An ensemble Kalman filter system for assimilation Doppler radar data for the initialization and prediction of convective storms is developed, based on a nonhydrostatic compressible model. The system has been tested with OSSE data sets. Issues including the ensemble size, covariance localization, the perturbation methods and their effects on analysis quality are examined. The relative role of radial velocity and reflectivity data in recovering the full flow and cloud fields will be compared. Hopefully, results from real data experiments will be also available by the due date of extended abstract or the time of conference.