Bias correction using analysis increments within an ensemble Kalman filter data assimilation

In the presence of model errors, the performance of the ensemble Kalman filter data assimilations can be degraded because the background uncertainty estimated by the ensemble forecast tends not to represent the systematic errors of the model completely. Thus, it is essential to estimate and to correct the systematic error of the forecast model in addition to inflate a background error covariance. We have developed a simple bias removal method using the analysis increment statistics within an ensemble Kalman filter data assimilation method and applied it to the carbon cycle data assimilation system under the imperfect model scenario with Observing Simulation System Experiments. This method does not require any other reference states such as a reanalysis data or other climatology data in order to estimate the model bias so that it allows us to correct the bias of the variables of which we do not have such a reference state. In this study, we assimilate the meteorological variables as well as atmospheric CO2 concentration and analyze not only these variables but also surface CO2 fluxes. Since the current knowledge of CO2 is not abundant enough to provide a definitive reference state of the carbon variables, we expect our new approach to reduce the model errors in the analysis of carbon variables besides the meteorological variables.n 8-9-2010-->