A Comprehensive Bayesian Treatment of the Universal Kriging model with Mat\'ern correlation kernels

TL;DR

This paper develops a Bayesian approach using Gibbs reference posterior distribution for Universal Kriging with Matérn kernels, ensuring proper inference and good frequentist coverage in Gaussian process prediction.

Contribution

It introduces a full-Bayesian solution for Universal Kriging with Matérn kernels, providing explicit conditions for posterior propriety and demonstrating practical effectiveness through examples and simulations.

Findings

Gibbs reference posterior is suitable for Universal Kriging with Matérn kernels.

The method has good frequentist coverage properties.

Explicit conditions ensure the existence and propriety of the posterior.

Abstract

The Gibbs reference posterior distribution provides an objective full-Bayesian solution to the problem of prediction of a stationary Gaussian process with Mat\'ern anisotropic kernel. A full-Bayesian approach is possible, because the posterior distribution is expressed as the invariant distribution of a uniformly ergodic Markovian kernel for which we give an explicit expression. In this paper, we show that it is appropriate for the Universal Kriging framework, that is when an unknown function is added to the stationary Gaussian process. We give sufficient conditions for the existence and propriety of the Gibbs reference posterior that apply to a wide variety of practical cases and illustrate the method with several examples. Finally, simulations of Gaussian processes suggest that the Gibbs reference posterior has good frequentist properties in terms of coverage of prediction intervals.