A study on the El Nino-Southern Oscillation-related seasonal climate predictability is conducted using a coupled Atmospheric General Circulation Model. Retrospective forecasts using the coupled ECHAM4.5 AGCM are evaluated. The oceanic component is mixed layer temperature model with spatially variable but temporally constant thickness. With a prescribed Q-flux, the observed Sea Surface Temperature climatology is successfully retained. With the historical record prescribed for the Eastern Tropical Pacific SST (150oE~eastern edges of Pacific, 30oS-30oN) during the period from the year 1981 to 2000, the AGCM is integrated out to 4-month lead time from initial conditions representing four different seasons each year.

Compared to the observed SST data (ERSST2), it is found out that the ensemble mean of six retrospective forecasts yields skillful prediction for SST in the Tropical Atlantic and the Indian Oceans. The seasonally averaged SST forecasts exhibit substantially higher Anomaly Correlation Coefficients with the observed SST than the damped persistence in the both oceans. ACC forms above 0.9 in the North Tropical Atlantic and Eastern North Atlantic from boreal winter until the following fall (NDJ-ASO), while ACC reaches the maximum in the South Tropical Atlantic from spring to summer (MJJ). The skillful SST forecasts with ACC above 0.7 are found also along the Eastern African coats and the marginal sea around India all the year round. Also, SST forecasts are found skillful around Philippines and in the west of Australia during boreal summer to fall (ASO).

The prediction skills of air temperature at 2m and precipitation are also examined by comparing with observed data sets (University of East Anglia) as well as six arbitrarily chosen AMIP history simulations performed using the same AGCM during the same period. Air temperature at 2m is found skillful in the central northern America and western central Africa, as well as Indonesia and Philippines especially from boreal winter to spring (FMA). The ACCs of air temperature at 2m are comparable with AMIP runs, indicating that these regions are highly sensitive to remote influence of SST in the Eastern Tropical Pacific. Overall, the air temperature is found least skillful from summer to the following fall (ASO), when the local SST forecasts are also least skillful. Precipitation forecasts exhibits good skill scores (>0.5) in northern Brazil all the year round. From spring to summer, high ACCs are also found in the southern Brazil. From October to January, it is found that precipitation forecasts are skillful in the North America, the North Eastern Africa and the Southern China.

One of the practical purposes of this study is to examine methods to minimize the random spread of predictions as well as the computational complexity with limited ensemble sizes, while keeping the physical essence of local ocean-atmosphere coupling. For this purpose, an atmospheric mixed layer model (Seager, 1995) is applied to drive SST anomalies for a set of experiments. In these experiments, only surface winds and cloud coverage are conveyed from AGCM to the local SSTs via the atmospheric mixed layer, while the whole surface heat fluxes are applied to the SST model for the control.

Compared to the control experiment, the SST prediction by the atmospheric mixed layer model exhibits substantial reduction of random spread. Also, anomaly correlation coefficients with observed records moderately increase in the tropical Atlantic and the Western African coasts. The spread in SST decreases most substantially in these regions from April to October.