Risk assessment of landslide and rockfall hazards in hilly region of southwestern China: a case study of Qijiang, Wuxi and Chishui

Geological events that cause casualties and huge property losses frequently occur in the hilly region of Southwest China. Especially for economically underdeveloped mountainous cities, the vulnerability of social and ecological systems greatly affects urban development. Therefore, it is necessary to conduct multi-hazard exposure research in mountain areas. The importance of this study is that the five machine learning algorithms: Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM) and Categorical Boosting (CatBoost) are first used to develop multi-hazard (specifically landslides and rockfalls) susceptibility map in three typical hilly areas in the Southwest China (Qijiang, Chuishui and Wuxi). The performances of the models are then compared using the F1-Score and the true skill score (TSS) respectively. Finally, the multi-hazard exposure maps of each district are developed by combining multi-hazard susceptibility and exposure factors. According to the results of this study, it can be found that the LightGBM and CatBoost algorithms are suitable for the development of landslide and rockfall hazard susceptibility maps in the hilly region of southwest China, and this type of gradient boosting algorithms have good development prospects. In addition, in future research on geological hazards in this hilly area, more attention should be paid to conditioning factors such as rainfall, altitude, slope, distance from the faults, and lithology. The results of this study can provide reference for the screening of landslide and rockfall conditioning factors in such mountainous areas, and can also provide certain support for the planning of the social ecosystem in mountainous areas and the formulation of natural hazard mitigation measures.