Ozone Content Variability in the Ground-level Atmosphere Layer in the Mazowieckie Voivodeship, Central Poland

Abstract

This publication presents an analysis of the variability of ground-level ozone in the Mazowieckie Voivodeship, Central Poland, in 2005−2012 and the forecast of ground-level ozone for the next day using artificial neural network models. The content of ground-level ozone in a given area is mainly determined by meteorological conditions and the presence of appropriate chemical compounds, i.e. ozone precursors. The average ozone mixing ratio is from 20 to 100 ppb, depending on the location of the measurement site. Despite its low concentration, ozone in the ground layer has a significant impact on natural environment, through the production of free radicals, shaping the greenhouse effect on the Earth and the formation of photochemical smog, to mention just a few effects. High contents of ground-level ozone may result in the occurrence of episodes that are harmful to human health. The analysis of ground-level ozone measurements in various time scales resulted in the following findings: (1) on an annual scale − the occurrence of a characteristic maximum in the spring-summer period and the minimum in the autumn-winter period, (2) on a daily scale − the occurrence of the highest ground-level ozone content in the afternoon and the lowest just before the sunrise, and (3) on a weekly scale − the existence of a weekend ozone phenomenon. The analysis of the long-term (1995-2016) ozone measurement series at Belsk gave grounds for distinguishing the three periods, representing an increase, decrease and re-increase of ozone content in the ground-level atmosphere in this locality. Models for forecasting the maximum 1-hour daily ozone concentration for the next day over the period of April–September 2015 were constructed using the Statistica 10 “Automatic Neural Networks” program package. The quality of the forecast based on neural models was verified for data from the same months of 2014. The results testify to the ability of the network to generalize the training-acquired knowledge for new, previously unexamined cases. A comparison of neural network modeling results with those of the Global Environmental Multiscale-Air Quality (GEM-AQ) troposphere chemistry model shows that the Unidirectional Multi-Layer Perceptron (MLP) type models applied in this study are an effective tool for the next-day ground-level ozone forecasting