A0520
Title: Matrix-free conditional simulations of Gaussian fields on a regular lattice
Authors: Somak Dutta - Iowa State University (United States) [presenting]
Debashis Mondal - Washington University in St Louis (United States)
Abstract: Conditional simulations of random fields given observations at finitely many sampled locations provide realistic depictions of their local variations. However, traditional methods using dense matrix computations are not scalable because their storage and computational complexity scale at least quadratically with the sample size. A new matrix-free algorithm for conditional simulation for a large class of spatial mixed models will be discussed, and its use in spatial mapping and downscaling will be demonstrated.