Comparing catchment sediment fingerprinting procedures using an auto-evaluation approach with virtual sample mixtures

Information on sediment sources in river catchments is required for effective sediment control strategies, to understand sediment, nutrient and pollutant transport, and for developing soil erosion management plans. Sediment fingerprinting procedures are employed to quantify sediment source contributions and have become a widely used tool. As fingerprinting procedures are naturally variable and locally dependant, there are different applications of the procedure. Here, the auto-evaluation of different fingerprinting procedures using virtual sample mixtures is proposed to support the selection of the fingerprinting procedure with the best capacity for source discrimination and apportionment. Surface samples from four land uses from a Central Spanish Pyrenean catchment were used i) as sources to generate the virtual sample mixtures and ii) to characterise the sources for the fingerprinting procedures. The auto-evaluation approach involved comparing fingerprinting procedures based on four optimum composite fingerprints selected by three statistical tests, three source characterisations (mean, median and corrected mean) and two types of objective functions for the mixing model. A total of 24 fingerprinting procedures were assessed by this new approach which were solved by Monte Carlo simulations and compared using the root mean squared error (RMSE) between known and assessed source ascriptions for the virtual sample mixtures. It was found that the source ascriptions with the highest accuracy were achieved using the corrected mean source characterisations for the composite fingerprints selected by the Kruskal Wallis H-test and principal components analysis. Based on the RMSE results, high goodness of fit (GOF) values were not always indicative of accurate source apportionment results, and care should be taken when using GOF to assess mixing model performance. The proposed approach to test different fingerprinting procedures using virtual sample mixtures provides an enhanced basis for selecting procedures that can deliver optimum source discrimination and apportionment.