An explainable percolation-based clustering framework for China's transport carbon emissions analysis

Mitigating carbon emissions requires coordinated actions across jurisdictions, as emissions often exhibit strong spatial-temporal synchronization and interdependencies. Understanding how emissions cluster spatially is therefore essential for designing collaborative mitigation strategies. However, existing studies on the spatial clustering of carbon emissions remain limited by subjective parameter settings and insufficient exploration of heterogeneous driving factors across clusters. Here, we integrate the percolation theory in physics with a spatial-temporal clustering algorithm to objectively delineate clusters of ground-transport-related carbon emissions (TCE) for 323 Chinese cities in 2019. Building on six identified spatial clusters at the regional level, we employ random forest models and interpretable machine learning techniques to examine the effects of various factors on TCE. The results further uncover regional heterogeneity in emission determinants: In the Yangtze River Delta cluster, per capita GDP plays a leading role. In contrast, emissions in the Beijing-Tianjin-Hebei and Central Plains clusters are primarily associated with population scale, transport provision, and intensive passenger mobility. A distinct pattern emerges in the Pearl River Delta, where emissions are overwhelmingly driven by road freight demand. Finally, in the Middle Reaches of the Yangtze River and the northern Jiangsu-Shandong coastal belt clusters, service-sector activity and temperature emerge as key contributing factors. This study contributes to the methodological frontier by introducing a percolation-based spatial clustering framework for emission analysis, which can be generalizable to broader contexts. Furthermore, it provides actionable insights for fostering cross-city coordination in emission mitigation, advancing China's decarbonization targets and sustainable transport goals.