The Analog Ensemble Kalman Filter and Smoother

In classical data assimilation using sequential Monte Carlo methods, a physical model is run at each time steps to simulate members corresponding to different forecast scenarios. In this paper, we propose to use statistical analogs provided by observational or model-simulated data to emulate the dynamical model and generate relevant forecast members. This new methodology is called AnEnKF/AnEnFS for Analog Ensemble Kalman Filter and Smoother. We test our methodology using the Lorenz-63 model. The simulations indicate that, for a rich analog database, the assimilation results with the AnEnKF/AnEnFS are comparable to those obtained using the Lorenz dynamical equations into a classical Ensemble Kalman Filter/Smoother.