Yun Liu

University of Wisconsin - Madison, Atmospheric & Oceanic Sciences

A Random Subgrouping Scheme for Ensemble Based Filters

Room 811 AO&SS, February 15, 2012, 2:25 PM

Abstract

Ensemble based filters can be divided into two categories: stochastic and deterministic. Both types of filters suffer from the problem of generating outliers in the ensembles produced in a nonlinear system. This is especially true for the deterministic filter with a big ensemble as the outliers can persist for a long time, develop into extreme values, and produce large analysis errors. To address the problem of outliers, a new technique is developed that uses deterministic filter algebra but adds stochastic information into the filter system through random subgrouping. Test results, using the random subgrouping technique on two low-order models (Lorenz-63 and Lorenz-96), show that the new scheme significantly improves performance compared to both stochastic and deterministic filters.