Variational assimilation of Geosat altimeter data into a two-layer quasi-geostrophic model over the Newfoundland ridge and basin

A variational data assimilation model is examined as a method for reconstructing ocean current fields from sea surface dynamic heights observed with the Geosat altimeter. A two-layer, eddy-resolving, quasi-geostrophic model with realistic bottom topography is used. The space of control variables is composed of initial as well as boundary conditions in both layers. A cost function is defined as a misfit between the altimeter data and the assimilation solution. Regularization terms are also included in the cost function in a form of second-order spatial and temporal derivatives of the stream function. A preparation step is taken to construct a first guess for the variational method; prior to the full assimilation, the lower-order assimilation is carried out with fewer control variables included. Focus is put on three potential problems: (1) the horizontal resolution of the data, (2) the vertical projection of sea surface information into the lower ocean: and (3) nonlinear behavior of the mesoscale variability. This method is applied to the energetic mesoscale features in the Gulf Stream extension over the Newfoundland ridge and basin. The ocean current field constructed in the model through the data assimilation is significantly correlated with the velocity field directly observed with current meters, both near the surface and at subsurface (near the bottom). The near-surface velocity is highly correlated between the data and the model, while the correlation coefficients of the subsurface velocity components are less than 0.5, with nearly unity regression slopes. Thus it is clear that the data contain part unexplained by the quasi-geostrophic model. The horizontal data resolution is adequate for describing the mesoscale variability near the surface, and it is feasible to extract the quasi-geostrophic component in the lower ocean velocity field. Once nonlinearity becomes very intense, local minima in the cost function prevent convergence to a unique solution. We show that remotely collected satellite data can be utilized for monitoring the global ocean at all depths.