Risk assessment of landslide and collapse disasters in mining area of Lishi using ensemble learning

Landslides and collapses are common geological hazards in loess hilly regions affected by underground coal mining, where mining-induced deformation and extreme rainfall jointly increase disaster risk. This study proposes a mining-specific risk assessment framework that integrates susceptibility, rainfall-triggered hazard, vulnerability, and comprehensive risk analyses for the urban area of Lishi, Shanxi Province, China. Based on 95 historical landslide and collapse events, nine conditioning factors were prepared, including topographic variables, lithology, hydrological proximity, distance to faults, a mechanically constrained distance-to-goaf factor, NDVI, and InSAR-derived surface deformation rate. A heterogeneous Stacking ensemble model was developed by coupling Random Forest and LightGBM as base learners with Logistic Regression as the meta-learner. The Stacking model achieved the highest predictive performance, with an AUC of 0.926, outperforming RF (0.857), LightGBM (0.876), Boosting-LightGBM (0.909), and Boosting-RF (0.918). Elevation, slope, and surface deformation rate were identified as the dominant factors controlling susceptibility. The high and extremely high susceptibility zones accounted for 27.9% of the study area and contained 94.79% of the historical hazard grid cells. Under rainfall frequencies of 10%, 5%, 2%, and 1%, the proportion of high and extremely high hazard zones increased from 16.6% to 44.4%, indicating the strong triggering effect of extreme rainfall. After incorporating socioeconomic vulnerability, high and extremely high risk zones expanded from 8.6% under the 10-year rainfall scenario to 25.8% under the 100-year scenario. The proposed framework provides a physically informed and data-driven tool for landslide and collapse risk management in mining cities.