Uncertainty and regional air quality model diversity: what do we learn from model ensembles?

Recently several Clean Air For Europe (CAFE) regional air quality projects were carried out in order to predict the impact of emission control policies, using an ensemble of models. After describing and intercomparing the skill of these models over long simulation periods, we examine whether this ensemble provides a spread of concentrations of ozone and aerosols that is representative of the simulated uncertainty. Using tools borrowed from the evaluation of ensemble weather forecasting, we analyze statistics of simulated concentrations over an entire summer season. Although the ensemble may have biases, the distribution of simulated concentrations is representative of the uncertainty. For ozone the ensemble spread is partly due to fluctuations resulting from different model formulations and input data, but also to the spread between individual model systematic biases. The variability of the uncertainty is fairly well reproduced by the ensemble. The skill of the ensemble in predicting uncertainty is also demonstrated by evaluating the reliability of probabilistic prediction of threshold exceedances. These results indicate that the ensemble can be used for evaluation of the impact of emission reduction policies and its uncertainty.