Landslide susceptibility assessment under future seismic and precipitation scenarios: a case study of the 2014 Mw 6.2 Ludian earthquake zone, Yunnan, China

The combined effects of earthquakes and precipitation will trigger more and larger landslides. However, how to quantify the probability of potential landslide occurrence under combined effects of the future seismic scenarios and precipitation changes remains a formidable challenge. In this study, the Ludian earthquake zone is selected as the research area. Multi-scenario seismic ground motion and projected future precipitation patterns are simulated. The eXtreme Gradient Boosting model optimized through Bayesian optimization (BO-XGB), interpreted with the SHAP model, is developed to evaluate landslide susceptibility (LS). The results show that the BO-XGB model has a high accuracy with an AUC value of 0.907; The scenario coupled rare ground motion and SSP5-8.5 (2030s) has the largest proportion of very high level LS, with a value of 41.12%. The dominant factors influencing LS are peak ground acceleration (PGA), distance from road, and precipitation in wettest month (PWM). The research framework provides a corresponding scientific theoretical basis for predicting the probability of landslides triggered by earthquakes and precipitation in the mountainous areas of southwestern China under various scenarios in the future.