Using multiple composite fingerprints to quantify fine sediment source contributions: A new direction

Abstract

Sediment source fingerprinting provides an essential means for estimating sediment source contributions, which are needed not only for soil conservation planning but also for erosion model evaluation/refinement. A single optimum composite fingerprint has been widely used in the literature to estimate sediment provenance. The objectives of this work are to (1) verify whether an optimum composite fingerprint exists, (2) present a new direction of using multiple composite fingerprints to improve the accuracy and reliability of source contribution estimation, and (3) evaluate the optimization model formulation and the validity of the tracer discriminatory weighting. This study shows that tracer selection greatly impacts the estimated source contributions. The optimum composite fingerprint may not exist, or at least cannot be identified simply based on ability of the tracer to discriminate sources because of the lack of correlation between ability of the tracer to discriminate and its rigor in estimating source contributions. The weak link is likely caused by (1) tracer conflicts, (2) differential tracer measurement errors, and (3) varying degree of the conservativeness of each tracer or lack of it. To overcome this shortcoming, a new approach of using multiple composite fingerprints was proposed. Contrary to the popular tracer reduction strategies, this new approach uses a maximum number of composite fingerprints, which contain non-contradictory tracers in each composite, to maximize the use of all tracer information. The new approach assumes that source proportions averaged over multiple composite fingerprints are more likely to be closer to the population means than any estimate using a single fingerprint alone. Such a ‘mean of the means’ approach has been shown to not only improve the accuracy but also reduce the uncertainty of the proportion estimates. The model of the absolute relative difference performed slightly better than the squared relative difference model in estimating source contributions, suggesting the former be preferred. The results also indicated that the tracer discriminatory weighting should be excluded as it tends to bias contribution estimates.