Statistical Sampling Strategies for Survey of Soil Contamination - Dealing with Contaminated Sites: From Theory Towards Practical Application - page 183

This technique is useful if the aim is to estimate the spatial mean, but for other
aims such as estimating a percentile, the entire SCFD, or the areal fraction exceed-
ing a threshold concentration it can not be used. The variation of the concentration
in the composite sample over repeated sampling will be small, because the spatial
variation between the aliquots of a composite sample is eliminated. The larger the
spatial variation between aliquots, the more precise the composite sample mean.
Bulking soil aliquots from different strata therefore pays. An example is where the
study area is geographically stratified (see Section 4.2.1.2 ), and one aliquot is taken
from each stratum to form a composite sample. Note that equal volumes of soil must
be collected at each location and the strata must have an equal surface area in order
to obtain unbiased estimates. If an estimate of the sampling variance of the esti-
mated mean is required, then several composite samples must be taken. If multiple
composite samples are taken, the spatial mean is estimated by the average of the
composite sample means

C

1
C

¯

y com = y i ; , (4.26)

i = 1

where C is the number of composite samples, and y i the (average) concentration of
composite sample i . The sampling variance of this estimated mean equals

V ( y i )
C

V (

y com ) ¯ = . (4.27)

Another application of composite sampling is group screening or group testing.
In group screening the aim is to determine whether, for instance, a contaminant or
a species of soil microbe, is present or not. The detection limit of the method for
(chemical) analysis must be low enough to detect the contaminant or species in a
(strongly) diluted sample.

4.2.5 Required Number of Sampling Locations

As stated in the Introduction (Section 4.1 ) the advantage of statistical methods for
survey is that the precision of the survey results (estimates) can be quantified. In the
formulas for the sampling variance, see Eqs. ( 4.4 ), ( 4.8 ), ( 4.15 ) and ( 4.18 ), spatial
variances appear, that are estimated from the same sample. If we are able, prior to
the sampling, to make a first guess at these spatial variances, we can compute in
advance the number of sampling locations required to achieve a given minimum
quality of the estimated target parameter. The computations described hereafter are
supported by the software Visual Sample Plan (VSP) (Matzke et al. 2007 ).

4.2.5.1 Constraint on Sampling Variance or Coefficient of Variation

The simplest situation is when the quality constraint is formulated in terms of the
sampling variance or standard deviation. For SI, the required number of sampling