Dynamic Landslide Susceptibility Assessment Using a PRSU Strategy and SBAS-InSAR Technology

Reconciling high-resolution statistical modeling with geomorphologically meaningful management units remains a challenge in landslide susceptibility assessment. Focusing on the landslide-prone Three Gorges Reservoir area in central China, this study evaluates a framework with three components: (i) a systematic evaluation of statistical metrics (mean, median, standard deviation, range, and IQR for continuous features; mode for categorical features) for feature upscaling within slope units; (ii) a comparative analysis between direct slope-unit modeling (Pixel Feature-to-Slope Unit, PFSU) and a strategy mapping pixel-level predictions to slope units (Pixel Result-to-Slope Unit, PRSU); and (iii) the integration of active deformation monitoring using 166 scenes of Sentinel-1A SBAS-InSAR data (2019–2025). PRSU-Q75-Stacking, combining Stacking with the 75th percentile (Q75) aggregation method, had the highest reported slope-unit AUC in the original evaluation. Coupling the selected susceptibility baseline with InSAR-derived deformation indicators produced a deformation-informed classification that may help prioritize field investigation and monitoring.