Unravelling changing sediment sources in a Mediterranean mountain catchment: a Bayesian fingerprinting approach

To determine the provenance of Holocene floodplain deposits of the Bugduz catchment in southwest Turkey, a Bayesian fingerprinting approach was used. An important requirement for any provenance study is that the potential sediment sources show sufficient spatial and compositional heterogeneity. The spatial distribution of potential sources, in this case the various lithologies present within the catchment, was mapped using field observations and ASTER and Quickbird satellite images. To distinguish the source lithologies, a set of geochemical tracers was identified using linear discriminant analysis. This optimum fingerprint was then used in the mixing model to determine the sediment provenance. The Bayesian mixing model uses Markov chain Monte Carlo random walks to determine the most probable source composition and mixing proportions. The uncertainty associated with the input data can be incorporated into the model through prior probability distributions. The spread of the posterior probability distributions represents the uncertainty associated with the mixing proportion calculation. The main contrasts in the provenance of the floodplain deposits reflect the spatial distribution of potential sediment sources throughout the catchment. There are, however, also important temporal variations in sediment provenance and lateral differences due to the nature of floodplain build-up. The observed spatial and temporal variability of sediment provenance gives a first indication that hill slope-channel and within-channel coupling relations are not uniform through the catchment and that different locations showed a distinct response to disturbances. Copyright © 2012 John Wiley & Sons, Ltd.