A Connectivity–Benefit Framework for Estimating Heterogeneous Urban Ecological Corridor Widths in Central Shanghai

Rapid urbanization intensifies habitat fragmentation, creating a need for spatially differentiated ecological corridor planning. Conventional delineation often relies on empirical symmetric buffers that do not represent urban matrix heterogeneity. Focusing on central Shanghai and using avian biodiversity as the conservation context, we integrated morphological spatial pattern analysis (MSPA) to identify ecological sources, raster-based principal component analysis (PCA) to construct a composite resistance surface, minimum cumulative resistance (MCR) to extract potential corridors, and circuit theory to estimate modeled connectivity. Corridors were divided using a nominal 300 m segmentation interval, and side-specific candidate widths were identified from inflection features in cumulative connectivity–benefit curves. XGBoost–SHAP was subsequently used to describe within-model associations for ten predictors: natural ecological factors (canopy height, NDVI, and tree cover), anthropogenic socio-environmental factors (architecture area proportion, distances to water and roads, and nighttime light), and avian diversity indices (richness, Simpson, and Shannon–Wiener). Key results were as follows: (1) 30 sources and 66 potential corridors were identified, with modeled connectivity showing a peripheral-high and central-low pattern; the Outer Ring Green Belt formed the principal network backbone. (2) Most candidate total widths (64.83%) were in the 100–200 m range, while 20.19% exceeded 200 m and 14.98% were below 100 m. (3) Within the fitted model, natural ecological predictors had higher mean absolute SHAP contributions than anthropogenic predictors and avian diversity indices, and canopy height had the largest contribution. Because seven environmental predictors also entered resistance-surface construction, these SHAP contributions are interpreted as model-dependent attributions rather than independent environmental drivers. The width estimates are conditional on the adopted resistance and parameter settings and have not been validated with independent bird movement or habitat-use data. They therefore provide planning hypotheses and candidate spatial estimates rather than validated ecological thresholds.