apart by chemical explosives to test in part for 1 "safety" against fission reactions. A consequence of these tests was the contamination of the immediately surrounding soil and vegetation with plutonium, americium, and/or uranium. The Area 13 data set is characterized by a single, very large plutonium concentration obtained near ground zero (GZ, point of detonation) with surrounding concentrations falling off rapidly in all directions from GZ in an unsymmetrical pattern. Hence, the "true" concentration surface has a definite structure or pattern with relatively low levels of contami- nation predominating within one or two thousand feet from GZ (depending on direction). A goal of the current NAEG sampling program is to evaluate the potential hazard to man from this contamination if these areas were ever released for habitation. This evaluation has included the collection of several thousand soil, vegetation, small vertebrate, and cattle tissue samples for radiochemical analysis (White and Dunaway, 1975; Dunaway and White, 1974). An important objective of this effort is to estimate the spatial pattern (concentration surface) of plutonium about GZ as it presently exists in surface soil (top 5 cm). Gilbert et al. (1975, 1976b) and Gilbert and Eberhardt (1977) have experimented with estimating the spatial pattern of plutonium in surface soil and vegetation at safety-shot sites using "nearest neighbor" and polynomial fitting routines in both original and logarithmic scales. John Tukey suggested (see discussion following the paper by Eberhardt and Gilbert, 1976) that better fits to the data might be obtained if the fitting routine was applied iteratively on residuals. For example, the residuals from the first fit would themselves be fitted and added to the initial fit of t original data. More generally, if observation ¥y (i = 1,2,...,n) and Ri; is the residual between the i the smoothed (fitted) value y., obtained on the ym iteration (j = 1,2,...,m), then *J aA Ys = 95, 7 Fey (lst iteration) (1) Riy = Yio + Rio (2nd iteration) (2) and so that Yi Var F%i2 FR (3) For m iterations, we may write m Rim ~ %n 7 2y jel m where > j= Yay > (4) . Vag is the final smoothed estimate of y,, and Ron the final residual. 321