Exploring key air pollutants and forecasting particulate matter PM10 by a two-step SARIMA approach

Abstract

This study explores hourly data for 9 air pollutants and 6 meteorological variables in Dimitrovgrad, Bulgaria for a period of one year. Factor analysis is applied for the classification of the variables. A grouping of pollutants and meteorological hazard is established in 7 artificial factors, which allows the explanation of relationships between variables and mutual sources of air pollution. As one of the key problematic air pollutant, PM10 is identified with an average value over the year of 58.7 micrograms per cubic meter. Using SARIMA method from the Box-Jenkins methodology with seasonal parameter s = 24 hours, models of PM10 with transfer functions are constructed and analyzed. For realistic forecasting of PM10 values at future times, it is necessary to know the data of both the meteorological variables and air pollutants that are precursors of PM10, on which its concentrations depend. For this purpose, SARIMA is applied in two successive steps. First, models are developed to forecast each precursor depending on the weather variables. In a second step, the obtained forecasted precursor and meteorological data are used to model and forecast PM10. The constructed models explain the considered data set with 68% to 86% and RMSE = 0.133. The forecast of PM10 pollution is calculated for 168 hours after the period with data not used in the modeling process.