Statistical Sampling Strategies for Survey of Soil Contamination - Dealing with Contaminated Sites: From Theory Towards Practical Application - page 168

mean, median or 90th percentile. In these so-called reconnaissance surveys, it often
suffices to know what concentrations occur in the area and how often, without know-
ing where these concentrations are. If the reconnaissance survey shows that the
actual concentrations frequently exceed a specific threshold concentration, i.e., over
a large proportion of the area, then we might decide to sample suspected areas addi-
tionally, so that we obtain more information on where these concentrations occur.
For estimating (parameters of) the SCDF, I generally prefer a design-based
approach over a model-based approach, because in estimating the parameters of
the distribution and their precision, postulation of a model of spatial variation (var-
iogram) is not needed (Brus and de Gruijter 1997 ). By avoiding such a model, the
quality of the estimates are independent of the quality of model-assumptions, simply
because such assumptions are not needed. This property of the results is called valid-
ity. Valid results may be especially important when the survey results are used to
check whether the status of the surveyed area complies with regulatory soil quality
standards.
In practice, probability sampling may be hampered by enclosures that are part
of the target population but are inaccessible for sampling, e.g. under sealed parts of
the area. If a randomly selected location happens to fall in such an enclosure, or if
in the field it appears that the location falls outside the target population, this loca-
tion must be eliminated from the list of sampling locations, and replaced by a new
sampling location from a reserve list in pre-determined order. Shifting such loca-
tions to nearby accessible locations inside the target population may cause biased
estimates.
Hereafter, I will describe several sampling designs (Section 4.2.1 ), and how
(parameters of) the SCDF can be estimated from the sample (Section 4.2.2 ). In
Section 4.2.3 . I will describe how ancillary information such as maps with covari-
ates, or existing measurements of the contaminant concentrations at point locations
can be used at the estimation stage of a sampling strategy. Section 4.2.4 describes
composite sampling, in which the soil taken at different sampling locations is mixed
(bulked) to save laboratory costs. The final Section 4.2.5 deals with the decision on
the number of sampling locations.
4.2.1 Sampling Designs
There are many sampling designs described in textbooks that are appropriate for
surveying contaminated soils, see for instance de Gruijter et al. ( 2006 ). Here, I
will describe only a few simple sampling designs that are applicable in many com-
mon situations. Section 4.2.1.4 touches on other, more advanced sampling designs,
and describes when these alternative designs can be advantageous. I prefer simple
sampling designs to keep estimation as simple as possible. Also, if one has many
soil contaminants to be surveyed, we may optimize the sampling design for one or
two contaminants, but this design can be far from optimal for other contaminants.
For instance, the precision of the estimated mean concentration of a contaminant
may be considerably increased by selecting sampling locations with probabilities