data fits. Using R2 as criterion, the antilog fits appear preferable to fits on untransformed data. In the next section, we examine residuals from the estimated surfaces in more detail. Examination of Residuals Residuals R 42? and R,.,, as defined by Equation 4 (also see Table 1 and Figure sy até plotted in Figures 13-15 for the untransformed and antilog fits for strata 1, 3, and 6. Figures 16 and 17 give these results for the log scale. Several summary statistics of the residuals for all six strata are given in Table 3. These data indicate that the iterative procedure is effective in reducing the mean and median size of the absolute values of the residuals in all three scales. The smaller residuals in strata 1 tend to approach zero with only three iterations, whereas larger residuals for the more heterogeneous data in stratum 6 near GZ tend to "bounce around" and approach zero more slowly. Figure 18 shows the percent reduction in the median of the absolute values of residuals that occur due to iterating two and three times (computed from Table 3). Percent reductions are highest in strata 1 and 2 and become smaller for the strata nearer GZ. The least reduction occurs in stratum 5 for the fit to untransformed data. The percent reduction between Iterations 2 and 3 (Figure 18) was consistently greatest for the fit on log units, followed by the antilog and untransformed data fits. Figure 18 indicates for this data set that the third iteration yielded a substantial improvement in fit over the second iteration. The squares (Ri) of the linear correlation coefficients between the observed 239°250py soil concentrations and the residuals from fitted surfaces are given in Figure 19. An R, = 1 would indicate a linear association between residuals and observed data such that large observed values would tend to be underestimated by the estimated concentration surface, and small observations would tend to be overestimated by the concentration surface. If this occurs for strata 1 and 2, it could indicate a tendency for the estimated low-level contours to be too far out from GZ if the negative residuals occur predominately around the outer edges of the strata. This does not appear to happen, however, since Figure 19 indicates that for the untransformed data fits, R,* on the first iteration is only 0.02 or 0.03 for strata 1, 2, and 3. This increases to about 0.20 for strata 4 and 5, and further increases to 0.82 percent for stratum 6. (This large Ry? for stratum 6 is caused by the datum 16,400 »uCi/m* in stratum 6 (see Appendix A) as indicated in Figure 19.) This is examined in more detail in Figures 20 and 21, where the residuals after three iterations on untransformed data are displayed on the estimated plutonium concentration contours for Iteration 3. Strata boundaries (from Figure 1) are also shown. There appears to be no obvious tendency for negative residuals to predominate around the edges of the map (Figure 21). Clusters of positive or negative residuals 340