SDG-Based Assessment of Urban Green Development and Identification of Nonlinear Coordination Drivers: An Explainable Machine Learning Analysis of the Pearl River Delta

Urban green sustainable development has become a key pathway for achieving high-quality regional growth under intensifying climate change and resource constraints. However, SDG-oriented localized assessments at the urban scale remain limited, particularly in identifying key factors and nonlinear relationships underlying multi-system coordination. This study develops an evaluation framework integrating the sustainable development goals (SDGs) with China’s new development philosophy, following a coordinated pathway of pollution reduction, carbon mitigation, green development, and economic growth. Using nine cities in the Pearl River Delta, we assess the spatiotemporal dynamics of green development and coordination, and apply explainable machine learning (SHAP) to identify key influencing factors and their nonlinear effects. The results show that the overall level of urban green development increased from 45.22 in 2015 to 55.54 in 2022. Innovation, green, and sharing subsystems exhibited the most significant growth, whereas the openness subsystem declined after 2019. Spatially, a clear core–transition–periphery structure emerged, with coordination levels decreasing from southeast coastal to northwest inland areas. SHAP-based analysis further reveals that innovation and openness dominate the explanation of coordination differences, jointly accounting for 77.95% of total feature importance. Moreover, key drivers exhibit pronounced nonlinear patterns characterized by threshold effects and diminishing marginal returns. Specifically, marginal contributions are weak or negative at lower levels, become positive after crossing empirical thresholds, and gradually attenuate or stabilize at higher levels. This study advances SDG-oriented urban assessment and provides robust evidence of nonlinear and context-dependent drivers of coordinated urban green development.