ABSTRACT
The kriging predictions and prediction error variances are derived from a statistical model of the spatial variation of the study variable. Computing the kriging predictor requires a model for the covariance or semivariance as a function of the vector separating two locations. Often, the covariance is modelled as a function of the length of the separation vector only, so as a function of the Euclidian distance between two locations. Kriging with a prediction support that is larger than the support of the sample data is referred to as block-kriging. The more extreme the covariate value at the prediction location, the larger the kriging variance with kriging with an external drift. This is analogous to the variance of predictions with a linear regression model. The estimated parameters of the semivariogram model are plugged into the kriging equations. There are two different approaches for estimating the semivariogram parameters from sample data: the method-of-moments and maximum likelihood estimation.
