A Parallel Implementation of the Ensemble Kalman Filter Based on Modified Cholesky Decomposition

摘要: This paper discusses an efficient parallel implementation of the ensemble Kalman filter based on modified Cholesky decomposition. The proposed starts with decomposing domain into sub-domains. In each sub-domain a sparse estimation inverse background error covariance matrix is computed via decomposition; estimates are concurrently separate processors. sparsity this estimator dictated by conditional independence model components for some radius influence. Then, assimilation step carried out in without need inter-processor communication. Once local analysis states computed, sub-domains mapped back onto global to obtain ensemble. Computational experiments performed using Atmospheric General Circulation Model (SPEEDY) T-63 resolution Blueridge cluster at Virginia Tech. number processors used ranges from 96 2,048. outperforms terms accuracy well-known transform (LETKF) all variables. computational time similar that LETKF method (where no performed). Finally, largest processors, 400 times faster than serial version method.