Spatiotemporal evolution and driving factors of urban carbon balance in China using interpretable machine learning approaches
摘要
Carbon balance plays a pivotal role in achieving China's "dual carbon" goals. Drawing on panel data from 270 Chinese cities spanning 2006–2021, this study constructs a carbon balance indicator and delineates its spatiotemporal heterogeneity. Cities are subsequently classified into four carbon balance functional zones using the K-Medoids clustering method. The XGBoost model, augmented by SHAP and ALE interpretability techniques, is then employed to identify key driving factors. The findings reveal: (1) Over the study period, the overall carbon balance capacity of Chinese cities exhibited a declining trend, manifesting a stepped spatial gradient of "low in the east and high in the west," with only the western region surpassing the national average; (2) Economic agglomeration level and resident consumption level emerge as the core driving factors, both exerting significant threshold effects on carbon balance; (3) Economic agglomeration dominates carbon balance dynamics in the carbon source-dominated zone, carbon source-transitional zone, and carbon sink development zone, whereas resident consumption level serves as the primary influencing factor in the carbon sink functional zone, underscoring marked regional heterogeneity in driving mechanisms.