A surrogate approach for estimating vehicle-related emissions under heterogenous traffic conditions

Vehicle emission analysis currently faces a trade-off between easy-to-use, low-accuracy macroscopic models, and computationally intensive, high-accuracy microscopic models. A surrogate model that leverages microscopic traffic and emission simulations to predict link-level emission rates was developed. The input variables were obtained by aggregating 1 Hz simulated vehicle trajectories into hourly traffic condition factors (e.g., link average/variation of speed, truck fleet percentage, road grade, etc.). The emission ground truth data were generated using the Motor Vehicle Emission Simulator opmode-based analysis module. Different parameter and machine learning model structures were examined to establish the statistical relationship of the input variables and the link-level emission rates. The ability of the model to accurately estimate vehicle-related emissions was demonstrated by using the Columbia, South Carolina road network as an example. This model served as a high-level planning tool to assess the impacts of emissions from transportation projects.