Multivariate extreme value modelling of air pollution concentrations and analysis of traffic and meteorological impact

János Gyarmati-Szabó, Haibo Chen, & Leonid V. Bogachev

Road traffic emissions are accounted for high proportions of many harmful pollutants. Severe congestion leads to the exacerbating traffic pollution, which is expected to worsen with the constantly growing traffic volume. Exact prediction of pollution episodes occurrence, strength and duration is a formidable problem due to the combination of many complex physical and chemical processes involved. This underpins the need for the development of sophisticated statistical methods in order to facilitate prediction of high pollution concentrations and to better understand their cause. In this paper, we investigate the joint extremes of ozone, nitrogen oxide and nitrogen dioxide. Extreme values of stationary processes are most commonly modelled by the Peaks Over Threshold (POT) technique. In the present paper, we extend the classic POT approach by assuming that the parameters of the Generalized Pareto Distribution are certain functions of traffic and meteorological factors such as traffic flow, speed, air temperature, relative humidity, and wind direction and speed. This enables us to investigate the impact of these factors on the distribution of extremes values. Moreover, the copula method helps to identify the combinations of traffic and meteorological conditions that determine persistent interdependence between extremes of different pollutants. Because of the complexity of the model, the parameter estimation was carried out using the Markov Chain Monte Carlo method. The appropriate goodness-of-fit tests confirm that our model provides an accurate estimation of the joint extremes in the air pollution concentrations.