A Bayesian Unmixing Model for Land-Use Fingerprinting Using Deltaσ⁵ C/N

In this study, a new unmixing model is presented for land-use fingerprinting eroded-soil using delta⁵N and C/N. The model is specified within the Bayesian Markov Chain Monte Carlo framework and is capable of unmixing erosion sources characterized by multiple erosion processes. A unique attribute of the formulation is the use of a truncation parameter to account for uncertainty during the erosion process. The model was applied for fingerprinting soils from forest and agriculture land-uses in Jerome Creek sub-watershed of the Upper Palouse Watershed, Northwestern Idaho. Multiple processes including upland rill erosion and floodplain headcut erosion are represented within the agriculture region of the Upper Palouse. A sensitivity analysis for the unmixing model and comparison of the model with sediment yield estimates for the land-uses add confidence to the accuracy of the model.