Statistical Sampling Strategies for Survey of Soil Contamination - Dealing with Contaminated Sites: From Theory Towards Practical Application - page 197

that the variogram is known. The optimized sampling pattern is optimal for spatial
interpolation with this prior variogram. In general this sampling pattern is far from
optimal for estimating the variogram. To estimate the nugget (intercept) of the vari-
ogram and to choose between alternative variogram model types, for instance, quite
a few pairs of points are needed with short mutual distances. Such pairs are missing
in grids, spatial coverage samples and geostatistical samples designed for spatial
interpolation only. This suggests that some compromise between uniform spatial
coverage (for interpolation) and spatial clustering (for estimating the variogram)
might be a good option. A simple, practical procedure is

=

n int + 1. split the total sample in two, n n var ;

2. select purposively n int locations on a grid, or design a spatial coverage or
geostatistical sample of n int locations;
3. supplement this first sample by n var locations at short distance of the locations of
the first sample;

To avoid spatial clustering of short distance locations in certain parts of the area,
we recommend selecting the grid nodes that will receive a short distance location
purposively and not at random, for instance by subsampling the regular grid system-
atically (see Fig. 4.13 ). Also, we recommend placing the additional locations on the
sides of the grid cells, so that the directions for the smallest lag coincide with those
of the larger lags.

gridpoints
short distance point

Fig. 4.13 Square grid sample with additional short distance locations for variogram estimation

4.4.2 Spatial Interpolation

Contrary to design-based sampling strategies in which the estimation is based on the
method used for selecting the sampling locations (sampling design), in model-based
sampling strategies, spatial interpolation is based on the model of spatial variation.