Long-range predictability in the tropics. Part II: 30-60-day variability

It is suggested that the slow evolution of the tropical Madden-Julian oscillation (MJO) has the potential to improve the predictability of tropical and extratropical circulation systems at lead times beyond 2 weeks. In practice, however, the MJO phenomenon is extremely difficult to predict because of the lack of good observations, problems with ocean forecasts, and well-known model deficiencies. In this study, the potential skill in forecasting tropical intraseasonal variability is investigated by eliminating all those errors. This is accomplished by conducting five ensemble predictability experiments with a complex general circulation model and by verifying them under the perfect model assumption. The experiments are forced with different combinations of initial and boundary conditions to explore their sensitivity to uncertainties in those conditions. When "perfect" initial and boundary conditions are provided, the model produces a realistic climatology and variability as compared to reanalysis, although the spectral peak of the simulated MJO is too broad. The effect of initial conditions is noticeable out to about 40 days. The quality of the boundary conditions is crucial at all lead times. The small but positive correlations at very long lead times are related to intraseasonal variability of tropical sea surface temperatures (SSTs). When model, initial, and boundary conditions are all perfect, the useful forecast skill of intraseasonal variability is about 4 weeks. Predictability is insensitive to the El Nino-Southern Oscillation (ENSO) phenomenon, but it is enhanced during years when the intraseasonal oscillation is more active. The results provide evidence that the MJO must be understood as a coupled system. As a consequence, it is concluded that further progress in the long-range predictability effort may require the use of fully interactive ocean models.