Quantifying the nonlinear effects of driving factors on grassland degradation in Inner Mongolia, China

Revealing nonlinear relationships and identifying high-risk grassland degradation (GD) areas are essential for grassland conservation and sustainable development in arid and semi-arid regions. Nevertheless, existing research mainly emphasize linear relationships, lacking systematic characterization of nonlinear responses, threshold effects and coupled interactions among driving factors. In this study, remote-sensing data were adopted to assess spatial patterns of GD. An interpretable artificial-intelligence framework was further applied to reveal nonlinear relationships, threshold effects and interactions among factors, identify pixel-scale dominant driving factors and conduct GD risk prediction. GD showed a stage-dependent transition from degradation expansion to substantial recovery, with degraded areas spreading mainly from the central grasslands toward the northeast. Negative SHapley Additive exPlanations (SHAP) contributions were dominated by natural factors in the first two stages (40.3% and 44.0%) but by anthropogenic factors in the third stage (44.1%). SHAP interaction analysis revealed that higher population density (PD) strengthened the negative effect of GDP, whereas DEM, relative humidity (RHU), and PD substantially modulated the nonlinear response of average temperature (Avg_TEM) across different temperature ranges. GD risk was predominantly low to moderate (40.7% and 53.5%), while high-risk areas accounted for only 5.7% and were mainly concentrated in parts of Xilingol, Hulunbuir, and Chifeng. This innovative analytical framework effectively interprets inner driving mechanisms of GD and provides scientific basis for region-specific ecological restoration and targeted grassland management.