Application of Iterative Ensemble Square-Root Filter in Storm-Scale Data Assimilation

An iterative ensemble square root filter (iEnSRF) is designed on the basis of the latest asynchronous algorithm. In this iterative scheme, the forecast backgrounds at the analysis time and an earlier time are synchronously updated; then, the ensemble forecast is launched from the analysis field at the earlier time; finally, these two steps are repeated to producean iterative analysis of the background at the analysis time. The performance of this iterative scheme is examined using simulated radar data assimilation with an idealized storm case. The iEnSRF results are compared to those yielded by a traditional EnSRF. In addition, the performance of iteration involving only the background at the analysis time is discussed. The results obtained using data from a single simulated radar show that iEnSRF can effectively retrieve the positive feedback between the vertical motion and the latent heat release in the presence of a poor initial condition that provides no storm information. This improvement significantly optimizes the balance between different variables in the initial analysis and increases the convergence speed of assimilation. Conversely, the traditional EnSRF is unable to retrieve this positive feedback relationship in the initial analysis with the same poor initial condition, resulting in slower convergence and a larger analysis error. Iterative analysis cannot outperform the traditional EnSRF if the iteration considers only the background at the analysis time, indicating that considering two backgrounds at different times is necessary for the iterative algorithm to produce improvement. Results using data from two simulated radars show that the iEnSRF still outperforms the traditional EnSRF, especially in the upper troposphere. A comparison of the results using single radar data and dual radar data indicates that the traditional EnSRF cannot effectively use more data to improve the initial analysis of non-observed variables such as temperature, whereas the iEnSRF can effectively use more data to further improve the initial analysis.