A Statistical Approach to Improve Air Quality Forecasts in the PREV'AIR System

Abstract

Since 2003, the national PREV'AIR system (www.prevair.org) has been delivering daily air quality forecasts of atmospheric pollutants (O3, NO2, PM10, PM2.5) over Europe and France. Those products are based on chemistry-transport modeling and in particular on the outputs from CHIMERE model. Analysed air quality maps of the previous day are also produced by mixing observed and simulated data through a geostatistical approach. More recently a methodology has been developed to improve the accuracy of CHIMERE forecasts. It consists of two main stages. In a preliminary stage, statistical short-term forecasting models are built and validated for each French and European rural or (sub)urban background monitoring site. The response variables are the O3 daily maximum, NO2 daily maximum and PM10 daily average concentrations of the current and the next 2 days. This part is based on the developments carried out and tested within CITEAIRII project (www.citeair.eu). In the operational stage, the statistical models identified as reliable enough are applied with a daily frequency to predict concentrations at the corresponding monitoring sites. Locally forecast concentrations are then combined with CHIMERE forecast concentration fields according to the same geostatistical approach as aforementioned. This has been tested so far for O3 and PM10. Validation against independent data shows a significant improvement of the forecasts compared with raw CHIMERE outputs. © Springer Science+Business Media Dordrecht 2014.