Modeling Background Error Covariance in Variational Data Assimilation with Wavelet Method

Abstract

Background error covariance (B) plays an important role in any meteorological variational data assimilation system, which determines how information of observations is spread in model space. In this paper, based on the WRF model and it’s 3D-Var system, an algorithm using orthogonal wavelet to model B-matrix is developed. Because each wavelet function contains both information on position and scale, using a diagonal correlation matrix in wavelet space can represent the anisotropic and inhomogeneous characteristics of B. The experiments show that local correlation functions are better modeled than spectral method, and the forecasts of track and intensity for typhoon Kaemi are significantly improved by the new method.