ABSTRACT
Landslides have emerged as a critical disaster type due to the increasing frequency of extreme rainfall events attributed to climate change. In recent years, research on landslide analysis and slope stability has grown. The most common research methods in landslide studies fall into two categories: numerical modeling and machine learning models. Both approaches hold potential for contributing to landslide disaster management, although research combining these methods remains limited. Therefore, the primary objective of this study is to harness the strengths of both approaches, which involves the establishment of landslide susceptibility maps and the subsequent development of probability-based numerical model, providing valuable insights for disaster prevention on high-risk areas. The results from this preliminary phase indicate that the categorization boundaries in the machine learning-based landslide susceptibility maps are primarily influenced by the proportion of historical landslide areas. However, given that this factor does not represent the main characteristics of the target objects, it should be excluded in future research. The forthcoming stages will emphasize the examination of slope and dip slope area proportions while incorporating additional landslide-related geomorphic and hydrological factors and enhancing data resolution to improve predictive model accuracy. In the context of probability-based numerical model, the research highlights the sensitivity of cohesion and the internal friction angle to the probability of slope system failure. Real-world slope and soil parameter experiments are planned for the next phase to establish a more realistic probability-based numerical model.
