Kriging and Cokriging: Optimal Interpolation

In order to make our maps, we decided upon the method of cokriging. This method is optimal among linear estimators in the sense that it minimizes the estimation variance, Var .

We did so after looking at the results of mapping nitrates in at least five different ways:

• Kriging (multiple models),
• Cokriging (multiple models, against multiple variables),
• Kernel methods (multiple models), and
• Radial Basis Functions (multiple models).
• Using commercial (SPlus) and non-commercial (GRASS) ``black box'' routines.

We made our choice after using cross-validation, a leave-one-out procedure whereby we compare estimated values to true values, and base our decision on a set of criteria.