Estimation of measurement uncertainty from field sampling: implications for the classification of contaminated land

Abstract

Methods have been devised for estimating measurement uncertainties due to field sampling. These methods have revealed very large discrepancies between the estimated levels of lead in one area of contaminated land. The uncertainty estimate for the estimated lead concentration at one sampling site, when sampled by one sampler (an individual collecting the samples) using one protocol, was very large (50%, for 95% confidence). The uncertainty on the estimated mean concentration was also large (55%) for nine different organisations using the same sampling protocol. These organisations were participating in the sampling trials by sending samplers to the sampling targets. This uncertainty was due entirely to the sampling, rather than the analytical measurement as all the analyses were carried out in random sequence within one batch. When sampling organisations used whatever sampling protocol that they considered appropriate, and analysed their own samples, the uncertainty on the mean concentration was 45.5%. The similarities in these estimates suggest that heterogeneity is a limiting factor in measurement uncertainty at this site. The existing criteria for the classification of contaminated land generally depend on a deterministic comparison between measured concentration of a contaminate and a threshold level. With such large values of measurement uncertainty in the estimated concentration of the contaminant due to field sampling, it is likely that land will be misclassified. This has legal, financial and possible health implications from both the unnecessary remediation of ‘uncontaminated’ land and regarding land as uncontaminated that is in fact contaminated. A new probabilistic classification scheme for contaminated land is proposed that allows for the overall measurement uncertainty as well as the estimated concentration of the contaminant. A term ‘possibly contaminated’ is given to locations where the measured concentration is below a threshold, but the uncertainty interval exceeds the threshold. The acceptable probability of misclassification can be selected to reflect the risks at that particular site.