The ultimate success of ensemble data assimilation techniques (e.g., ensemble Kalman Filtering) depends on the existence and accurate estimation of error covariances that provide spatial correlations between values of the same variable at different grid points and those between different variables at the same or different spatial locations. For the problem of assimilating radar data, the basic observed variables are reflectivity and the radial wind component, while the most important unobserved variables are the tangential and vertical wind components. A successful ensemble data assimilator would be able to estimate the tangential wind component given the radial wind component and covariance statistics. Such a method could then be employed as a Single-Doppler Velocity Retrieval (SDVR) method. Ensemble data assimilation algorithms have been proposed as a general solution to the 4-D data assimilation (4DDA) problem. However, we attempt SDVR as a potentially more telling test of the method as we are directly trying to retrieve something which is completely unobserved using ensemble statistics. The same codes developed for SDVR could be employed for doing 4DDA. Doing SDVR by ensemble data assimilation still requires the running of a complex 4-D model (the ARPS) in order to generate the covariance statistics, but the mean flow field is steady-state.

This paper reports on recent results implementing ensemble data assimilation with particular attention to the accuracy and computational effort of the method and the necessary size of the ensemble.