Gaël Forget, Bruno Ferron, Herlé Mercier, Laboratoire de Physiques des Océans, CNRS/IFREMER/UBO, Plouzané, France.

In the context of the ARGO project, we study how an idealised network of profiling floats can constrain the thermohaline structure of the upper ocean in a primitive equation model of the North Atlantic. We use a 4D-varationnal assimilation formalism (strong constraint adjoint method) with a model resolution of 1° and an assimilation window of one-year. The synthetic data are T/S profiles, measured in a reference trajectory (target) of the model, to which some noise was added. The first guess trajectory is identical to the target, except it has wrong T/S initial conditions. The purpose of he 4D-assimilation is to modify those initial conditions to minimise the misfit between the model trajectory and the data. We have generated various observing systems with networks of different densities and shapes, different amount of data noise. In the present formulation of the estimation problem, and in the limits imposed by the idealised situation we consider, 4D variationnal assimilation of a one year ARGO dataset (T/S profiles) is able to significantly improve a low resolution model trajectory (with better results for the last 6 months). We find large sensitivity to the spatial coverage of the domain by the float array, indicating that (in the present context) increasing the number of floats would be worth it. Sensitivity to the level of noise injected in the synthetic data seems weaker, what is encouraging for real data assimilation that could large representativity errors. Finally, for our specific estimation problem, we obtain similar error residuals for a drifting or an eulerian array.

See also.

Poster