On the construction of the prediction error covariance matrix
Takuji Waseda, Leland M. Jameson, Max I. Yaremchuk, H. Mitsudera · University of North Texas Digital Library (University of North Texas) · 2001
Implementation of a full Kalman filtering scheme in a large OGCM is unrealistic without simplification and one generally reduces the degrees of freedom of the system by prescribing the structure of the prediction error. However, reductions are often made without any objective measure of their appropriateness. In this report, we present results from an ongoing effort to best construct the prediction error capturing the essential ingredients of the system error that includes both a correlated (global) error and a relatively uncorrelated (local) error. The former will be captured by an EOF modes of the model variance whereas the latter can be detected by wavelet analysis.