The traditional susceptibility evaluation of geological hazards usually comprises a global susceptibility evaluation of the entire study area but ignores the differences between the local areas caused by spatial non-stationarity. In view of this, the geographically weighted regression model (GWR) was used to divide the study area at regional scale. Seven local areas were obtained with low spatial auto-correlation of each evaluation factor. Additionally, 11 evaluation factors, including the aspect, elevation, curvature, ground roughness, relief amplitude, slope, lithology, distance from the fault, height of the cut slope, multiyear average rainfall and the normalized difference vegetation index (NDVI) were selected to establish the evaluation index system of the geological hazard susceptibility. The Pearson coefficient was used to remove the evaluation factors with high correlation. The global and seven local areas were evaluated for susceptibility using the information value model and the global and regional division susceptibility evaluation results were obtained. The results show that the regional division information value model had better prediction performance (AUC = 0.893) and better accuracy. This model adequately considers the influence of the geological hazard impact factors in the different local areas on geological hazard susceptibility and weakens the influence of some factors that have higher influence in the global model but lower influence in local areas on the evaluation results. Therefore, the use of the regional division information value model for susceptibility evaluation is more consistent with the actual situation in the study area and is more suitable for guiding risk management and hazard prevention and mitigation.

The appropriate selection of machine learning samples forms the foundation for utilizing machine learning models. However, in landslide susceptibility evaluation, discrepancies arise when non-landslide samples are positioned within areas prone to landslides or demonstrate spatial biases, leading to differences in model predictions. To address the impact of non-landslide sample selection on landslide susceptibility predictions, this study uses the western region of Henan Province as a case study. Utilizing historical data, remote sensing interpretation, and field surveys, a sample dataset comprising 834 landslide points is obtained. Ten environmental factors, including elevation, slope, aspect, profile curvature, land cover, lithology, topographic wetness index, distance from river, distance from faults, and distance from road, are chosen to establish an evaluation index system. Negative sample sampling areas are delineated based on the susceptibility assessment outcomes derived from the information value model. Two sampling strategies, whole-region random sampling (I) and partition-based random sampling (II), are employed. Random Forest (RF) and Back Propagation Neural Network (BPNN) models are used to forecast and delineate landslide susceptibility in the western region of Henan Province, with prediction accuracy evaluated. The model prediction accuracy is ranked as follows: II-BPNN (AUC = 0.9522) > II-RF (AUC = 0.9464) > I-RF (AUC = 0.8247) > I-BPNN (AUC = 0.8068). Under the Receiver Operating Characteristic (AUC) curve and accuracy, the II-RF and II-BPNN models exhibit increases in the region by 12.17% and 15.61%, respectively, compared to the I-RF and I-BPNN models. Moreover, the II-BPNN model shows improvements over the I-BPNN model with increases in AUC and accuracy by 14.54% and 16.52%, respectively. This indicates enhancements in model performance and predictive capability. In terms of recall and specificity, the II-RF and II-BPNN models demonstrate increases in recall by 15.09% and 17.47%, respectively, and in specificity by 15.80% and 14.99%, respectively. These findings suggest that the optimized models have better predictive capabilities for identifying landslide and non-landslide areas, effectively reducing the uncertainty introduced by point data in landslide risk prediction.

The process of reservoir impoundment poses a significant threat to the stability of reservoir bank slopes, potentially triggering new landslides or reactivating ancient ones. Consequently, long-term and stable monitoring of surface deformation in reservoir areas is essential for ensuring safe reservoir operation. SBAS-InSAR technology—characterized by its high precision, multi-temporal capability, and wide spatial coverage—offers an effective means of comprehensively characterizing landslide deformation in such environments. In this study, SBAS-InSAR is applied to monitor landslides in the Xiluodu Reservoir area using Sentinel-1A imagery. Ascending and descending orbit data are jointly inverted to reconstruct the two-dimensional (2D) surface deformation time series. The deformation patterns and their spatiotemporal evolution are analyzed in conjunction with remote sensing imagery, topographic and geological data, and reservoir water level fluctuations. The integrated analysis identifies 10 and 12 significant deformation zones in the vertical and east–west directions, respectively—demonstrating improved detection accuracy compared to single-orbit approaches. Two representative landslides, the Mixiluo and Huanghua landslides, are selected for detailed investigation. Their toe deformation exhibits a pronounced response to both rainfall and reservoir water level variations. These findings provide valuable reference data and technical support for the early identification of reservoir bank landslides and the safe operation of reservoirs in this and similar engineering contexts.

Reservoir bank collapse is a critical geological hazard during the operation of large-scale water conservancy projects, controlled by unique hydrodynamic mechanisms induced by reservoir impoundment, and differs significantly from ordinary landslides. Traditional risk assessment methods, however, often struggle to achieve effective integration between macro-regional zoning and micro-mechanical analysis. Against this limitation, this study proposes a GIS-integrated multi-scale risk screening framework to achieve the preliminary integration of qualitative regional evaluation and quantitative site-specific analysis. Compared with traditional multi-scale studies, the innovations of this research are as follows: (1) a customized GIS component was developed to realize semi-automatic profile extraction from high-resolution DEMs and batch Bishop stability calculations, overcoming the bottleneck of spatializing micro-models over large areas; (2) a “bottom-up” dynamic feedback mechanism was established, utilizing the quantitative safety factor from site-specific evaluations as an explicit indicator for the conservative screening correction of the macro-regional risk map. Applied to the Xiluodu Reservoir, this framework illustrates a potential multi-scale approach for cross-scale risk screening driven by physical–mechanical mechanisms. This provides both a global perspective and a localized physical basis, offering a strategic screening tool for reservoir management. By linking failure mechanisms directly to spatial impacts, the framework provides a plausible conservative feedback rule for risk-informed decision-making in complex reservoir settings.