Introduction Habitat degradation in semi-arid coal-mining landscapes is strongly conditioned by geomorphology, yet restoration planning often relies on habitat maps that insufficiently capture landform-specific constraints, shifting drivers, and connectivity bottlenecks.Methods We developed a geomorphology-informed, multi-source framework to assess habitat quality (HQ) dynamics and restoration priorities in the Yushenfu mining area from 1995 to 2023 by integrating InVEST-based HQ assessment, spatial autocorrelation and trend analysis, structural equation modeling, and connectivity modeling based on MSPA, graph indices, and circuit theory.Results HQ exhibited persistent geomorphology-dependent heterogeneity, with loess hilly-gully terrain maintaining comparatively cohesive high-quality habitat mosaics, whereas aeolian sandy terrain remained dominated by aggregated low-quality patches. Regionally, the area share of very low-quality habitats increased from 18.01% to 26.13%, while high-quality habitats declined from 60.34% to 54.59%. Driver analyses indicated a shift from vegetation-mediated regulation in earlier stages to stronger cumulative human modification and topographic control in later stages. Core habitat area contracted from 648.39 km2 to 349.01 km2, indicating progressive simplification of ecological network structure and increasing dependence on a limited set of sources, corridors, and bottlenecks.Discussion Restoration should shift from patch-based intervention to landform-stratified, connectivity-oriented prioritization, emphasizing source protection and corridor consolidation in loess terrain and resistance reduction through substrate stabilization and disturbance mitigation in aeolian sandy terrain.