Since the development of synoptic, mesoscale, and cloud-scale research and prediction models, a fundamental goal has been to incorporate observations into these models such that more accurate atmospheric state estimates and forecasts may be produced. On the thunderstorm scale, early attempts to retrieve wind and thermodynamic fields from Doppler velocity and reflectivity observations were based on a static retrieval concept. More recently, four-dimensional variational (4DVar), or adjoint, schemes have been applied to convective-scale observations. Although 4DVar appears to be an appropriate method for cloud-scale state estimation, the expense of developing and modifying a 4DVar system and the difficulties involved with nonlinear moist processes make this approach problematic. A potential alternative being evaluated is the ensemble Kalman filter (EnKF).

We will report on preliminary results of the application of an EnKF data assimilation methodology to dual-Doppler observations of the 17 May 1981 Arcadia, Oklahoma tornadic supercell thunderstorm. Since this is the first attempt to use an EnKF approach to retrieve wind and temperature from real observations of deep, moist convection, we will address fundamental questions about this assimilation methodology:

* How do the results depend on the technique used to initialize the ensemble members?

* What are the apparent strengths and weakness of the EnKF method, compared to traditional retrievals and 4DVar?