Measurement Uncertainty Arising From Sampling: Implications for the Objectives of Geoanalysis†

Abstract

Four new methods for the estimation of the measurement uncertainty due to field sampling have been evaluated, by an application to soil sampling on contaminated land. The methods range in cost and complexity from one designed for a single sampler using one sampling protocol, to another designed for groups of different sampling organisations applying different sampling protocols. The size of the uncertainty arising from sampling is often large, ranging up to 55%, and this is much larger than the uncertainty generated by the analytical determinations in the laboratory (e.g., 5%). Although further applications of these methods to different geochemical sampling media are needed to confirm the generality of this finding, two implications seem important to the objectives of geoanalysis. Firstly, rather than simply measuring concentration values, geoanalysts need also to estimate the uncertainties on these measurements. Users of the concentration measurements can then propagate these uncertainty values through geochemical models, to give an estimate of the uncertainty in the interpretation. An example of this is given for the classification of contaminated land. Secondly, geoanalysts need to begin the estimation of uncertainty at the stage of field sampling, which is the first stage of the measurement. This will require the geoanalyst to be involved in the design of sampling as well as analytical protocols. The objective of geoanalysis should therefore be: ‘to determine the composition of the earth and its parts, with realistic estimates of measurement uncertainty arising from both sampling and analysis’. Given this information it will then be possible for geoanalysts to decide objectively whether the quality of the measurements are fit-for-purpose. This will enable measurements with relatively high uncertainty (e.g., field sensors, laser ablation) to be shown to be fit for some purposes, and considered as equally valid when compared with traditional laboratory-based techniques with lower uncertainty.