Multiscale coordination dynamics in large freshwater lake basin ecosystems: machine learning reveals spatiotemporal variations and driving mechanisms of risk–quality–service process interactions

Large freshwater lake basins face dual pressures from human activity and ecological degradation. However, the mechanisms driving coordinated ecosystem development through interactions among ecological risk (ER), environmental quality (EQ), and ecosystem services (ES) remain unclear. To address this gap, this study developed a composite index of ecosystem coordinated development level (ECDL) by integrating the ER–EQ–ES subsystems using the coupling coordination degree Model (CCDM). Integrating exploratory spatiotemporal data analysis (ESTDA), multiscale geographically and temporally weighted regression (MGTWR), Pearson correlation, and machine learning algorithms, spatiotemporal variations in ECDL and its multidimensional environmental drivers were explored. A multiscale analysis across the Dongting Lake basin, sub-basins, and grid levels between 2000, 2010, and 2020 revealed that ER increasingly concentrated in lake-dense areas, while EQ and ES showed sustained growth in inland forest regions. ECDL remained stable in upstream sub-basins, whereas downstream urban clusters exhibited lower and more fragmented coordination. Coldspots in suburban agricultural belts and lakefront plains consistently suppressed ECDL. ER imposed negative constraints on ECDL, with sub-basin MGTWR coefficients as low as –0.26. EQ was enhanced in high ECDL clustering, with sub-basin coefficients reaching 0.34. ES showed strong local sensitivity. Pearson correlation revealed generally positive associations between natural environments (NE) and ECDL (r > 0.60), but negative correlations with anthropogenic (AE) and composite environments (CE). XGBoost–SHAP analysis uncovered nonlinear interactions where NE promoted, AE suppressed, and CE constrained ECDL. Response mechanisms of individual factors varied significantly across sub-basins. NE dominated in western ecological barrier zones, modulated by CE. Leaf area index and land diversity showed threshold effects at normalized values of 0.30 and 0.50. In southern urban–agricultural transition zones, AE progressively inhibited NE’s regulatory role. In eastern water-intensive zones, AE’s suppressive effect weakened as NE’s regulatory function recovered. This study overcomes single-scale or model limitations by highlighting scale-dependent and environmental threshold effects of ECDL, offering insights to support SDGs 6,11, and 15.