Nonlinear health benefits of public green space: evidence from a nationwide machine learning study in China.

Introduction: Urban greening is widely recognized as an important factor in human health. However, existing studies have yielded inconsistent conclusions regarding its health benefits, partly due to divergent greening metrics and the prevalent assumption of linear relationships. Methods: This study investigated the associations between three types of urban greening indicators -green cover (GC), general green space (GS), and active public green space (PGS) --and the self-rated physical and mental health of urban residents across China. We matched individual-level health data from the 2020 China Family Panel Studies (CFPS) with county-level greening indicators derived from national statistical yearbooks. To account for potential nonlinearities and to evaluate feature importance, we employed explainable machine learning models (XGBoost) combined with SHapley Additive exPlanations (SHAP). Results: The results indicated that GC and GS had no significant associations with physical health, and their associations with mental health were inconsistent. In contrast, PGS and the ratio of PGS to GS (PGSRatio) demonstrated robust, significantly positive associations with both physical and mental health, with slightly stronger effects observed for physical health. SHAP-based analyses further revealed nonlinear threshold effects: PGS and PGSRatio offered limited health benefits at lower levels, but their impacts increased sharply once baseline thresholds of 12.4 and 36.3% were exceeded. Ideal health-promoting thresholds were identified at 18% for PGS and 45% for PGSRatio. Discussion: These findings emphasize that not all green space yields equivalent health benefits; rather, the provision of sufficient, accessible, and active public green space is critical for maximizing the dual health benefits of urban greening.