Short - Term PM10 Concentration Forecast Modelling in the MARQUIS-Service

Abstract

The information system MARQUIS will provide to the public short-term air quality forecasts for selected areas in Europe. Basis for the forecasts are the current data gained by the air quality monitoring stations in the region and forecasts of the meteorological data. For the service concerning PM10, there is a need to do the forecast in the morning, forecasting the daily mean of the PM10 concentration for the current and the following days. The paper presents the examined methods for this short term PM10 forecast modelling at two open country monitoring stations in Germany on the basis of the results of EURAD (classical emission and dispersion modelling based on the area of Europe), Machine Learning, Multiple Linear Regression and Neural Networking. Advantages and disadvantages of the methods are discussed and some of the validations and of the future plans are presented. Additionally, the paper presents the status of the ProFet-System based operational modelling for three vehicle traffic dominated monitoring stations, using a Multiple Linear Regression model. The results of ProFet are foreseen for the real time information of the public and the triggering of measures for ad hoc reduction of vehicle traffic induced PM10 concentrations. Some of the validations and of the future plans are presented for ProFet as well.