Graph Neural Networks for Hourly 1 km Urban CO2 Emissions Estimation Using Real-World Data

Due to urbanization, people’s daily commutes can emit huge amounts of CO2. Therefore, an accurate picture of the number of vehicles moving through a city is necessary for traffic management and effective emissions control. The development of intelligent transportation systems has enabled real-time traffic data collection. For example, Seoul has hourly traffic speed data for almost all of its road network and traffic volume data measured by 139 in-situ sensors. Telecom operators also process cell phone data to provide traffic mobility data. In this study, we developed a graph neural network model to estimate city-wide traffic volumes from traffic speeds, reflecting the actual road network connectivity. Our approach learns the relationship between speed and volume and then extrapolates this relationship for periods when city-wide traffic data did not exist. To train the model, we used hourly full-coverage speed data from the Seoul Traffic Information Center and volume data from a telecom company for a limited period from April to September 2024. We then used the trained model and full-coverage speed data to construct traffic volume for a longer period from January 2018 to December 2023. The model’s estimated traffic volume was evaluated against in-situ traffic volume, achieving an R² of 0.888 and an RMSE of 446.30 vehicles per hour on average over the 6-year period. Next, we calculated road-scale CO2 emissions at an hourly timescale using country-specific emission factors based on the estimated traffic volumes. Our estimates, which show the spatial distribution of large emissions on urban highways and main arterials, can provide more spatio-temproal variability compared to global OC2 emission inventories such as EDGAR and ODIAC, which provide smoother patterns. By constructing reliable city-wide traffic volume data, this study supports more precise CO2 emission assessments and decision-making process for urban transportation management.