The first coupled atmosphere-ocean forecast of short term (i.e., months to years) climate was made soon after the development of a simplified coupled model to simulate the ENSO phenomenon. This forecast, which predicted the growth of warm SST in the tropical Pacific one year before the onset of the 1986/7 warm phase of ENSO [ Cane et al., 1986], is as important for climate forecasting as the first prediction using the barotropic vorticity equation was to weather forecasting. Since that time, the field of short term climate forecasting, especially as related to ENSO, has grown and prospered.
We now know, from numbers of statistical and dynamical forecasts, that
there is skill in forecasting the SST in the central and eastern Pacific up
to a year in advance (see Barnston et al. [1994] and Latif et
al. [1994] for up-to-date reviews). The skill depends on the model and
the methods of initialization. Models that are initialized solely by blowing
winds over the ocean until the initial time and then coupling to the
atmosphere generally have nowcast (i.e., initial) correlation skills of
about 0.8 for the NINO3 index
and then maintain useful correlation skills out to about a year.
Figure 1 shows the time dependent forecast skill for the Zebiak and Cane [1987] coupled model. Skills drop off considerably when forecast through the spring regardless of the month begun: this drop of skill through the spring has been called the ``predictability barrier.'' Webster and Yang [1992] suggest that this is an intrinsic barrier to climate predictions in the tropics. Recently, however, Rosati et al. [1995] have successfully scaled the barrier in limited numbers of forecasts by assimilating ocean thermal data into the initial conditions for the forecast. And Chen et al. [1995] find no spring barrier when they ``nudge'' the Zebiak and Cane coupled model winds toward the observed winds in generating forecast initial conditions, whereby NINO3 skill correlations greater than 0.6 were obtained for lead times up to 12 months, independent of the month from which the forecast is initiated.
Large, robust changes in the precipitation and wind patterns are concurrent with SST changes in the tropical regions. We may therefore confidently say that skill in predicting tropical SST implies skill in predicting precipitation and winds in the tropics. But extratropical effects of tropical SST changes are far more problematic. For example, Mo and Kalnay [1991] examined the impact on the midlatitude forecast skill of tropical SST anomalies from one cold (1988) and one warm (1987) ENSO event. They report only modest improvements in the ensemble long-range (10-30 day) forecast skill for the midlatitude Northern Hemisphere early summer circulation. While there are discernible and undeniable correlations of extratropical precipitation with tropical Pacific SST ( Ropelewski and Halpert [1987], for example), these correlations have not yet led to usable skills in prediction.
A novel approach to this problem is the so-called ``two tiered'' forecasting system [ Bengtsson et al., 1993; Hunt et al., 1994; Graham and Barnett, 1994]. SST predictions are made with relatively simple models and the SST so predicted is then used as the lower boundary conditions for a high resolution atmospheric GCM. While the forecast of 500 mb height, say, shows apparent skill, the sensitivity of the midlatitude response is such that ensembles of atmospheric forecasts using slightly different initial conditions must be made to define the envelope of the predicted variability [ Barnett et al., 1994]. The technique has been applied in a nowcasting mode by Graham [1994], where the current SST anomaly in the Pacific and Atlantic is used as a fixed boundary condition for high resolution atmospheric GCM simulations of the rainfall over east-northeast Brazil.