Integrating susceptibility and effective rainfall thresholds for regional landslide warning in Wushan

This study delves into the early warning system for regional rainfall-induced landslide hazards. By integrating landslide susceptibility mapping and critical rainfall thresholds, a spatiotemporal probabilistic prediction framework for landslide hazards is established. Taking Wushan County as the study area, the authors utilised 313 recorded rainfall-triggered landslide events and meteorological rainfall data spanning from 2010 to 2024. Initially, a landslide susceptibility prediction model was constructed using the Random Forest algorithm, with model accuracy validated by the area under the curve (AUC) metric (AUC = 0.85). Subsequently, critical rainfall thresholds were determined through a comprehensive analysis of antecedent effective rainfall, employing three models: effective rainfall intensity-duration (EI-D), cumulative effective rainfall-duration (EE-D), and cumulative effective rainfall-effective rainfall intensity (EE-EI). Finally, a landslide hazard warning map was generated by spatially superimposing susceptibility classifications and rainfall threshold exceedance probabilities, accompanied by corresponding disaster prevention strategies. The results demonstrate that this integrated methodology enhances the precision of regional landslide early warnings, as supported by quantitative validation metrics (probability of detection = 0.76, true skill statistic = 0.44), providing a scientifically robust decision making framework for natural resource management authorities in disaster risk mitigation.