Monte Carlo Climate Forecasting

Abstract

Ensemble forecasts of global climate conditions during the seven largest Pacific warm/cold events of the last 23 years have been made with a new two-tiered climate forecast technique. The internal model variability (IMV) within the atmospheric model used in the forecasts was large, approximately ? to ½ the variability obtained from the same model forced by observed SST over the last 20 yr. This variability can lead to a wide range of realistic-looking forecasts, even when the equatorial SST forcing was nil. For example, single model forecasts forced by climatological SST produced excellent simulations of the extreme cold event of 1988/89 and the large warm event of 1982/83. Clearly, single simulations are totally inadequate for forecasting and sensitivity studies. The 10 member ensembles used in this study appear to be the minimum for reliable forecasts/sensitivity studies in midlatitudes, although fewer realizations may be needed in the Tropics. The long-range forecasts are shown to be insensitive to initial conditions (which the model ?forgets? after several months), but critically dependent on the nature of the SST forecast in the central equatorial Pacific. The spectrum of IMV is described quantitatively. A regional phase space representation of this spectrum is obtained and used to demonstrate a new significance test for model-derived climate forecasts. The results presented suggest the traditional atmospheric GCM may not be the best tool with which to make long-range climate forecasts.