Data and code for: Drivers and spatial spillover of grassland degradation and recovery in Inner Mongolia, 2000–2024

Data and code for the paper "Drivers and spatial spillover of grassland degradation and recovery in Inner Mongolia, 2000–2024: implications for targeting restoration". Data (CC BY 4.0): fused 14-class land-cover maps of Inner Mongolia for 2000, 2005, 2010, 2015 and 2020 (1 km, Albers Equal-Area), produced by a hierarchical two-tier Bayesian fusion of eight land-cover products, with per-pixel confidence, decision-uncertainty and inter-product agreement layers; the ecological zones and zone-mean NDVI and driver series of the regional causal screening; Granger-causality results (with Toda–Yamamoto and separate-differencing sensitivity analyses) and convergent-cross-mapping results; per-pixel CCM maps; spatio-temporal graph-neural-network outputs on a 3-km graph (spillover influence, 2025–2027 NDVI projections and a relative degradation-risk index); and the transition, validation and attribution tables. DATA_README.md describes every file; MANIFEST.csv lists file sizes and SHA-256 checksums. Code (MIT): the complete processing pipeline (IM_code_v1.1.zip), from land-cover harmonisation and fusion to attribution, causal screening, the graph-neural-network analysis and the figures.