Deep Learning-Based Mesoscopic Pollutant Emissions Modeling in Road Traffic Networks

Air pollution is a pressing global concern, particularly in urban environments, where transportation systems contribute significantly. Enhancing air quality calls for effective policies, yet evaluating their impact remains challenging. Emission models emerge as pivotal tools in this endeavor. Nevertheless, microscopic emission models face resource and time limitations, especially when applied to large road networks, while macroscopic models sacrifice spatiotemporal resolution and accuracy. This paper presents a new deep learning-based approach that leverages the growing wealth of data made available through recent advances in intelligent transportation systems. It introduces an efficient mesoscopic emission model designed to provide accurate estimates of CO2 and NOx emissions, which are key drivers of air quality degradation. The model considers a number of vehicle attributes, average speeds, and road characteristics, and is trained on a comprehensive and diverse dataset. The results showcase its capability to capture emissions at the link level accurately. This approach offers a scalable and efficient solution to enhance urban air quality models when integrated with mesoscopic traffic simulations.