Uncertainty-based assessment of tracer selection, tracer non-conservativeness and multiple solutions in sediment fingerprinting using synthetic and field data

Purpose: Modification of sediment properties used in fingerprinting applications occurs along transport pathways as a result of particle size and organic matter enrichment/depletion, and geochemical transformations. Statistical approaches have been widely used to correct for enrichment and depletion, but detection of, and the un-mixing errors and uncertainties that arise from non-conservative behaviour remains under-recognised. Additionally, the over-determined nature of sediment fingerprint un-mixing models results in a range of potential solutions which are yet to be formally assessed.

Materials and methods: Synthetic source data comprising 50 tracers and four sources were ‘mixed’ to generate known target tracer compositions. Firstly, both conservative and deliberately corrupted tracer behaviours were processed by repeated un-mixing from the minimum permissible number of tracers (n=3) to the maximum (n=50) using the FR2000 model. Secondly, using a smaller synthetic dataset, one tracer was deliberately corrupted in a controlled way to determine the impact on results and the ability of the permutation version of the Monte-Carlo FR2000 un-mixing model to detect non-conservative behaviour. Finally, this approach, and the particular case of near equivalent (or equifinal) solutions, was applied to data from on-going sediment provenance studies in Ireland.

Results and discussion: Uncertainty in source predictions was better reduced by increasing, rather than decreasing the number of tracers, therefore questioning the justification for tracer reduction strategies. Non-conservative behaviour negatively affected the accuracy of mean source predictions but had no significant effect on uncertainty. The degree of tracer corruption (−90 to +155 %) from the ‘perfect’ target value resulted in a wide range of source predictions. The applied permutation un-mixing model was successful at detecting and rejecting the corrupted tracer below −50 % and above +20 % corruption. The true corruption (the uncertainty bounds reported by prediction at the upper and lower levels) was, therefore, significantly improved. The methodology to examine multiple solutions identified reasonably consistent source predictions when applied to field data. The suitability of this technique on data with limited tracers and no particle-size or organic matter correction is, however, questionable and warrants further investigation.

Conclusions: Tracer selection is a key stage in reliable sediment fingerprinting applications. Non-conservative behaviour results in inaccurate source group prediction. Existing studies may therefore require critical evaluation, particularly where small sample numbers are collected in systems where enrichment/depletion of source group signatures (particle size, organic effects and geochemical alteration) results in non-conservative tracer behaviour (corruption) during entrainment and transport or storage within sediment sinks.