Improved reliability in the interpretation of geochemical measurements by the quantification of uncertainty from sampling

All geochemical measurements require the taking of field samples, but the uncertainty that this process causes is often ignored when assessing the reliability of the interpretation, of the geochemistry or the health implications. Recently devised methods for the estimation, optimisation and reduction of this uncertainty have been evaluated by their application to the investigation of contaminated land. Uncertainty of measurement caused by primary sampling has been estimated for a range of six different contaminated land site investigations, using an increasingly recognized procedure. These site investigations were selected to reflect a wide range of different sizes, contaminants （organic and metals）, previous land uses （e.g. tin mining, railway sidings and gas works）, intended future use （housing to nature reserves） and routinely applied sampling methods. The results showed that the uncertainty on measurements was substantial, ranging from 25% to 186% of the concentration values at the different sites. Sampling was identified as the dominant source of the uncertainty （〉70% of measurement uncertainty） in most cases. The fitness-for-purpose of the measurements was judged using the optimized contaminated land investigation （OCLI） method. This identifies the optimal level of uncertainty that reduces to overall financial loss caused by the measurement procedures and the misclassification of the contamination, caused by the uncertainty. Generally the uncertainty of the actual measurements made in these different site investigations was found to be sub-optimal, and too large by a factor of approximately two. The uncertainty is usually limited by the sampling, but this can be reduced by increasing the sample mass by a factor of 4 （predicted by sampling theory）. It is concluded that knowing the value of the uncertainty enables the interpretation to be made more reliable, and that sampling is the main factor limiting most investigations. This new approach quantifies this problem for the first time, and allows sampling procedures to be critically evaluated, and modified, to improve the reliability of the geochemical assessment.     

Fitness-for-Purpose of Reference Material Reference Values in Relation to Traceability of Measurement, as Illustrated by USGS BCR-1, NIST SRM 610 and IAEA NBS28

As in all fields of sample analysis, reference materials play a large role in supporting measurements in the geosciences. While a rather large number of materials are in distribution (> 380), not all are equally effective or fit-for-purpose in supporting laboratory data quality and thereby assuring the desired comparability of measurements between laboratories. Equally important, reference values that are not fit-for-purpose cannot be used effectively to establish traceability links between laboratory measurements and national and international standards. The needed fitness-for-purpose is not achieved for reference values either when more than one reference value has been proposed and a consensus does not exist among users as to which should be used by all, or when reference value uncertainties are too large in comparison to those of routine laboratory measurements. The focus of this review will be, first to outline the current reality, and second to suggest ways in which certifications of RMs can be improved to provide reference values that are universally accepted and more fit-for-purpose in general laboratory use. The discussion will be illustrated largely by current uses of USGS BCR-1, NIST SRM 610 and IAEA NBS28, as these three materials are those for which the largest body of newly published data exists, according to recent bibliographies of the geoanalytical literature published annually in Geostandards Newsletter: The Journal of Geostandards and Geoanalysis.     

Determination of Trace Elements (Ag, B, Ge, Mo, Sn, W, Tl) in Reference Samples of Soils and Bottom Sediments by Atomic Emission Spectrometry: A Traceability and Fitness-for-Purpose Study

Reference samples of soils from the Institute of Applied Physics, Irkutsk (RIAP), the Institute of Geochemistry, Irkutsk (IGI) and the United States Geological Survey, Reston (USGS) were analysed with the aim of determining Ag, B, Ge, Mo, Sn, Tl and W abundances by an atomic emission method with air-stabilised D.C. arc excitation. Two series of reference samples of soils and bottom sediments, GSS-1-8 and GSD-1-12 (IGGE), were used to ensure the traceability link for the analytical results. Traceability was also demonstrated through the comparison of measured results by AES and ICP-MS methods. It is shown that the reference samples GSS-1-8 and GSD-1-12 satisfied the "fitness-for-purpose" criterion (uncertainty U of the certified value should be one-third to one-tenth the magnitude of routine laboratory data uncertainty S (S/U > 3-10)) and can be applied for calibrating AES techniques.     

