be within a factor of 1.5 and differences greater than a factor of 2 should prompt
further inquiry into the method used to prepare the composites.
9. Short- and long-range spatial heterogeneity
The results required in section 8 provide at least eight pairs of samples separated
by 1 m. This will yield eight estimates of the relative standard deviation (RSD)
associated with short-range heterogeneity. When these estimates are all less than
100%, which would represent reasonably good homogeneity for munitions resi-
dues, they can be pooled to give an overall RSD. It would not be unusual for pairs
of core samples to have RSD values greater than 100% because they represent such
small volumes (Jenkins et al. 1997b). If one unit or one pair within a unit gives an
RSD estimate that is much larger than the others, and the concentrations are mod-
erate or high, reanalysis of new subsamples may be appropriate. When reanalysis
confirms an atypical result, that area may require more intensive sampling than the
other sections. However, when concentrations are low relative to quantitation lev-
els, larger RSDs are common and reanalysis is probably not necessary. Further-
more, when "area integrated" samples are employed, the heterogeneity will be
greatly reduced, usually by a factor of 10 or more, depending on how many aliquots
are included.
This same set of analyses also yields information on long-range spatial heteroge-
neity. The means of pairs within a unit can be compared and the unit means should
also be compared. Such results might lead to changes in unit assignment or they
may call for further preliminary samples as noted in section 2 above.
Concluding remarks
The array of results described above can be used to help plan grid layouts and
compositing strategies for a comprehensive sampling plan. Although we can not
assume a one-to-one correspondence of RSDs from core samples to larger surface
samples, the preliminary results are essential for choosing sampling depths and
extraction times, and for validating on-site compositing and analysis procedures.
Obviously, this preliminary plan is not a trivial exercise but, measured against the
cost of conducting a poorly designed full-scale sampling plan with costly off-site
analysis, we believe the expense is entirely justified and represents a cost-effective
approach. Such preliminary data should result in a full-scale plan that requires the
fewest possible analyses to produce reliable results. Further, the savings in analysis
costs and the timeliness of having results available offer tremendous advantages.
Specific guidance for compositing can only be given after data quality objectives
are specified for a site. For example, if concentration distribution within a unit is
required, compositing would be done within grids. These data might be used to
change remediation unit boundaries. It would also be feasible to study units
sequentially because of the fast turnaround with on-site analysis. In contrast, veri-
fication of the effectiveness of remediation might dictate that only a mean and
upper confidence limit is needed for each unit.
Consider the example cited earlier of a 50- 100-m remediation unit. Suppose
the preliminary study indicated the use of surface (0- to 5-cm) samples. The RSD
estimates for core samples separated by 1 m averaged 80%. If we use area inte-
grated composite samples containing 16 aliquots, we would expect the RSD for the
(
)
composite to be about 20% 80%/ 16 . This assumes that analytical error is small
compared to sampling error, a condition we have found in our studies (Jenkins et
al. 1997b). This also assumes perfect mixing of the 16 aliquots, which we know is
29
Previous Page
