Investigating Sources of Inaccuracy in the Analysis and Forecast of a Real Tornadic Thunderstorm Case with the EnKF Method through OSS Experiments

The ensemble Kalman filter (EnKF) algorithm was applied to assimilate real radar data for the central Oklahoma tornadic thunderstorm case of 29-30 May 2004. Although the algorithm works almost perfectly when applied to the assimilation of simulated radar data under the perfect model assumption for modeled convective storms, the results of this real case study are, however, much less satisfactory. The main problem with the analysis is that the cross-beam wind component can not be retrieved accurately when data from a single WSR-88D radar are assimilated. The predicted storm propagates faster than the true storm. No significant improvement was obtained in the analysis or forecast with various tuning to the filter, while efforts to improve the storm environment by analyzing all available conventional observations using 3DVAR did not help much either.

The lack of low-level data coverage due to the relatively large distance of the storm from two closes WSR-88D radars; inevitable model errors, especially those related to resolution and microphysics; and possible errors in the environmental condition can all contribute to the inaccuracy in the analysis and forecast. But the major challenge with real cases is that we do not have a good knowledge of the different sources of errors, especially when they co-exist, and there is no truth for analysis verification. To investigate and understand the sources of inaccuracy in the analysis and forecast of this case, a simulated storm was first produced, which propagates roughly at the same speed and direction of the real storm, and a set of OSS experiments were designed to mimic the setup of the real data experiment, and the impacts of different errors or factors on the analysis are examined separately. It is hoped that the results of these OSS experiments will help us find ways to improve the analysis and forecast of the real case.