Exploring the influencing factors of urban carbon emissions with explainable machine learning: evidence from China's first-tier cities

Abstract To meet China's strategic goal of "tailored and categorized approaches" to carbon reduction, systematic analysis is essential for formulating city-specific mitigation pathways.Utilizing data from 19 China's first-tier cities spanning 2002 to 2023, this study employs the XGBoost-SHAP model to investigate eight key factors influencing carbon emissions: economic development level (PGDP), population size (POP), industrial structure (IS), technological innovation (TI), energy intensity (EI), urban form (D), public transportation (PT), and new digital infrastructure (DI)..Furthermore, K-means clustering classifies the cities into five distinct clusters, enabling an in-depth analysis of heterogeneous drivers across city types.The main results are as follows:(1) POP, EI, TI, PT, PGDP were significant factors influencing carbon emissions across major Chinese cities.(2) The influence of individual drivers exhibited heterogeneity among city types. While EI exerted a significant impact on emissions within all five city clusters, the effects of POP, TI, and PT varied considerably.Based on these findings, we propose policy suggestions focusing on system governance key elements, differentiated emission reduction strategies, and collaborative governance system construction to facilitate urban green transformation.