Forecasting daily maximum ground-level ozone concentrations using stochastic models

Abstract

Analysing and predicting air quality parameters is an important task due to the health impact caused by air pollution. Ground ozone levels are a topic of considerable environmental concern, since excessive levels of ozone indicate high pollution. We develop stochastic models that estimate the ground-level ozone concentrations in air at Pavlovo station in Sofia for the period 2012 - 2016. First, we use the binary logistic regression to predict the probability of occurrence or nonoccurrence of ozone concentrations above or below a threshold (100 µg/m3). Second, in order to model the daily maximum ozone concentrations the Student’s t and gamma distributions are used. The parameters of these distributions are modeled as linear parametric or nonparametric functions. The input predictors include surface and upper air temperature, atmospheric pressure, relative humidity, wind speed and direction, solar radiation, nitric oxide and nitrogen dioxide. The developed models explain the variation in daily ozone maxima and provide reliable tool to predict ozone levels exceeding a relevant threshold.