Abstract The process of combining models of the ocean circulation with large data sets is known in meteorology as model initialization and data assimilation. This process is new to oceanographers, who only now are on the verge of having available world-wide synoptic maps of dynamic variables. In this paper we carry out a series of idealized initialization/assimilation experiments with a primitive equation (PE) model, which constitute a first step in developing a realistic process model and data assimilation techniques for the Gulf Stream system. The PE model is used in a spin-down mode and initialized with an analytic jet profile with geostrophically balanced fields. Two major questions are addressed in the present study. The first concerns the initialization process of a PE model during which internal/inertial gravity wave noise is produced. We ask: are the initialization shocks equally crucial for ocean models as they have been for their atmospheric counterparts? The results of an extensive series of balanced versus unbalanced initializations indicate that, for a PE model with a rigid lid, a brutally unbalanced initialization is required to produce strong internal gravity wave shocks. A geostrophically balanced initialization is sufficient to ensure smooth jet evolutions, with no apparent gravity waves, over long time durations in the spin-down mode. No sophisticated initialization procedures seem, therefore, to be required. The second question addressed is: which component of the flow is the most important in data assimilation to drive the model response towards a baseline reference ocean? We specifically compare the knowledge of the depth-integrated flow only, corresponding to measurements of the total transport, with the knowledge of the density field only, or equivalently the velocity shear. The knowledge of the interior density field is much more effective in decreasing the root-mean-square (r.m.s.) errors relative to the reference ocean. If the baroclinic structure is known, coarse horizontal resolutions of data insertion can be reached before significantly worsening the model estimates. If only the depth-averaged flow is known, a decrease in the horizontal resolution of data assimilation has an immediate effect: the r.m.s. errors sharply increase and the assimilation run diverges from the reference ocean. In the assimilation of the barotropic flow alone, even with dense resolution, the errors in the deep layers always show an increasing trend. The relative effectiveness of baroclinic versus barotropic data insertion can be rationalized in the context of geostrophic adjustment theory.
Read full abstract