It is well recognized that the three-dimensional data assimilation (3DVAR) generally utilizes an isotropic and static background error covariance, which does not reflect the day-by-day variation of the background forecast error covariance structure. A method to incorporate the flow-dependent ensemble covariance to the 3DVAR data assimilation system for the Weather Research and Forecasting model (WRF) is proposed. In the new data assimilation system, the initial ensemble perturbations are produced by the computationally efficient ensemble transform Kalman filter (ETKF). The extended control variable method is used to incorporate the ensemble covaraince into WRF-3DVAR when updating the ensemble mean.

The performance of the new method, i.e., the hybrid ETKF-3DVAR, are compared with the 3DVAR. The experiments are conducted by running WRF on a CONUS domain and assimilating rawinsonde observations during Jan. 2003. Results have shown that the extended control variable method can successfully incorporate the flow-dependent estimate of the background error covariance from the ensemble into the WRF 3DVAR system. Compared to the 3DVAR covariance, the ETKF ensemble covariance can more realistically reflect the synoptics of the background flow. The ETKF ensemble variance can distinguish forecasts with large error variance from forecasts with small error variance much better than the 3DVAR variance.