Y. Liu, X. Rong, Z. Liu, S. Wu, S. Zhang, and R. Jacob, "A Random Subgrouping Scheme for Ensemble Based Filter (sEnSRF)," Preprint ANL/MCS-P1978-1111, November 2011. [pdf]
Ensemble based filters can be divided into two categories: stochastic filter and deterministic filter. Both suffer an outlier problem when they are applied to a nonlinear system, especially for the deterministic filter. A nonlinear system generates outliers in an ensemble. For a deterministic filter, the outliers can persist for a long time and develop into extreme outliers, which tend to generate large analysis errors. To address this problem, a random subgrouping technique is developed here to overcome the effect of outliers in deterministic filters. The new technique uses deterministic filter algebra but adds stochastic information into the filter system through random subgrouping. Test results using random subgrouping technique with two low-order models (Lorenz-63 and Lorenz-95) show that the new scheme dramatically improves performance compare to both stochastic and deterministic filters.