfall exceeds 12.38 mm (the regional average value of the SF1d index) or the cumulative snowfall within the previous 30 days exceeds 64.85 mm (the regional average value of the three indices of SF90p, SF95p, and SF99p), it can be considered that an extreme snowfall event has occurred. At the same time, the threshold of I = 9.29 × D^−2.27 can be used to forecast avalanches triggered by extreme snowfall events in the entire region.

1. Introduction

Southeastern Tibetan Plateau (SETP) is located in the southeastern part of the Qinghai-Tibet Plateau, encompassing the eastern segment of the Himalayas, the lower reaches of the Yarlung Tsangpo River, and parts of the Nyainqêntanglha and Hengduan mountain ranges. The study area features significant topographic relief, with deep valleys, interlacing mountain ranges, and a general northwest-southeast elevation gradient—higher in the northwest and slightly lower in the south. This region serves as a critical gateway for moist South Asian monsoon air to enter the Qinghai-Tibet Plateau, where the Yarlung Tsangpo Basin forms the largest water vapor corridor on the plateau. During the snow season, warm and moist air currents from the Indian Ocean trigger widespread snowfall here [1]. It is also one of the most climate-sensitive and responsive regions on the Qinghai-Tibet Plateau, exhibiting pronounced impacts of climate change [2,3].

With the deepening implementation of China’s Western Development Strategy and the vigorous growth of tourism in southeastern Tibet, the region faces severe avalanche risks, intensifying conflicts between avalanche disasters and human life and property safety. Additionally, in recent years, ongoing climate warming has led to frequent extreme snowfall events in the study area, further escalating avalanche risks. Avalanches in southeastern Tibet have caused significant casualties: the massive Ranwu avalanche on 24 March 1996, resulted in 64 fatalities; and in January 2023, a severe avalanche triggered by extreme snowfall at Duoxionglashan Tunnel Pass (on the Pai-Mo Highway at the junction of Mainling and Medog Counties, Nyingchi City) caused 28 deaths [4,5,6]. Meanwhile, avalanches severely disrupt the region’s transportation links with the outside world. Prior to 2009, annual traffic availability in Medog County was limited to just 2–3 months due to avalanches; from November 2018 to April 2019, more than half of the period saw road closures, drastically constraining local socioeconomic development [7]. However, despite these severe challenges, few scholars have combined research on extreme snowfall and avalanche events. Therefore, it is crucial to investigate the patterns of extreme snowfall in southeastern Tibet and conduct early-warning research on avalanche disasters triggered by such events. This aim is to improve the precision of regional avalanche forecasting and provide a theoretical reference and practical basis for disaster prevention and mitigation in alpine mountainous areas.

In the field of avalanche disaster research, scholars at home and abroad have made notable progress since last century. Qiu (1985) summarized the avalanche geomorphic characteristics of the Nyainqêntanglha Mountains in southeastern Tibet, finding that the transport capacity and scale of avalanches can rival those of floods [8]. Wang (1992) analyzed the geographical environmental conditions for avalanche distribution on the Qinghai-Tibet Plateau, noting that avalanches primarily occur in fragmented terrains along the plateau’s margins and adjacent areas. He also explained that avalanches along the southern Sichuan-Tibet Highway and in western Sichuan are influenced by wind-driven snow and solid precipitation [9]. Wei et al. (2004) evaluated the avalanche risk for railways entering Tibet using critical snow thickness and slope safety angles [10]. Zhao et al. (2017) conducted field investigations of 90 avalanches along the Anjiu La Mountain to Guxiang section of the Sichuan-Tibet Highway, using statistical analysis to determine avalanche distribution patterns in the region [11]. Chen et al. (2018) established a hazard zoning model for avalanches in the same section of the Sichuan-Tibet Railway using field surveys, the analytic hierarchy process (AHP), and fuzzy comprehensive evaluation [12]. Gao et al. (2003) used GIS technology to optimize transportation routes in disaster-prone areas of the Palong Tsangpo River basin [13]. Duan et al. (2016) statistically analyzed the relationship between equivalent friction coefficients and formation area sizes for 36 typical wet avalanches in the Ranwu-Tongmai section of the Palong Tsangpo River, conducting predictive simulations and regression analyses for avalanches along transportation routes [14]. Wen et al. (2021) analyzed avalanche factors and development characteristics in the Palong Tsangpo River basin through large-scale data statistics, summarizing spatio-temporal distribution patterns and conducting avalanche susceptibility zoning [15]. With technological and societal advancements, remote sensing techniques have been increasingly combined with field observation sites for avalanche risk assessment, zoning, early warning, and mitigation. Some other countries, such as Switzerland, Austria, France, Norway, and the United States, have developed comprehensive avalanche prevention and control standards based on their national spatio-temporal avalanche patterns [16,17]. Keylock et al. (1999) created an automatic avalanche risk prediction model to calculate risk probabilities [18]; while Blagovecsenskii et al. (1994) classified low, medium-, and high-risk avalanche zones in the Tianshan Mountains based on topographic relief and valley cross-sectional morphology [19]. Owens et al. (1989) produced avalanche susceptibility maps for traffic routes by integrating elevation, vegetation density, snowfall, and other factors [20]. Current European research on avalanche distribution, focused mainly on the Alps, emphasizes avalanche arrival probability and periodic patterns—approaches suitable for regions with long-term, extensive observational data [21,22,23,24]. However, these methods are less applicable to SETP, which has limited historical data records. Consequently, studies on snowfall thresholds triggering avalanches in SETP have been scarce and methodologically simplistic, with even fewer investigations addressing avalanches under extreme snowfall conditions.

