Response of coseismic landslide susceptibility to the temporal scales and interactions of deformation and rainfall

This study examines how the time scales of rainfall and deformation and their interaction affect coseismic landslide susceptibility assessment. Based on 5007 landslides triggered by the 5 September 2022, Luding earthquake, we integrated 18 causative factors, combining static causative factors with rainfall and deformation derived at the monthly scale and the annual scale. We trained ensemble decision tree classifiers with five-fold cross-validation and used Shapley additive explanations (SHAP) to quantify main and interaction effects. Incorporating rainfall and deformation data solely from the month preceding the earthquake outperformed models relying on annual-scale data. In this case study, the coefficient of variation of elevation, distance to faults, and rainfall was dominant. The contribution of short-term deformation is substantial, whereas the contribution of annual deformation rate is below 1%. The interaction analysis shows that moderate downward deformation combined with heavy rainfall markedly increases susceptibility, while under extreme rainfall the additional influence of deformation weakens. Our work provides an interpretable and reusable workflow for rapid coseismic landslide susceptibility assessment. Furthermore, the experimental results highlight the potential significance of monthly rainfall–deformation coupling, which is critical for accurate evaluation.