ABSTRACT
Consider the random field { Z(s) : s ∈ D ⊂ ℝ d } observed at locations s 1, ···, s n , and the corresponding data vector Z(s) = [Z(s 1), ···, Z(s n )]′. The domain D is fixed and continuous, hence we are dealing with geostatistical data. Our sample is an incomplete observation of the surface Z(s,ω) that is the outcome of a random experiment with realization ω. One of the pervasive problems in spatial statistics is the prediction of Z at some specified location s 0 ∈ D. This can be a location that is part of the set of locations where Z(s,ω) has been observed, or a new (= unobserved) location.
