Scholarlink

<p>Landslide susceptibility mapping (LSM) is vital for mitigating landslide initiation in mountainous urban areas. However, heterogeneous triggering factors, rapid land use and land cover (LULC) changes, and limited interpretability of data-driven models hinder its practical application under dynamic urban development. This work proposes a novel framework for dynamic LSM considering different landslides triggers and development strategies, leveraging interpretable machine learning and patch-generating land use simulation (PLUS). Initially, a detailed field investigation was conducted to optimize the selection of conditioning factors for distinct landslide types and to divide the study area into accumulation and rock areas. Optimal adjacency intervals for LULC projection were determined and used to construct various development scenarios, which were then incorporated into different ensemble learning algorithms for strategy optimization in LSM. The SHapley Additive exPlanation (SHAP) technique completed the interpretation of factors and the identification of determinants in high susceptibility zones. The results indicate that stacking models outperformed other models in both accumulation and rock areas, with the area under curve (AUC) of 0.979 and 0.971. Soil thickness and relief are suggested to be the most influential predictors of landslide susceptibility. The adoption of an ecological protection scenario (EPS) within the accumulation area and a cropland protection scenario (CPS) elsewhere to minimize future susceptibility escalation in 30 years. Besides, accounting for the heterogeneity of conditioning factors and the incorporation of shorter-interval LULC changes markedly enhances the accuracy of future susceptibility forecasts. The proposed framework offers a robust approach for dynamic LSM in evolving mountainous urban environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integration of interpretable machine learning and PLUS model for dynamic landslide susceptibility mapping of urban settlements in the Three Gorges Reservoir area

  • 期刊论文

  • 发表日期：

    2026年4月20日

查看全文
  • Fancheng Zhao,
  • Fasheng Miao,
  • Yiping Wu,
  • Zhao Qian,
  • Guyue Zheng

摘要

Landslide susceptibility mapping (LSM) is vital for mitigating landslide initiation in mountainous urban areas. However, heterogeneous triggering factors, rapid land use and land cover (LULC) changes, and limited interpretability of data-driven models hinder its practical application under dynamic urban development. This work proposes a novel framework for dynamic LSM considering different landslides triggers and development strategies, leveraging interpretable machine learning and patch-generating land use simulation (PLUS). Initially, a detailed field investigation was conducted to optimize the selection of conditioning factors for distinct landslide types and to divide the study area into accumulation and rock areas. Optimal adjacency intervals for LULC projection were determined and used to construct various development scenarios, which were then incorporated into different ensemble learning algorithms for strategy optimization in LSM. The SHapley Additive exPlanation (SHAP) technique completed the interpretation of factors and the identification of determinants in high susceptibility zones. The results indicate that stacking models outperformed other models in both accumulation and rock areas, with the area under curve (AUC) of 0.979 and 0.971. Soil thickness and relief are suggested to be the most influential predictors of landslide susceptibility. The adoption of an ecological protection scenario (EPS) within the accumulation area and a cropland protection scenario (CPS) elsewhere to minimize future susceptibility escalation in 30 years. Besides, accounting for the heterogeneity of conditioning factors and the incorporation of shorter-interval LULC changes markedly enhances the accuracy of future susceptibility forecasts. The proposed framework offers a robust approach for dynamic LSM in evolving mountainous urban environments.