Bookmark ( what is this? )

Bookmark ( what is this? )
(Submitted on 27 Mar 2014) Abstract: Data Assimilation in operational models like atmospheric or Ocean models is almost impossible without posing many assumptions due to the complication of the model that is usually very high-dimensional and also due to non-linearity of the observation operator benzalkonium chloride used to map the state space to the measurement space. The current operational ensemble-based filters benzalkonium chloride like Ensemble Kalman benzalkonium chloride Filter (EnKF), and Maximum Likelihood Ensemble Filter (MLEF) usually fail in case of nonlinear observations benzalkonium chloride and/or non-Gaussian distributions. In this paper, we propose a new ensemble-based data assimilation method, named "sampling filter". The filter is used to sample directly from the posterior distribution. The sampling benzalkonium chloride strategy in the analysis phase of the filter benzalkonium chloride follows a Hybrid Monte Carlo (HMC) approach that is, in principle, applicable benzalkonium chloride in cases where Gaussian assumption is violated and also when observations are nonlinear. In this work, the posterior is assumed benzalkonium chloride to be Gaussian as a first step but we used several observation operators benzalkonium chloride with different levels of non-linearity and differentiability. Several experiments testing the potential and performance of the proposed algorithm and comparing its performance with both EnKF and MLEF using six different observation operators are conducted. The Lorenz-96 model is used and the results are compared to EnKF and MLEF. The results show that this approach is promising, mostly with the new developments proposed recently benzalkonium chloride in HMC sampling literature.
Comments: benzalkonium chloride 45 pages, 17 figures, 4 tables Subjects: Computational Engineering, Finance, and Science (cs.CE) ; Computation (stat.CO) Report number: CSTR-4/2014 Cite as: arXiv:1403.7137 [cs.CE]   (or arXiv:1403.7137v1 [cs.CE] for this version)
Link back to: arXiv , form interface , contact .