Landslide Hazard Warning Based on Information Value- Random Forest Model

Predominantly triggered by rainfall, landslides stand as a significant natural hazard. Extensive research has focused on determining critical rainfall thresholds for landslide occurrence. However, a crucial gap remains in identifying specific locations vulnerable to these events. This study addresses this gap by integrating landslide susceptibility evaluation and rainfall threshold analysis under probabilistic scenarios to develop a comprehensive warning framework. The research was conducted in Pingwu County, Mianyang City, employing a multi-step methodology. Initially, the study area was segmented into slope units, whereafter landslide susceptibility was assessed through an innovative integration of the information value model and random forest algorithm. Subsequently, effective antecedent rainfall calculations were performed to establish probability-based rainfall threshold models. The final phase involved synthesizing susceptibility and rainfall threshold analyses to generate a systematic hazard warning framework. The research findings unveiled a series of significant outcomes, which are as follows: First, the coupled model outperformed single methods in susceptibility assessment compared to individual methods. Second, through in-depth statistical analysis and field data calibration, this study determined the optimal parameters for the study area through statistical calibration: a three-day rainfall duration and an effective rainfall coefficient of 0.7, which most accurately characterize rainfall-induced landslide patterns in Pingwu County. Third, among the evaluated approaches, the EE-D model exhibited the highest predictive accuracy. Based on these findings, a five-level warning system was developed, incorporating both rainfall conditions and regional characteristics. Validation of the warning system using 16 historical landslide events showed a clear correlation between warning levels and actual occurrences, with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$12.5 \%, 37.5 \%$</tex>, and 50 % of landslides corresponding to Level 3, Level 4, and Level 5 warning zones, respectively. The warning zones showed significant dynamic responses to rainfall variations, consistent with field observations, confirming the model's effectiveness in reflecting real-world landslide occurrence patterns.