A Bayesian Mixing Model Framework For Quantifying Temporal Variation In Source Of Sediment To Lakes Across Broad Hydrological Gradients Of Floodplains

Paleolimnological reconstructions provide insights into hydrological variability of dynamic floodplain lakes. However, spatial and temporal integration of multiple reconstructions often remains underdeveloped because the efficacy of different paleolimnological measurements varies among lakes due to gradients in energy of floodwaters and sediment composition. Here, we use linear discriminant analysis to identify 10 significant elemental concentrations in sediment obtained from multiple sampling campaigns that distinguish among three end-member allochthonous sources for lakes in the Peace-Athabasca Delta (PAD; Canada): Athabasca River, Peace River, and local catchment runoff. Over 90% of sediment samples were correctly classified into original groups after cross-validation due to the distinctiveness of the three end-members, which permitted development of a robust Bayesian mixing model to discern the relative contributions of sediment from the three sources. We evaluate performance of the mixing model via application to sediment cores from two adjacent lakes in the Athabasca sector of the PAD and demonstrate its effectiveness to discriminate three known hydrological phases during the past 300 years. Notably, model results indicated that similar to 60% of the sediment originated from the Peace River during the largest ice-jam flood event on record (1974), which was unrecognized by other methods. The approach provides a new, universal method that can be applied across the full range of sediment composition to quantify changes in source, frequency, and magnitude of sediment delivery by river floodwaters to lakes and is transferable to other dynamic floodplain landscapes where broad range of sediment composition challenges application of other approaches.