Spatiotemporal dynamics and urban-rural disparities of household carbon emissions in China: A prefecture-level analysis using interpretable machine learning

With the advancement of socioeconomic development and evolution of urban spatial forms, household carbon emissions (HCE) have emerged as a significant contributor to China's total carbon emissions, yet their spatiotemporal characteristics and urban-rural disparities at the prefecture level remain insufficiently explored. This study developed an innovative methodological framework where the “Driver-State-Constraint” analytical approach provides a theoretical structure for feature selection in an interpretable machine learning model, allowing for systematic investigation of the spatiotemporal evolution, urban-rural transition patterns, and underlying driving factors across 337 prefecture-level cities in China from 2010 to 2019. Our findings revealed three key insights. First, HCE displayed distinct urban-rural differentiation and followed a Northeast-Southwest gradient pattern, with spatial autocorrelation intensifying over time; notably, urban areas manifested characteristic “local lock-in” effects, while rural areas presented stronger diffusion tendencies. Second, regional imbalance was predominantly driven by intra-provincial disparities (82.72%), with western regions experiencing significantly higher imbalances than eastern regions. Spatial Markov chain analysis further revealed strong state dependence, with rural areas demonstrating particularly high sensitivity to urban development. Third, the influence patterns reflected significant nonlinearity and regional heterogeneity, with per capita consumption expenditure emerging as the dominant explanatory power, while spatial structural elements demonstrated distinct phase-dependent characteristics in their impacts. These insights suggest the necessity for implementing differentiated low-carbon strategies tailored to local development stages, urban-rural linkages, and spatial structural conditions.