An Iterative Ensemble Transform Kalman Filter Based on the Smoothing and the Filtering Steps

Li Chen,Xiao Ding,Dunbiao Niu,Zhujun Cao,Enbin Song

DOI: https://doi.org/10.1109/iccais52680.2021.9624649

2021-01-01

Abstract:For nonlinear systems, the performance of traditional filtering algorithms is generally not very good, so we consider using the observation information to implement a smoothing step before the forecasting step. In this article we propose a smoothing iterative formulation of the ensemble transform Kalman filter (SIETKF) in the perfect-model, which is similar to the iterative ensemble Kalman filter (IEnKF -S) by Sakov and Oliver. First we propose a two-step (the smoothing and the filtering steps) framework of the data assimilation. Specifically, we collect the observation information at the current moment, which is used to smooth the state at the previous moment (i.e., the smoothing step), and then complete the filtering based on the observation information and the new smooth state (i.e., the filtering step). Next, inflation is further implemented to improve the robustness of the algorithm and to decrease the sample errors. We compare the performance of the proposed algorithm with other algorithms in 3-element and 40-element Lorenz models. The experimental results show that our algorithm performs better than two traditional algorithms. In particularly, for the strong nonlinear model, it outperforms the algorithm that uses the observation information at the next moment (IEnKF -S).