Using multiple composite fingerprints to quantify source contributions and uncertainties in an arid region

Abstract

Purpose: Developing targeted protection measures at a watershed scale requires spatially distributed information of sediment sources. Therefore, the objectives of this study are to (1) test and evaluate the ability of multiple composite fingerprints (MCF) to quantify sediment provenance using multiple particle size classes in an arid region; (2) quantify uncertainty of the estimated proportional contributions of sediment sources; and (3) provide decision support information for sediment control in the Danghe Reservoir Watershed. Materials and methods: In total, 66 samples were collected from north alluvial fan, south alluvial fan, and high mountains, and all samples were divided into six particle size groups. A multistep test was used to remove the tracers that were non-conservative, unable to differentiate sources, or highly variable within a source. Based on geochemical properties of distributed source samples and a linear mixing model, a MCF method with multiple particle size tracking was used to estimate proportions of three potential source contributions. More importantly, the uncertainty of sediment source contributions was quantified using the Gaussian first-order approximation. Results and discussion: The results showed that the MCF method with multiple particle size tracking could obtain relatively accurate estimates of the contributions with an overall mean absolute relative error of 3.5% and a relatively narrow 95% confidence interval. The major contributions were consistently coming from the high mountains for all six particle groups. During these runoff events, the overall estimated mean proportions were 49.0%, 26.5%, and 24.5% from the high mountains, south alluvial fan, and north alluvial fan, respectively. Furthermore, the Gaussian first-order approximation revealed that more than 60% of the total uncertainty contribution was a byproduct of the downstream sediment mixture, while each individual sediment source produced less than 15% of the absolute uncertainty. Conclusions: Acquiring watershed scale sediment source information is challenging and the MCF method proved accurate. A majority of the contribution uncertainties were associated with the downstream sediment mixture, which is because the sediment sink inherited both spatial and temporal variations of all contributing sources. Consequently, a larger sample size is recommended for sediment mixtures, compared to each sediment source, in order to increase the accuracy of the source proportion estimation.