Machine learning reveals divergent drivers and migrating thresholds of cropland degradation in Southern and Northern China

Under the dual pressures of rapid urbanization and climate change, cropland systems in China are facing escalating degradation risks, posing a critical threat to national food security. However, the compound driving mechanisms and nonlinear threshold effects underlying cropland degradation remain insufficiently understood, particularly across the contrasting contexts of southern and northern China. Therefore, this study constructed a multidimensional cropland degradation index and integrated interpretable machine learning (XGBoost–SHAP) with Generalized Additive Model (GAM) to identify dominant drivers and nonlinear thresholds from 2000 to 2020. Results indicated that, although cropland degradation in China generally followed a fluctuating mitigation trend, degradation levels remained consistently higher in the south (0.435) than in the north (0.398). Socioeconomic pressures dominated cropland degradation in southern China, whereas climatic stressors were more influential in northern China. Slope, temperature, and population density emerged as the primary drivers of cropland degradation at the nationwide, northern, and southern levels, respectively. At the county level, dominant drivers displayed marked spatial heterogeneity characterized by both longitudinal zonation and terrain dependence. GAM results further revealed complex nonlinear responses of cropland degradation to major drivers, with clear sensitivity intervals and time-varying thresholds. These thresholds differed substantially between southern and northern China. At the national scale, the cropland degradation index reached its peak at a population density of 75 persons/km², a temperature of 20°C, and a slope of 8°. Meanwhile, degradation risk shifted progressively from steep to moderate slopes and from sparsely populated to more densely populated areas. These findings provide a scientific basis for threshold-based early warning and regionally differentiated cropland protection policies, thereby supporting land degradation neutrality and sustainable development.