An intercomparison of the Environment Canada variational (3D-Var and 4D-Var) and ensemble Kalman filter (EnKF) data assimilation systems is being conducted in the context of producing global deterministic numerical weather forecasts. Both 3D-Var and 4D-Var experiments are considered that each use either the background error covariances similar to those used operationally, which are nearly static with horizontally homogeneous and isotropic correlations, or flow-dependent covariances based on the EnKF background ensembles. An EnKF experiment, run with the same horizontal resolution as the 4D-Var inner loop, uses the mean of each 96-member analysis ensemble to initialize the higher resolution deterministic forecasts. In addition, the Ensemble-4D-Var approach is evaluated. This approach uses 4D flow-dependent background error covariances estimated from the EnKF ensembles to produce a 4D analysis with the variational data assimilation system, but without the need of tangent-linear or adjoint versions of the forecast model. All experiments assimilate the same full set of meteorological observations and use the same configuration of the forecast model to produce medium-range forecasts.
Results show that use of the 4D-Var analysis, with the background error covariances similar to those used operationally, or the EnKF ensemble mean produces forecasts of comparable quality. A positive impact is obtained from using the EnKF flow-dependent error covariances in the variational systems (gain of ~10 hours at day 5 in southern extra-tropics vs. standard 4D-Var). Finally, three distinct configurations of the variational data assimilation system were compared with each using EnKF background-error covariances: the Ensemble-4D-Var approach produces improved forecast quality relative to 3D-Var, but not better than 4D-Var.