École Nationale Supérieure de
Techniques Avancées

Abstract

The marine circulation dynamics in coastal zones can be assessed by observation and modelisation. If coastal domains are quantitatively observed (routine remote sensing, periodic in situ surveys, long term measurements), consistent model set up remain relatively elusive, especially for the prescription of appropriate initial and open boundary conditions. This issue raised by the specification of model boundary conditions is proposed to be posed in the optimal control framework, which leads to a data assimilation problem solved by an adjoint method: these model parameters are determined according to information from observations. A four-dimensional variational (4DVAR) assimilation method for open boundary control is then developed and implemented on the basis of the primitive equation circulation model OPA and its adjoint. The algorithm is validated with three academic twin experiments: (i) the detection of perturbations on the barotropic forcing along the open boundary of a non-stratified model are obtained by assimilating surface currents taken inside the domain; (ii) the development of a density current inside a semi-infinite homogeneous channel is retrieved from a synthetic data set figuring density profiles distributed downstream; (iii) the observation of the spin up of a coastal circulation induced by a baroclinic Kelvin front incident to a shallow shelf topography permits to identify the upstream boundary conditions linked to the separation at the shelf break. The Gulf of Lions, which is largely opened to the north-western Mediterranean basin, appears to be an appropriate study case to investigate mechanisms and magnitudes of coastal lateral export to the open ocean. The assimilation method is applied in situations of regional influence on the gulf circulation. It is assessed in the hind cast mode for monitoring applications, especially on the data related aspect. The regional forcing provided by the Liguro-Provençal current flowing along the shelf break is controlled to study the ability of the available observational networks to detect transient cross shelf intrusions according to their spatial and temporal coverage.