A global oceanic four-dimensional data assimilation system has been developed for use in initializing coupled ocean-atmosphere general circulation models and also to study interannual variability. The data inserted into a high-resolution global ocean model consist of conventional sea surface temperature observations and vertical temperature profiles. The data are inserted continuously into the model by updating the model's temperature solution every time step. This update is created using a statistical interpolation routine applied to all data in a 30-day window for three consecutive time steps and then the correction is held constant for nine time steps. Not updating every time step allows for a more computationally efficient system without affecting the quality of the analysis.
The data assimilation system was run over a 10-yr period from 1979 to 1988. The resulting analysis product was compared with independent analysis including model-derived fields like velocity. The large-scale features seem consistent with other products based on observations. Using the mean of the 10-yr period as a climatology, the data assimilation system was compared with the Levitus climatological atlas. Looking at the sea surface temperature and the seasonal cycle, as represented by the mixed-layer depth, the agreement is quite good, however, some systematic differences do emerge.
Special attention is given to the tropical Pacific examining the El Nin˜o signature. Two other assimilation schemes based on the coupled model using Newtonian nudging of SST and then SST and surface winds are compared to the full data assimilation system. The heat content variability in the data assimilation seemed faithful to the observations. Overall, the results are encouraging, demonstrating that the data assimilation system seems to be able to capture many of the large-scale general circulation features that are observed, both in a climatological sense and in the temporal variability.