SEDIMENT FINGERPRINTING EXPERIMENTS TO TEST THE SENSITIVITY OF MULTIVARIATE MIXING MODELS TO VARIABILITY IN THE PARTICLE SIZE OF SOURCE MATERIALS

Abstract

The sediment fingerprinting technique can provide information to help identify and quantify the source of mobilised sediments in agroforestry catchments. Discriminating the potential contribution from sediment sources is necessary for the understanding of soil redistribution processes and, to support soil and water resources conservation and catchment management strategies. Sediment fingerprinting experiments have been used to demonstrate the sensitivity of numerical mixing model outputs to different particle size distributions in source materials and experimental sediment mixtures. Relationships between tracer properties and particle size are not always consistent between fingerprint elements due to different controls (e.g. mineralogy versus adsorption). Results suggested that particle size correction procedures require careful scrutiny in the context of variable source characteristics and that fractionation of source and receptor materials to compare like-with-like across the size range might improve mixing model output. Accordingly, the influence of source material particle size distribution on mixing model output and uncertainty need to be tackled. To explore this factor in more detail, experimental sediment mixtures were prepared by combining different proportions of the <63 µm fraction of experimental source soils with significantly different geochemistry. Subsequently, both experimental sources and sediment mixtures were dry sieved at <40 µm and <20 µm. In an attempt to understand if particle-size correction factors may be appropriate under certain conditions, the specific relationships between individual major and minor element geochemical constituents and particle size fractions were explored. The results from mixing model comparisons of like-with-like (at <63, <40 and <20 µm) for both source materials and the different sediment mixtures are discussed in the context of the variable impact of fluvial sorting on end-member signatures used in mixing models.