Construction of China’s comprehensive food supply system and nonlinear effects of driving factors: based on XGBoost-SHAP

Food quality and safety are fundamentally shaped by environmental conditions and human activities. This study investigates China’s food supply system by constructing a comprehensive Food Supply Index (FSCI) based on provincial panel data (2007–2023) covering five key food categories. Using the entropy method for objective indicator weighting and the XGBoost-SHAP model, which outperforms traditional linear models and alternative machine learning methods in capturing nonlinear relationships while providing interpretability, the nonlinear effects of environmental and socioeconomic drivers were identified and explained. Key results indicate that the FSCI shows a generally upward yet fluctuating trend over time, alongside pronounced regional disparities-the top eight provinces together account for 52.7% of the total contribution. All drivers exhibit significant nonlinearity. Resource drivers (total agricultural water use, agricultural fiscal spending) and urbanization rate show clear threshold effects. Infrastructure drivers (agricultural machinery power, freight volume) display accelerated or differentiated impacts after certain levels. Digital and innovation drivers (internet access, patent applications) follow three-stage or interval-specific patterns. The proposed framework and findings provide methodological and empirical insights for enhancing food system resilience in northern regions under analogous climate constraints.