Bayesian approach to OSL dating of poorly bleached sediment samples: Mixture Distribution Models for Dose (MD2)

Highlights: • Mixture of a parametric distribution and a non parametric distribution is used to describe poorly bleached OSL measurement. • This model is developed in a Bayesian framework. • A JAGS implementation of the model is provided to be usable. Optically Stimulated Luminescence (OSL) is commonly used to date the last exposure of grains extracted from sediments to sunlight. However, it is frequent that some of the measured grains were not sufficiently exposed to light before burial. Such samples are said to be poorly bleached. We propose a new statistical model based on a Bayesian approach to analyse OSL measurements performed on poorly bleached sediment samples. For such data, we propose a mixture model of Gaussian distributions to analyse equivalent doses ($D_e$) distributions. This model can either be applied directly to the observed $D_e$ values, or after a log-transformation. Bayesian analysis requires numerical approximation, to do this we use the JAGS (Just Another Gibbs Sampler) programme to run models using Markov Chain Monte Carlo simulations. We apply the model to synthetic datasets and real samples.