quantile method overestimates m and s (average of + 48 percent, range:
.
to + 160 percent, O-15 cm soil).
For
137
.
Cs with few measurements below
-
the MDA, the differences between methods are smaller.
For
137
Cs 0-15
cm soil measurements the arithmetic mean, x, underestimates -5 percent
(range: 32 percent to +6 percent), and for the quantile method, x
averages 21 percent lower (range -50 percent to 0 percent).
As we can
see (Table 46), radionuclide concentration data above the MDA the
arithmetic averages make“a good approximation to the Krige mean for
coefficient of variation c, averaging 0.9 (range: 0.6 to 1.5).
Recently,
White4] found the arithmetic mean to have a 75 percent efficiency for
‘coefficients of variation, c, less than 2.
Tnis efficiency is also show
by Aitchison and Brown .34
The value of the
shifting parameter can be seen fror Tables 45 anc
46 to be roughly equal to the MDA velues of Z4las (0.2 to 1.5 pci/gz)
and 137¢e (0,1 pCi/gm).
MDA,
set
to MDA.
Both of these data sets have values less than
Similar analysis
negative values of T.
The 1376.
on unaltered data exhibits
lower
or
values (Table 46) are seen on
samples Enjebi (Janet) NE and Kidrinen (Lucy) to be unreasonably large,
and without physical basis.
The improvement in lognormal fit wes
marginal and the T could have been set to zero; however,
the Krige method
is fairly insensitive to T as illustrated by example, 29 anc as our
tests have also shown.
The computer computational codes were tested with artificial lognormal samples.
A 105 term approximation4? generated by a rectangular
distributed psuedo-random generator produced these artifical sarcpies.
Testing samples numbers ranged from 4 to 255.
41 -
a
5011120
Ps
OF D>
eat
4]
/
e k yb eee
ec KY
Pia vys
Z