Trade-offs and synergies of ecosystem services in high density cities: Revealing nonlinear driving mechanisms through machine learning

Metropolises are confronted with ecosystem degradation driven by rapid urbanization and continuously intensified human activities, posing significant challenges to human well-being and urban sustainable development. Identifying the trade-offs and synergies among ecosystem services (ESs) and their underlying driving mechanisms forms the foundation for implementing effective ecological management strategies. Taking the central urban areas of Chongqing as a case study, we integrated Spearman correlation analysis, Geographically Weighted Regression (GWR), Self-Organizing Map (SOM), and XGBoost-SHAP methods to analyze the interrelationships and driving mechanisms of five typical ESs habitat quality (HQ), soil retention (SR), food production (FP), net primary productivity (NPP), and water yield (WY). The results revealed that: (1) (HQ), (SR), (FP), and (NPP) predominantly exhibited synergistic relationships, while trade-offs were commonly observed with (WY), especially in HQ–WY; (2) The GWR model indicated that spatial nonstationarity relationships occur in different ESs and were closely associated with urbanization levels; (3) Vegetation have dominant impact on ESs relationships, with vegetation canopy height (VCH) demonstrating the strongest positive impact on ESs, while the normalized difference vegetation index (NDVI) exhibited an inhibitory effect on HQ characterized by a distinct threshold; (4) Ecosystem service bundles (ESBs) were identified through SOM and we subsequently classified the study area into five types of zones: urban development zone, ecological buffer zone, core ecological zone, agricultural potential zone, and ecological conservation zone. Finally, corresponding ecological management and control strategies were proposed for different zones to harmonize urbanization with ecological protection in high density city. • Urbanization and land-use competition shape ES trade-offs, synergies, and spatial heterogeneity in high-density cities. • Interpretable machine learning reveals nonlinear ES responses and key socio-ecological thresholds. • Vegetation structure dominates ES dynamics, with canopy height exerting the strongest influence. • A multi-dimensional diagnosis framework couples ES bundles with threshold analysis for high-density urban contexts.