The focus of this study is the control run performance of four general circulation models (GCMs): the Oregon State University (OSU) two-layer atmospheric GCM (AGCM), the OSU coupled ocean-atmosphere model (CGCM), the Goddard Institute for Space Studies (GISS) nine-layer AGCM, and the European Centre for Medium-Range Weather Forecasts (ECMWF) T21 model. The analysis variable is monthly mean sea level pressure (MSLP), and model validation is performed for a limited domain (North America/Atlantic/Europe). The first part of the investigation deals with the magnitude and gross spatial structure of model errors in means and interannual variability (for January and July only). These errors are examined with the aid of maps of time-mean MSLP, difference fields, and local variance ratios. The significance of the local (grid point by grid point) differences in means and variances is then determined by performing univariate t- and F-tests. This information on the spatial structure of large-scale systematic errors is important for understanding the results of significance tests performed on the overall fields. In the second part of the investigation, the statistics recommended by Wigley and Santer (this issue) for use in model validation are applied to test the overall significance of observed/simulated differences in means, variances, and spatial patterns over the entire annual cycle.

Significance levels are determined with the pool permutation procedure (PPP) introduces by Preisendorfer and Barnett (1983). Results indicate that all four models have highly significant errors in the mean field and spatial pattern over the entire annual cycle. Errors in the temporal variance are generally less significant, and significance levels for variance tests can depend critically on the choice of averaging period for observed validation data. The actual test statistic values show that there are considerable differences in model performance. The ECMWF T21 model simulated the spatial pattern and time-mean MSLP field with greater fidelity than the other models considered here. ¿ American Geophysical Union 1990