A ChiMerge–WOE Ensemble Learning Framework for Landslide Susceptibility Assessment in Jiuzhaigou County, China

A novel framework combining ChiMerge discretization, Weight of Evidence (WOE) transformation, and tree-based ensemble learning achieves high accuracy in mapping landslide susceptibility in Jiuzhaigou Valley, China. The approach outperforms conventional methods by identifying statistically meaningful thresholds in conditioning factors and quantifying their association with landslides, leading to an AUC of 0.9432 for the best-performing model (WOE-CatBoost). High-risk zones, covering 28.59% of the study area, account for 85.96% of observed landslides, primarily in steep valleys, fractured lithological areas, and regions affected by human activities such as road construction and tourism development. This method enhances both prediction accuracy and interpretability, offering critical spatial support for landslide prevention and sustainable land management.

Liu, Y., Zhang, L., Zhang, Y., Yao, Y., & Bao, Z. (2026). A ChiMerge–WOE Ensemble Learning Framework for Landslide Susceptibility Assessment in Jiuzhaigou County, China. Sustainability, 18(13), 6488. https://doi.org/10.3390/su18136488