Integrating causal inference and explainable AI to reveal the driving factors of provincial carbon emissions in China

China’s vast territory exhibits substantial disparities in natural endowments, industrial structures, economic development levels, and energy utilization patterns across provinces, posing significant challenges to achieving the national “dual-carbon” goals. Although extensive research has examined the determinants of carbon emissions, the causal effects of key driving factors and their heterogeneous mechanisms across regions and development stages remain insufficiently understood. To address this gap, this study takes Chinese provinces as the unit of analysis and integrates a causal random forest algorithm with Explainable Artificial Intelligence (XAI) techniques to investigate the spatiotemporal evolution of carbon emissions, identify the causal effects of major driving factors on carbon intensity (CI), explore regional heterogeneity, and project future emission trajectories under different development scenarios. The results indicate that provincial carbon emissions in China exhibit a pattern of aggregate growth accompanied by declining carbon intensity, while interprovincial disparities gradually converge, with high-emission clusters concentrated in industrially and economically developed provinces such as Shandong, Hebei, and Henan. Energy efficiency, economic development level, and energy structure are identified as the most influential determinants of CI, forming a hierarchical causal transmission pathway. Moreover, the driving mechanisms display significant regional and stage-specific heterogeneity, with economic and technological factors exerting stronger effects in the central and western regions, energy structure playing a dominant role in the eastern region, and industrial structure emerging as the primary driver in the northeastern region. Scenario simulations further reveal that the enhanced policy scenario achieves the greatest emission reduction potential, whereas the extensive development scenario entails substantial high-carbon risks. These findings advance understanding of the causal mechanisms underlying provincial carbon emissions and provide empirical support for differentiated carbon reduction strategies and regional low-carbon transition pathways in China.