Using Machine Learning to estimate the impact of different modes of transport and traffic restriction strategies on urban air quality

Abstract

Classical pollutant dispersion models, based on the numerical resolution of some approximate form of the momentum, energy and chemical species conservation equations, are usually limited by incomplete descriptions of the atmospheric boundary layer hydrodynamics, partial characterizations of the emission inventories and, often, high computational costs. Using the metropolitan area of Barcelona as benchmark, the Machine Learning aproach presented here alleviates these limitations providing very accurate local predictions of key pollutant concentrations. Originating mostly from Open Data sources, time-series data on road, maritime and air traffic along with meteorological records from October 2017 to June 2021, have allowed, by means of Machine Learning techniques, to create a model capable of estimating the individual contributions of each mode of transport to worsened Air Quality. Also, when used to investigate the impact of recently implemented mitigation measures, model results predict a reduction of approximately 8 μg·m−3 for CO and NOx. In contrast, O3, PM10 and SO2 are found to be unaffected. The COVID-19 lockdown provided an accidental opportunity to improve the model's robustness and predictive capability through unusually low emission rates from transportation.