Climate-driven thaw slump susceptibility on the Qinghai–Tibet plateau using geographically explainable machine learning

Retrogressive thaw slumps (RTS) are a thermokarst landform increasingly found on the Qinghai–Tibet Plateau, characterized by progressive headwall retreat and exceptional sensitivity to thermal change. However, static assessments fail to spatially interpret their driving mechanisms. This study examines current and future susceptibility and shifts in driving factors. A multi-model ensemble with area under the receiver operating characteristic curve values above 0.96 was developed from an inventory of 3,693 RTSs, and Shapley additive explanations together with GeoShapley were used to attribute model predictions to environmental drivers and their spatially varying effects. The models indicate that by mid-century (2041–2060), the high-susceptibility area will increase from 5.76 % to 10.80 % to 6.01 %–9.76 % under SSP1-2.6, 8.01 %–11.64 % under SSP3-7.0, and 8.41 %–10.34 % under SSP5-8.5, with the northeastern plateau a primary zone of heightened risk. According to the GeoShapley framework, thawing degree days will increase in influence along the southeastern margins, freezing degree days will intensify their stabilizing effect in the northwest, and the influence of precipitation will stabilize or decrease across most of the plateau. These findings provide mechanistic insights and support targeted risk management strategies.