Abstract: To address the difficulty of quantifying future climate impacts in static landslide assessments,this study proposes a dynamic evaluation method integrating multi-source factors with climate scenarios.Taking the eastern section of the Qilian Mountains as the study area,we firstly construct an indicator system by integrating multi-source data such as topography and geology.Multiple machine learning models are then compared to select the optimal algorithm.Finally,we couple scenarios from the sixth coupled model intercomparison project (CMIP6) (SSP1-2.6,SSP2-4.5,SSP5-8.5) to predict landslide susceptibility in different future periods.The results show that the random forest model performs best.In the baseline period (1991—2020),medium-high susceptibility areas were concentrated in low-slope valleys.Under the SSP5-8.5 scenario,the proportion of high and very high susceptibility areas is projected to increase to 25.4% by the end of the 21 century,whereas SSP1-2.6 could limit it to around 20%.This study overcomes the limitations of static assessments and reveals that,under high-emission pathways,climate change will significantly exacerbate landslide risks in the eastern Qilian Mountains.Emission reduction is therefore crucial.

Key words: eastern Qilian Mountains, landslide susceptibility, random forest, CMIP6, climate change