Spatiotemporal heterogeneity analysis of multi-type clean energy consumption and carbon dioxide emissions in Chinese cities: Integrating multiscale geographically and temporally weighted regression with machine learning

As a pervasive global challenge, carbon dioxide emissions are intrinsically related to energy consumption. However, the importance and spatiotemporal heterogeneity of the impact of various types of energy consumption, including clean energy and fossil fuels, on carbon emissions remain insufficiently investigated. Drawing on remote sensing data from 329 Chinese cities spanning 2005 to 2017, this study integrates SHAP-interpreted eXtreme Gradient Boosting with the Multiscale Geographically and Temporally Weighted Regression (MGTWR) model to elucidate the key contributors to carbon dioxide discharge and further investigate the spatiotemporal non-stationarity of the effect of energy consumption on carbon emissions. The results identify GDP, coal, oil, and electricity as key drivers of CO2 emissions, with a 1% GDP increase in developed regions raising emissions by up to 65%. Temporally, coal, natural gas, wind, and solar exerted short-term effects (under three years), nuclear power showed medium-term influence, while hydropower and oil exhibited persistence over a decade. Spatially, clean energy exhibited an east–west divergence, with the solar power’s emission-reduction coefficient in eastern regions being twice that in the west. These findings indicate that optimizing energy efficiency and fulfilling carbon reduction targets necessitate strategically tailored policies, which must be precisely aligned with the unique characteristics of specific regions and energy sources.