The asymptotic variance of the maximum likelihood estimate is proved to decrease when the maximization is restricted to a subspace that contains the true parameter value. Maximum likelihood estimation allows a systematic fitting of covariance models to the sample, which is important in data assimilation. The hierarchical maximum likelihood approach is applied to the spectral diagonal covariance model with different parameterizations of eigenvalue decay, and to the sparse inverse covariance model with specified parameter values on different sets of nonzero entries. It is shown computationally that using smaller sets of parameters can decrease the sampling noise in high dimension substantially.
<p>A widely popular group of data assimilation methods in meteorological and geophysical sciences is formed by filters based on Monte-Carlo approximation of the traditional Kalman filter, e.g. <span>E</span><span>nsemble Kalman filter </span><span>(EnKF)</span><span>, </span><span>E</span><span>nsemble </span><span>s</span><span>quare-root filter and others. Due to the computational cost, ensemble </span><span>size </span><span>is </span><span>usually </span><span>small </span><span>compar</span><span>ed</span><span> to the dimension of the </span><span>s</span><span>tate </span><span>vector. </span><span>Traditional </span> <span>EnKF implicitly uses the sample covariance which is</span><span> a poor estimate of the </span><span>background covariance matrix - singular and </span><span>contaminated by </span><span>spurious correlations. </span></p><p><span>W</span><span>e focus on modelling the </span><span>background </span><span>covariance matrix by means of </span><span>a linear model for its inverse. This is </span><span>particularly </span><span>useful</span> <span>in</span><span> Gauss-Markov random fields (GMRF), </span><span>where</span> <span>the inverse covariance matrix has </span><span>a banded </span><span>structure</span><span>. </span><span>The parameters of the model are estimated by the</span><span> score matching </span><span>method which </span><span>provides</span><span> estimators in a closed form</span><span>, cheap to compute</span><span>. The resulting estimate</span><span> is a key component of the </span><span>proposed </span><span>ensemble filtering algorithms. </span><span>Under the assumption that the state vector is a GMRF in every time-step, t</span><span>he Score matching filter with Gaussian resamplin</span><span>g (SMF-GR) </span><span>gives</span><span> in every time-step a consistent (in the large ensemble limit) estimator of mean and covariance matrix </span><span>of the forecast and analysis distribution</span><span>. Further, we propose a filtering method called Score matching ensemble filter (SMEF), based on regularization of the EnK</span><span>F</span><span>. Th</span><span>is</span><span> filter performs well even for non-Gaussian systems with non-linear dynamic</span><span>s</span><span>. </span><span>The performance of both filters is illustrated on a simple linear convection model and Lorenz-96.</span></p>
<p style='text-indent:20px;'>We present an ensemble filtering method based on a linear model for the precision matrix (the inverse of the covariance) with the parameters determined by Score Matching Estimation. The method provides a rigorous covariance regularization when the underlying random field is Gaussian Markov. The parameters are found by solving a system of linear equations. The analysis step uses the inverse formulation of the Kalman update. Several filter versions, differing in the construction of the analysis ensemble, are proposed, as well as a Score matching version of the Extended Kalman Filter.</p>
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.