A comparison of geological and statistical approaches to element selection for sediment fingerprinting

Elevated sediment loads reduce reservoir capacity and significantly increase the cost of operating water treatment infrastructure making the management of sediment supply to reservoirs of increasing importance. Sediment fingerprinting techniques can be used to model the relative contributions of different sources of sediment accumulating in reservoirs. The goal of this research is to compare geological and statistical approaches to element selection for sediment fingerprinting modelling. Time-integrated samplers (n = 45) were used to obtain source samples from four major subcatchments flowing into the Baroon Pocket Reservoir in South East Queensland, Australia. The geochemistry of these potential sources was compared to sediment cores (n = 12) sampled in the reservoir. Elements that provided expected, observed and statistical discrimination between sediment sources were selected for modelling with the geological approach. Two statistical approaches selected elements for modelling with the Kruskal-Wallis H test and discriminatory function analysis (DFA). In particular, two approaches to the DFA were adopted to investigate the importance of element selection on modelling results. A distribution model determined the relative contributions of different sources to sediment sampled in the Baroon Pocket Reservoir. Elemental discrimination was expected between one subcatchment (Obi Obi Creek) and the remaining subcatchments (Lexys, Falls and Bridge creeks). Six major elements were expected to provide discrimination. Of these six, only Fe2O3 and SiO2 provided expected, observed and statistical discrimination. Modelling results with this geological approach indicated that 36 % (+/- 9 %) of sediment sampled in the reservoir cores were from mafic-derived sources and 64 % (+/- 9 %) were from felsic-derived sources. The geological and the first statistical approach differed by only 1 % (sigma 5 %) for five out of six model groupings with only the Lexys Creek modelling results differing significantly (35 %). The statistical model with expanded elemental selection differed from the geological model by an average of 30 % for all six models. Elemental selection for sediment fingerprinting therefore has the potential to impact modelling results. Accordingly, we believe that it is important to incorporate both robust geological and statistical approaches when selecting elements for sediment fingerprinting. For the Baroon Pocket Reservoir, management should focus on reducing the supply of sediments derived from felsic sources in each of the subcatchments.