Impact of Estimates of Initial Ocean State on ENSO Forecasts

We use the UCLA AGCM coupled with the MIT global OGCM in order to obtain ensemble seasonal forecasts for the years 1993-2001, which include strong El Niņo and La Niņa events. The AGCM has a new PBL parameterization, and the coupling to the OGCM is implemented without any flux corrections. The forecasts start from either early March or early June and cover the period until the end of the following northern winter (December-January-February). The initial conditions for the AGCM are taken from a long uncoupled simulation (with prescribed SST). We consider two kinds of initial conditions for the OGCM: a) the estimates provided by the JPL's Estimating the Circulation and Climate of the Ocean (ECCO) project, and b) the fields obtained from a long uncoupled OGCM simulation (with prescribed SST and wind stress from observations) for a period that includes 1993-2001. There are five members in each ensemble from slightly different AGCM conditions, for a total of 180 forecasts. The forecast skill is relatively high in all cases, particularly for the 1997 El Niņo and 1998/1999 La Niņa events with higher values for initial conditions in June. Forecast skill is much higher than persistence in the March cases, and higher than persistence in the June cases. In general, the forecasts obtained from ocean estimates provided by ECCO show errors about 0.5C smaller in the central-eastern equatorial Pacific (150W-120W) than those from ocean states provided by the uncoupled OGCM run. According to these results, initial ocean states can have a significant influence on the skill of seasonal forecasts with a coupled atmosphere-ocean CGM.