Nonlinear response of vegetation restoration to mining intensity in an arid coal base: evidence from Shendong coalfield, China

Coal mining in drylands is generally expected to degrade vegetation, yet the Shendong coalfield presents a Restoration Paradox: intensively mined ares showed faster surface regreening than less disturbed landscapes. We examined this paradox from 2000 to 2023, a period spanning early restoration institutionalization, rapid mining expansion, formalized reclamation liability, and localized green-mine governance. Geographically and temporally weighted regression with difference-in-differences (GTWR-DID) estimated the modeled net anthropogenic contribution (Δhuman), while a Relative Restoration Index (RRI) measured structural regreening against site-specific baselines and untapped references. Geographically weighted correlation (GWC) evaluated local RRI–Δhuman coupling, and XGBoost-SHAP resolved nonlinear interactions among climatic, mining, and socioeconomic predictors. RRI improved more rapidly in moderate- and ultra-high-intensity zones than in untapped and low-intensity zones, and Good and High RRI classes increased by 32.89% points. Δhuman was predominantly positive, with positive local associations persisting in core high-intensity corridors despite a contraction of strong-coupling areas. Predictor importance diversified over time, and the GDP density–mining intensity interaction contribution crossed zero near 94.01 million CNY/km 2 (sensitivity range, 93.84–95.54 million CNY/km 2 ). Together, these findings link the Restoration Paradox to spatially concentrated human-associated contributions and conditional economic capacity, rather than to GDP or mining intensity acting independently. The results characterize surface structural greening, not comprehensive recovery of soil, groundwater, biodiversity, or ecosystem function. By separating restoration outcomes from modeled attribution and nonlinear interactions, the framework supports differentiated planning for Land Degradation Neutrality.