Rainfall-conditioned landslide susceptibility mapping using an FR–CART hybrid framework in southern Thailand

Rainfall-induced landslides in the steep tropical terrain of southern Thailand pose significant challenges to communities, infrastructure, and regional development. This study presents a rainfall-responsive landslide susceptibility framework that integrates static spatial conditioning factors with dynamic rainfall information. A Frequency Ratio (FR) model was first applied to generate a static Landslide Susceptibility Index (LSI) using 17 geospatial variables representing topographic, geological, hydrological, and anthropogenic characteristics. To incorporate the temporal influence of precipitation, a Classification and Regression Tree (CART) model was employed using short-term rainfall metrics, including daily, 3-day, 7-day, and 15-day cumulative rainfall, to derive rainfall-triggering thresholds that conditionally refine existing susceptibility classes rather than generate a new susceptibility index or estimate landslide hazard probability. Model validation shows that the standalone FR model achieved an Area Under the Curve (AUC) of 0.74, while the integrated FR–CART framework improved performance to an AUC of 0.82. Among the rainfall indicators, the 15-day cumulative rainfall index exhibited the strongest association with landslide occurrences, emphasizing the role of antecedent moisture conditions. In this study, rainfall-conditioned susceptibility is explicitly defined as a rule-based upward adjustment of static LSI classes (e.g., from High to Very High) under real-time or forecasted rainfall scenarios. This conditioning process does not constitute the derivation of a new susceptibility index, nor does it imply probabilistic landslide hazard or risk assessment. The resulting rainfall-conditioned susceptibility outputs support susceptibility-informed interpretation within early-warning applications for monsoon-prone regions of southern Thailand. • FR–CART framework is developed to map rainfall-conditioned landslide susceptibility. • Daily and cumulative rainfall indices (3-, 7-, and 15-day) refine susceptibility classes. • Antecedent vs. triggering rainfall roles are explicitly differentiated and validated.