The standard error (SE) for the total (strata 3-6) kriging inventory
estimate in Table 3 is a "ball-park" estimate obtained as the square
root of the approximate expression:
Var(I-I)
~
«2
v2
= s¢M
2
2
t/a,
ta 252
(oF M w2
a
2
ts/yy)
.
(19)
This was derived using stratified random sampling theory with a constant
variance of s* = 0.09 per stratum (sill of the sperical variogram of
residuals; Figure 7).
Other parameters in Equation 19 are:
a = 1.287,
the slope of the regression between log Pu and log FIDLER (see Equation 2); a2 = 0.00373 is the extension variance of the center of a.grid
cell to the cell for a linear variogram; and M = £n 10 = 2.3026.
I,,
ns and v, are the estimated inventory, number of Pu congentrations, and
number of 100- x 100-foot cells, respectively, for the i
strata.
The
expression in parentheses in the second term of Equation 19 is the
contribution to the total variance due to the FIDLER.,
The first term is
the contribution due to the "block correction term" T (v).
Using the
values of s*, M, a, and of given above, Equation 19 becomes:
T_T) 0.4769
Var(I-I)
>
f2
72
7 Ty/n,
+ (1.287) 2 £(0.0198) >»
7 Is/u,
.
Hence, the contribution to total yariance due to the FIDLER is negligible
compared to that contributed by T (v).
From Table 3, Part A, we see that for strata 3, 4, and 5, the kriging
and strata mean inventory estimates differ by only 0.5, 0.3, and 0.3
curies,
respectively.
However,
for stratum 6,
the innermost strata
surrounding GZ (see Figure 1), the kriging estimate of inventory is
slightly less than half that obtained using strata means
curies).
(8.5 versus 19
There is little doubt that the kriging estimate for this
stratum_is too low since one of our approximations was to ignore the
term 107/10"? , which is always greater than 1. We have noted above
that our results may be biased downward, especially near GZ where the
drift m(x,y) may change by several orders of magnitude within short
distances.
More FIDLER readings near GZ are required before the change
in drift near GZ can be estimated with much assurance.
We should not necessarily assume, however,
that the 20 curies estimated
by Gilbert (1977) or the 19 curies given in Table 3 are closer to the
true inventory than the kriging estimate. As discussed by Gilbert and
Essington (1977), the estimate of 20 curies has a large standard deviation (8.2 curies) that results from the highly skewed distribution of Pu
data obtained for stratum 6. These authors illustrate, using a hypothetical example, the extreme instability of inventory estimates using the
estimated stratum mean when sampling from a highly skewed distribution.
If more precise estimates of inventory are needed for stratum 6, a reasonable approach might be to take FIDLER readings on a much finer grid and
400