Machine learning based estimation of urban on-road CO₂ concentration in Seoul

The urban on-road CO₂ emissions will continue to increase, it is therefore essential to manage urban on-road CO₂ concentrations for effective urban CO₂ mitigation. However, limited observations of on-road CO₂ concentrations prevents a full understanding of its variation. Therefore, in this study, a machine learning-based model that predicts on-road CO₂ concentration (CO₂^traffic) was developed for Seoul, South Korea. This model predicts hourly CO₂^traffic with high precision (R² = 0.8 and RMSE = 22.9 ppm) by utilizing CO₂ observations, traffic volume, traffic speed, and wind speed as the main factors. High spatiotemporal inhomogeneity of hourly CO₂^traffic over Seoul, with 14.3 ppm by time-of-day and 345.1 ppm by road, was apparent in the CO₂^traffic data predicted by the model. The large spatiotemporal variability of CO₂^traffic was related to different road types (major arterial roads, minor arterial roads, and urban highways) and land-use types (residential, commercial, bare ground, and urban vegetation). The cause of the increase in CO₂^traffic differed by road type, and the diurnal variation of CO₂^traffic differed according to land-use type. Our results demonstrate that high spatiotemporal on-road CO₂ monitoring is required to manage urban on-road CO₂ concentrations with high variability. In addition, this study demonstrated that a model using machine learning techniques can be an alternative for monitoring CO₂ concentrations on all roads without conducting observations. Applying the machine learning techniques developed in this study to cities around the world with limited observation infrastructure will enable effective urban on-road CO₂ emissions management.