Quantifying nonlinear thresholds of grey-green space morphology on urban carbon metabolism across local climate zones: a machine learning study in Changchun city

Urban grey-green spaces play a critical role in regulating carbon metabolism by influencing atmospheric CO₂ concentration (XCO₂) and net primary productivity (NPP). However, the heterogeneous drivers of carbon metabolism across different Local Climate Zones (LCZs) remain insufficiently understood. This study investigated Changchun, China, by integrating LCZ classification, multidimensional morphological indicators, and interpretable machine learning. Based on typical urban development patterns and morphological contrasts, LCZ 1–3 were aggregated into a compact built-up group (hereafter the LCZ 1 group), while LCZ 6 (open low-rise) was retained as an independent group; these two groups were selected as representative zones. CatBoost-SHAP models were developed to reveal nonlinear relationships between urban morphology and carbon metabolism. Results showed that grey-space characteristics (B_CORE, BD, BH) dominated carbon metabolism in LCZ 1, contributing over 55%, whereas green-space and climatic factors (VFC, ACH, MAT) played greater roles in LCZ 6. Context-dependent nonlinear turning-point characteristics were identified: in LCZ 1 group, response changes occurred around B_CORE ≈ 2.0 and over BH range of 10–40 m; in LCZ 6, VFC around 0.65 and B1 around − 0.25 showed distinct turning points. Identical indicators showed different association across LCZs, highlighting context-specific planning. This study provides an integrated LCZ-based interpretable machine-learning framework to characterize localized nonlinear responses and support low-carbon urban planning.