Universal Kriging - 12.1 Introduction

In Chapter 1 we mentioned that up to three components are considered to represent the variability of regionalized variables : a deterministic component, a correlated component, and a random component. In Chapter 2 we considered some of the models based on the assumption of randomly distributed values: the random component was assumed to be the only one present. In Chapters 3 to 11, more general but also more complex models were described, based on the assumption that the values were correlated with the possible presence of random noise. In this chapter we introduce a model used to analyse non- stationary data, for which a deterministic component (or drift) must be assumed. This model will be used for estimation purposes, and the corresponding optimal linear estimation procedure is known as universal kriging. The theory of universal kriging is much more complex than the theory of kriging. Fortunately, in the majority of mine valuation problems the drift can be ignored and the simpler form of kriging can be used.