The effects of sampling design on estimating the magnitude and distribution of contaminated sediments in a large reservoir

Abstract

We simulated the sampling of 137Cs inventory (pCi/cm2) in Lower Watts Bar Reservoir to evaluate the effects of different sampling designs on the estimation of the magnitude of 137Cs contamination in the reservoir and to determine if the spatial pattern of contamination affected the performance of alternative sampling designs. Sampling designs evaluated were random, stratified random, systematic, and transect. These sampling designs were evaluated by creating multiple realizations of each design and sample size, and comparing the bias, variance, mean squared error (MSE), and per cent error in estimates of mean 137Cs inventory, 95th percentile of 137Cs inventory, and percentage of bottom sediments having 137Cs level > 400 pCi/cm2. Two alternative spatial maps of 137Cs inventory were used because the spatial distribution of 137Cs is uncertain. Increasing sample size beyond 250 for random and systematic designs resulted in little further reduction in MSE, variance, or bias of estimates of the three parameters. Variance generally contributed to MSE more than bias for random and systematic designs, whereas bias was important with transect sampling. The performance of random and systematic sampling varied minimally between the two maps. Transect sampling yielded inconsistent results due to bias not decreasing with increasing sample size. Stratification did not result in substantial reductions in MSE compared to simple random or systematic sampling. The per cent error of estimates of the percentage > 400 pCi/cm2 were larger than estimates of the mean or 95th percentile of 137Cs inventory. A staged sampling approach is recommended, with the first stage being determination of the spatial heterogeneity of the parameter of interest. Copyright © 2001 John Wiley & Sons, Ltd.