Comparisons of Storm-Scale Doppler Radar Data Assimilation using LETKF and EnSRF

Ensemble Kalman Filtering (EnKF) data assimilation using Doppler radar observations has been proven useful for examining the atmospheric state of a convective storm and initialization of storm-scale forecasts. The Ensemble Square Root Filter (EnSRF) and the Local Ensemble Transform Kalman Filter (LETKF) are two popular variations of EnKF. EnSRF method requires the observations to be processed one at a time to avoid the computation of matrix square roots. LETKF uses all observations (within the localization region) to simultaneously update the state variables. In practice, both methods restrict the update to state variables that are within a certain radius of the observation location because model variables far away from the observation are assumed to have no covariances with the observation and the sampling error associated with estimating the covariances from a finite sample become large. LETKF localizes the impact of observations by increasing the observational error, instead of directly localizing background error covariances, which is done in the EnSRF. The effects of such differences have not been studied in storm scale radar data assimilation.

This study uses an Observing System Simulation Experiment (OSSE) of a supercell storm to compare the LETKF and EnSRF methodology for storm-scale analysis and forecast. Initial results show similar storm–scale analyses from both methods. We will continue to explore the potential benefits of a simultaneous update on improving dynamic balance and investigate the impacts of localization. Both filters will also be applied to real data from the May 8, 2003 supercell to further compare the similarities and differences of the methods. The scalability of the two methods will also be examined.