Assimilation of Oceanographic Data.

Available from UMI in association with The British Library. Requires signed TDF. The oceans play a very important role in low frequency (months to decades) climate variability and an accurate knowledge of the ocean state is an essential requirement to produce a good climate forecast in those time-scales. Data assimilation in numerical ocean models has the objective of combining models and observations to produce a better estimate of the ocean state. A variational assimilation technique is used to assimilate simulated and real oceanic data into numerical ocean models. In contrast to other statistically-based assimilation methods, the variational technique directly incorporates dynamics into the assimilation problem and produces a best-fit phase-space model-trajectory by finding the model initial conditions which minimize the "distance" between model and observation fields, while constraining the model variables to exactly satisfy the equations of motion. Experiments with a reduced-gravity model of the tropical Pacific show that, by using the variational technique, the data do not only have a local impact but spread their influence to vast regions of the model domain. The method is particularly valuable at highlighting model-forcing -data inconsistencies which appear when the model cannot fit the data throughout the whole assimilation period. Being dynamically based, the variational method can be used to give a good insight on how to correct those model or forcing deficiencies. Forecasts performed from analyses obtained by assimilating simulated data, suggest that the assimilation of more data from the regions where inconsistencies are particularly serious can improve the forecast results since the model retains the information given by the data for a longer period. This, however, represents only a partial improvement and biases and deficiencies in the model of the forcing should be corrected as much as possible. Encouraging results are obtained from experiments which assimilate simulated acoustic tomography data. A full 4-dimensional variational technique and a variationally formulated sequential assimilation method are applied using an advection-diffusion numerical model. Accurate analyses of the correct (simulated) ocean state are obtained from a relatively few number of travel-time observations. It is shown that, variational methods can be very effective for the assimilation of this kind of indirect measurements which are usually nonlinearly related to the variables of the model. In addition to the data assimilation experiments carried out in this thesis, simple experiments are performed to investigate the usefulness of variational methods in accelerating convergence to equilibrium of numerical models, a subject which is of interest in climate modelling studies. We are not able to accelerate convergence to steady state in the case examined, (a frontogenetic situation for the advection-diffusion equation), but mechanisms for improvement and other possible applications are suggested.