Total cloudiness of 29 models participating in the Atmospheric Model Intercomparison Project is compared with the ISCCP C2 as well as the Nimbus-7 and Meteor observational estimates. The root-mean-square differences between the annual means of the model calculations and the C2 observations after global means are removed vary from about twice to nearly four times the difference between the C2 and Meteor observations. The large differences are in some cases due to the fact that although a model qualitatively has patterns of spatial variations similar to those of the observations, the magnitude of those variations is much too small. In other cases the models have produced the approximate magnitude of the spatial variability of the observations but display sizable errors in the pattern of that variability.
Deficiencies with respect to the model simulations of the mean seasonal cycle are also pronounced. For instance, the differences between the zonal averages of total cloudiness for contrasting seasons suggest that near 60° most models predict minima in cloudiness in summer, whereas observations strongly suggest the opposite. In addition, smoothed seasonal cycle analyses suggest that a portion of these deficiencies in some models is the result of a simulated seasonal cycle that leads that of the observations by about two months. However, some models, which appear to have the proper phase of the seasonal cycle, still show large root-mean-square differences and small correlations when compared with the smoothed seasonal cycle of the C2 observations. The C2 and Meteor observations show a modest signal in total cloudiness for the only important interannual variation during the July 1983 through June 1988 observation period—the 1986/87 ENSO event. A few models reproduce this event about as well as do the Meteor observations, whereas many models fail to show any evidence of it.
Overall, models that better reproduce the ENSO results also tend to do well with seasonal variations. No specific differences are evident in the physical characteristics of models that are relatively adept at reproducing seasonal and interannual variations and those that perform more poorly. However, there is the general conclusion that models that have more sophisticated physical processes tend to better simulate the cloud observations.