Routine assimilation of radial velocity and reflectivity from Doppler radars and very short-term prediction at convective scales remains an outstanding problem in NWP. Application of four-dimensional variational schemes (4DVar) at these scale has shown promise, but there are a number of difficulties still to overcome, such as need for a covariance estimate for the background forecast, inherent nonlinearities in the dynamics and microphysics, and the need to maintain linearized and adjoint versions of the forecast model.

An alternative to 4DVar are filtering techniques that use ensembles of nonlinear forecasts to provide the necessary estimates of the background uncertainty, such as the ensemble Kalman filter and its variants. Although they are unproven, these techniques address many of the difficulties encountered with 4DVar. We will report results from preliminary tests of such techniques at the convective scales using simulated data (for example, from a numerical simulation warm thermal or spreading cold pool). These experiments also contain a wealth of information related to the dynamics and statistics of forecast errors at the convective scales.