Statistical classification of the observation of nuggetless spatial Gaussian process with unknown sill parameter

The problem of classification of spatial Gaussian process observation into one of two populations specified by different regression mean models and common stationary covariance with unknown sill parameter is considered. Unknown parameters are estimated from training sample and these estimators are plugged in the Bayes discriminant function. The asymptotic expansion of the expected error rate associated with Bayes plug-in discriminant function is derived. Numerical analysis of the accuracy of approximation based on derived asymptotic expansion in the small training sample case is carried out. Comparison of two spatial sampling designs based on values of this approximation is done.