Assessment of the Applicability of the Mixing ModelsUsed in the Sediment Fingerprinting of Different SourcesDeposited in the Lavar Fin Reservoir, the Province of Hormozgan

Identification of the erosion-pron parts of a watershed is of at most importance if the soil conservation activities are to be implemented on it to mitigate sedimentation into the flood-receiving reservoir. Sediment fingerprinting is one of the most common methods used for quantifying source contributions of the suspended load. As the mixing models with different structures of sediment fingerprinting method are implemented, their advantagesand disadvantages should be identified. The applicability of eight mixing models,namely: Collins, Hughes, Motha, Slattery, Landwher, Modified Landwher, and Bayesian with the CLR transformation and the Dirichlet distribution were investigated in order to quantify source contributions of sediment deposited in the Lavar Reservoir, the Province of Hormozgan. Twenty-three soil samples were collected from the contributing watersheds, 9 sediments samples were extracted from the reservoir, and concentration of 56 elements were measured in each of the samples. The optimum composite fingerprints were identified by statistical methods and the mixing models were executed. Based on the results, four geochemical properties,namely Mn, La, Nd and Th were selected as optimum fingerprints. The results obtained by all of the mixing models were similar when the values of tracer concentrations in the sediment samples fall inside of those ranges in the source samples. When the values of tracers in the sediment samples fall outside of those values in the source samples, the mixing models with the same objective functions presented similar results. The results of Collins̛, model were similar to those of Hughes, and the results of Bayesian models were similar to those of Hughes the with the CLR transformation;the results calculated by the Motha were similar to those presented by Slattery, and results of Landwher were similar to the modified Landwher. Generally, applicability of the various mixing models in fingerprinting are different, as their outputs are dependent on the target functions, which are minimized in optimization.