Air pollution has serious environmental and human health-related consequences; however, little work seems to be undertaken to address the harms in Middle Eastern countries, including Saudi Arabia. We installed a continuous air quality monitoring station in Jeddah, Saudi Arabia and monitored several air pollutants and meteorological parameters over a 2-year period (2018-2019). Here, we developed two supervised machine learning models, known as quantile regression models, to analyze the whole distribution of the modeled pollutants, not only the mean values. Two pollutants, namely NO2 and O3, were modeled by dividing their concentrations into several quantiles (0.05, 0.25, 0.50, 0.75, and 0.95) and the effect of several pollutants and meteorological variables was analyzed on each quantile. The effect of the explanatory variables changed at different segments of the distribution of NO2 and O3 concentrations. For instance, for the modeling of O3, the coefficients of wind speed at quantiles 0.05, 0.25, 0.5, 0.75, and 0.95 were 1.40, 2.15, 2.34, 2.31, and 1.56, respectively. Correlation coefficients of 0.91 and 0.92 and RMSE values of 14.41 and 8.96, which are calculated for the cross-validated models of NO2 and O3, showed an acceptable model performance. Quantile analysis aids in better understanding the behavior of air pollution and how it interacts with the influencing factors.
Keywords: air pollution; climate change; extreme value analysis; nitrogen oxides; ozone; quantile regression; supervised machine learning.