ISO Best Practices in Reference Material Certification and Use in Geoanalysis

The International Organisation for Standardisation (ISO) has published many guides, or technical standards, of great value to analytical geochemists. Two of particular importance are Guide 33 (Uses of Certified Reference Materials) and Guide 35 (Certification of Reference Materials). Both were first developed in the 1980s and undergo regular review and updating by the Reference Materials Committee (REMCO) that operates within ISO. Recent revisions have focused on adding statistical rigour to both guides. Although this offers significant advantages for use by professional metrologists, there are consequent issues of comprehension by the analytical chemists who in fact have the greatest need of them. A major focus of Guide 35 is the development of reference material uncertainties that are in full compliance with the Guide on Measurement Uncertainty (GUM), jointly issued by ISO, IUPAC and others. Guide 35 details handling of uncertainty due to (1) degradation on the shelf and in transport, (2) sample heterogeneity and (3) inter‐method and inter‐laboratory bias, as well as within‐laboratory repeatability. The International Association of Geoanalysts has developed a protocol for reference material certification that applies Guide 35 to the specific needs of the geoanalytical community. The approach being taken by the IAG in developing GUM‐compliant uncertainties for its certified values is presented. Recommendations made in Guide 33 for how a laboratory should compare its own results with certified values in assessing laboratory accuracy are outlined. Additionally, the subject of misusing reference materials is discussed. The apparent misuse occurs because so few CRMs exist that meet critical measurement needs of geoanalytical laboratories and that also meet the rigorous metrological demands of the latest editions of the ISO Guides. All of the focus of the IAG certification programme has been to undertake certifications that would fill gaps in CRM availability and thus serve to limit this misuse.     

New Approach to Geochemical Measurement: Estimation of Measurement Uncertainty from Sampling,rather than an Assumption of Representative Sampling

It is argued that the current division between field sampling and chemical analysis is counterproductive in terms of ensuring that geochemical measurement results are fit for their intended purpose. An integrated approach to the whole measurement process has many advantages including no dependence on the two assumptions that either the samples are necessarily representative if taken with a correct protocol, or that the measurement results can be assumed to be true values of chemical concentration. The measurement results then require values of measurement uncertainty, including that from sampling as well as from chemical analysis. This enables the user of the measurement results, rather than the producer, to judge their fitness for a specific purpose. Case studies are used to illustrate the practicality and benefits of this new approach, including the use of measurement results with optimal, but relatively high, levels of uncertainty to make reliable decisions. This contrasts with the traditional assumption that pursuit of the lowest possible measurement uncertainty is the best approach.     

How Fit Are Your Data?

Most laboratories aim to produce data of the highest quality. Trying to lower uncertainties to infinitesimal figures and push detection limits even lower are valid goals. However, is it possible to overachieve? Are old data still of good enough quality to be usable? In a geochemical context, the main goal of producing analytical results is to answer geological or environmental questions. Not all scientific problems require the same data quality. What is really required are data of adequate quality – i.e., ‘fit-for-purpose’– to ensure that the geological problem at hand can be solved. Furthermore, it is doubtful that uncertainties and reproducibilities associated with field sampling are better than those from laboratories. It is thus proposed that, as geoanalysts, we encourage data users (students, colleagues or referees) to ensure that their analytical results are of sufficient quality to solve the problem. However, authors have to demonstrate, through the use of reproducibility testing, reference and quality control materials, that the quality of their results is sufficient to solve the problem. Uncertainties and detection limits in publications should not only be evaluated with respect to a set value, such as 10%, but also with regard to the geological problem to be solved.     

Estimating and Optimising Analytical and Sampling Uncertainty in Environmental Investigations: Application and Evaluation

Measurements taken to characterise environmental contamination contain uncertainty, which is generated by both field sampling and chemical analyses. Recently devised techniques have been applied for the first time to estimate this uncertainty in the commercial monitoring and assessment of contaminated land. The uncertainty reduces the reliability of the classification of the land that is made following a site investigation. The possible misclassification of areas of land, as a result of measurement uncertainty, can lead to substantial financial penalties, resulting from litigation or unnecessary remediation. Previous studies have developed methods for the estimation and financial optimisation of measurement uncertainty. These methods have now been applied to a series of six contrasting site investigations, which were conducted by various commercial organisations. The previous uses of these sites included a gas works, a tin mine and railway sidings. The measurement uncertainty was successfully estimated for each of the six investigations, showing its applicability to a wide range of different sampling methods, such as trial pits, window sampling and augering. The measurement uncertainty ranged widely between sites from 25% to 158%, indicating that investigations can differ widely in their reliability. The field sampling tended to generate the largest component of the measurement uncertainty when compared to the contribution from the chemical analysis. The Optimised Contaminated Land Investigation (OCLI) method was applied to each site, with the initial aim of estimating the financial losses that could be incurred as a result of misclassifying the land, due to the uncertainty. It showed that the expectation of loss value per sampling location ranged from only £58 at one site to over £ 11 000 at another. The optimal level of uncertainty that produced the minimal financial loss was then calculated for each site. It provided a reduction in the expectation of loss for the whole site of over £ 10 000 at two of the sites and over £90 000 at two others. These findings demonstrate that implementing concepts of uncertainty can have practical benefits in environmental monitoring, and can enable improvements to be made in the quality of sampling and hence of measurements in general.