An ensemble four-dimensional variational (4DVAR) data assimilation algorithm will be presented. The method is applied to the NCEP's limited area eta-model in coarse resolution (160km/38lyrs), using operationally available observations.

The algorithm optimally combines nonlinear ensemble forecasts with the 4DVAR data assimilation. Some issues regarding its practical applications, such as error covariance localization, computational cost, need for the adjoint model, etc., will be discussed.