Clustering Of Calibrated Radiocarbon Dates: Site-Specific Chronological Sequences Identified By Dense Radiocarbon Sampling

Calibrated radiocarbon (C-14) determinations are commonly used in archaeology to assign calendar dates to a site's chronological phases identified based on additional evidence such as stratigraphy. In the absence of such evidence, we can perform dense C-14 sampling of the site to attempt to identify periods of heightened activity, separated by periods of inactivity, which correspond to archaeological phases and gaps between them. We propose a method to achieve this by hierarchical cluster analysis of the calibrated C-14 dates, followed by testing of the different clustering solutions for consistency based on silhouette coefficient and statistical significance using randomization. Separate events identified in such a way can then be regarded as evidence for distinct phases of activity and used to construct a site-specific sequence. This can be in turn used as a Bayesian prior to further narrow down the distributions of the calibrated C-14 dates. We assessed the validity of the method using simulated data as well as real-life archaeological data from the Bronze Age settlement of Troy. A Python implementation of the method is available online at https://github.com/demjanp/clustering_14C.