Evaluating the Impact of High Source Variability and Extreme Contributing Sources on Sediment Fingerprinting Models

Abstract Sediment fingerprinting is a powerful tool used in drainage basin analysis to identify and quantify sediment sources, crucial for effective water management strategies. However, methodological debates persist regarding the influence of tracer type, tracer selection, and source dominance on fingerprinting model accuracy. This study introduces a novel linear variability propagation analysis (LVP method) to address and quantify potential bias in fingerprinting model outcomes, particularly when dealing with dominant or non-contributing sources and high source variability. We compare the results from two different models, Frequentist and Bayesian, to assess these effects using two datasets: the first one which was synthetically generated, and the other, obtained from a published laboratory study. Both datasets consisted of virtual mixtures. In such a way, uncertainties related to physical processes were eliminated, leaving only those which were introduced by mathematical or statistical methods. The comparison between theoretical and estimated apportionments from the synthetic dataset reveals systematic discrepancies in the results of both models when dominant or non-contributing sources coexist with high source variability. We analytically demonstrated that these deviations arise from the classical variability analysis used in both models. The proposed LVP method provides a means to quantify and mitigate these biases, offering a significant advancement for field fingerprinting studies where direct comparison with theoretical apportionments is not feasible. The laboratory dataset further validates these findings, revealing systematic deviations when non-contributing or dominant sources are present. Increasing the number of sources from 2 to 4 further enhanced the discrepancies that were observed.