Short-range heterogeneity is estimated using the

the sample size study. Here, we are looking at our

RSDs from the 10 pairs (two pairs at each depth)

ability to obtain replicate analytical-size

separated by 1.0 m. These RSDs range from 7.3 to

subsamples from a carefully homogenized bulk

127%, with a pooled RSD of 71.8%. Although we

sample. Each subsample goes through the entire

expect RSDs to increase as concentrations decrease

extraction and HPLC analysis procedure. The

(Horowitz 1982), that pattern is not apparent for

RSDs for HMX for the 12 duplicates ranged from

these samples. What is clear is that close-lying

1.6 to 99.7%, with a pooled RSD of 55.4%. In this

samples can produce widely disparate results, and

case, RSDs do appear to vary with concentration,

our ability to estimate mean concentration for a

although the pattern is not completely consistent.

grid using a single discrete sample is very poor.

For samples with concentrations above 10 mg/kg,

While the concentrations of RDX, TNT, 4-AmDNT,

the pooled RSD is 32%, compared to 74% for

and 2-AmDNT are quite low relative to HMX at

samples with concentrations below 10 mg/kg.

Over all, these RSDs are considerably larger than

dard deviations are similar to their means, and

those we have observed for similarly processed

hence our ability to estimate mean concentrations

samples at other explosives sites, which ranged

from a single discrete sample for these compounds

from 4.5 to 13.5% (Jenkins et al. 1996).

is also very poor (Table 3).

We conducted a sample size study to assess the

In sampling area 1-48, there are an insufficient

reproducibility of analysis as a function of the size

number of cases where measured values were

of subsample used. This study (Table 5) yielded

above MDLs to obtain good estimates of standard

some unexpected results. We expected to see mea-

deviations (Table 4). Even so, inspection of the data

surable reduction in RSD estimates as sample size

tells us that HMX and RDX concentrations for the

increased, but that trend was not present. The RSD

four discrete samples at the same depth for a given

estimates are much lower than those observed for

grid are so disparate, that, here again, it would be

the 12 duplicates discussed earlier; they are all in

impossible to obtain good estimates of mean con-

the 1020% range, except for the 0.5-g sample size

centrations from analysis of single discrete

and one 2.0-g ground sample. The mean concen-

samples.

tration estimates from different sample sizes

Finally, the very-short-range heterogeneity is

showed excellent consistency, but they are consid-

examined using the 12 sets of duplicate samples

erably lower than the original estimates (Table 2).

from location 1-44, coupled with the results from

The initial HMX estimate for sample 1-44-8 was

1448 a*

47.9

29.0

39.5

33.6

33.1

1448 b

41.1

26.9

51.2

67.8

57.8

1448 c

54.4

36.7

36.7

56.8

19.2

1448 d

42.9

30.0

35.6

47.5

65.9

1448 e

38.9

44.3

42.9

22.8

43.5

1448 (X%RSD)†

45.1 13.7

33.4 21.3

41.2 15.2

45.7 39.3

43.9 42.7

14411 a

168

171

132

113

73.5

14411 b

128

120

133

145

99.3

14411 c

143

155

135

118

65.2

14411 d

129

133

169

138

274

14411 e

136

139

129

124

105

14411(X%RSD)

14111.5

14313.9

14011.7

12810.4

12469.5

* Letters indicate different replicate samples.

† Mean and percent relative standard deviation.

16