Kriging Versus Alternative Interpolators: Errors and Sensitivity to Model Inputs

Abstract

This study deals with interpolations through time and space. Two data sets are utilized, soil water content values at the 25 cm depth for a 45 hectare and an artificially-generated set of random but spatially correlated values. Interpolators include punctual kriging, a nonparametric estimator and co-kriging. Results show that although the model parameters depicting spatial interdependence (such as range and sill for a variogram) are highly dependent on the sample, the interpolations resulting from use of such models are not very sensitive to the model parameters. Co-kriging predictions were shown to be effective by using moisture content values for two different times as covariates. Additionally, an example illustrates the variogram for moisture content measured at one time can be adjusted and used for interpolations at a second time.