Matteo Corazza, Eugenia Kalnay, DJ Patil and others
University of Maryland
We use the QG simulation system based on the NCEP 3D-Var data assimilation (Morss et al, 2000). It has "errors of the day" comparable to those observed in the NCEP system. We show that bred vectors are very robust with respect to the norm used, and that they have strong similarities to the background forecast errors. In this simulation system, augmenting the background error covariance with the bred vectors reduces the analysis errors by about 30-40% and almost eliminates the day-to-day variability in the errors.