Applicability evaluation and driving factor analysis of ecosystem health classification in the Yangtze River Delta

A scientifically rigorous assessment of ecosystem health and the development of effective conservation strategies are critical for achieving regional sustainable development and human wellbeing. Using the Yangtze River Delta (YRD), China, as a case study, we applied the Vitality-Organization-Resilience-Services model to evaluate ecosystem health from 2010 to 2020. The Ecosystem Health Index was classified into health grades using a Self-Organizing Map (SOM) neural network, and Random Forest (RF) analysis was used to identify key driving factors. The results revealed a progressive improvement in ecosystem health, with a spatial pattern of higher values in the southwest and lower values in the northeast. High-value areas were concentrated in the mountainous regions of western Zhejiang and southern Anhui, whereas low-value areas were primarily clustered in the North Anhui Plain and the urbanized belt along the Yangtze River. The SOM results demonstrated strong statistical separability and spatial consistency. One-way ANOVA η 2 values ranged from 0.935 to 0.957, and Tukey's post-hoc test confirmed significant differences among health grades. Global Moran's I values exceeded 0.93, indicating strong spatial clustering. External validation using species richness data from nationally protected wildlife habitats showed that the SOM-identified Excellent level substantially overlapped with ecological hotspots. RF-SHAP identified SLOPE as the most important factor during all three periods. Land use/land cover increased in importance after 2015, whereas DEM, PRE, and POP consistently ranked among the top five factors. Overall, this study provides a generalizable analytical framework for ecosystem health classification and differentiated ecological management in rapidly urbanizing regions.