SOURCE OF SEASONALITY AND SCALE DEPENDENCE OF PREDICTABILITY IN A COUPLED OCEAN-ATMOSPHERE MODEL

The seasonality of predictability of ENSO (related to the so-called sp ring predictability barrier) is investigated using the Cane-Zebiak cou pled model. Observed winds are used to force the ocean component of th e model to generate analyzed initial conditions. It is shown that decr ease of predictability during Northern hemispheric spring is due to fa st error growth (with a doubling time of small errors of about seven m onths) bring associated with many, but not all, spring analyzed initia l conditions. With winter analyzed initial conditions, errors always g row more slowly (with a doubling time of about 15 months). The fast gr owth rate of errors seen in the dominant empirical orthogonal function (EOF) in spring is present in ail smaller scales of motion (higher EO Fs) in all seasons. The coupled model allows initial errors in smaller scales to quickly cascade to the dominant scale in spring of certain years, while it does nor allow this in winter. Further, if the initial conditions are generated from a long coupled run (coupled initial con ditions as opposed to analyzed initial conditions), then errors in the dominant mode grow slowly both in spring and winter. These results es tablish that the origin of the seasonality of predictability lies in t he use of observed winds to create initial conditions. The authors pro pose that the analyzed initial conditions have an ''imbalance'' that a rises from the fact that the variability of observed winds has a much larger small-scale high-frequency component than model winds. Such imb alances in the spring initial conditions in certain years quickly affe ct the evolution of the dominant mode, leading to loss oi predictabili ty. Even though such imbalances may be present in the winter initial c onditions, they take a much longer time to influence the dominant mode , thus accounting for the greater predictability in winter.