Estimation of High-Resolution CO₂ Emissions by Road Segment in South Korea Using Machine Learning Model Analysis and Applications

In 2021, the global greenhouse gas (GHG) emissions totaled 49.55 gigatons of carbon dioxide equivalent (GtCO₂eq), with the transport sector accounting for 15.8% of the total. Among these, road transport was the dominant source. Therefore, reducing carbon dioxide (CO₂) emissions from the road transport sector is critical for achieving carbon neutrality in transportation, which requires a high-resolution emissions inventory. However, current CO₂ emission calculations are typically conducted at the national level, limiting spatial accuracy. This study aims to develop a road-level CO₂ emissions inventory across the Republic of Korea. To achieve this, high-resolution traffic volume data is essential. Given the lack of observed traffic data on many roads, we first developed traffic volume estimation models using machine learning algorithms, including Random Forest (RF), LightGBM (LGBM), XGBoost (XGB), and Deep Neural Networks (DNN). The models demonstrated strong performance, with R² values of 0.9404 (MSE: 94,331) in Seoul, 0.9490 (MSE: 26,929) in Daejeon, and 0.8619 (MSE: 40,293) in Incheon. Furthermore, we applied clustering techniques and model diagnostics to construct optimal region-specific and variable-specific models, allowing us to quantify the uncertainty of the estimated traffic volumes. Using emission factors, we estimated road-level CO₂ emissions from the predicted traffic volumes and indirectly derived the uncertainty in CO₂ emissions based on traffic volume uncertainties. Additionally, CO₂ observations from select roadside monitoring sites were used for further validation. In future work, we plan to enhance the traffic volume estimation models and develop a direct CO₂ emissions prediction model. These efforts are expected to support evidence-based policymaking and enable more effective CO₂ emission reduction strategies in the road transport sector.