Introducing Climatological Flow-Dependence in the WRFVAR Background Error Model For Variational Data Assimilation. Application to Antarctica

The structure of the analysis increments in a variational data assimilation scheme is strongly driven by the formulation of the background error covariance matrix, especially in data-sparse areas such as the Antarctic region. The grid-point modeling in this study makes use of regression-based balance operators between variables, empirical orthogonal function decomposition to define the vertical correlations and high order efficient recursive filters to impose horizontal correlations. A particularity is that the regression operator and the recursive filters have been made spatially inhomogeneous. The computation of the background error statistics is performed with the Weather Research and Forecast (WRF) model from a set of forecast differences. The mesoscale limited area domain of interest covers the Antarctica, where the inhomogeneity of background errors are expected to be important due to the particular orography, physics, and contrast between ice, land and sea.