Table 7. HMX concentrations from on-site colorimetric method for duplicate
subgrid subsamples.
HMX concentration (mg/kg)
Replicate
Variance RSD
Subgrid
Replicate 1
Replicate 2
Mean
variability*
%
D2C
210
337
273
1.60
8097
32.9
D2D
111
165
138
1.49
1446
27.7
D4A
341
444
393
1.30
5321
18.5
D7C
649
845
747
1.30
19281
18.6
D9B
116
120
118
1.03
9
2.4
D9C
1.4
51
26
6.4
1209
135
D10A
4.2
4.1
4.2
1.02
0
1.7
C4A
1170
1260
1220
1.08
4151
5.3
C4C
984
871
928
1.13
6432
8.6
Mean replicate variability (all values)
5.15
Mean replicate variability (values >50 mg/kg)
1.24
Mean RSD (all values) = 27.9%
Mean RSD (conc. > 50 mg/kg) = 16.3%
* Replicate variability is the higher value/lower value for replicates.
error, and the sampling error due to spatial het-
combination of spatial heterogeneity of analyte
erogeneity in analyte distribution within the
distribution, subsampling error, and error due to
subgrid. To assess the magnitude of these various
analysis. Likewise the mean RSD in Table 7 for
contributions to the total uncertainty, a series of
duplicate subsamples for samples with concen-
replicate subsamples were collected. For the nine
trations above 50 mg/kg was 16.3% and is due to
subgrids that were resampled, duplicate pile
the contribution of subsampling error and analy-
subsamples were collected and analyzed by the
sis error. From a series of measurements on a
on-site colorimetric method (Table 7). For eight of
single extract, the RSD for analysis was only 3.2%,
the nine subgrids, the ratios (higher duplicate
so the major portion of this 16.3% estimate is as-
value divided by the lower value) ranged between
sociated with subsampling and extraction. Based
1.02 and 1.60, with a mean of 1.24 indicating that
on a comparison of the variances associated with
we were able to reproducibly obtain subgrid
RSDs of 30.1% and 16.3%, only about 30% of the
subsamples. The replicates for subgrid D9C dif-
total variance is associated with subsampling and
fered by a factor of 36.4, but the concentrations
analysis while 70% is due to spatial heterogeneity
for both replicate samples were low. This anoma-
within the subgrid. It is interesting to compare
lous result appears atypical. With the contribu-
this result with those from the seven core samples
tions of spatial heterogeneity excluded as a source
in a wheel pattern, both here and in the earlier
of uncertainty, the results were in very good agree-
studies (Jenkins et al. 1992). The mid-range spa-
ment. An ANOVA was not conducted for this
tial heterogeneity error for area integrated samples
data set since variances were not homogeneous;
was only a little more than twice as large as the
however, RSD values were much more normally
subsampling plus analytical error, whereas the
distributed and were once again inversely related
comparable relationship for the core samples
to concentration.
yielded a difference of about 10 times. Clearly the
Taking the results from Tables 6 and 7 together,
area integrated samples do a better job of mini-
mizing error due to spatial heterogeneity. We
certainty due to spatial heterogeneity and
should note that area integrated samples can be
subsampling error. The mean RSD estimate when
obtained by a variety of protocols which might be
sampling a subgrid with concentrations above 50
equal to or better than the circular path used in
mg/kg was 30.1% (Table 6). This can be consid-
this study.
ered an estimate of total uncertainty due to the
17