B0567
Title: Spatial factor models based on fractional Gaussian fields
Authors: Somak Dutta - Iowa State University (United States) [presenting]
Subrata Pal - Iowa State University (United States)
Abstract: Factor models with spatially autocorrelated factor scores have become a popular tool for analyzing and predicting multivariate spatial data. However, estimating and prediction methods have largely focused on stationary spatial random fields for modelling the factor scores. The focus is on fractional Gaussian fields that cover a large class of spatial models, including intrinsic random fields. By embedding the spatial locations in a regular rectangular grid, a monotonic stochastic EM algorithm is proposed for maximum likelihood estimation of parameters. The approach is matrix-free and suitable for large spatial data with spatial misalignments. The methodology is illustrated on a dataset on groundwater mineral concentrations in Bangladesh.
