A long-term effort to develop a parallel Ensemble Kalman Filter (EnKF) to assimilate sea surface height (SSH) and temperature observations into the Global Modeling and Assimilation Office's (GMAO) global quasi-isopycnal OGCM has recently been completed. The EnKF relies on multivariate forecast-error covariances between the observables and the model prognostic variables to compute the Kalman gain matrix and analysis increments. The forecast-error covariances are estimated adaptively from an ensemble integration of the OGCM and are used to update each ensemble member. This is in contrast with more conventional assimilation algorithms such as optimal interpolation (OI), 3DVAR or 4DVAR, in which the error covariances do not vary with time.

One of the GMAO's objectives is to use the EnKF in its routine seasonal-to-interannual ensemble forecasting procedure to process SSH and temperature observations with the OGCM prior to coupling with the NSIPP-1 AGCM. An ensemble of coupled-model forecasts can thus be initialized with the final state of each ensemble member at the end of the EnKF run. The procedure is discussed in this talk, with focus on hindcasts of the 2002 El Nino event. The impact on the coupled-model-hindcast skill of assimilating (1) in situ temperature profiles from the TAO array and from expendable bathythermographs and (2) remotely sensed SSH anomalies from the TOPEX/Poseidon altimeter is discussed. The skill of the EnKF assimilation system with simultaneous processing of the temperature and SSH data is contrasted to (1) that of a EnKF system with sole processing of the temperature data, (2) to that of processing the SSH data only, also with the EnKF, and (3) to that of the temperature OI system which is used in the Office's current routine forecasting procedure and is expected to be soon replaced with the multivariate EnKF.