In this paper, Pavel Sakov, Jean-Matthieu Haussaire and I generalize the Iterative ensemble Kalman filter (IEnKF) in the presence of additive model error, yielding the IEnKF-Q. The IEnKF is known as the most accurate scalable data assimilation method to address a mild nonlinearity in the dynamics of chaotic models. However, model error was not accounted for in the model so far. The generalization that we propose is rigorous and based on the IEnKF variational analysis. We also test the IEnKF-Q on the Lorenz 40-variable model. It consistently outperforms the EnKF which accounts for additive model error, and the IEnKF with a straightforward but approximate accounting of model noise.

The paper, simply entitled An iterative ensemble Kalman filter in presence of additive model error, will soon appear in the Quarterly Journal of the Royal Meteorological Society. Its preprint can be found here.