To improve hydrographic forecasts in the North and Baltic Seas, the HIROMB-BOOS Model (HBM) has been coupled with the Parallel Data Assimilation Framework PDAF (http://pdaf.awi.de). The forecast system assimilates satellite sea surface temperature as well as in situ data of temperature and salinity profiles to initialize forecasts up to 5 days. The assimilation uses a fully-featured ensemble Kalman filter, which dynamically estimates the uncertainty of the state estimate with an ensemble model model states and applies spatially localized updates to improve the ocean state. We discuss the structure of the assimilation system, which can analogously be used to extend other forecast models for data assimilation. Further, the forecast improvements caused by the data assimilation are discussed.