Comparison of statistical methods for estimating plutonium inventories in soil

Several statistical methods have been used to estimate total plutonium inventory in contaminated surface soil on the Nevada Test Site, including systematic sampling, importance sampling, stratified random sampling, and two forms of kriging. In computer simulation experiments to compare these methods, systematic sampling using regular and preferential grids gave accurate and precise inventory estimates. However, the difficulty in estimating sampling variance from a single sample is a major drawback in field applications. Although importance sampling and stratified random sampling allowed estimation of sampling error, these methods gave less precise results than grid samples and were harder to use. Inventory estimates from kriging were biased, and kriging variances were not representative of true sampling variability. Kriging may not be appropriate where contaminant levels change rapidly over short distances