Presentation PDF (886.1 kB)
The bias was estimated by averaging the differences between the first guess forecasts and the analyses for a winter month. Two bias correction paradigms have been tried. In the conventional paradigm, the estimated bias was removed from the background forecast before the assimilation. In the second paradigm, instead of removing the bias from the forecast, the observations were shifted toward the model attractor by adding the bias to the observations and the mapped observations were used in the assimilation. The latter paradigm aims to reduce shift-induced forecat errors due to intializing an imperfect model that is systematically different from the nature by using initial conditions that are close to the nature. The impacts of the two bias correction methods for the ensemble data assimilation system will be presented in the meeting.