STATISTICAL-ANALYSIS WITH BOOTSTRAP DIAGNOSTICS OF ATMOSPHERIC POLLUTANTS PREDICTED IN THE APSIS EXPERIMENT

Predictions of O-3 and NO2 delivered by three eulerian models from the APSIS study are compared with observed levels of the phenomen recorde d on 25 May 1990, over the greater Athens area, Greece. A variety of m easures are used to test mcdel performances, including normalised mean square error, bias and correlation, for which confidence intervals ar e constructed. Causes of differences in the model differences are brie fly discussed. A jackknife diagnostic check is performed on the adequa cy of the bootstrap procedure used to make the intervals, which are fo und to be suitably robust. Similarities in model behaviour, and specif ic performancies, are detected. Some statistical criteria seem to give a slight advantage to one of the three sets of model results, althoug h important biases can occur locally, between observations and predict ions. On the whole, no one of the set of model results emerges as sign ificantly superior to the others.