Assessment of Landslide Susceptibility and Risk in Tengchong City, Southwestern China Using Machine Learning and the Analytic Hierarchy Process

Southwestern China, characterized by highly undulating terrain and mountainous areas, faces frequent landslide disasters. However, previous studies in this region mostly neglected the role of extreme rainfall in landslide susceptibility assessment and the socio-economic risks threatened by landslides. To address these gaps, this study integrated 688 recorded landslides for Tengchong City in the southwest of China and 10 influencing factors (topography, lithology, climate, vegetation, and human activities), particularly two extreme precipitation indices of maximum consecutive 5 day precipitation (Rx5day) and maximum length of wet spell (CWD). These influencing factors were selected after ensuring variable independence via multicollinearity analysis. Four machine learning models were then built for landslide susceptibility assessment. The Random Forest model performed the best with an Area Under Curve (AUC) of 0.88 and identified elevation, normalized difference vegetation index (NDVI), lithology, and CWD as the four most important influencing factors. Landslides in Tengchong are concentrated in areas with low NDVI (<0.57), indicating increased vegetation cover might reduce landslide frequency. Landslide risk was further quantified via the Analytic Hierarchy Process (AHP) by integrating multiple socio-economic factors. High-risk zones were pinpointed in central-southern Tengchong (e.g., Heshun and Tuantian townships) due to their high social exposure and vulnerability. Overall, this study highlights extreme rainfall and vegetation as key modifiers of landslide susceptibility and identifies the regions with high landslide risk, which provides targeted scientific support for regional early-warning systems and risk management.