Susceptibility assessment of freeze-thaw erosion induced debris flow using random forest, Eastern Tibetan Plateau

Machine learning algorithms have shown excellent results in susceptibility assessment of debris flow hazards in different areas. These results depend on selecting control factors that align with the actual conditions of the study area. Due to the hazard’s formation conditions, alpine experience significantly advanced freeze-thaw erosion, yet current research seldom considers this as a controlling factor. Consequently, this study selects the northern area of the Gongjue Basin in the Eastern Tibetan Plateau, where the freeze-thaw erosion plays a controlled driving force for debris flow. The primary emphasis is on investigating the influence of freeze-thaw erosion on the debris flow susceptibility assessment model. To this end, a statistical analysis was performed on the frequency and overall performance of control factors chosen in relevant literature on debris flow susceptibility assessment using machine learning. Control factors with high frequency and performance were selected from the perspectives of material sources, dynamic conditions, and hydrological factors, leading to an optimized selection strategy, and the Random Forest Algorithm was employed for susceptibility assessment (No Freeze-thaw erosion model, NFEM). Subsequently, the freeze-thaw erosion index, a new control factor gauging the intensity of freeze-thaw erosion in the study area, was incorporated, and the susceptibility assessment was also conducted using the Random Forest Algorithm (Freeze-thaw erosion model, FEM). The results show that FEM improved accuracy by 0.457 and AUC by 0.0541 compared to NFEM, indicating enhanced predictive performance. Nevertheless, when comparing watershed samples, both models demonstrated limited predictive power. In terms of susceptibility outcomes, FEM yielded more precise assessment results based on the available data.