Evaluation of long-term ozone simulations from seven regional air quality models and their ensemble

Long-term ozone simulations from seven regional air quality models, the Unified EMEP model, LOTOS-EUROS, CHIMERE, RCG, MATCH, DEHM and TM5, are intercompared and compared to ozone measurements within the framework of the EuroDelta experiment, designed to assess air quality improvement at the European scale in response to emission reduction scenarios for 2020. Modelled ozone concentrations for the year 2001 are evaluated. The models reproduce the main features of the ozone diurnal cycle, but generally overestimate daytime ozone. LOTOS-EUROS and RCG have a more pronounced diurnal cycle variation than observations, while the reverse occurs for TM5. CHIMERE has a large positive bias, which can be explained by a systematic bias in boundary conditions. The other models and the ¿ensemble model¿, whose concentrations are by definition averaged over all models, represent accurately the diurnal cycle. The ability of the models to simulate day-to-day daily ozone average or maxima variability is examined by means of percentiles, root mean square errors and correlations. In general, daily maxima are better simulated than daily averages, and summertime concentrations are better simulated than wintertime concentrations. Summertime correlations range between 0.5 and 0.7 for daily averages and 0.6 and 0.8 for daily maxima. Two health-related indicators are used, the number of days of exceedance of the threshold for the daily maximal 8-h ozone concentration and the SOMO35. Both are well reproduced in terms of frequency, but the simultaneity of occurrence of exceedance days between observations and simulations is not well captured. The advantage of using an ensemble of models instead of a single model for the assessment of air quality is demonstrated. The ensemble average concentrations almost always exhibit a closer proximity to observations than any of the models. We also show that the spread of the model ensemble is fairly representative of the uncertainty in the simulations