It was pointed out in Gilbert et al.
(1975, Fig. 28) that the contour
lines obtained to date for Area 13 and GMX are biased,
it is believed,
due in part,
to the nearest-neighbor estimatiom method and to the
use of soil samples collected at random locations within strata
(stratified random sampling) resulting in inadequate coverage of
portions of the field.
One approach that may reduce the bias in
these contours is to fit contours to the log-transformed data.
This
might be effective since the large concentration peaks would be
scaled down,
it is presumed,
less variable data.
for making more accurate fits to the
This bias might also be reduced by the use of
more strata or the division of strata into substrata within which
samples would still be taken at random.
This approach would tend to
reduce the size of areas where no samples were collected,
and hence
reduce the bias due to such gaps in information.
Another approach would be to increase the number of locations at
which samples are taken.
This could be done as part of a "double
sampling'' scheme for estimating plutonium inventory in soil so that
costs would not be expected to be greater than for presently used
sampling designs.
The general idea, as it might be applied to a
safety-shot site, would be to define strata using the FIDLER and to
collect a relatively large number of soil samples within strata.
All
241
samples would be analyzed for
Am using the Ge(Li) system and a
small proportion would also be analyzed for 239-240Pu using wet
chemistry techniques (both Pu and Am analyses being done on the same
aliquot from the soil sample).
known as double sampling.
This kind of sampling procedure is
Gilbert and Eberhardt (1975) evaluated its
effectiveness for Area 13 and GMX and concluded that except for the
two lowest plutonium concentration strata, the technique can yield
estimates of plutonium inventory with greater precision than possible
with plutonium analyses alone for a fixed cost.
This result has
implications relative to the estimation of unbiased concentration
contours since Gilbert and Eberhardt
(1975)
estimated that about
twice as many soil samples would need to be collected using double
sampling than were used to estimate Pu inventory using stratified
random sampling.
All of these soil samples would be Ge(Li) scanned
for 241am, but less than 20% would also need to be analyzed for 2392405p, (keeping total cost equal to that in estimating inventory using
104