Though it is always important to remember that making climate predictions is probabilistic, it is often useful to summarize predictability using the degree of match of the mean rainfall predicted by the GCM versus the observed (e.g. solid lines in Fig. 2.2 for East Africa). The correlation between the best estimate prediction (red line) and observed (blue line) is an indication of the extent to which the model is able to anticipate the observed rainfall anomaly. Many analyses like that in Fig. 2.2 have been made for various regions and models. Summary maps of skill for a number of models are posted on the IRI web site. One difficulty in interpreting the result in Fig. 2.2 and the general maps of skill, is that there is always the possibility that the skill is a random fluke. We know that if we make 100 correlations amongst random time series, 5% of the correlations will show as statistically significant at the 5% level, even though there is no physical basis for the correlation. Thus, it is valuable to supplement the type of analysis in Fig. 2.2 with an analysis of the GCM circulations and associated SST forcing, to evaluate if the skill is physically plausible and if mechanisms like those discussed earlier appear to be operating i.e. can we trace the chain from SST forcing, to atmospheric circulation anomalies to rainfall anomalies. In the absence of evidence for physically plausible mechanisms linking the SST to the atmospheric circulation and rainfall, any simple correlation between GCM rainfall and observed rainfall needs to be treated with caution. Figure 2.5 shows the GCM circulation and rainfall anomaly fields simulated during the 5 wettest observed rainfall seasons in East Africa. Also shown is the mean SST forcing that was operating in those five years. It is clear that the GCM circulation across the tropical Pacific and Indian Oceans looks consistent with the SST forcing, and there is high confidence that the ability of the GCM to accurately represent the observed rainfall is physically based, and, together with the diagnostic results of observed fields (Fig. 1.11), this creates a strong scientific basis for making climate forecasts based for the October-December rainfall season in East Africa, based on SST forcing.