Landslide Susceptibility Assessment in Badong County, Hubei Province, Based on Machine Learning and SHAP Interpretability Analysis
Year: 2026
Type: article
Abstract: Badong County is located in the hinterland of the Three Gorges Reservoir Area and is characterized by complex geological conditions and highly rugged terrain, making landslide susceptibility assessment particularly important. Based on an inventory of 384 landslides, this study retained 11 conditioning factors and constructed a balanced dataset with a 1:1 ratio of landslide to non-landslide samples. Sensitivity analysis of 500, 1000, and 1500 m buffer distances showed only small differences in non-landslide sampling, with the 1000 m buffer retained for subsequent analysis. The dataset was evaluated using both a conventional 7:3 random training–test split and five-fold spatial block cross-validation, with the spatially independent results used as the primary basis for model evaluation. Four machine learning models, namely RF, MLP, SVM, and XGBoost, were compared. Under spatial block cross-validation, XGBoost achieved the highest mean AUC of 0.8662, compared with 0.8751 under the random split, and also showed the best overall performance among the four models. Calibration analysis further indicated good agreement between the predicted susceptibility scores and the observed landslide proportions. Bootstrap resampling was used to quantify prediction uncertainty, and the integration of susceptibility and uncertainty identified areas characterized by both high susceptibility and high uncertainty. Independent validation using spatially withheld landslides showed that the high- and very-high-susceptibility zones accounted for 29.25% of the study area and contained 71.61% of the withheld landslides. SHAP analysis indicated that elevation, distance to road, slope, and NDVI were the principal contributors to the XGBoost model output. Overall, the results provide a spatial reference for landslide susceptibility zoning, preliminary engineering-site screening, and the prioritization of field investigation and monitoring in Badong County.
Source: Land
DOI: 10.3390/land15101972
Language: English
Subfield: Management, Monitoring, Policy and Law
Field: Environmental Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,137
OpenAlex ID: W7222066211