Table 2.
.
239-240
:
7
2
Estimated
Pu Concentrations (uNCi/m™” per 5-cm Depth) at Grid
Nodes Near GZ Using Untransformed and Log-Transformed
39-240py
Observations.
Iteration
Untransformed
B72g208
20
252
1
2
3
*
354
629
210
8
-434
92
223
210
618
Log Transformed*
B721409
-
62
11?
1045
2171
291
~=10340
126
664
175
- 156
~
25
100
512
E721600
97
223
135
15
198
-
E721200
68
N936400
«(13
199
N936100
34
1804
272
113
527
660
508
382
564
682
265
239
280
100
109
76
78
84
84
N
41
108
77
186
E721600
77
129
251
963
259
5058
105
128
76
166
79
186
216
18
114
202
58
255
281
77
125
77
248
288
378
53
83
76
418
173
62
475
50
52
941
167
82
373
392
183
15535
- -460
372
638
477
385
644
782
366
219
53
697
1014
18130
1438
236
83
396
469
206
81
108
~ 304
1121
211
140
- lll
288
248
282
18856
67
59
55
65
133
~ 135
-1019
283
785
632
86
189
236
191
375
801
1016
69
194
181
717
295
290
54
9
125
137
859
937
35
44
43
317
277
37793
1459
311
91
78
123
63
96
398
544
237
371
128
- 152
88
254
39
647
217
151
- 382
-
a
1137
96
45
76
93
248
621
1787
135
23
-430
190
N935800
96
172
8721800
738
39
43
104
361
301
278
206
76
77
41
241
381
298
246
A
Tabled values are exp(Zce r), where zor is the estimated plutonium concentration (log scale)
at grid coordinate (XY).
;
The goodness-of-fit of the estimated surface to observed concentrations
at sample collection points can be measured by squaring the linear
correlation coefficient between the observed data and the estimated
surface at the sampling locations.*
This statistic (R2, the percent
of the total variation at sample locations explained by the fitted
surface) is plotted in Figure 12 for each stratum, iteration, and fitting
scale.
The fits are very good for strata 1 and 2 for the untransformed
as well as for the log and antilog fits.
R* tends to decrease for
strata near GZ particularly for the untransformed and antilog scales.
The effect of iterating is to increase R* in all strata for all three
*Pierre Delfiner pointed out (personal communication) that R2 will tend to
be overoptimistic as a measure of goodness of fit since y, and ¥. tend to
be correlated when Vy is obtained using a nearest neighbor approach.
338