Combining spectroscopy and magnetism with geochemical tracers to improve the discrimination of sediment sources in a homogeneous subtropical catchment

An important step in the sediment source fingerprinting approach is the selection of the appropriate tracing parameters to maximize source discrimination. The use of multiple tracing properties may reduce uncertainties and increase discrimination between sources. Accordingly, this study investigates the discrimination and quantifies the contribution of sediment sources delivering sediment to a river draining a homogeneous subtropical agricultural catchment based on the combination of ultraviolet-visible spectra derived parameters (UV), magnetic (M), and geochemical tracers (GEO). The investigated catchment (Conceicao River - 804 km(2)), located in Southern Brazil, has predominantly deep and strongly weathered Ferralsols. The main land-uses found in the area are cropland (89%), pasture (5%) and forest (5%). A total of 187 samples were collected to characterise the five main sediment sources, including cropland, pastures, unpaved roads, gullies and stream banks. A total of 53 tracers, including 21 geochemical tracers, two magnetic properties and 30 parameters derived from UV spectra, were analysed. Tracers were selected following a three step procedure, including: (i) an interquartile range test, (ii) a Kruskal-Wallis H test, and (iii) a linear discriminant function analysis (LDA). The LDA was performed using six different sets of variables: (i) GEO only; (ii) UV only; (iii) M + UV (MUV); (iv) GEO + UV (GUV); (v) GEO + M (GM) and (vi) GEO+ M+ UV (GMUV). The selected tracers were introduced into a mass balance mixing model to estimate the source contributions to in-stream sediment by minimizing the sum of square residuals. Most geochemical tracers were considered not conservative by using the interquartile range test in this catchment with highly weathered soils. The GM approach resulted in the highest percentage of samples correctly classified (SCC), with 74%, followed by the approaches with GMUV and GUV, with 73%. Alternative tracers, UV individually or combined with M tracers, correctly classified only 59 and 60% of the samples, respectively. Moreover, they did not provide significant additional discrimination power even when combined with the GEO tracers. The apportionment model resulted in similar source contribution results for all approaches, with the absence of significant difference when comparing the mean source contributions obtained for the entire set of sediment samples (Cropland: 17-23%; Pastures: 24-34%; Unpaved Roads: 3-12%; Stream Banks: 26-31%; Gullies: 14-19%). Due to the strong homogeneity of soil types found in the Conceicao catchment, these differences in source contributions remained very low and the results of the mixing model were impacted by the high number of potential sources and the relatively limited quality of the sediment source discrimination. According to the model results, the low discrimination between the potential sediment sources illustrates the difficulties for discriminating land-used based sediment sources, with more than three potential sources, in homogeneous catchments with highly weathered soils (e.g. Ferralsols, Nitisols) under tropical conditions.