A1467
Title: Physics-guided space-time mapping of sea surface height: Novel covariance models and methods
Authors: Arthur Guillaumin - Queen Mary University of London (United Kingdom) [presenting]
Cimarron Wortham - NorthWest Research Associates (United States)
Early Jeffrey - NorthWest Research Associates (United States)
Cheuk Yan Cliff Lo - Queen Mary University of London (United Kingdom)
Abstract: The purpose is to design a space-time covariance model of sea surface height of the oceans by leveraging large gridded simulation data from an idealized ocean model run in various physical regimes. Two approaches are considered: one parametric, which relies on a mixture of Lagrangian space-time covariance models, and another one akin to a non-parametric approach, where a deep neural network is trained to directly infer the frequency-wavenumber spectrum of the process. Some novel approaches are also discussed for efficient approximate inference, both for the estimation of parametric covariance models and for Gaussian prediction. The models and methods are applied to the mapping of sea surface height measurements: Firstly, on simulation data from a coupled climate model, and secondly, to real satellite data.
