Exploration of the robustness of some common statistical measures for evaluating air quality models

The utility and meaning of the standards for evaluating regional and urban air quality models are explored by comparing ozone predictions from several forecast procedures to observations from the ozone episodes of August 3, 1990 to August 6, 1990 in the San Joaquin Valley, California. The forecast procedure considered are: SARMAP air quality model simulation; persistence forecasting; random selection forecasting; and constant prediction forecasting. Peak ozone and hourly predictions are evaluated by using 13 statistical quantities which include measures of bias, error, correlation, and scatter and for various subset of data.