GIS-based landslide susceptibility assessment using a random forest model: a case study of Ganzi and Aba, China

Landslides pose a significant threat to human life and regional development in Ganzi and Aba Prefectures due to their complex topography. Therefore, this paper aims to enhance landslide hazard prevention by conducting assessment and landslide susceptibility mapping (LSM) using four models: random forest (RF), information value (IV), logistic regression (LR), and a combined IV‒LR approach. These models facilitate the generation of susceptibility maps, which are crucial for forecasting potential landslide zones in the area. Based on geological data and landslide information, 13 conditioning factors were selected to evaluate landslide susceptibility. The results revealing the areas with the highest susceptibility are primarily located in Wenchuan County, Mao County, Jiuzhaigou County, and Luding County; elevation and distance to roads play a critical role in determining the possibility of landslide events. Model performance was validated using receiver operating characteristic (ROC) curve analysis, which demonstrated that the RF model achieved superior predictive accuracy (AUC = 0.972) compared to other methods. The findings provide actionable insights for landslide risk management and mitigation strategies in the region.