Assimilation of meteorological data using an optimal control technique

The adjoint methods for data assimilation are applied recently for large-scale numerical weather forecast. They lead to the minimization of a nonlinear cost function defined as a discrete sum in space and time difference between the model forecast and the assumed known observations. Our aim is to use such methods for mesoscale model data assimilation. For that reason, after a brief description of the model being used and the choice of cost function, we show the adjoint model deduction. An application using the collected sounding during the Pyrenees experiments in 1990 (PYREX 90) is made. The numerical results obtained are good illustrations of developed algorithms and demonstrate the feasibility of the method. The interest of this study is its ability to work with temporal and spatial data and to restitue a dynamical analysis of meteorological fields.