查看更多>>摘要:With the intensification of extreme climate change globally, typhoon-induced landslides have become a serious threat to people's property and lives. Although some studies have identified wind, rainfall, and vegetation as contributing factors to landslides, challenges still remain in incorporating these factors into susceptibility mapping. This study aims to establish a foundation for selecting key factors that affect the susceptibility of typhoon-induced landslides. Specifically, this study innovatively employs the interpretability of machine learning, including Partial Dependence Plots (PDP) and Shapley values, to analyze the correlation between landslides and factors, and compares this with qualitative analysis. It assesses the significance of static factors, incorporates typhoon-related factors, and examines their collective impact. The critical static factors of typhoon-induced landslides identified by Shapley values, including elevation, normalized difference vegetation index (NDVI), road, slope, land use, river, aspect of slope, and vegetation. The k-fold cross-validation was utilized for computation of average descent accuracy, and facilitated the selection of optimal combination of dynamic and static factors, the effectiveness of which was confirmed through ROC (Receiver Operating Characteristic curve). The optimal combination of dynamic factors was determined by average descent accuracy: maximum sustained wind speed, 24 h pre-rainfall, distance from the landslides to the typhoon center and wind circle radius of near gale. Through rigorous verification, it was determined that optimizing factor combinations could increase the accuracy of evaluations by 1.5 %-3.5 %, thereby enhancing both the precision and reliability of susceptibility assessments for typhoon-induced landslides.