Interpretable Machine Learning for Diagnosing Remote Sensing Ecological Index-Derived Ecological Quality Dynamics in the Yangtze River Delta

Fine-scale evidence remains scarce regarding where ecological quality has improved or deteriorated in the Yangtze River Delta (YRD) and which landscape conditions are associated with these trajectories. We developed a 1 km2 hexagon-based diagnostic framework integrating the Remote Sensing Ecological Index (RSEI), Sen–Mann–Kendall trend analysis, Local Moran’s I clustering, recurrence-based ecological stress typology, and XGBoost–SHAP interpretation for 2000–2025. Annual RSEI was standardized by year to capture relative trajectories of ecological quality rather than absolute change under a fixed loading system. The regional mean RSEI fluctuated markedly and declined only slightly, from 0.639 in 2000 to 0.632 in 2025, suggesting that long-term ecological change was nonlinear. At the hexagon scale, 64.77% of valid units showed positive RSEI trends, with significant improvement covering 15.08% of units and significant degradation covering 5.47%. Local Moran’s I identified distinct High–High and Low–Low clusters; persistent low-quality clusters and stable high-quality areas accounted for 10.0% and 7.8% of valid hexagons, respectively. XGBoost–SHAP results indicated statistical associations between RSEI trends and soil moisture, elevation, impervious surface change, and nighttime light change, rather than direct causal effects. This framework provides a spatially explicit basis for identifying priority monitoring areas, ecological stress zones, and differentiated land management units across rapidly urbanizing megaregions.