Achieving measurement comparability in mercury speciation analysis in seawater: Key requirements and best practices

The comparability of measurement results is an important issue in contemporary mercury (Hg) speciation in seawater. Sampling campaigns must be properly designed to determine significant differences on spatial and temporal scales, considering two major parameters: the variability of expected data at a given sampling point/transect and variability in the results due to the intrinsic properties of specific analytical methods, particularly the measurement uncertainty. This study assessed the required sample size, considering several aspects of data variability when determining total Hg, dissolved gaseous Hg, and methylated Hg species in seawater. The required sample sizes were calculated using (1) the measurement uncertainty of a single-laboratory measurement of analytical methods used; (2) performance of the laboratories in interlaboratory comparison exercises; and (3) natural variability in Hg species/fractions in a selected case study in the Central Adriatic Sea. It was shown that the measurement uncertainty of a particular method and interlaboratory variability among laboratories have significant influence on data interpretation in case natural variability of Hg fractions is relatively small, such as for example the open seawater depth profiles. In contrary, in areas with large natural variability of Hg contractions, such as coastal and contaminated sites, their influence on data interpretation is negligible. The present paper introduces the importance of proper estimation of measurement uncertainty in international programs, such as GEOTRACES, where data comparability is of fundamental importance to assess temporal and spatial trends of Hg measurements in the marine environment. • Required sample size was based on several aspects of data variability. • Sample size depends on sample variability, uncertainty and interlaboratory comparison. • Uncertainty is important for sample size only when natural variability is low. • Log-normal distribution requires geometric means and geometric standard deviations. • Measurement uncertainty can be reduced by applying an appropriate calibration curve.