Ecological Security Assessment and Multi-Scenario Early Warning in Black Soil Basins Based on the MESI–XGBoost–BN Integrated Framework

Black soil degradation poses critical threats to agricultural sustainability and global food security, yet systematic frameworks integrating ecological security assessment, driver identification, and forward-looking early warning remain underdeveloped for black soil watersheds. This study develops and implements a comprehensive assessment–interpretation–early warning framework for the Xingkai Lake Basin, a representative black soil region at the China–Russia border. We developed a multivariate ecological security index (MESI) to describe the spatiotemporal dynamics between 2000 and 2023. An XGBoost–SHAP framework was applied to quantify dominant drivers of ecological security spatial heterogeneity and examine synergistic effects among driving factors through geographical detector interaction analysis. A Bayesian network model was subsequently employed to simulate the probability of warning grade occurrence under multiple univariate and multivariate scenarios. Findings revealed the following: (1) Spatial analysis revealed persistent north–south differentiation, with high spatial association zones contracting from 25% to 21% despite strengthened global spatial auto correlation. (2) XGBoost–SHAP driver analysis quantified that land use intensity and landscape fragmentation collectively explained over 75% of spatial heterogeneity in MESI. (3) BN models demonstrated greater sensitivity in simulating no warning and severe warning levels. This study provides scientifically rigorous insights into the sustainable management of ecosystems in black soil river basins and offers a generalizable decision-support framework for conducting ecological safety early warning research in other regions facing similar agricultural pressures.