Plenary Lecture
Background Error Estimation in Sequential Data Assimilation Methods
Professor Elias D. Niño-Ruiz
Department of Computer Science
School of Engineering
Universidad del Norte
Barranquilla, Colombia
E-mail: enino@uninorte.edu.co
Abstract: Data Assimilation methods are based on adjusted forecasts of imperfect numerical models making use of noisy, real observations. In practice, many challenges are presented during the assimilation of observations: background error estimates are approximated based on few samples of the numerical model, observational operators target just a few components of the model state (sparse observational operators), the number of model components ranges in the order of millions, and Gaussian assumptions on background and observation errors are may not satisfied. This talk covers different methodologies in order to overcome these situations. Efficient and practical covariance matrix estimators are presented for their use in the context of high-dimensional probability error distributions and even more, parallel resources can be exploited in order to speed-up algebraic computations. We highlight two well-known covariance matrix estimators from the statistical literature: the Rao-Blackwell Ledoit and Wolf estimator and the modified Cholesky decomposition for inverse covariance matrix estimation. Some comparative examples for the discussed methods are presented making use of an Atmospheric General Circulation Model. The numerical results reveal that the use of covariance matrix estimation in the context of data assimilation can improve the quality of background error estimates and therefore, the impact of spurious correlations can be mitigated. Even more, the proposed methods are attractive since their practical and parallel implementations are straightforward when observational operators are linear.
Brief Biography of the Speaker: Elias D. Niño-Ruiz obtained the Diploma in System Engineering (2007) from the Universidad del Norte (UniNorte) Barranquilla, Colombia, M.S. in System Engineering (2009) and M.S. in Industrial Engineering (2010) from UniNorte, and Ph.D. in Computer Science and Applications (2015) from the Virginia Polytechnic Institute and State University (Virginia Tech) Blacksburg, USA. He worked as summer-intern at Argonne National Laboratory, USA (2013) and Lawrence Livermore National Laboratory, USA (2014). He served as researcher (2011-2015) in the Computational Science Laboratory as well as instructor of Numerical Methods (2015) at Virginia Tech. In 2016, he joined UniNorte's Department of Computer Science. Niño's research interests are in the area of computational science and engineering.