Abstract
In March 2020, the World Health Organization declared a pandemic due to the rapid and worldwide spread of the SARS-CoV-2 virus. To prevent spread of the infection social contact restrictions were enacted worldwide, which suggest a significant effect on the anthropogenic emission of gaseous and particulate pollutants in urban areas. To account for the influence of meteorological conditions on airborne pollutant concentrations, we used a Random Forest machine learning technique for predicting business as usual (BAU) pollutant concentrations of NO(2) and PM(10) at five observation sites in the city of Berlin, Germany, during the 2020 COVID-19 lockdown periods. The predictor variables were based on meteorological and traffic data from the period of 2017-2019. The differences between BAU and observed concentrations were used to quantify lockdown-related effects on average pollutant concentrations as well as spatial variation between individual observation sites. The comparison between predicted and observed concentrations documented good overall model performance for different evaluation periods, but better performance for NO(2) (R(2) = 0.72) than PM(10) concentrations (R(2) = 0.35). The average decrease of NO(2) was 21.9% in the spring lockdown and 22.3% in the winter lockdown in 2020. PM(10) concentrations showed a smaller decrease, with an average of 12.8% in the spring as well as the winter lockdown. The model results were found sensitive to depict local variation of pollutant reductions at the different sites that were mainly related to locally varying modifications in traffic intensity.