A spatially explicit optimization framework for synergistic water-food-carbon management in agricultural watersheds

Balancing food security with water conservation and carbon mitigation is a critical challenge for sustainable agricultural watershed management. However, the spatiotemporal dynamics, nonlinear responses, and management pathways of the agricultural water–food–carbon (WFC) nexus remain insufficiently understood. Here, we develop an integrated framework combining interpretable machine learning, scenario simulation, and multi-objective optimization to examine WFC dynamics and identify adaptive management pathways across China's third-level watersheds. We find that grain production remains strongly differentiated by water availability and regional resource endowments, while agricultural carbon emissions show a gradual transition toward stabilization with substantial differences in emission sources. Overall WFC coordination has improved, but regional disparities persist, particularly in water-limited regions. The XGBoost–SHAP analysis identifies agricultural water productivity, grain production capacity, and resource-use intensity as important determinants of WFC performance and reveals nonlinear responses with identifiable transition boundaries that vary with local resource and production conditions. Future projections show increasingly divergent WFC trajectories across SSP–RCP pathways, with high-emission pathways associated with greater hydrological pressure in water-limited agricultural regions. Changes in production and demand structures further shape the trajectories of grain supply–demand balance and agricultural emissions across scenarios. Multi-objective optimization indicates that coordinated adjustments in resource allocation and cropping systems can improve WFC performance, increasing irrigation water productivity by 2.4% while reducing agricultural carbon intensity and grain carbon intensity by 9.2% and 5.4%, respectively. These findings advance understanding of nonlinear WFC responses and provide evidence-based pathways for differentiated agricultural resource management across heterogeneous basins under future socio-environmental change.