Explorative forecasting of air pollution

In the paper a model to predict immission concentrations of PM10, SO2, O3 for a selected number of forward time steps is proposed. The proposed model (e-APFM) is an extension of the Air Pollution Forecasting Model (APFM). APFM requires historical data for a large number of points in time, particularly weather forecast, meteorological and pollution data. e-APFM additionally requires information about the wind direction in sectors and meteorological station. This information also permits pollution at meteorological stations for which we do not have the necessary data (in particular the data about pollution) to be forecast. The experimental verification of the proposed model was conducted on the data from the Institute of Meteorology and Water Management in Poland over a period of two years (between January 2011 and December 2012). Experiments show that the e-APFM method has lower deviations between the measured and predicted concentrations compared to the APFM method for the first day and similar deviations for the next two days (for hourly values) and for the first day and mostly worse for the second and third day (for daily values).