A1612
Title: Bayesian inference for geometrically anisotropic spatial random fields on regular lattices
Authors: Fan Dai - North Dakota State University (United States) [presenting]
Somak Dutta - Iowa State University (United States)
Abstract: Geometric anisotropy arises when the dependence structure of a spatial random field varies with direction. A Bayesian inference framework is developed for a class of geometrically anisotropic random fields defined on regular lattices. The models are constructed through fractional Laplacian differencing on the lattice, which provides a flexible way to capture directional dependence and spatial smoothness. It is further shown that, as the lattice spacing decreases, this class of lattice-based models converges to a continuum family of anisotropic Matern-type random fields, thereby establishing a theoretical link between discrete and continuous spatial models. The framework enables principled uncertainty quantification for anisotropy and other dependence parameters. The methodology is illustrated through an analysis of ocean chlorophyll concentration data from NASA's MODIS-Aqua project.
