CONTROL OF A FREE-SURFACE BAROTROPIC MODEL OF THE BAY OF BISCAY BY ASSIMILATION OF SEALEVEL DATA IN PRESENCE OF ATMOSPHERIC FORCINGS

摘要

The purpose of this study is to assimilate satellite altimetry and tide-gauge data in the barotropic, free-surface, finite element model MOG2D, covering the Bay of Biscay and nested in a North East Atlantic domain. In a first step, we explore the error sub-space of the model in presence of forcing uncertainties, and especially in presence of high frequency atmospheric forcing errors. This is done by an ensemble modelling approach (Monte-Carlo) in which the atmospheric fields are perturbed in a multivariate way: by generating an a priori ensemble of perturbed atmospheric forcing fields (10-meter wind and surface pressure from ARPEGE the meteorological model), and computing the corresponding a posteriori ensemble of model states, one can approximate the forecast errors of the model by ensemble spread statistics. These statistics are shown to be neither homogeneous over the domain, nor stationary, since they are very dependent on the meteorological forcing. Then, the forecast covariance matrix is modelled through forecast error Ensemble EOFs. These statistics, in form of 3D multivariate EOFs (Sea Level Anomaly, barotropic velocities, surface pressure and wind-stress components), are used in a reduced-order sequential scheme, SEQUOIA, set up in an Optimal Interpolation configuration with the MANTA kernel developed at LEGOS/POC (De Mey, 2005), to constrain the model forecast in the framework of twin experiments. In a reference experiment, the data assimilation system is calibrated and sensitivity tests are conducted. The system provides significant error reduction for all state vector variables, but appears to be sensitive to configuration parameters: particularly, one need to constrain atmospheric forcing fields to achieve an efficient control of the model errors. Finally, the capability of realistic observing networks to reduce the model errors are compared. Frequent and regularly spaced observations, such as tide-gauges (SLA) or HF radars and buoys (velocity), appeared to be more adapted to the present data assimilation configuration than altimetry data.