Sediment source fingerprinting: Testing hypotheses about contributions from potential sediment sources

Techniques involving composite fingerprints and multivariate mixing models are being increasingly used in catchment studies to establish the relative contributions of potential sources to the suspended sediment output. Such information is important both for understanding the fine sediment dynamics of a catchment and for targeting remediation measures required to reduce sediment-related environmental problems. A multivariate mixing model algorithm, which compares the concentrations of a range of geochemical properties of the suspended sediment load with those of a number s of potential sources, is commonly used to provide estimates of the relative contributions (P₁, P₂, ⋯, P_s) of those sources to the suspended sediment load. However, such models do not provide measures of the uncertainty associated with the P-values. This paper describes how the usual mixing model can be modified, such that the optimization procedure used to estimate the sediment proportions P₁ P₂⋯ contributed by different sources also provides measures of their uncertainty. This approach allows hypotheses concerning the P values to be tested, such as: (i) whether the individual P-values differ significantly from zero, and (ii) whether the P-values change significantly between events. To calculate the uncertainty associated with a P-value, a statistical model which considers the correlation between the tracer variables is used. This approach has been tested using data from a small rural catchment in southern Brazil where a sediment source investigation is in progress. Sediment samples collected during 48 storm events were used to establish the source contributions.