Forecasting of particulate matter with a hybrid ARIMA model based on wavelet transformation and seasonal adjustment

Particulate matter is one of the primary atmospheric pollutants with significant effects on human health. Accurately and reliably forecasting air quality for future horizons makes it possible to take the necessary precautions to minimize potential risks. In this study, monthly PM10 concentration forecasts were made for Erzurum in Turkey. The first ten years of monthly data between 2006 and 2018 were used for training of the model, and the last two years were used to test predictions with the model. PM10 data had trends and seasonal effects removed with seasonal adjustment and were decomposed to three levels with MODWT. For each subseries obtained, modelling was performed with appropriate coefficients chosen with ARIMA. Particulate forecasting was performed with wavelet reconstruction for the approximate and detail series. According to the experimental results, the wavelet-transform based hybrid WT-ARIMA model was more successful than the traditional ARIMA model with regard to the RMSE, R-2, IA, MAE and MAPE. The developed model had values of RMSE 1.50, R-2 0.99, IA 99.92%, MAE 1.26 and MAPE 3.02%. The proposed model may be used as reference for early warning in regions with high air pollution observed due to accurate forecasting capability for particulate matter pollution.