AIR QUALITY FORECASTING FOR ALL SEASONS IN LARGE GEOGRAPHICAL AREAS

Abstract

Today, air quality monitoring plays a vital role due to increasing number of pollutants that threaten human health. Importance of providing accurate information on air quality for forthcoming times is therefore very high. For this purpose, many studies have been carried out to develop air quality forecasting models. However, most of these studies focus on a particular season and relatively small geographical areas. In this paper, unlike the previous ones, an air quality forecasting model is proposed for all seasons in large geographical areas. Turkiye that is a pretty large country where there are seven distinct regions with different geographical and meteorological characteristics is selected to apply the forecasting model. The proposed model categorizes the upcoming 6-hour air quality level as “healthy”, “moderate” and “unhealthy”. The model utilizes low and high order statistical features extracted from the measurements of air quality monitoring stations covering most parts of the geographical regions of Turkiye. The features are then fed into both linear and non-linear classifiers including artificial neural networks, Fisher’s linear discriminant analysis, nearest neighbor and Bayes classifier. Results of the experimental study indicate that the proposed forecasting model is a promising candidate to predict air quality through all seasons at relatively large geographical areas with varying characteristics.