Sampling

Abstract

Reliable analysis and kriging demand sound sampling, which must be sufficient and have an acceptable configuration. Sampling to estimate the variogram is problematic because the spatial scale of variation is often unknown, yet there must be numerous pairs of sampling points within the correlation range, if it exists. One might determine the spatial scale from visible features such as landforms and vegetation on the ground or from remote sensing. If that is not possible a nested survey and hierarchical analysis, by either analysis of variance or residual maximum likelihood (REML), can provide a first approximation to the variogram and a guide for subsequent sampling. Variograms from previous surveys or from ancillary data, in particular aerial image data, may also be used to guide sampling. Once a variogram with known parameters is available sampling for kriging can be optimized so that some tolerable kriging error is met but never exceeded. Alternatively, if the budget for sampling is set the kriging equations can be solved to determine the kriging errors everywhere within the region of interest and in particular the maximum absolute error.