Background error covariances for a BlendVar assimilation system

Abstract

We propose a new climatological background error covariance matrix suitable for the so-called BlendVar scheme, which deals with a problem on how to best preserve large-scale information of the global coupling system in the high-resolution limited area model (LAM) analysis. The BlendVar scheme is composed from a Digital Filter (DF) Blending step, treating the inclusion of the global model analysis, and from high resolution 3D-Var. The new background error covariance matrix forces 3D-Var to act mainly at smaller scales. We created a LAM assimilation ensemble forecasting system, where the DF Blending step is present, to sample the new matrix. To build and demonstrate properties of such a background error covariance matrix, we use the high-resolution model ALADIN coupled to the global model ARPEGE. The DF Blending step is taking advantage of ARPEGE 4D-Var assimilation system while 3D-Var is improving the small-scale part of ALADIN analysis. We assess the impact of using the new background error covariances in the BlendVar scheme with the full data assimilation cycle over the period of one month. We also compare performance of the new BlendVar set-up with respect to DF Blending and 3D-Var used alone. Objective scores with respect to radiosonde and aircraft observations favour the BlendVar scheme with the newly specified background error statistics.

Keywords

• digital filter blending
• 3D-Var
• background error covariances
• LAM assimilation
• initial conditions