This thesis is concerned with methods of combining remotely sensed sea surface temperature (SST) data, sea surface height (SSH) data and models on the ocean eddy or meso-scale (10-100km), for example, in the Gulf Stream. The method used is adjoint or four-dimensional variational assimilation (4D-Var).

Initially, we concentrate on setting up the assimilation system. Adjoint sensitivities are crucial in variational assimilation, so we begin with an exploration of adjoint sensitivity and the limitations in its use, using the Lorenz (1963) equations; a simple chaotic analogue to an ocean model. We show that applied to trajectories which are long relative to the predictability timescales of a chaotic system, cumulative error growth results in adjoint sensitivities diverging exponentially and thus becoming useless for data assimilation. In the Lorenz system, an intermediate timescale is found on which an ensemble of adjoint gradients can give a reasonably accurate (O(10%)) estimate of the sensitivity to finite amplitude perturbations. These results are then extended to a chaotic ocean
circulation model. We discover that the ensemble-adjoint technique can be used in this relevant geophysical system, in some cases.

Then, we consider assimilation experiments using synthetic observations with sampling and errors similar to real satellite data. An eddy-permitting reduced gravity shallow water one and a half layer model with a double-gyre wind-stress forcing is
used. This model advects a passive SST field. Initially, we consider identical twin experiments where the model producing the synthetic observations is identical to that used in the
assimilation. We consider the relative utility of SST and SSH observation. The information given by SSH is more useful, but there are significantly more SST data. Together they allow a superior estimate of the ocean circulation to either used alone. Finally, we consider fraternal twin experiments, where the model producing the observations is different to that used in the assimilation. It is found to be straightforward to reproduce the jet region SSH even with model error, but the SSH away from the jet region
is less well reproduced. This suggests that this technique will be particularly applicable to the
jet regions, such as the Gulf Stream. In future work, we intend to assimilate real satellite SST and SSH data into a regional model of the Gulf Stream area.

More information available at
http://www.atm.ox.ac.uk/user/lea/

Email: Daniel.Lea@jhuapl.edu