A1377
Title: Matrix-free conditional simulations of Gaussian random fields
Authors: Somak Dutta - Iowa State University (United States) [presenting]
Debashis Mondal - Washington University in St Louis (United States)
Abstract: In spatial analysis, conditional simulation of spatial variables at unobserved locations given observations at sampled locations is fundamental for statistical inference, yet existing methods based on explicit matrix factorization often face severe computational challenges for large datasets. A matrix-free approach for conditional simulation based on novel decompositions of the inverse covariance matrix enables scalable computation without explicit matrix factorization. The proposed framework applies to a broad class of spatial models, including Gaussian Markov random fields, fractional Gaussian fields, and Matern covariance models. Practical utility of the method is demonstrated through an application to mapping groundwater arsenic exceedance regions.
