Abstract
The predictability of the operational ECMWF forecast model in the tropics has been studied using the archived data for 1983 and 1984. The predictability of the very large scale quasi-stationary motions for the time scale of longer than 30 days and the transient motions for the time scale of 3-10 days are examined separately.
The very large scale motion is predictable only up to 2 days in the tropics. This is very short compared to about 6 days in the northern hemisphere mid-latitudes and about 4 days in the southern hemisphere. The largest part of the error for 7 day forecasts is found to be due to the systematic error; about 70-90% of the total error can be explained in this way. The systematic error has a structure similar to the gravest internal symmetric Rossby mode of zonal wavenumber 1, but it grows without propagation. Several forecast experiments have been performed to isolate the cause of the systematic error. It has been shown that the error is most sensitive to the convective heating distribution in the tropics, indicating a major weakness of the convective parameterization in the model.
The predictability of the transient disturbances in the tropics is examined using the time filtering technique. The results indicate that the disturbances are predictable up to or beyond 4 days over the central Atlantic and over the central Pacific. However, the forecast skill is low over the far eastern parts of the Pacific ocean where the disturbance develop into tropical storms. This is probably due to the resolution of the model being too coarse for predicting tropical storm development. The forecast is also not skillful in the Indian Ocean where satellite wind data is not available.
