Estimation of sampling bias between different sampling protocols on contaminated land

Abstract

A method has been devised for the estimation of sampling bias between different sampling protocols on contaminated land. A pilot study was conducted at one site of contaminated land near Wirksworth in Derbyshire. Four sampling schemes (simple regular grid taking single auger samples, regular grid taking five-fold composite samples, herringbone and stratified random) were applied to sample topsoil. The samples were decomposed with nitric and perchloric acids and analysed using ICP-AES for 24 elements. The interpretation of the results was focused on the study of two elements with contrasting properties: lead, which shows high geochemical variability across the field and copper, with very low geochemical variation. Although the mean lead concentration vary markedly between the different sampling protocols (e.g., 5680–6868 µg g^–1) the differences are not statistically significant owing to the high geochemical variability of lead, quantified as geochemical variance. For copper with much lower geochemical variance a significant sampling bias of 2.45 µg g^–1 was detected between two of the sampling designs (i.e., the simple regular grid and the stratified random design). The methodology could be applied for more demanding studies such as interlaboratory trials for estimating sampling bias between different sampling organizations at this site.