A1569
Title: Matrix-free conditional simulation of Gaussian random fields
Authors: Somak Dutta - Iowa State University (United States) [presenting]
Debashis Mondal - Washington University in St Louis (United States)
Abstract: Conditionally simulating Gaussian random fields given a large set of observations presents significant computational challenges. These challenges stem from the necessity to compute and store large matrices or their explicit factorizations. Typically, these requirements grow at a super-linear rate with the number of observations. A novel approach that relies on matrix-free rectangular roots of precision matrices is introduced. This approach's applicability is demonstrated in a widely used class of spatial models and outperforms existing methods that use sparse matrix factorizations or other techniques. The effectiveness of the method is illustrated by applying it to analyze groundwater arsenic contamination in Bangladesh and ocean temperature measurements from Argo floats.
