Thermokarst hazards susceptibility mapping on the Qinghai-Tibet plateau based on a collaborative IV-machine learning architecture

Accelerating climate change intensifies permafrost degradation and thermokarst hazards on the Qinghai-Tibet Plateau (QTP), threatening ecosystems and infrastructure. To address the scarcity of susceptibility studies, we integrated the Information Value (IV) method with machine learning models to assess thaw slump and thermokarst lake susceptibility along the Qinghai-Tibet Highway (QTH). The results revealed that soil type and rainfall dominated thaw slump susceptibility, while slope and soil type were the primary controls for thermokarst lake formation. IV-optimized models, particularly IV-RF (with AUC of 0.9793 and 0.9885 for thaw slumps and thermokarst lakes, respectively), outperformed single models. High to very-high susceptibility areas for thaw slumps and thermokarst lakes are predominantly located in the central and northern regions of the study area, accounting for 20.2% and 27%, respectively. InSAR-derived subsidence rates correlated significantly with susceptibility levels, and high-subsidence areas spatially coincide with high-susceptibility zones. This framework provides a quantitative basis for prioritizing thermokarst mitigation in vulnerable alpine corridors.