ronmental conditions and was deposited on site
Duplicate Method 8330 results by commercial
as particles.
laboratory analysis were much less precise (RSD
Table 3 also includes statistical analyses for the
= 19.2%) than those from the colorimetric and
composite sample. The mean values from the
acetone HPLC methods. Apparently any addi-
tional homogenization of samples by the com-
analysis of the composite are within 10% of the
mercial laboratory was inadequate to overcome
mean values for the seven discrete samples that
the effect of using a much smaller sample size (2
make up the composite for each method. The RSDs
g) for Method 8330. The correlation coefficient for
for the on-site colorimetric and acetone HPLC
on-site colorimetric and Method 8330 results was
were 7.3% and 2.8%, respectively. These findings
only 0.945 and the slope of the zero intercept
reinforce our earlier conclusion that preparing a
regression relationship was 0.813. Some decrease
homogeneous and representative composite for a
in the correlation coefficient might be expected
set of discrete samples is feasible and does not
due to the relatively large random error associ-
require sophisticated equipment or exceptional
ated with the Method 8330 results. However, the
effort or time when 20-g subsamples are used for
major reason for the poorer correlation coefficient
analysis. The ability to prepare adequate compos-
for this relationship is that the two sets of analy-
ites for on-site analysis documented here, and
ses were done on different subsamples of soil,
elsewhere (Jenkins et al. 1996), is particularly im-
unlike the on-site colorimetric and acetone HPLC
portant. Providing adequate characterization of
results, which were obtained from the same ac-
the mean concentration for even a small geo-
etone extract. This is an important observation
graphical area using discrete samples would re-
since validation of results from on-site methods is
quire large numbers of samples and analyses, gen-
invariably conducted with different subsamples
erally beyond the financial resources available for
of soil (splits), and differences observed may be
a specific investigation. However, the use of com-
due to actual differences in the analyte content in
posites and on-site analysis can effectively deal
the subsamples analyzed rather than differences
with this problem.
due to the analytical methods.
The much higher RSD for the commercial
Mean concentrations for the seven discrete
Method 8330 results (47.1%) agrees with similar
samples analyzed by the on-site colorimetric
results for the discrete samples and is likely due
method ranged from 16 mg/kg to 324 mg/kg.
in large part to the 2-g subsample size, which is
Analysis of variance (ANOVA) indicated that
inadequate to moderate the remaining heteroge-
there was a significant difference among samples
neity within the bulk sample. The practice of
at the 95% level for all three methods of analysis
scooping a subsample from the top of a sample
with F ratios of 257, 478, and 27.1, respectively,
bottle, common in many laboratories, may have
for the on-site colorimetric, acetone HPLC and
exaggerated the heterogeneity problem. Because
Method 8330 (Table 3). The results from ANOVA
segregation of particles can occur due to vibra-
were used to partition the error from these three
tion during shipping and storage (even for previ-
sets of data into analytical error (SA ) and sam-
ously homogenized samples), careful subsampling
pling error (SS) associated with short-range spa-
requires rehomogenization of the entire sample
tial heterogeneity (Table 3). RSDs due to sam-
prior to subsampling.
pling were much higher than those due to analysis
and were in quite good agreement for the three
Results for characterization
sets of data (sampling RSDs ranged from 68.2% to
of grid-sized areas for HMX
77.9%). For the on-site colorimetric data, sampling
In the next phase of the study, the site was
divided into sixteen 6- 6-m square grids be-
error was over 11 times the error due to analysis.
These findings are in excellent agreement with
tween target tanks C and D, and extending in
those obtained in our earlier study where the con-
front and beyond the target tanks as shown in
Figure 10. Each grid was then divided into 3-
taminants present were TNT, DNT or ammonium
picrate. In that study, four sampling locations
3-m quadrants that were labeled subgrids AD;
yielded data from the seven discrete samples that
letter designations were assigned in a clockwise
were sufficiently homogeneous to allow partition
fashion starting with the upper left quadrant. In-
of variances into sampling and analytical error,
dividual analyses for surface soil samples obtained
and the SS/SA ratio ranged from about 6 to over
in each subgrid and for grid composites are pre-
22. This similarity is not surprising since HMX,
sented in Appendix A. A diagram showing an
like TNT, 2,4-DNT and picrate, is a solid at envi-
overview of the HMX concentrations obtained
13
Previous Page
