The ensemble Kalman filter (EnKF) and its variants are promising candidates for a variety of assimilation problems in atmospheric science. The EnKF employs flow-dependent forecast error covariances (estimated from an ensemble of forecasts), assimilates asynoptic observations at their proper time, and produces an estimate of analysis uncertainty in the form of an ensemble of analyses. At the same time, the EnKF is relatively simple to implement for large atmospheric models, as it does not require minimization techniques or the model's adjoint.

We have implemented an EnKF for the Weather Research and Forecasting (WRF) model. The feasibility of the scheme has been explored in experiments with simulated observations for both synoptic-scale domains covering the continental United States and much smaller domains in which moist convection is explicitly resolved. The EnKF provides excellent estimates of the atmospheric state given, in the former case, observations broadly representative of the North-American radiosonde network or, in the latter case, a single Doppler radar. We will report results from ongoing experiments comparing the EnKF to the three-dimensional variational assimilation scheme (3DVar) available for WRF using both simulated and actual observations.