The ENSO cycle of 2005-07 proved again that there is still outstanding challenges for successful ENSO prediction. Most models failed to predict the cooling of 2005 and almost all models predicted lower temperatures than observed during the warming of 2006. As of late July 2007 there is significant uncertainty in ENSO forecasts for the remainder of the year, although La Niņa conditions have been predicted to develop for the past several months. The memory for the ENSO cycle resides in the ocean, so the goal of our study is to understand the ocean's role in the NCEP ENSO Climate Forecast System (CFS). Using TAO/TRITON and Argo observing system, we derived a near real time gridded monthly temperature (T) and salinity (S) product for the upper tropical Pacific. These T and S fields are compared to those from GODAS, which provides the initial condition for the CFS.

We find that there is a phase bias in GODAS upper ocean heat content anomalies averaged over 5S-5N during the 2005-07 ENSO cycle. Salinity is hypothesized to play a role in this bias, since GODAS assimilates Argo temperatures but uses a climatological T-S relation to update salinity. Compared to observations, GODAS significantly underestimates salinity variability leading to biases in mixed layer depth, especially near the eastern edge of the western Pacific warm pool. The implications of these biases for the CFS ENSO prediction will be discussed.