Deep learning approach to predict microscopic pollutant emissions from mesoscopic traffic simulations

Urban air pollution, primarily driven by vehicular emissions, remains a major concern for major European cities striving to meet air quality standards. Traditional traffic and emission models face challenges in large urban environments. Macroscopic models, due to their aggregate approach, often underestimate emissions, while microscopic models, although accurate, require extensive data and computational resources. This research introduces a novel methodology that couples a mesoscopic traffic model with a microscopic emission model by generating vehicle speed profiles based on deep learning techniques. The deep learning component produces instantaneous speed profiles from mesoscopic model outputs, which are essential for precise emission estimation using microscopic emission models. This approach improves the accuracy and granularity of road traffic emission estimation, capturing local emission peaks. Applied to the Île-de-France region, home to over 12 million people, the method demonstrates its scalability while maintaining high precision, providing a robust tool for managing urban air quality.