RESULTS Concentration Surfaces (Estimated Grid Values) Estimated *39°249py concentration contours and three-dimensional plots are displayed in Figures 3 through 11 for Iterations 1, 2, and 3 for both untransformed and log-transformed data. The observed plutonium data are displayed at their collection points in Figures 3, 4, 6, and 7. Note that Figures 4 and 7 are enlargements of the GZ area in Figures 3 and 6. As expected, contours drawn from estimated grid matrices in both scales indicate greatly elevated concentrations in the GZ area. Note, however, that contours for untransformed data on Iteration 1 (Figures 3 and 4) show regions of low concentrations (< 1 uCi/m*) to the north, northwest, and east of GZ in regions where no data were collected. Indeed, much of the < 1 wCi/m? region in Figures 3 and 4 (untransformed fits) consists of grid estimates that are negative and hence spurious. The negative concentration contours for Iteration 3 are displayed in Figure 5. These negative estimates can also be seen in the three-dimensional representation of the estimated grid node concentrations for Iteration 3 (Figure 10). The largest negative grid estimates for Iterations 1, 2, and 3 are ~1435, -1666, and -1786, respectively. They occur to the south, southwest of GZ (Figure 5). The examination of residuals in the next section indicates that the average absolute (mean, median) size of residuals for these untransformed fits decreases with each iteration. Hence, the estimated concentration surface using untransformed data is becoming distorted even though the residuals at sample locations are decreasing. the same extent for fits in the antilog scale. This does not happen to This is seen, for exam- ple, by examining Table 2, which gives the estimated grid node concentrations in the immediate GZ area for the untransformed and antilog scales. These show the changes that occur in the grid node estimates due to iterating on the residuals. In the untransformed scale grid, estimates surrounding the peak at GZ (location N936092, E721352) tend to decrease with each iteration. This effect is not as evident for the antilog scale. Note the presence of negative grid estimates in the untransformed scale and that they tend to grow larger with each iteration (also see Figures 8 and 9). The estimated antilog surface is not free from bias, however, as can be seen in Figures 6-9, where the 100 to 1,000 uci/m2 region is estimated to extend several hundred feet south of GZ for both the untransformed and antilog scales. This does not agree with the 241am (FIDLER) contours for this region (Figure 1) and probably results from too few samples being collected in that area. 328