ETL 1110-1-175
30 Jun 97
portrayed on the maps that is available to the
concern contamination, there can be large ranges
investigator. The maps of kriged standard devia-
of values for data involving contaminant concen-
tions can be used to determine where there are
trations, and these large ranges have a tendency to
large areas of uncertainty in the kriging estimates.
increase the incidence of data that may seem to be
statistical outliers. Even more complicating is the
f. Finally, the variogram and kriging algo-
presence of high concentrations of organic mate-
rithms are most useful as interpolation rather than
rials that may create challenging analytical prob-
extrapolation tools. Once the application extends
lems in laboratory determinations that also may
to areas beyond the geographic extremes of the
lead to reported values that seem to be statistical
measured data, or perhaps those extremes plus the
outliers. In either case, the kriging practitioner is
range, there needs to be some qualification of the
likely to find that the apparent outliers have a
area of extrapolation. For instance, in universal
strong effect on the results of the kriging
kriging, the practitioner would need to have some
application.
assurance that the conditions of drift defined in the
c. When HTRW site investigations find data
study area continue into the area of extrapolation.
that seem to be outliers, the data need to be very
carefully evaluated before removal is seriously
6-3. Errors in Measured Data
contemplated. Automated outlier detection tools,
as suggested in section 4-8, may best be used to
a. Data associated with HTRW site investi-
identify points that may be outliers and warrant
gations have the same opportunities for errors that
further investigation. Often data that appear to be
most investigations do. The errors may involve,
outliers may be the most important and meaningful
among others, bias, inaccuracy, or lack of repre-
data of all measurements. For example, in the first
sentativeness. The classical nature of these errors
case described in the preceding paragraph, appar-
is described in EM 200-1-2, "Technical Project
ent outliers often are representative values. In the
Planning," (U.S. Army Corps of Engineers 1995),
second case, the reported value may be an errone-
which describes HTRW data-quality design.
ous determination that has been affected by the
extremely contaminated nature of the sample
matrix. The investigator needs to either possess or
plicate the function of errors in HTRW site
have access to qualitative or institutional knowl-
investigations. Because these investigations often
edge that will aid in outlier interpretation.
6-3
Previous Page
