GEOSTATISTICAL ANALYSIS OF VALIDATION DATA OF AN AIR POLLUTION SIMULATOR

Jean-Paul Chilès (1) , Serge Antoine Séguret (1) , Pierre-Marc Riboud (2)

1 GEOSCIENCES - Centre de Géosciences
2 EDF R&D - EDF R&D

Chemistry-transport models for air quality forecasting are affected by the uncertainty on the input data (emissions of pollutants and meteorological conditions), the approximations in the modelling of the physicochemical reactions, and numerical approximations (space and time discretization). The validation of the accuracy of these simulators can be done by comparing predictions with actual measurements. This exercise has been carried out for a model for daily forecasting at the scale of Europe, with reference to daily measurements at about one hundred stations over one year. A thorough variographic analysis shows that the error field cannot be characterized independently of the predicted and observed fields. Indeed the forecasts usually display space and time variations similar to those of the measurement data, up to a multiplicative factor, but are often poorly correlated with the reality. These results can be used to define priorities in the improvement of the chemistrytransport model. The presentation is focused on sulphates and nitrogen dioxide.

Domaines: Sciences de l'environnement; Statistiques [math.ST]; Théorie [stat.TH]

Jean-Paul Chilès, Serge Antoine Séguret, Pierre-Marc Riboud. GEOSTATISTICAL ANALYSIS OF VALIDATION DATA OF AN AIR POLLUTION SIMULATOR. 8th International geostatistics Congress, Dec 2008, Santiago, Chile. pp.861-870. ⟨hal-00776934⟩