Snowfall is a critical component of the Earth system and is highly sensitive to climate change [25]. Changes in snowfall amount play a vital role in regulating surface snow cover and water distribution. Additionally, excessive or extreme snowfall serves as a major hazard-inducing factor [26], often triggering various cryosphere disasters—especially avalanches—that cause substantial economic losses and pose significant threats to transportation, energy supply, infrastructure, agricultural production, and even human lives [27,28,29,30]. Over the past half-century, rising temperatures have accelerated hydrological processes, enhanced the water cycle, induced significant changes in precipitation patterns, and exacerbated the occurrence of extreme snowfall events [31,32]. In particular, the last decade has seen frequent extreme weather events, with extreme heavy snowfall garnering widespread attention [33,34,35]. Against the backdrop of global warming, the frequency and intensity of extreme snowfall events continue to exhibit an increasing trend, and in some regions where total snowfall has decreased, this upward trend in extreme snowfall is expected to persist within a certain range of temperature increases [36,37,38,39].

Some scholars have already conducted a series of studies on snowfall amounts. Gao et al. (2023) found that snowfall in the Qinghai-Tibet Plateau below 5000 m exhibits significant altitude dependence under climate change [40]. Li et al. (2024) revealed that while overall snowfall and extreme snowfall in southeastern Tibet show a downward trend, extreme snowfall along the Brahmaputra-Yarlung Zangbo River Valley has increased, greatly intensifying avalanche threats in this region [41,42]. However, research on extreme snowfall events that trigger avalanche disasters remains scarce. Additionally, few studies have linked avalanche disasters with pre-event snowfall characteristics, indicating that threshold research on avalanche initiation under extreme snowfall conditions requires further exploration.

As a core region of Tibet, southeastern Tibet is influenced by both the westerly circulation and Indian Ocean warm currents. Combined with its dramatic topographic relief and extensive marine glaciers and snow cover, the region is highly sensitive to climate change. Serving as the source of many major Asian rivers, snowmelt in southeastern Tibet is a crucial water supply for river systems. Due to glacier retreating and increased non-monsoon precipitation, the contribution of snowmelt to runoff has increased and now exceeds that of glacial meltwater. In recent years, against the backdrop of climate change, snow disasters on the Qinghai-Tibet Plateau have become increasingly frequent, posing significant threats to major engineering projects such as the Sichuan-Tibet Railway [43]. Current research on extreme snowfall in southeastern Tibet mainly focuses on the changing trends of extreme snowfall under climate change and the controlling effects of complex mountainous terrain on snowfall amounts. However, few studies have focused on the impact of increasing extreme snowfall events on avalanche disasters. Building on previous research into the spatial differentiation of avalanche hazards in southeastern Tibet, this study generates gridded data of extreme snowfall using historical records, constructs an I-D (intensity-duration) threshold model for extreme snowfall in conjunction with existing scientific experimental platforms, and uses GIS technology to conduct visual analysis of extreme snowfall thresholds triggering avalanche disasters in the region. This study is to provide a reference for disaster prevention and mitigation strategies against avalanches in southeastern Tibet amid climate change and increasing extreme events.

The remaining content of this paper is as follows: Section 2 introduces the study area, methods, and data used in this research. Section 3 presents the spatiotemporal distribution characteristics of extreme snowfall indices in southeastern Tibet and the calculation results of I-D thresholds. Section 4 discusses in detail the relationship between extreme snowfall indices and climate change in southeastern Tibet, as well as the correlation between I-D threshold results and topographic factors.

2. Materials and Methods

2.1. Study Area

Southeastern Tibet is in the southeastern part of the Qinghai-Tibet Plateau (as shown in Figure 1), characterized by dramatic topographic relief where mountain ranges and valley basins interweave, exhibiting typical alpine valley topography. Under the combined influence of the westerly circulation, moist Indian Ocean monsoon, and complex mountainous terrain, the region receives abundant precipitation, with extremely heavy snowfall during the snow season [44]. Against the backdrop of climate warming, extreme snowfall events are frequent here. Meanwhile, intense snowmelt processes in the snow-covered areas of southeastern Tibet are prone to destabilizing the structure of ice and snow masses on mountain slopes. Combined with the influence of the Himalayan tectonic fault zone, it leads to instability in snow mass structure on hillsides, exacerbating the occurrence of avalanche events.

2.2. Data Source

2.2.1. Avalanche Datas

The historical avalanche location data in our study are mainly collected from government reports and news media coverage, with their timing and locations listed below (

Table 1. Specific historical avalanche location information used in the study.

