A Bayesian approach to geostatistical interpolation with exible variogram models

Keywords

Statistical learning
Code:
21/2009
Title:
A Bayesian approach to geostatistical interpolation with exible variogram models
Date:
Thursday 20th August 2009
Author(s):
Bonaventura, Luca; Castruccio, Stefano; Sangalli, Laura M.
Download link:
Abstract:
A Bayesian approach to covariance estimation and geostatistical interpolation based on flexible variogram models is introduced. In particular, we consider black-box kriging models. These variogram models do not require restrictive assumptions on the functional shape of the variogram; furthermore, they can handle quite naturally non isotropic random fields. The proposed Bayesian approach does not require the computation of an empirical variogram estimator, thus avoiding the arbitrariness implied by the construction of the empirical variogram itself. Moreover, it provides a complete assessment of the uncertainty in the variogram estimation. The advantages of this approach are illustrated via an extensive simulation study and by application to a well known benchmark dataset.
This report, or a modified version of it, has been also submitted to, or published on
Journal of Agricultural, Biological, and Environmental Statistics, Vol. 17, pp. 209-227, 2012