PROSPECTS AND PROBLEMS FOR MULTI-SEASONAL PREDICTIONS - SOME ISSUES ARISING FROM A STUDY OF 1992

Global climatic predictions for 12 months in advance have been made fo r 1992. A two-phased model approach was used. The first phase was the generation of predicted sea-surface temperature anomalies for the low- latitude Pacific Ocean obtained from the Cane and Zebiak coupled model . The second phase involved superimposing these anomalies upon the cli matological sea-surface temperatures used in a global climatic model i n order to force climatic perturbations unique to 1992. The difference s between the outputs of this model and the same model run with climat ological sea-surface temperatures permit identification of possible cl imatic changes for 1992. An ensemble approach was used involving 10 ru ns each of 1 year's duration for both climatological and perturbed con ditions. This procedure was used to allow for chaotic influences to be included in the model outputs. Such influences in the model arise fro m the divergence of forecast projections in phase space rather than be ing attributable to stochastic processes. The impact of chaos on the p redictions is illustrated in a number of cases. A succinct summary of the predictive skill for rainfall anomalies, probably the most difficu lt climatic variable to predict satisfactorily, is provided. A number of regional situations is examined in some detail to indicate the pote ntial skill that exists with this two-phased approach for situations o f both above and below average rainfall. Some noticeable failures also occurred that were attributed to the underlying climatology of the gl obal climatic model being inadequate, rather than to a deficiency in t he technique. As would be expected from the limited oceanic region for which sea-surface temperature anomalies were predicted, the climatic predictions were restricted basically to areas influenced by ENSO even ts. Although there is room for considerable improvements in the result s presented here, the two-phased technique represents a viable, simple , and relatively economic interim method for generating multi-seasonal predictions for the immediate future.