Peut-on optimiser les constantes d'un modèle de turbulence marine par assimilation d'observations ?

Abstract The system of equations describing ocean circulation is statistically averaged and consequently is not closed. The turbulent diffusive effects need to be parameterised via a turbulence model. Some empirical constants are then necessary to close the system. This work aims to show the feasibility of applying the data assimilation method of optimal control to identify these constants. A prototypic model with two equations of vertical diffusion for the velocity and the salinity is considered. The turbulent viscosity and diffusivity coefficients are computed with a k model. The experiment consists of optimising the two constants of this model by assimilation of velocity and salinity data. These observations have at first been numerically constructed by the simulation of the break of a halocline between two layers, the upper one being in movement. The cost function measuring the discrepancies between the simulation results and the data is minimised iteratively in the direction of its gradient computed by an adjoint model. In spite of the shape of the cost function surface with a valley and a plateau, the method converges. The difficulties which could be encountered in the parameter optimisation in such non-linear models of marine turbulence, are finally discussed.