Bayesian sediment fingerprinting provides a robust tool for environmental forensic geoscience applications

Abstract

Abstract Sediment fingerprinting is an approach for the quantitative determination of sediment provenance (both spatial sources and types of sediment supply) over a range of temporal and spatial scales. Though widely adopted, studies often vary in their attention to the underlying assumptions and in their treatment of modelling uncertainty. A Bayesian approach to the multivariate problem of ‘unmixing’ sediment sources is reported, showing the significance of source group variability and source group sampling density to the accuracy of model output. The model produces results as median source group contributory coefficients (and associated 95% quantiles). The model was applied to environmental data obtained from selected soil erosion studies reported within the peer-reviewed literature. Good correspondence (r 2 =0.89) between reported mean source group contributory coefficients and median values were found when recalculated using the Bayesian analysis. However, confidence levels are highly variable, ranging from 2% to 97%. The robustness of any unmixing solution depends on factors such as the number of samples, the number of source groups and the variance of source group properties. It is concluded that ‘forensic-style’ investigations must recognize these uncertainties and be appropriately resourced to achieve tolerable accuracy and precision. The discussion considers additional confounding factors such as non-conservative tracer behaviour and enrichment/depletion during the sediment delivery process.