Statistical Sampling Strategies for Survey of Soil Contamination - Dealing with Contaminated Sites: From Theory Towards Practical Application - page 204

Fig. 4.16 Example of
two-way composite sampling
on a square grid. At the nodes
of the grid, two soil aliquots
are collected, one for a row
composite sample, one for a
column composite sample

sweep-out methods (Gore et al. 1996 ) require measuring individuals of a composite
as long as there is a possibility that there is a larger value among the unmeasured
individuals of that composite. After each analysis of an individual, the total of the
remaining, unmeasured individual samples is updated by subtracting the concentra-
tion of the measured individual sample from the previous total. This updated total
is used as the upper bound of the maximum of the “residual” composite, i.e., the
composite that could be formed by mixing the remaining individuals. The proce-
dure stops when the tentative global maximum exceeds the updated upper bound of
the maximum of all (residual) composites.
In the sweep-out methods described above, the individual to be measured in a
selected composite is selected randomly, which is reasonable because there is no
information available on the individual values. When composites are formed by
a two-way compositing design some information is available, and it is possible to
select the individual with the highest probability of having the maximum (Gore et al.
2001 ). Figure 4.16 shows an example of a two-way composite sampling design.
From every individual aliquot, two portions are taken, one of which is combined
with the portions from the same column to form a column composite, the other
is combined with the portions from the same row to form a row composite. The
individual that contributes to the column composite with the largest value and to
the row composite with the largest value has the largest probability of having the
maximum value, and will therefore be selected for measurement. For a sweep-out
method for this two-way compositing design, I refer to Gore et al ( 2001 ).
In the above methods, all values above a threshold concentration are identified
with certainty. The alternative is to accept some uncertainty, and to continue measur-
ing individuals until, for all (residual) composites, the probability that they include
an individual with a value above the threshold concentration is smaller than some
chosen limit (Carson 2001 ). Correll ( 2001 ) proposed to measure the individuals of

all composites having a concentration higher than z t / √ k , where z t is the threshold
concentration for the individual values, and k is the composite size. This modified
threshold concentration gave few false negatives (i.e., cases where the composite
is below the modified threshold concentration, while at least one of the individuals