Regional road transportation emissions in China: A spatial machine learning study on carbon mitigation strategies

China’s regions along the Belt and Road have experienced rapid growth in road transport activity, accompanied by increasing carbon emissions and spatial disparities. This study analyzes the spatiotemporal evolution and regional predictors of road transportation carbon emissions across Chinese provincial-level regions along the Belt and Road from 2012 to 2022 using a spatial machine learning approach. A Geographically Weighted Random Forest (GWRF) model is employed to capture both nonlinear effects and spatial heterogeneity in emission predictors. Results indicate that annual emissions increased from 313.21Mt in 2012 to 625.27Mt in 2022, with an average annual growth rate of 9.96%. Inland provinces showed shifting emission centers and expanding spatial distribution, while coastal regions remained more stable. Key predictors include registered population, energy intensity, freight volume, and land-use structure, jointly accounting for over 80% of overall importance in the GWRF importance analysis. The GWRF model outperformed conventional methods (R2 = 0.98) and provided accurate regional insights into emission dynamics. By integrating refined traffic activity data with spatial analytics, the study offers a decade-long assessment of regional emission trends under rapid economic transformation. The findings support regionally differentiated mitigation strategies. For the Belt provinces, where energy intensity is dominant, priority interventions include freight fleet energy-efficiency improvement and accelerated retirement of high-emission vehicles. For the Road provinces, where population-driven travel demand is more prominent, measures should emphasize public transport prioritization and demand management. For freight-corridor provinces with high freight volume, strategies should strengthen multimodal freight and reduce empty running through logistics optimization. For provinces with rising land-use effects, integrated transport-land use planning should curb car-dependent expansion. This work contributes to the literature by advancing the application of spatial machine learning in transport emission modeling and providing evidence-based implications for low-carbon transport planning in large, heterogeneous regions.