9 WIND AND THERMODYNAMIC RETRIEVALS IN A SUPERCELL THUNDERSTORM : ENSEMBLE KALMAN FILTER RESULTS

Atmospheric state estimation on the scales of deep, moist convection will be an important component of operational high-resolution numerical weather prediction. For the foreseeable future, radar measurements of Doppler velocity and reflectivity will continue to be the primary source of volumetric observations on these scales. A variety of techniques have been developed for retrieving the wind, temperature, and moisture fields from radar observations of convective storms (e.g., Armijo 1969; Hane et al. 1981; Brandes 1984; Roux 1985; Ziegler 1985; Shapiro et al. 1995; Sun and Crook 1997, 1998; Weygandt et al. 2002a,b). These retrieval techniques range from direct solutions of limited sets of governing equations to data assimilation methods employing a 3D cloud model. Recently, Snyder et al. (2001) evaluated a relatively new data assimilation method – the ensemble Kalman filter (Evensen 1994; Houtekamer and Mitchell 1998) – for convective scale retrievals and forecasts. In tests on a simulated supercell thunderstorm, the ensemble Kalman filter scheme was stable and was able to reproduce rather accurately the original model fields from a limited number of observations (i.e., perturbed samples from the control simulation) (Snyder et al. 2001). An attractive feature of the ensemble Kalman filter approach is that once a forward model has been developed, relatively little additional coding is necessary to assimilate observations into the model. Therefore, this approach is being considered as an alternative to four-dimensional variational (4DVar) data assimilation, which requires the design and coding of an adjoint model. In this paper, we describe the first attempts to use the ensemble Kalman filter to retrieve the wind, temperature, and other fields from radar observations of a real thunderstorm. For the retrievals, we selected the 17 May 1981 Arcadia, Oklahoma tornadic supercell case, which includes observations from two 10-cm research Doppler radars that were separated by 40 km and that were 25-55 km from the target storm (Dowell and Bluestein 1997). Although our focus here is on the basic issues of using the ensemble Kalman filter on real data, we also plan to evaluate for the same case the strengths and weaknesses of the ensemble Kalman filter compared to the 4DVar method (Crook et al. 2002) and a single-Doppler retrieval method (Weygandt et al. 2002a, b).