Role of Topo-Climatic Factors in Landslide Susceptibility Mapping: A Machine Learning Perspective

With the rapid growth of machine learning (ML) applications in landslide susceptibility mapping (LSM), many studies have relied on large and diverse datasets to improve predictive performance. However, in data-scarce regions, model transferability becomes challenging, and not all input variables contribute meaningful information to landslide prediction, leading to the well-known problem of “garbage in, garbage out”. Therefore, careful selection of physically meaningful predictors is essential. In this study, we develop a Random Forest–based framework for landslide mapping using topographic and climatic factors only as inputs. Notably, the model incorporates effective rainfall (ER), derived from a process-based hydrological model, and a recently proposed topographic metric, topographic source density (D_ts), that were not used in previous LSM studies. Results demonstrate that incorporating these predictors substantially improves model performance, achieving a receiver operating characteristic area under the curve (ROC-AUC) of 0.972. Overall, the findings suggest that a physically meaningful machine learning approach can reliably predict regional landslide susceptibility.