Improving the singular evolutive extended Kalman filter for strongly nonlinear models for use in ocean data assimilation (Q2764820)

The authors are interested in the use of the singular extended evolutive Kalman (SEEK) filter in the context of a strong reduction of rank and with strongly nonlinear models. The objectives of the article are to understand the difficulties which may arise in using the SEEK filter with a nonlinear model even if the model and the observations are perfect. After a review of the basic principles of the SEEK filter algorithm they propose two different algorithms capable of retrieving the useful information available in the tangent linear model, in spite of its time evolution. These different new versions of the SEEK are tested with a time-varying linear model and compared with the initial SEEK algorithm. These new algorithms are tested in a strongly nonlinear data assimilation experiment using the Lorenz model.