Evaluation of five different sediment fingerprinting approaches for estimating sediment source contributions in an arid region

• Published in: Geoderma Vol. 427; p. 116131
• Main Authors: Niu, Baicheng, Liu, Benli, Zhang, Xunchang (John), Liu, Fenggui, Zhou, Qiang, Chen, Qiong, Qu, Jianjun, Liu, Bing
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2022
• Subjects: Arid region; Multiple composite fingerprints; Optimal composite fingerprint; Optimal solution; Optimal tracer number; Sediment fingerprinting methodology; Optimal composite fingerprint; Sediment fingerprinting methodology; Optimal tracer number; Arid region; Multiple composite fingerprints; Optimal solution
• ISSN: 0016-7061; 1872-6259
• DOI: 10.1016/j.geoderma.2022.116131

•More composite fingerprints (CFs) produce better estimation based on the law of large numbers.•Methods of optimal mathematical solutions depend on the number of CFs.•Optimal number of tracers in a CF depends on several factors (not the more the better)•CFs with greater discriminant ability may not necessarily translate to better estimation. Fingerprinting methods are widely used to quantify sediment provenance at a watershed scale. However, different fingerprinting methods often yield different estimates for the same watershed. The objectives are to discuss in detail the performance of five fingerprinting approaches, compare the efficiency of various solutions, and investigate the effects of the number of composite fingerprints and the number of tracers in a composite fingerprint on the estimation accuracy. Source samples were collected from three typical geomorphic areas of the Dune, Gobi, and Mountains in the Danghe Reservoir watershed in Northwest China, and sediment samples from the Danghe reservoir. Overall results suggested that a multiple composite fingerprints approach was superior to a single or few composite fingerprints approach. Generally, increasing the number of composite fingerprints tended to improve the estimation accuracy. The preferred methods of solving the mixing model should depend upon the number of composite fingerprints. When the number of composite fingerprints is large, analytical solutions, compared to numerical solutions, are preferred as the computation is simple while providing an exact solution and better estimation by eliminating fitting errors. For a single composite fingerprint or only a few composite fingerprints, Monte Carlo (MC) simulation, which can statistically increase sample number, is favored to obtain a reliable estimation. However, for a moderate number of composite fingerprints, direct optimization using average tracer concentrations without MC simulation may be desirable, because it tends to yield similar accuracy to a full MC simulation while avoiding intensive computation. There was no clear relationship between the number of tracers in a composite fingerprint and the estimation accuracy, as the optimal number of tracers is situation dependent. Results also indicate that a composite fingerprint with higher discriminant ability may not necessarily translate to better estimation, as it is also affected by tracers’ conservatism and measurement errors.