Unveiling nonlinear drivers of ecosystem service bundles and dynamic buffer zones for precise watershed governance: A coupling framework of SOM and XGBoost-SHAP in the Yellow River Basin.

Precise ecological functional zoning and an understanding of its underlying drivers are fundamental for sustainable watershed management, yet traditional static zoning often fails to capture the spatial transitions and nonlinear feedbacks within social-ecological systems. This study develops an integrated framework to identify ecosystem service bundle (ESB) and quantify ecological buffer zones in the Yellow River Basin (YRB). By coupling the Self-Organizing Map (SOM) with the eXtreme Gradient Boosting-SHapley Additive exPlanations (XGBoost-SHAP) machine learning model, we analyze the nonlinear effects of socio-ecological drivers across multiple scales and project ESB dynamics toward 2030 under Shared Socioeconomic Pathway-Representative Concentration Pathway (SSP-RCP) scenarios.Our results reveal six spatial ESBs, with the SOM algorithm achieving a 93.3% efficiency improvement over K-means in handling high-resolution data. We quantitatively delineate ecological buffer zones (3.7%-9.3% of the basin), revealing a consistent northwestward migration trend driven by shifting hydro-thermal gradients. Driver effects exhibit significant scale dependence: while precipitation remains the dominant factor for water yield, socioeconomic factors like GDP increasingly govern carbon sequestration and habitat quality under future high-emission pathways. To bridge research with practice, we propose a transition from rigid boundaries to differentiated spatial governance, including cross-regional collaborative committees and adaptive planning for migrating buffer zones. This framework provides a quantitative decision-making basis for balancing ecological conservation and high-quality development in major river basins.