For example, if a criterion required cleanup of any region with TRU activity greater than 80 pCi/g, averaged over 0.5 hectare (ha), the criterion could be applied to the 0.5 s (s is the standard deviation) upper bound on the estimated average. That is, if the estimate plus 0.5 .s exceeded 80 pCi/g, soil might be removed. if soil was not removed because the estimate plus 0.5 s was less than 80 pCi/g, probability is .69 that the true average was in fact less than 80 pCi/g, under the assumption of normality. On the other hand, this approach results in some soil being removed that really has lower TRU activity than 80 pCi/g. The other estimates that were required for surface and subsurface characterization and cleanup were almost all made using standard techniques. Some of these, for example the method used to estimate the ratio of TRU to 241Am, were changed based on experience with actual data, but they were changed to other standard methods. Classical approaches were also used for analyzing data from other programs such as the plowing experiment on Janet (see Section 6.7). In all cases, however, both with kriging and more classical methods, consideration was taken and adjustments made for unique aspects of the Enewetak situation. Some of the considerations and alterations are discussed in Section 5.2.6. The greatest adjustments were required in experimental and sampling design. For example, the subsurface sampling methodology underwent considerable alteration before a satisfactory approach was found. In some cases, such as the plowing experiment and in sampling the Aomon Crypt, special sampling methods were designed to fit the situation. Even the collection of the soil samples for determining the ratio of TRU to 241 4m wasspecifically designed to allow valid comparison with the IMP 241 Am data from the same locations. The general approach used for the surface cleanup was to obtain preliminary estimates using kriging and data from a 50 meter (m) grid, then collect additional data on a small grid in and around areas that did not meet the applicable criterion. Arithmetic means of adjacent IMP measurement values were then used to provide estimates of activity and boundaries for cleanup areas. After a soil lift, the area would be remeasured at the closer spacing so arithmetic means could again be used for determining if the lifted area met the criterion, and the process was repeated if necessary. A similar approach was used for subsurface cleanup. Once the excision boundaries were determined from soil samples and the soil had been removed, additional soil samples and IMP measurements were taken to check if another iteration would be required. By using an iterative approach, less data were needed and the initial data collection for both surface and subsurface characterization could be speeded up. Yet, the cleanup was still done conservatively, because contamination above the cleanup criterion would be detected and removed on the nextlift. This iterative process along with the kriging technique used for the initial characterization was quite effective during the cleanup. 5.2.1 Surface Characterization Kriging. The kriging technique, originally developed at the School of Mines in Paris, France, Matheron, 1967), was inspired by certain estimation problems in mining. It was named by Matheron in honor of D. G. Krige, a South African mining engineer who pioneered the use of weighted averages in ore reserve estimation. Many of the terms defined below, such as "nugget effect" and "zones of influence," reflect the mining heritage of kriging. However, the method has been successfully applied to petroleum exploration, meteorological variables, seafloor mapping, water table mapping, and other geoscience applications. The kriging estimator is a weighted moving average of the data with the weights determined using a function called the variogram. The variogram mathematically relates the variability of the difference between the values at two points to the distance between the points. The variogram is estimated from a set of data values, but the task is simplified because most variograms fit one of a few commonpatterns. 138