Statistical Validation of GCM-Simulated Climates: The Time Domain Approach

Major statistical properties (probability densities, mean values, variances, medians, spectral densities, persistence parameters) of the observed global annual temperature (1861-1976) were compared with respective properties of the global annual temperature simulated with the models CGCM3.1 (T63), GFDL-CM2.0, GFDL-CM2.1, GISS-AOM and INM-CM3.0 (project 20c3m) at different scales of spatial averaging, from global to sub-continental. Most models overestimate the linear trend in the data. For individual models, the differences between the properties of the observed and simulated data were found to be statistically significant in most cases. The temperatures averaged over the ensemble of simulated data, though statistically significant in most cases, differ from respective observed temperatures by not more than 1°C at almost all scales of spatial averaging. The width of the 95% confidence interval for such average temperature is 0.5 °C –1.4 °C. The estimates of ensemble-averaged variances of modeled temperature are satisfactory at all scales of spatial averaging except the oceanic and sub-continental. The frequency domain and persistence properties generally agree with respective properties of the observed data.