OSL signal saturation and dose rate variability: Investigating the behaviour of different statistical models

Abstract

Statistical models derived from two inferences, Frequentist and Bayesian, are compared for OSL data analysis. Controlled laboratory experiments were designed to investigate: (1) dose recovery behaviour and (2) dose variability. (1) Dose recovery tests were performed on single grains of quartz close to saturation. We test whether the Central Dose Model (CDM, Galbraith, 1999), which is the most commonly used model, is capable of converging towards the given dose. Furthermore, a set of new Bayesian age models developed by Combès and Philippe (2017) is applied to determine the dose recovery ratio. The results suggest that the CDM underestimates the given dose if no D0 criterion is applied. (2) Bayesian models and frequentist models (CDM and Average Dose Model: ADM, Guérin et al., 2017) are used in dose variability experiments to compare the estimated average dose. To mimic natural beta dose heterogeneity to single grains of quartz, log-normal dose distributions with different dispersions were created artificially. The results indicate an underestimation of the average dose by at least 10% for CDM, lognormal-median and Cauchy for dispersion values greater than 40%. Conversely, we show that the ADM, the Bayesian lognormal-average and Gaussian models converge towards the average of the distribution and display almost no underestimation, for a significant gain in accuracy.