Landslide Susceptibility Prediction using Random Forest model based on remote sensing and GIS

Landslide susceptibility prediction (LSP) modeling provides an effective means for exploring the spatial distribution patterns of landslides in the study area. However, challenges such as the difficulty in achieving precise and efficient slope unit division and the complexity of selecting appropriate prediction models remain. To address these issues, this paper focuses on colluvial landslides in Huangtan Town, Zhejiang Province. Firstly, a Multi-scale Segmentation (MSS) algorithm is utilized to partition Dingnan County into 100,000 slope units, thereby ensuring the efficiency and accuracy of the slope unit division. Secondly, 18 environmental factors are selected as input variables for the LSP model. Subsequently, landslide susceptibility is predicted using the Random Forest (RF) model. Finally, the LSP results are evaluated through a comprehensive approach that includes the AUC value and characteristics of the landslide susceptibility index. The research outcomes indicate that the MSS algorithm can efficiently and accurately divide slope units. The landslide susceptibility in the study area is well predicted, demonstrating the superiority of the RF model in terms of prediction accuracy and providing more reliable technical support for LSP.