Numerical weather prediction at the convective scale remains an outstanding problem. At convective scales, numerical models are typically initialized with some initial disturbance, perhaps derived from Doppler radar observations, superposed on a horizontally uniform, environmental sounding. Radiosoundings are seldom available at the time and location of interest, and this is particularly problematic because of the well-known sensitivity of the resulting simulations to small changes in the sounding.

We investigate the possibility that the required soundings can be inferred or estimated given observations of radial velocity and reflectivity from Doppler radars, together with a forecast model and a reasonably sophisticated data assimilation scheme. It is clear that such observations contain significant, but indirect, information about the sounding: most simulations of observed convective storms involve some subjective manipulation of the sounding to improve the agreement of the simulations with radar observations. Parameter estimation within the context of an ensemble Kalman filter provides a more objective means of achieving the same end. We will report results from preliminary tests of such techniques using simulated data from numerical simulations of supercells and squall lines.