A multivariate ensemble Kalman filter (EnKF) with covariance localization is used to assimilate in situ temperature and salinity observations and remotely sensed altimetry anomalies into a global OGCM with more than 30-million prognostic state variables. An online bias correction algorithm is built into the system to account for the fact that the assimilation continuously alters the model climatology. Following the data assimilation, the ensemble of OGCM states is used to initialize the oceanic component of the GMAO coupled forecasting system (GCFS). The GCFS is then run in hindcast mode without any data assimilation to assess the impact of the ocean initialization on its seasonal forecasting skill. To this end, the hindcast skill is compared to that of the production forecasting system in which optimal interpolation is used to initialize the OGCM prior to coupling it to the other CGFS components.

Other issues addressed in this talk include (1) the effect of ensemble size on the EnKF performance, (2) the importance of online bias correction when sea surface height anomalies are assimilated,(3) whether the time-dependency of the EnKF covariance model has a measurable impact on the assimilation performance and (4) the feasibility of assimilating the ocean observations in coupled mode rather than processing them with the standalone OGCM prior to coupling to the other GCFS components.