Trend and causes of snow cover in the headwaters of the Yangtze River from 1980 to 2020

The Qinghai-Tibet Plateau, a region highly sensitive to global climate change, hosts the headwaters of the Yangtze River. The dynamics of its cryosphere, particularly snow cover, are critical to regional water resources and ecological security. Given the observed high variability of snow cover under climate warming, a quantitative analysis of its response mechanisms to temperature and precipitation anomalies at the sub-basin scale is essential for accurate hydrological projections. To better reveal the spatial heterogeneity and climate-driven mechanisms of snow cover change in the source area of the Yangtze River, the region was divided into four sub-regions based on sub-basins: Tuotuo River, Dangqu, Chumar River, and other sub-regions, and the variation characteristics of snow cover indices in each sub-region and their relationships with climatic factors were explored. Based on snow cover, temperature, and precipitation data from 1980 to 2020, the spatiotemporal variation characteristics of snow cover in sub-regions were analyzed using the Mann-Kendall test to detect monotonic trends and the ESMD method to identify nonlinear and non-stationary oscillations in the time series. The relationship between snow cover and climatic factors was analyzed using the partial correlation method to isolate the individual influence of temperature and precipitation by controlling for the other. To move beyond traditional statistical correlations and explore the physical mechanisms, the XGBoost-SHAP model was employed, leveraging its strong predictive performance and the SHAP framework’s interpretability to reveal nonlinear thresholds and interaction effects. The results showed that snow cover changes exhibited significant spatial heterogeneity: during the snow accumulation period in the source area of the Yangtze River, average snow depth and snow water equivalent showed an increasing trend in southeastern and high-altitude peripheral areas, while other areas showed a decreasing trend. Snow depth and snow water equivalent increased in high-altitude areas of the Tuotuo River, the margins of Dangqu, the northwest of the Chumar River, and small southeastern parts of other regions, while most other areas generally decreased, and snow cover extent was mostly declining. This spatial pattern can be attributed to the fact that high-altitude and southeastern peripheral regions, with persistently lower temperatures, are less susceptible to warming-induced melt, allowing potential precipitation increases to dominate snow accumulation trends, whereas lower-lying and central areas are more vulnerable to temperature-driven ablation. Correlation analysis showed that snow cover changes in the four sub-regions were not consistent with changes in precipitation and temperature: overall, snow cover was positively correlated with precipitation; snow depth and snow water equivalent were negatively correlated with air temperature in the Tuotuo River and Dangqu, while a few areas in the northwest of the Chumar River and southeastern parts of other regions showed positive correlations, and most other regions showed negative correlations; snow cover extent was mostly negatively correlated with temperature, indicating spatial differences in the impact of climate warming and humidification on snow cover in cold regions. The XGBoost-SHAP model shows that historical precipitation conditions have a greater effect on current snow depth than instantaneous precipitation. Precipitation in low-temperature ranges contributes positively to snow cover, but as temperature rises, this positive contribution weakens until it disappears, with SHAP values approaching 0 at −15 ℃ to −10 ℃. Thereafter, the interaction between temperature and precipitation leads to snow reduction, and even increased precipitation cannot further increase snow. At −5 ℃ to 0 ℃, snow change reaches another critical point, beyond which further temperature increases no longer significantly intensify snow reduction. The stronger influence of historical precipitation indicates that snow has a “memory effect,” whereby prolonged wet or dry conditions affect subsequent accumulation or melting. The nonlinear relationship between temperature and snow characterizes phase change processes more accurately than linear assumptions, helping to improve traditional hydrological models, enhance snow prediction accuracy, and better assess the impacts of climate change on the cryosphere.