Four Decades of Modeled Landscape Connectivity Change for Sustainability Assessment Across the Three-North Shelterbelt Forest Program Region

Large-scale ecological restoration is often evaluated by increases in vegetation area, although such increases do not necessarily coincide with stronger landscape connectivity. We assessed modeled omnidirectional connectivity across the Three-North Shelterbelt Forest Program (TNSFP) region at nine dates from 1985 to 2025. Morphological spatial pattern analysis identified foreground landscape elements, and Omniscape.jl (v0.6.2) mapped cumulative current density at approximately 300 m. Theil–Sen estimation and the Mann–Kendall test measured temporal change, while spatially validated Light Gradient Boosting Machine (LightGBM) models and SHapley Additive exPlanations (SHAP) values identified the main land-cover, climate, and terrain predictors in five analysis zones. Mean cumulative current density increased from 6.76 to 7.55 model units (+11.7%). The Loess Plateau Zone had the largest increase (+23.0%), whereas the Xinjiang Desert–Oasis Zone had a smaller, non-significant increase (+5.9%). Forest area increased by 31.0%, but total foreground area increased by only 1.79%. Significant increasing trends occurred in 39.3% of foreground-gain areas, whereas significant decreasing trends occurred in 28.1% of foreground-loss areas. SHAP analysis showed that the dominant predictors differed among zones, with slope, temperature, forest, and grassland showing distinct regional importance. These results show that increases in vegetation area alone are insufficient for evaluating restoration outcomes because connectivity responses depend on the location and spatial arrangement of land-cover turnover. The modeled connectivity indicator can complement metrics based on vegetation area as a spatial monitoring component of sustainability assessment, but it does not represent observed ecological connectivity or species movement.