Position	Date	Longitude	Latitude
Linzhi	24 March 1996	96.74 E	29.51 N
Changdu	3 March 2002	96.60 E	31.21 N
Changdu	13 April 2002	88.63 E	29.14 N
Linzhi	19 April 2005	95.77 E	29.86 N
Changdu	17 May 2007	96.92 E	30.05 N
Changdu	20 January 2008	97.18 E	31.14 N
Linzhi	25 January 2008	95.32 E	29.33 N
Changdu	25 February 2008	98.22 E	31.50 N
Changdu	26 October 2008	96.91 E	30.50 N
Linzhi	27 October 2008	95.77 E	29.86 N
Linzhi	28 October 2008	95.5 E	29.90 N
Linzhi	28 October 2008	95.77 E	29.86 N
Changdu	26 February 2009	96.92 E	30.05 N
Linzhi	21 March 2011	96.03 E	30.14 N
Linzhi	21 March 2011	95.77 E	29.86 N
Linzhi	24 March 2011	90.77 E	29.13 N
Linzhi	25 March 2011	95.77 E	29.86 N
Changdu	15 April 2015	94.71 E	30.93 N
Changdu	26 February 2016	95.83 E	30.74 N
Linzhi	2 November 2016	95.77 E	29.86 N
Linzhi	2 November 2016	94.82 E	30.28 N
Linzhi	2 April 2018	97.47 E	28.66 N
Changdu	16 May 2018	97.21 E	31.50 N
Linzhi	16 May 2018	97.47 E	28.66 N
Linzhi	16 May 2018	97.04 E	29.32 N
Linzhi	19 May 2018	97.47 E	28.66 N
Naqu	3 November 2018	93.69 E	31.48 N
Linzhi	28 February 2019	95.77 E	29.86 N
Changdu	12 April 2019	93.89 E	30.69 N
Changdu	13 April 2019	94.71 E	30.93 N
Linzhi	17 March 2020	96.92 E	30.05 N
Changdu	17 March 2020	95.77 E	29.86 N

).

2.2.2. Snowfall Datas and Other Datas

The data used in this study include the China Regional Ground Meteorological Element Forcing Dataset V2.0 [45,46,47,48], Digital Elevation Model (DEM), and historical avalanche location data. The China Regional Ground Meteorological Element Forcing Dataset V2.0 primarily provides snowfall, precipitation, and temperature data, while the DEM data are used to examine the relationship between snowfall intensity, its duration, and the topographic environment.

The China Regional Ground Meteorological Element Forcing Dataset V2.0, the second-generation version of this dataset, covers ground meteorological data for China from 1951 to 2020 with a spatial resolution of 0.1°. The DEM (Digital Elevation Model) data originate from the GDEM V3 dataset jointly developed by Japan’s Ministry of Economy, Trade and Industry (METI) and the U.S. National Aeronautics and Space Administration (NASA), which has a resolution of 30 m.

2.3. Methods

2.3.1. Extreme Snowfall Event Diagnosis

Currently, extreme snowfall indices are mostly defined using threshold methods, including absolute thresholds and percentile threshold methods, where snowfall events exceeding the threshold are identified as extreme snowfall events [49,50,51]. Since the study area is primarily a monsoon-influenced region with a wide geographical scope, the percentile threshold method was adopted for extracting extreme snowfall events. The percentile threshold is defined as follows: for each snow season (October to April), the subsample of daily snowfall data is sorted in ascending order, and the long-term averages of the 90th, 95th, and 99th percentiles are defined as the thresholds for extreme snowfall events in the snow season. An extreme snowfall index is defined when the cumulative snowfall exceeds the 90th, 95th, or 99th percentile of the total snowfall for that snow season. Additionally, this study uses the maximum daily snowfall amount within the snow season as another threshold for defining extreme snowfall events. The extreme snowfall indicators used in this study are listed in

Table 2. Extreme indices.

Indices	Implication	Unit
SF1d	Maximum daily snowfall of a snow season	mm
SF90p	More than 90% of the total snowfall in a snow season	mm
SF95p	More than 95% of the total snowfall in a snow season	mm
SF99p	More than 99% of the total snowfall in a snow season	mm

.

2.3.2. I-D Thresholding Method

Strictly speaking, the snowfall thresholds determined by single or multiple extreme snowfall events represent results derived from a black-box model, which to some extent overlooks the physical processes underlying avalanche occurrence.

The I-D snowfall threshold is typically plotted within a double logarithmic plane coordinate system, determined by a series of data points. The fitted threshold curve represents the optimal separation between triggering and non-triggering conditions, where values above the threshold indicate a potential for avalanche triggering. Depending on differences in the threshold-curve fitting formulas, the physical quantities represented by the horizontal and vertical axes of the coordinate system vary slightly. In this study, the I-D threshold is determined using the formula proposed by Caine (1980) [52]:

$$\mathrm{I}=\mathsf{\alpha }\times {\mathrm{D}}^{-\mathsf{\beta }}$$

(1)

Among them, I and D represent snowfall intensity (mm/day) and duration (days), respectively, while α and β are empirical parameters determined based on the data distribution.

2.3.3. Trend Test Analysis Methods

To investigate the interannual variation trends and trend significance of extreme snowfall indices from 1951 to 2020, this study uses the Theil-Sen method to perform regression calculations [53] and Mann-Kendall (M-K) tests for each pixel unit [54]. Compared with traditional linear regression, the Theil-Sen method has the advantages of handling non-normally distributed data and being robust to outliers [55]. Its main formula is as follows:

$$\mathsf{\beta }=\mathrm{m}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\left(\begin{array}{c}{\mathrm{x}}_{\mathrm{j}}-{\mathrm{x}}_{\mathrm{i}}\\ \mathrm{j}-\mathrm{i}\end{array}\right),1<\mathrm{i}<\mathrm{j}<\mathrm{n}$$

(2)

Among them, β is the trend slope: a positive value indicates that the extreme snowfall index trends toward delay/prolongation, while a negative value indicates the trend toward advance/reduction; in the formula, n = 70, where x_i represents the ith value over the 70-year period, and x_j corresponds to the jth value (similar to x_i and the jth value).

The Mann-Kendall test (K-test) is a nonparametric method for verifying monotonic trends and has been widely used in trend detection and analysis of hydrological and meteorological time series [56]. The formula is as follows:

$$\mathrm{Z}=\left\{\begin{array}{c}{\displaystyle \frac{\mathrm{S}-1}{\sqrt{\mathrm{V}\mathrm{a}\mathrm{r}(\mathrm{S})}}},\mathrm{if} \mathrm{S}>0\\ 0,\mathrm{if} \mathrm{S}=0\\ {\displaystyle \frac{\mathrm{S}+1}{\sqrt{\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{S}\right)}}},\mathrm{if} \mathrm{S}<0\end{array},\mathrm{V}\mathrm{a}\mathrm{r}(\mathrm{S})={\displaystyle \frac{\mathrm{n}(\mathrm{n}-1)(2\mathrm{n}+5)}{18}}\right.$$

(3)

$$\mathrm{S}={\sum }_{\mathrm{i}=1}^{\mathrm{n}-1}{\sum }_{\mathrm{j}=\mathrm{i}+1}^{\mathrm{n}}\left\{\begin{array}{c}1,\mathrm{if} {\mathrm{x}}_{\mathrm{j}}>{\mathrm{x}}_{\mathrm{i}}\\ 0,\mathrm{if} {\mathrm{x}}_{\mathrm{j}}={\mathrm{x}}_{\mathrm{i}}\\ -1,\mathrm{if} {\mathrm{x}}_{\mathrm{j}}<{\mathrm{x}}_{\mathrm{i}}\end{array},1<\mathrm{i}<\mathrm{j}<\mathrm{n}\right.$$

(4)

where the Z value is used to test whether the time trend is statistically significant, and S represents the direction and magnitude of the trend.

2.3.4. Threshold Performance Evaluation Methods

This study introduces a confusion matrix to analyze the prediction accuracy and reliability of optimal snowfall thresholds under various extreme snowfall indices [57,58,59,60]. In the validation system, the period from 1963 to 2018 was used as the calibration period, and historical extreme snowfall events were classified into four categories: (1) True Positive (TP): snowfall intensity exceeds the threshold and triggers an actual avalanche; (2) True Negative (TN): snowfall intensity is below the threshold and no avalanche is triggered; (3) False Positive (FP): snowfall intensity exceeds the threshold but no avalanche is triggered; (4) False Negative (FN): an avalanche is triggered but snowfall intensity is below the threshold. Evaluation metrics include sensitivity (Se, which measures the proportion of correctly predicted positive samples (avalanche-triggering extreme snowfall events) relative to all actual positive samples, and specificity (Sp), which reflects the proportion of correctly predicted negative samples (non-triggering events) relative to all actual negative samples—higher values of Se and Sp indicate better model accuracy and predictive capability for avalanches triggered by extreme snowfall. Additionally, positive predictive power (PPP) assesses the proportion of true positive predictions among all positive classifications, negative predictive power (NPP) measures the proportion of true negative predictions among all negative classifications, and efficiency (E) calculates the ratio of correctly predicted samples (both positive and negative) to the total sample size; these three metrics evaluate the reliability of the model’s predictions, collectively providing a comprehensive assessment of its performance in distinguishing avalanche-triggering conditions from non-triggering ones under extreme snowfall scenarios.

$$\mathrm{S}\mathrm{p}=\mathrm{T}\mathrm{N}/(\mathrm{T}\mathrm{N}+\mathrm{F}\mathrm{P})$$

(5)

$$\mathrm{S}\mathrm{e}=\mathrm{T}\mathrm{P}/(\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{N})$$

(6)

$$\mathrm{P}\mathrm{P}\mathrm{P}=\mathrm{T}\mathrm{P}/(\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{P})$$

(7)

$$\mathrm{N}\mathrm{P}\mathrm{P}=\mathrm{T}\mathrm{N}/(\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{P})$$

(8)

$$\mathrm{E}=(\mathrm{T}\mathrm{P}+\mathrm{T}\mathrm{N})/(\mathrm{T}\mathrm{P}+\mathrm{T}\mathrm{N}+\mathrm{F}\mathrm{P}+\mathrm{F}\mathrm{N})$$

(9)

2.3.5. Partial Correlation Analysis

Partial correlation is a statistical method used to measure the relationship between two variables while controlling for the influence of one or more other variables [61]. This research method can help us understand the direct relationship between variables. In this study, this method is mainly used to investigate the changes in extreme snowfall indices under climate change. Therefore, extreme snowfall indices, precipitation, and temperature are the main variables considered. The core idea is to eliminate the interference of another variable when clarifying the relationship between two variables. In the analysis and calculation of raster data, each pixel can be regarded as an observation point. In this study, a pixel-by-pixel analysis strategy is adopted for partial correlation calculation, that is, pixel to pixel analysis is carried out for each pixel, rather than analyzing and calculating the entire image as a whole. The formula is as follows:

$$R_{xy,z}={\displaystyle \frac{R_{xy}-R_{xz}{R}_{yz}}{\sqrt{(1-R_{xz}^2){ (1-R}_{yz}^2)}}}$$

(10)

In Equations (10) and (11), R_xy,z denotes the first-order partial correlation coefficient between x and y when the control variable z is held constant; R_xy,zm denotes the second-order partial correlation coefficient between x and y when the control variables z and m are held constant.

3. Results

3.1. The Spatiotemporal Distribution Characteristics of Extreme Snowfall Indices

Studying the temporal changes in extreme snowfall indices can provide in-depth insights into the trends, magnitudes, and potential for abrupt or discontinuous changes in extreme snowfall events. Based on statistics of the regionally averaged extreme snowfall indices in the study area from 1951 to 2020 (as shown in Figure 2), SF1d showed an upward trend, while SF90p, SF95p, and SF99p exhibited downward trends (p < 0.01). This result is highly consistent with previous studies, though the magnitudes of change for these indices differ. SF1d maintained a high rate of increase (slope = 9.81 mm/day), SF90p showed the most pronounced decline (slope = −0.16 mm/day), while the declines for SF95p (slope = −0.04 mm/day) and SF99p (slope = −0.02 mm/day) were smaller. These findings indicate that over the past 70 years, the daily extreme snowfall values in the study area have trended toward increased, whereas the cumulative snowfall amounts corresponding to the 90th, 95th, and 99th percentiles of total snowfall in the snow season have shown decreasing trends. This suggests that under the broader context of climate warming, the instability of snowfall in the study area has intensified, making extreme snowfall events more likely to occur. Additionally, this study plotted the distribution of the 1951–2020 average extreme snowfall indices in the study area based on the calculated historical values (as shown in Figure 3). The figure clearly shows that the four extreme snowfall indices in southeastern Tibet are primarily concentrated in the Yarlung Tsangpo Basin, mostly along the water vapor channels of the Qinghai-Tibet Plateau, extending northward from the southern monsoon direction into the plateau.

The results of the extreme snowfall index trend test (as shown in Figure 4) indicate that all four extreme snowfall indices near the Hengduan Mountains show significant increasing trends. Similarly, near the Tanggula Mountains (Tanggula Mountains) and Nyainqêntanglha Mountains in the northern part of the study area, these four indices also exhibit significant increase, though SF1d and SF90p show this trend over a larger area than SF95p and SF99p. This suggests that extreme snowfall events in these two regions have trended upward over the past 70 years, and future precautions against snow disasters triggered by extreme snowfall in these areas are warranted. Notably, in the eastern part of the Himalayas and the Yarlung Tsangpo Basin, the four extreme snowfall indices show significant decreasing trends, with SF1d, SF95p, and SF99p exhibiting this trend over a wider range than SF90p. This indicates a downward trend in snow season snowfall in the eastern Himalayas and Yarlung Tsangpo Basin over the past seven decades. The reduction in snow season snowfall suggests that rising average temperatures in this region under global warming have affected snow formation, highlighting the significant impact of climate change on extreme snowfall events here [62]. Historical data (as shown in Figure 1) show that avalanche events in southeastern Tibet primarily occur in the eastern part of the study area. Given the significant increase in extreme snowfall indices in the east and the intensified disturbance of weather systems caused by climate warming, the eastern part of the study area poses a high risk of avalanches.

3.2. Study on I-D Thresholds for Extreme Snowfall Events Triggering Avalanche Disasters

3.2.1. Analysis of Extreme Snowfall Events Triggering Avalanches

Starting from the date of an avalanche event, a backward search analysis was conducted on the input cumulative snowfall. Critical extreme snowfall events associated with each avalanche were extracted in the order of 1, 3, 5, 7, 15, and 30 days before the event. Avalanches triggered by non-extreme snowfall events were filtered out, and the intensity (I) and duration (D) of the snowfall were calculated. As shown in Figure 5, snowfall within the 30 days prior to the avalanche event exhibited high variability, with alternating periods of heavy snowfall, low-intensity snowfall, and snow-free intervals. Snowfall thresholds, as part of the final regional avalanche disaster early warning system, require a standardized definition of extreme snowfall events to facilitate the system’s automated operation. To this end, this study defined critical extreme snowfall events triggering avalanches by setting a minimum no-snow time gap (NSG). When the snow-free period exceeds this minimum gap, snowfall below the minimum daily intensity is considered to have no impact on avalanche events, equivalent to no snowfall occurring. Based on extreme snowfall indices and the above-defined no-snow window, the research process (as shown in Figure 5) may still include short-duration but high-intensity snowfall events. Due to the discontinuous nature of snowfall processes, these events cannot be directly incorporated into calculations but play a critical role in avalanche occurrences.

3.2.2. Statistics of I-D Thresholds Under Extreme Snowfall Events

Through the above calculations, multiple I-D thresholds are matched for each extreme snowfall index, with their magnitudes and interval duration primarily related to the number of input snowfall values. Considering that snowfall amounts and avalanche occurrences are influenced not only by altitude, slope aspect, and gradient but also by the recording of historical data (limited by data quality), the snowfall closest to an avalanche event does not always provide the most relevant effective values for avalanche occurrence. To avoid unnecessary data waste in calculations using limited historical avalanche events, this study set a maximum search radius, designating all snowfall values within a given range for each avalanche event as research targets and selecting the critical snowfall event with the highest return period as the one triggering the avalanche disaster. Finally, I-D points were plotted on a double logarithmic coordinate system, empirical parameters α and β were obtained through fitting, and the required I-D thresholds for the study were solved.

By setting the minimum no-snow time gap and maximum search radius for snowfall amounts near avalanche event locations and coupling extreme snowfall indices, this study identified a series of extreme snowfall events; this process is not only simple and time-efficient but also enables multiple calculations with different configurations to ultimately output I-D thresholds under multiple parameters. The performance of these thresholds was evaluated by backward-identifying snowfall events defined by I-D thresholds, and using a confusion matrix to assess all extreme snowfall events during the calibration period (1963–2018).

Since the confidence levels of correct predictions for all I-D thresholds are the same, and considering that issuing incorrect avalanche warnings could trigger a series of chain reactions and cause significant social impacts, this study only selects the results with the lowest FP as the final threshold outcomes for different extreme snowfall indices (as shown in

Table 3. I-D thresholds for different parameters of each extreme snowfall index.

Extreme Snowfall Indices	NSG (Days)	Search Radius (m)	I-D Threshold	FP	Option
SF1d	1	9000	I = 14.88 × D−2.81	226	Yes
	3	18,000	I = 10.68 × D−1.75	302	No
	1	27,000	I = 8.51 × D−2.95	270	No
	1	36,000	I = 27.63 × D−0.5	344	No
SF90p	3	9000	I = 5.79 × D−2.81	146	Yes
	7	18,000	I = 4.9 × D−2.95	375	No
	5	27,000	I = 8.51 × D−1.52	357	No
	5	36,000	I = 0.6 × D−1.93	384	No
SF95p	3	9000	I = 5.79 × D−2.81	332	Yes
	4	18,000	I = 0.4 × D−1.3	451	No
	7	27,000	I = 0.6 × D−1.25	637	No
	4	36,000	I = 0.42 × D−4.74	347	No
SF99p	3	9000	I = 10.68 × D−0.66	303	Yes
	5	18,000	I = 10.68 × D−0.45	514	No
	7	27,000	I = 14.88 × D−0.41	706	No
	3	36,000	I = 0.46 × D−3.09	892	No

). The parameters of snowfall thresholds in the study area range from 0.4 to 14.88 for α and from −3.09 to −0.41 for β.

3.2.3. Analysis of I-D Thresholds for Extreme Snowfall-Triggering Avalanches

Based on the previously calculated I-D thresholds with different parameter configurations for various extreme snowfall indices, this study analyzed 33 avalanche events in the study area. The calculated results were compared with actual snowfall and cumulative snowfall amounts to correct errors, after which the optimal parameters for snowfall intensity and duration were screened and plotted on a double logarithmic coordinate system. The resulting I-D thresholds for the four extreme snowfall indices—SF1d, SF90p, SF95p, and SF99p—are shown in

Table 4. I-D thresholds and their prediction performance for each extreme snowfall index.

Extreme Snowfall Indices	I-D Threshold	TP	FN	TN	FP	Se	Sp	PPP	NPP	E
SF1d	I = 14.88 × D−2.81	28	1	3	1	0.96	0.75	0.97	0.75	0.94
SF90	I = 5.79 × D−2.81	22	1	6	4	0.95	0.60	0.85	0.86	0.85
SF95	I = 5.79 × D−2.81	16	2	8	7	0.89	0.53	0.69	0.80	0.73
SF99	I = 10.68 × D−0.66	20	1	9	3	0.95	0.75	0.86	0.90	0.88

. The snowfall threshold parameter α ranges from 5.79 to 14.88 (average = 9.29), and parameter β ranges from −2.81 to −0.66 (average = −2.27). The table indicates that while the I-D thresholds identified in this study can distinguish between ordinary extreme snowfall events and those triggering avalanches, they cannot eliminate FP and false negatives.

Specifically, the threshold under the SF1d index show the best warning performance (E = 94%, Se = 96%), but its low specificity (Sp = 75%) indicated a risk of over-warning. The threshold under the SF90p index had good warning performance (E = 85%), but its specificity (Sp = 60%) was the lowest among the four thresholds, with FP = 4 revealing a significant tendency toward over-warning. Although the threshold under the SF95p index had the lowest overall accuracy (E = 73%), its negative predictive power (NPP = 80%) showed a good effect in excluding non-avalanche events. The threshold under the SF99p index achieved a better balance between specificity (Sp = 75%) and negative predictive value (NPP = 90%), making it suitable for risk-averse warning scenarios. The positive predictive powers (PPP = 69%–97%) of the models were significantly higher than the negative predictive values (NPP = 75%–90%), reflecting that the threshold system had better discriminative ability for avalanche events than non-avalanche events.As shown in Table 4, the I-D thresholds in this study exhibit high sensitivity, specificity, and efficiency (with efficiency ranging from 73% to 94%), indicating that the calculated I-D thresholds can effectively distinguish between avalanche events triggered by extreme snowfall and those triggered by non-extreme snowfall. Considering the four indices collectively, the threshold I = 9.29 × D^−2.27 can be applied to the entire study area.

3.2.4. Validation of the I-D Threshold Model

To ensure the reliability of the I-D thresholds determined in this study, the model was calibrated using snowfall and avalanche event data from 1961 to 2017 and validated with data from 2018 and 2020. The validation process was identical to the calibration process: a warning is triggered when the snowfall amount under the four extreme snowfall indices exceeds the corresponding I-D threshold. Using the same minimum no-snow time gap and search radius defined for threshold development, snowfall events under each index were identified and compared with the occurrence times of avalanche events to calculate TP, TN, FP, and FN.

As shown in

Table 5. I-D threshold verification table for each extreme snowfall index.

Extreme Snowfall Indices	I-D Threshold	TP	FN	TN	FP	Se	Sp	PPP	NPP	E
SF1	I = 14.88 × D−2.81	10	0	1	1	1	0.50	0.91	0.10	0.92
SF90	I = 5.79 × D−2.81	3	4	2	3	0.43	0.60	0.60	0.60	0.50
SF95	I = 5.79 × D−2.81	5	2	2	3	0.72	0.60	0.71	0.43	0.67
SF99	I = 10.68 × D−0.66	7	2	0	3	0.78	1	1	0.42	0.83

, the I-D thresholds determined in this study exhibited results similar to those of the calibration phase during the validation phase, effectively identifying avalanche events with a model accuracy ranging from 50% to 92%.

4. Discussion

4.1. Relationship Between Extreme Snowfall Indices and Climate Change

Partial correlation is a statistical method used to measure the relationship between two variables while controlling for the effects of one or more other variables, helping to understand the direct relationship between them. In this study, which focuses on changes in extreme snowfall indices under climate change, the core variables analyzed are extreme snowfall indices, precipitation, and air temperature. The key idea is that when examining the relationship between any two variables, the influence of the third variable must be excluded (controlled for).In the analysis of raster data, each pixel can be treated as an observation point. This study employs a pixel-by-pixel analysis strategy for partial correlation calculations, meaning that analyses are conducted on a pixel-to-pixel basis rather than treating the entire image as a single entity.

Additionally, to explain the causes of the correlation distribution between extreme snowfall indices, air temperature, and precipitation in the study area, this study conducted MK trend tests on precipitation and air temperature during the same period.

The results of precipitation analysis are shown in Figure 6. The extreme snowfall indices exhibit significant positive correlations with corresponding precipitation in the eastern Himalayas, and all four indices also show significant positive correlations in the northeastern part of the study area. This is primarily because the eastern Himalayas, located in the study area, act as the windward zone for warm and moist Indian Ocean maritime airflows, making them more prone to precipitation; the northeastern study area, part of the Western Sichuan Plateau, is influenced by the East Asian monsoon, leading to strong associations between extreme snowfall indices and precipitation in these two regions. Notably, SF1d shows a negative correlation in the northwestern study area compared to the other three indices. This is because the SF1d index measures the maximum snowfall during the snow season, which places less emphasis on cumulative snowfall, whereas the other three indices are calculated based on cumulative amounts at specific percentiles. Additionally, the southern and southeastern study areas, characterized by subtropical climates, have precipitation concentrated in summer with almost no snowfall events, resulting in strong negative correlations here. Combined with precipitation trend tests (as shown in Figure 7), precipitation in the study area shows a decreasing trend only in parts of the south eastern Himalayas, while increasing trends are observed elsewhere. This is primarily due to enhanced disturbance intensity of weather systems in the region under climate warming. Over the past four decades, the diurnal temperature range in the northern Himalayas has narrowed due to faster nighttime warming compared to daytime warming, whereas the southern Himalayas showed negligible nighttime warming trends. This divergence in temperature trends has amplified extreme climate indices in the north, particularly in winter [21,63].

The results of air temperature analysis are shown in Figure 8. The extreme snowfall indices exhibit significant positive correlations with corresponding air temperatures in the eastern study area (near the Hengduan Mountains and Nyainqêntanglha Mountains) and the eastern Himalayas. This is primarily due to the massive barrier formed by these three major mountain ranges, which creates a moisture channel. When warm, moist air from the Indian Ocean is forced to rise over the mountains, it generates orographic snowfall during the snow season. As temperatures increase, the atmosphere’s water-carrying capacity strengthens, further intensifying convection and leading to increased snowfall. In contrast, the western study area shows significant negative correlations between extreme snowfall indices and air temperature, mainly due to low water vapor content and inherently low snowfall in this region. Combining the air temperature trend tests (as shown in Figure 9), temperatures in the eastern and southern study areas exhibit decreasing trends, while a distinct increasing trend is observed near the Hengduan Mountains. Rising temperatures accelerate the melting of glacial snow cover, increasing local water vapor content and maintaining high relative humidity, which in turn promotes more frequent extreme snowfall events [64]. Overall, these trends will lead to accelerated snow and ice melt in southeastern Tibet, accompanied by more frequent extreme snowfall in localized areas [50,65].

Extreme snowfall indices exhibit significant positive correlations with both precipitation and air temperature at historical avalanche occurrence sites. Climate warming and changes in snowfall amount significantly alter the hydrological cycle in the study area, leading to earlier snowmelt and faster snowmelt rates, which highly increases the likelihood of avalanches, ice collapses, and snowmelt floods [51,66,67]. Notably, in recent years, rain-snow events have increased and extended toward high-altitude, snow-rich, and steep-slope areas. This process not only couples with extreme rain-snow and warming events but also involves the transport of substantial heat, accelerating the imbalance of snow mechanical fields in alpine regions and destabilizing the force equilibrium [68,69,70]. As climate warming continues, avalanche risks in the study area are expected to intensify.

4.2. The Relationship Between I-D Thresholds and Terrain Environmental Backgrounds

Statistical analysis of the thresholds in the study area shows that the snowfall threshold parameter α for the four extreme snowfall indices ranges from 5.79 to 14.88 (mean = 9.29), and parameter β ranges from −2.81 to −0.66 (mean = −2.27). The parameters α and β of the I-D thresholds in this study exhibit a certain correlation (R² = 0.17): the larger the intercept on the y-axis (the larger α) in the logarithmic coordinate system, the steeper the slope (the more negative β). Since α reflects the proportional relationship between snowfall intensity and duration, while β determines the rate at which snowfall intensity decreases as duration increases, we can conclude that: in regions with threshold curves featuring larger α and more negative β values, short-duration intense snowfall rarely triggers avalanches, but long-duration moderate-intensity cumulative extreme snowfall events may still exceed the threshold and induce avalanches. Given that the overall threshold calculated from the four extreme snowfall indices is I = 9.29 × D^−2.27, it can be inferred that avalanches triggered by extreme snowfall in southeastern Tibet are primarily caused by cumulative snowfall rather than short-duration intense daily snowfall. Previous studies have shown that topographic factors play a significant role in avalanche occurrence. To further analyze how physical environmental factors influence the snowfall threshold formula, this study calculated the snowfall “I” and “D” for all pixels from 1951 to 2020 based on the overall I-D thresholds of the four extreme snowfall indices in the study area, and discussed their correlations with topographic and physical environmental variables.

As shown in

Table 6. Correlation Statistics Table of I, D and Topographic Factors.

	Extreme Snowfall Indices	Elevation	Terrain Relief	Surface Cutting Depth	Slope	Curvature	Elevation Coefficient of Variation	Terrain Roughness
I	SF1d	−0.13	0.16	0.16	0.14	0.07	0.07	0.14
	SF90p	−0.12	0.15	0.15	0.13	0.06	0.06	0.13
	SF95p	−0.11	0.12	0.12	0.11	0.04	0.05	0.12
	SF99p	−0.10	0.10	0.10	0.09	0.03	0.04	0.10
D	SF1d	0.04	−0.02	−0.02	−0.02	−0.02	−0.04	−0.01
	SF90p	0.04	−0.02	−0.02	−0.02	−0.02	−0.04	−0.01
	SF95p	0.05	−0.05	−0.05	−0.04	−0.04	−0.05	−0.04
	SF99p	0.05	−0.06	−0.05	−0.05	−0.04	−0.05	−0.04

, the relationships between “I” and the three topographic indices—terrain relief, surface cutting depth, and slope gradient—decrease gradually from SF1d to SF99p. This is primarily because extreme snowfall events in the study area are dominated by large-scale weather systems, weakening the role of local orographic lifting. Additionally, the correlations between curvature/elevation coefficient of variation and I also diminish from SF1d to SF99p. Since curvature and elevation coefficient of variation reflect the degree of surface deformation, this trend indicates that avalanches triggered by extreme snowfall in the study area are less influenced by micro-topographic features and more governed by macro-scale disaster-inducing environments. Notably, the elevation factor exhibits a negative correlation with “I”, mainly due to the attenuation of water vapor as it rises with increasing altitude.

Among all extreme snowfall indices, the duration of extreme snowfall shows no significant correlations with topographic factors (correlation coefficients are close to 0), and this pattern remains stable across all extreme percentile events. This indicates that the influence of topographic characteristics on snowfall duration in the study area is negligible.

In southeastern Tibet, the duration of extreme snowfall events is primarily governed by the dynamic characteristics of large-scale meteorological systems. While topographic characteristics influence snowfall intensity near the surface, they have limited ability to regulate the timing of meteorological systems over large areas. Therefore, under the broader context of climate warming, greater attention should be paid to the intensity of extreme snowfall events in southeastern Tibet to prevent avalanche disasters triggered by such extreme events.

5. Conclusions

Although many studies on extreme snowfall and avalanche disasters have been conducted in southeastern Tibet, the coupling relationship between the two has received little attention. Understanding the spatiotemporal variation of extreme snowfall in southeastern Tibet and its relationship with climate change, and establishing I-D thresholds to explore avalanche disasters triggered by extreme snowfall events, are of critical importance for early warning research. This serves as a prerequisite for government authorities and policymakers to address climate change risks and challenges and implement pre-disaster prevention measures. Here, we determined the I-D thresholds for avalanche events triggered by extreme snowfall in southeastern Tibet using the newly released China’s ground meteorological element dataset combined with historical avalanche disaster records. This study focuses on early warning research for avalanche disasters in alpine regions under climate change. The main conclusions of this paper are as follows.

(1)

Collectively, an extreme snowfall event in southeastern Tibet can be identified when either the daily snowfall exceeds 12.38 mm (the regional average of the SF1d index) or the cumulative snowfall over the preceding 30 days exceeds 64.85 mm (the regional average of the SF90p, SF95p, and SF99p indices).

(2)

In the I-D thresholds determined in this study, the parameter α ranges from 5.79 to 14.88 (mean = 9.29), and the parameter β ranges from −2.81 to −0.66 (mean = −2.27). The four defined thresholds effectively distinguish between avalanche events triggered by extreme snowfall and ordinary extreme events, with the corresponding I-D threshold forecasts achieving an accuracy of 50% to 92% and a Percentage of PPP ranging from 60% to 100%. The threshold I = 9.29 × D^−2.27 can be used to predict avalanches triggered by extreme snowfall across the entire study area.

(3)

This study conducted partial correlation analysis based on extreme snowfall indices and historical precipitation/temperature data. The results reveal that extreme snowfall indices exhibit positive correlations with precipitation in the southern and northwestern parts of southeastern Tibet, but show negative correlations near the Hengduan Mountains. Regarding temperature, significant positive correlations are observed in the eastern region of southeastern Tibet, while negative correlations prevail in the western region.

(4)

In southeastern Tibet, the duration of extreme snowfall events is primarily determined by the dynamic characteristics of large-scale synoptic systems. While topographic characteristics influence near-surface snowfall intensity, their influence on the temporal evolution and progression over large areas is limited.