University of Wollongong
Browse

File(s) not publicly available

Bayesian analysis of De distributions in optical dating: Towards a robust method for dealing with outliers

journal contribution
posted on 2024-11-17, 16:22 authored by Bo Li, Zenobia Jacobs, Richard G Roberts
In optical dating, especially single-grain dating, various patterns of distributions in equivalent dose (De) are usually observed and analysed using different statistical models. None of these methods, however, is designed to deal with outliers that do not form part of the population of grains associated with the event of interest (the ‘target population’), despite outliers being commonly present in single-grain De distributions. In this paper, we present a Bayesian method for detecting De outliers and making allowance for them when estimating the De value of the target population. We test this so-called Bayesian outlier model (BOM) using data sets obtained for individual grains of quartz from sediments deposited in a variety of settings, and in simulations. We find that the BOM is suitable for single-grain De distributions containing outliers that, for a variety of reasons, do not form part of the target population. For example, De outliers may be associated with grains that have undesirable luminescence properties (e.g., thermal instability, high rates of anomalous fading) or with contaminant grains incorporated into a sample when collected in the field or prepared in the laboratory. Grains that have much larger or smaller De values than the target population, due to factors such as insufficient bleaching, beta-dose heterogeneity or post-depositional disturbance, may also be identified as outliers using the BOM, enabling these values to be weighted appropriately for final De and age determination.

Funding

Australian Research Council (CE170100015)

History

Journal title

Quaternary Geochronology

Volume

67

Language

English

Usage metrics

    Categories

    No categories selected

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC