spatial pattern as contrasted with estimating a total were discussed by Eberhardt and Gilbert (1976) at the lst ERDA Statistical Symposium held at Los Alamos, November 2-5, 1975. The topic was presented as a "problem" for discussion, and many valuable suggestions were offered by the statisticians in attendance, particularly Dr. John W. Tukey of Princeton University and Bell Laboratories. We hope to continue working on this design problem in conjunction with continuing experimentation with computer contouring algorithms. In preparation for the IAEA/ERDA and the ERDA symposia, we estimated 239Pu concentration contours in surface soil of the GMX site using algorithms QUAD Since most of these could not be published and TREND available on SURFACE II. in the Proceedings of the IAEA/ERDA symposium due to space limitations, we All of the contours shown here are include them here as Figures 9 through 14. in units of nCi (of 2739Pu) per gram (dry) of surface soil. Figure 8 shows questionable contours west of GZ obtained using NEAR as mentioned above. Figure 9 was obtained using NEAR on the logarithms of the data. As mentioned above, the use of logarithms eliminated some of the doubtful contours obtained on the untransformed data (Figure 8). More samples must be collected just west of GZ, however, to obtain accurate contours for that region. The algorithm QUAD, which searches for a minimum number of data points in each quadrate within a specified distance! 3 of the point to be estimated, was also applied to the logarithms of the data (Figure 10). Note that contours are not estimated west of ground zero. This occurred because of too few data points in this area. Hence, QUAD can indicate when insufficient data are a problem. Figures 11 and 12 show contours resulting from fitting 4th and 6th degree polynomials to the logarithms of these GMX soil data using the algorithm TREND. These contours show considerably less detail than those obtained above using NEAR or QUAD. The 6th degree fit to the data is somewhat better than that provided by the 4th degree equation. However, the black line along the left margin of the 6th degree fit (Figure 12) indicates severe edge effects. That is, the regression equation is giving grossly inaccurate estimates in this area where no data were collected. Figure 13 shows contours obtained by using TREND to fit a 6th degree polynomial to the untransformed data. Note that the area included within the 0.5 nCi/g contour is much larger for the original scale (Figure 13) than for the log-transformed scale (Figure 12). Serious edge effects are also present in Figure 13. These are plotted in Figure 14 to illustrate the wild Pu estimates obtainable using polynomial fits in regions of the study site where few data have been collected. Of the contour methods applied thus far to the soil Pu data collected via simple random sampling within strata, it appears that QUAD or NEAR applied to the logarithms of the data are the most promising (see Gilbert et al., 1976, Table IIT for a statistical evaluation). Improved estimates appear to result if an iterative fitting procedure is applied to the residuals of previous (log) fits. This iterative approach was suggested by Tukey (see discussion after paper by Eberhardt and Gilbert (1976)) and has been investigated in part by Gilbert (1976) using NEAR on SURFACE II. 13For Figure 8, we specified (a) maximum distance to nearest data point must be < 100 feet, (b) maximum search radius = 200 feet. 262