ment systems that have been evaluated for treat-
a number of variables such as waste disposal his-
ing explosives-contaminated soils include anaero-
tory, the physical and chemical properties of the
bic bioslurry, aerobic bioslurry, white rot fungus,
specific explosive, and the soil type. The conclu-
and landfarming (Craig et al. 1995, Sundquist et
sion was that, to improve the quality of site char-
al. 1995).
acterization data, the major effort should be placed
on the use of higher sampling densities and com-
posite sampling strategies to reduce sampling
error.
UNIQUE SAMPLING DESIGN CONSIDER-
In a subsequent study at an HMX-contaminated
ATIONS FOR EXPLOSIVES
antitank firing range, similar short-range hetero-
geneity was observed (Jenkins et al. 1997). At both
Heterogeneity problems and solutions
The heterogeneous distribution of explosives in
the short- and mid-scale, sampling error over-
soil is often alluded to but seldom quantified. The
whelmed analytical error. It was observed that the
problem is probably considerably greater for ex-
particulate nature of these contaminants in near-
plosives residues in soil than for most other or-
surface soils is a major contributor to substantial
ganic waste. According to available Superfund site
spatial heterogeneity in distribution.
data, the median coefficient of variation (CV) (stan-
There are several practical approaches to reduc-
dard deviation divided by the mean) for volatiles,
ing overall error during characterization of soils
extractables, pesticides/polychlorinated biphe-
contaminated with explosives, including increas-
nyls (PCBs), and tentatively identified compounds
ing the number of samples or sampling density,
in soils ranges from 0.21 to 54% for individual con-
collecting composite samples, using a stratified
taminants (EPA 1992b). Data from 11 munitions
sampling design, and reducing within-sample
sites show the median CV for TNT was 284%, and
heterogeneity. Because explosives have very low
the TNT CV ranged from 127% to 335% for indi-
volatility, loss of analytes during field preparation
vidual sites. Comparable data for RDX show a
of composite samples is not a major concern.
median CV of 137% with a range of 129% to 203%;
the median CVs for 2,4-DNT, AP/PA, and PETN
Increasing the number of samples
were 414%, 184%, and 178%, respectively. If the
One simple way to improve spatial resolution
natural variability of the chemicals of potential
during characterization is by collecting more
concern is large (e.g., CV >30%), the major plan-
samples using a finer sampling grid, such as a 5-
ning effort should be to collect more environmen-
m grid spacing instead of a 10-m spacing. Though
tal samples (EPA 1992b).
desirable, this approach has been rejected in the
Jenkins et al. (1996a, b) recently conducted stud-
past because of the higher sampling and analyti-
ies to quantify the short-range sampling variabil-
cal laboratory costs. When inexpensive on-site
ity and analytical error of soils contaminated with
analytical methods are used, this approach be-
explosives. Nine locations (three at each of three
comes feasible. The slightly lower accuracy asso-
different facilities) were sampled. At each location,
ciated with on-site methods is more than compen-
seven core samples were collected from a circle
sated for by the greater number of samples that
with a radius of 61 cm: one from the center and
can be analyzed and the resultant reduction in
six equally spaced around the circumference. The
total error.
individual samples and a composite sample of the
seven samples were analyzed in duplicate, on-site,
Collection of composite samples
using the EnSys RISc colorimetric soil test kit for
The collection of composite samples is another
TNT (on-site method) and later by Method 8330
very effective means of reducing sampling error.
at an off-site laboratory. Results showed extreme
Samples are always taken to make inferences to a
variation in concentration at five of the nine loca-
larger volume of material, and a set of composite
tions, with the remaining four locations showing
samples from a heterogeneous population pro-
more modest variability. For sites with modest
vides a more precise estimate of the mean than a
variability, only a small fraction of the total error
comparable number of discrete samples. This oc-
was because of analytical error, i.e., field sampling
curs because compositing is a "physical process
error dominated total error. For the locations show-
of averaging" (adequate mixing and subsampling
ing extreme short-range heterogeneity, sampling
of the composite sample are essential to most
error overwhelmed analytical error. Contaminant
compositing strategies). Averages of samples have
distributions were very site-specific, dependent on
greater precision than the individual samples.
5
Previous Page
