Machine learning-based high-resolution estimation and explainable analysis of road transport CO2 emissions across South Korea.

To support targeted mitigation of road-transport carbon dioxide (CO2) emissions, this study constructs a machine-learning-based high-resolution CO2 emissions inventory for South Korea at the hourly road-link level. The results show that roads with similar monthly total emissions can exhibit substantially different hourly emission patterns, which were systematically classified into three major types-Daytime-Steady, Commuting-Peak, and Mixed/Weekend-High-according to road class and regional characteristics. To examine regional predictive associations, shapley additive explanations (SHAP) analysis was conducted using emissions and land-cover data. In metropolitan areas, road-cover contribution was dominant, reflecting road saturation associated with high population density, whereas building-cover contribution was highest in non-capital regions, highlighting the importance of managing activity-centered emission hotspots rather than focusing solely on road infrastructure alone. Overall, the results suggest that road CO2 emissions are associated with interactions between temporal demand and spatial context, providing a basis for targeted mitigation, atmospheric and climate modeling, and carbon monitoring and verification.