The conflict between sampling resolution and stratigraphic constraints from a Bayesian perspective: OSL and radiocarbon case studies

Abstract. Bayesian modelling is often implemented in geochronology and its applications to geomorphology, archaeology, etc. The rationale behind such practices is the aim to improve robustness, precision and accuracy thanks to the use of prior knowledge regarding the studied sites, and in particular the order of samples constrained by stratigraphy. All chronological models tested in this study (OxCal, Chronomodel and BayLum) use the same mathematical model to handle stratigraphic constraints. However, this model has been shown to lead to estimation biases. First, this bias is illustrated with BayLum modelling on a high-resolution OSL dataset. Then, this paper compares statistical inferences obtained with the three above-mentioned modelling software on the Neolithic East mound of Çatalhöyük (Turkey). For this site, 49 radiocarbon ages were obtained with the aim to determine the start of occupations at this locality. Interestingly, age uncertainties are rather large, because of calibration curve plateaus. Therefore, the conditions for estimation biases are met. We discuss the behaviour of the different models and show that caution must be taken when modelling results are at odds with measurements. While OxCal, Chronomodel and BayLum are all affected by a spread in ages resulting from their common model of stratigraphic errors, Chronomodel suffers from a great loss of precision and OxCal, through the phase model, concentrates ages undesirably. We also conclude that the onset of occupations at Çatalhöyük was probably earlier than previously thought based on the OxCal model.