Comparing alternative tracing measurements and mixing models to fingerprint suspended sediment sources in a mesoscale Mediterranean catchment

Purpose Knowledge of suspended sediment provenance in mesoscale catchments is important for applying erosion control measures and best management practices as well as for understanding the processes controlling sediment transport in the critical zone. As suspended sediment fluxes are highly variable in time, particularly given the variability of soil and rainfall properties in mesoscale catchments, knowledge of sediment provenance at high temporal resolution is crucial. Materials and methods Suspended sediment fluxes were analyzed at the outlet of a 42-km(2) Mediterranean catchment belonging to the French critical zone observatory network (OZCAR). Spatial origins of the suspended sediments were analyzed at high temporal resolution using low-cost analytical approaches (color tracers, X-ray fluorescence, and magnetic susceptibility). As the measurements of magnetic susceptibility provide only one variable, they were used for cross-validation of the results obtained with the two alternative tracing methods. The comparison of the tracer sets and three mixing models (non-negative least squares, Bayesian mixing model SIMMR, and partial least squares regression) allowed us to estimate different sources of errors inherent in sediment fingerprinting studies and to assess the challenges and opportunities of using these fingerprinting methods. Results and discussion All tracer sets and mixing models could identify marly badlands as the main source of suspended sediments. However, the percentage of source contributions varied between the 11 flood events in the catchment. The mean contribution of the badlands varied between 74 and 84%; the topsoils on sedimentary geology ranged from 12 to 29% and the basaltic topsoils from 1 to 8%. While for some events the contribution remained constant, others showed a high within-event variability of the sediment provenance. Considerable differences in the predicted contributions were observed when different tracer sets (mean RMSE 19.9%) or mixing models (mean RMSE 10.1%) were used. Our result shows that the choice of the tracer set was more important than the choice of the mixing model. Conclusions These results highlighted the importance of using multi-tracer multi-model approaches for sediment fingerprinting in order to obtain reliable estimates of source contributions. As a given fingerprinting approach might be more sensitive to one type of error, i.e., source variability, particle size selectivity, multi-tracer ensemble predictions allow to detect and quantify these potential biases. High sampling resolution realized with low-cost methods is important to reveal within- and between-event dynamics of sediment fluxes and to obtain reliable information of main contributing sources.