In Part 2, we noted that in order to utilize the SST forcing of the atmosphere for prediction purposes, it would be necessary to make a prediction of the SST itself. Since SST anomalies change relatively slowly over a period of several months, models capture some short-lead time skill simply by assuming that there is no change in the anomalies during the period being forecast for. However, ideally we would like to be able to forecast developments in the SST anomalies. These come about through a combination of ocean and atmosphere processes, and some of the developments do have a degree of predictability. In lecture 2 we discussed atmospheric GCMs. There have also been developed conceptually similar models that encompass the ocean as well as the atmosphere. These models are often referred to as coupled General Circulation Models (CGCMs). They allow predictions of ocean and atmosphere developments simultaneously, and are perhaps the theoretically most appealing approach to seasonal forecasting.

(i) Coupled GCMs (CGCMs)

The model is formulated as described for the atmospheric GCM discussed in Part 2, except now the layers and grid extend into the ocean. To make a forecast, the model is given the best estimate of the current state of the climate system at all its grid-points and allowed to step forward in time (that is, simulate the time evolution every few minutes of all model variables) to naturally generate a forecast for the coming season.

In the model experiments described in Part 2, at each time step we specified the observed SST at the model atmosphere's lower boundary. Now, the CGCM has its own ocean. So at each time step, new conditions are predicted for both the atmosphere and the ocean. The predicted values at the surface of the ocean and the atmosphere determine the exchanges of heat and momentum between the ocean and the atmosphere. Within the model, as it steps forward, there is a slow evolution of the ocean state over the coming seasons. A key question is the extent to which this slow evolution represents an accurate prediction of the ocean development over that period. A further question is then the extent to which the model's atmosphere responds to the model's predicted ocean surface temperatures, to accurately depict the coming season's atmospheric climate features, such as regional rainfall and temperature patterns.

It has turned out that the tropical Pacific El Niñosea-surface temperature anomaly pattern in Fig. 1.3c can be predicted with good skill typically a couple of seasons ahead of time, if a CGCM is given the current ocean conditions in the tropical Pacific down to a few hundred meters below the surface. This represents a substantial discovery, since El Niñois the single most important factor in climate on seasonal to interannual timescales.