Physically based data-driven analysis for large-scale investigation of the July 2025 rainfall-induced landslide in Sancheong, South Korea

Abstract

The July 19, 2025, landslide event in Sancheong County, South Korea, was triggered by extreme monsoonal rainfall, causing widespread landslides and severe damages. This study investigates the spatiotemporal characteristics, conditioning factors, and triggering mechanisms of the event using a combination of remote sensing data, field surveys, rainfall analysis, and physically based data-driven modeling. A detailed landslide inventory was constructed and validated through field verification, comprising 568 initiation locations. Rainfall analysis revealed cumulative rainfalls (CRs) of 498–619 mm over a 53–57-h period, with a mean return period of approximately 145 years, confirming the exceptional intensity of the event. The I-D threshold comparison indicated that the 2025 event far exceeded the national rainfall-landslide thresholds, while 3-day antecedent rainfall (12–19 mm) has minimal influence. Landslide susceptibility was modeled using a genetic algorithm-optimized extreme gradient boosting model, integrated with SHapley Additive exPlanations to enhance interpretability. The model achieved excellent performance (AUC = 98.4% for training, AUC = 96.2% for validation, AUC = 98.5% for testing) and identified slope, geology, digital elevation model, effective accumulated rainfall (EAR) for the day of landslide initiation (EAR0), EAR for 1 day prior to the landslide initiation (EAR1), and CR as the dominant influential factors. Failures were concentrated on moderate to steep slopes (25–35°), within Precambrian metamorphic and Cretaceous lithologies, and at mid-elevation zones (240–500 m). Rainfall thresholds of EAR0 > 115 mm and CR > 450 mm were found critical for slope failures. These findings demonstrate that landslide initiation in Sancheong is primarily governed by the interaction between lithological weakness, steep topography, and extreme rainfall, providing a quantitative foundation for regional hazard mapping and early-warning systems under intensifying monsoon conditions.