Fall 2000 Seminar Series - September 19, 2000
University of Minnesota
School of Statistics
College of Liberal Arts

Bayesian Spatial Interpolation

Benjamin Kedem
Math Department
University of Maryland, College Park

Tuesday, September 19, 2000
4:00 PM
217 Lind Hall
Minneapolis, East Bank Campus
Social at 3:30 PM, 300 Ford Hall

Abstract

Optimal statistical prediction (interpolation) in spatial random fields is an important problem in computer imaging, epidemiology, geology, hydrology, and meteorology, to mention a few areas. Most methods assume explicitly or implicitly the fields are Gaussian, however geophysical data are skewed more often than not. We present a Bayesian method for spatial prediction/interpolation that takes into account skewness, and compare it to kriging and trans Gaussian kriging using real and simulated data. New software will be described.