A sparse matrix formulation of model-based ensemble Kalman filter

Abstract We introduce a computationally efficient variant of the model-based ensemble Kalman filter (EnKF). propose two changes to original formulation. First, we phrase setup in terms precision matrices instead covariance matrices, and new prior for matrix which ensures it be sparse. Second, split state vector into several blocks formulate an approximate updating procedure each these blocks. study simulation example computational speedup approximation error resulting from using proposed approach. The is substantial high dimensional vectors, allowing run on much larger problems than can done with In introduced block negligible compared Monte Carlo variability inherent both procedures.