广东省韶关市"4·20"极端降雨诱发滑坡发育特征及其主控因子分析OA

[Objective]On April 20,2024,an extreme rainstorm event occurred in Shaoguan City,Guangdong Province,South China.The 24-hour rainfall in Jiangwan Town reached a historical maximum value of 206 mm,which triggered a large number of landslides.These hazards caused serious damage to residential buildings,road blockages,and widespread social concern.Timely acquisition of landslide inventories,understanding their development distribution patterns,and identifying main controlling factors are crucial for post-disaster emergency response and reconstruction.[Methods]Based on high-resolution Planet remote sensing images,the normalized difference vegetation index(NDVI)difference method combined with terrain correction and morphological post-processing was adopted to automatically extract landslide areas.A complete landslide inventory was compiled.Meanwhile,the spatial distribution patterns and causal factors of the landslides were analyzed,combined with topographic,rainfall,and geological environmental factors.The SHapley additive exPlanations(SHAP)method was applied to quantitatively identify the dominant controlling factors of landslide occurrence.[Results]The results showed that the extreme rainfall event triggered 1 426 landslides in total,with a total area of 4.56 km2,mainly small to medium scale in size.Landslides predominantly clustered along rivers in a Northeast-Southwest orientation,forming belt-like distributions,with notable group-occurring effects.Spatial statistical analysis revealed that landslides were intensively distributed in slope areas with elevations of 200-300 m and slopes of 20°-30°.Four machine learning models,namely logistic regression(LR),support vector machine(SVM),random forest(RF),and eXtreme gradient boosting(XGBoost),were used to evaluate the accuracy of landslide susceptibility mapping.The results showed that random forest and eXtreme gradient boosting models performed best,identifying highly susceptible areas mainly on mountain slopes on both sides of the river valleys.Through quantitative analysis of the main controlling factors of landslides using the SHAP method,it was found that elevation,rainfall,profile curvature,and topographic wetness index(TWI)were the key driving factors for landslide occurrence.[Conclusion]This study provides reliable technical approaches,refined data support,and practical reference for rapid identification of rainfall-induced group-occurring landslides and machine learning-based susceptibility evaluation in similar mountainous areas.