An ocean prediction model was driven by observations via assimilation. The best forecast was obtained using a smoothing scale of 12.5 km and a time window of 24 h for data selection. Mostly, the forecasts were better than that of a run without assimilation, the skill score increased with increasing forecast range, and the score for temperature was higher than the score for salinity. It is shown that a vast number of data can be managed by the applied method without data reduction.