A global, eddy-resolving capability to predict the ocean circulation is technologically feasible within the next decade. For this purpose, the satellite altimeter is the most promising operational source of oceanic data with global coverage, but only at the surface. A comparable source of subsurface data is not on the horizon. This investigation demonstrates the ability of a numerical ocean model to dynamically transfer simulated altimeter data into subsurface information. This was done for a variety of dynamical regimes with (1) barotropic, baroclinic, mixed, and episodic instabilities, (2) flat bottoms of large-amplitude topography, (3) relatively vigorous or gentle exchange between the layers, and (4) major time scales that are short (~60 days), long (~1 year), or both. In all cases the pattern of the deep pressure field is much different from the one for the current-related variations in the sea surface elevation, and sometimes not obviously related to it. The model was able to reconstruct the deep pressure field even in situations with energetic shallow and deep circulations, baroclinic instability, and a vigorous vertical exchange of energy. However, in such experiments the frequency of updating for the free surface elevation was critical. In this study the maximum update interval that allowed successful dynamic surface to subsurface transfer was about half the shortest major time scale (SMTS, e.g., 50--60 days in the experiments with baroclinic instability). Without knowledge of the deep pressure field, numerical predictions of the surface pressure field and the depth of the pycnocline typically were better than climatology for 1/4 to 1/2 the SMTS, but with successful dynamic surface to subsurface transfer, forecasts without updating were better than climatology for the SMTS or more. The time scale for predictive skill can be substantially longer than the maximum updata interval permitted because the update interval must be short enough to allow decreasing error in the deep pressure field from one free surface updata to the next until the error asymptotes at some acceptable level, approximately 30--50% in these results.

Forecasts of nondispersive isolated eddies with beta Rossby number O(1) demonstrated predictive skill for 3 months or more even when the subsurface initial state was unknown. A serendipitous result of the study is some progress in understanding the generation of cyclonic eddies studies associated with the Loop Current in the eastern Gulf of Mexico (section 5.5).