A machine learning approach to investigate the build-up of surface ozone in Mexico-City

Ground-level ozone is an important pollutant regarding air quality and climate. Mexico City frequently experi-ences severe ozone episodes due to a combination of strong ozone precursor emissions and its specific topo-graphical environment which critically impacts meteorological conditions. High ozone levels during these episodes cause harmful effects to the public health and the environment. This necessitates ranking air quality and meteorological variables according to their contributions towards the build-up of ozone. In this study, three machine learning models are used to learn a prediction function with hourly data of eight predictors as input and hourly ground-level ozone mixing ratios as output. One-year hourly data of eight predictors collected in Mexico -City from March 2015 to February 2016 is employed to train and test the models. The best model, capturing ozone peak levels with 92% accuracy during 6-18 March 2016, is used to rank the predictors according to their importance in the build-up of ozone applying a shapley additive explanations approach based on the game theory shapley values. This 6-18 March 2016 period encompassed different meteorological and emission conditions and included a severe ozone smog episode from 12 to 17 March 2016. Such ranking of the air quality and meteo-rological variables is crucial for policy-making decisions regarding the prevention and mitigation of ozone detrimental effects during severe ozone episodes and provides insight into the functional dependency of ozone on its predictors. The proposed approach showcases Mexico City, but its principles can be applied for ozone episodes at any other location.