Comparing Radiocarbon Age Modeling Routines for Ocean Sediment Cores: Bacon, Undatable, and BIGMACS

Multiple methods are available to generate radiocarbon age models for ocean sediment cores. These methods may produce different age models and ultimately affect the interpretation of paleoclimate data. Here we compare the results of three different software packages applied to data from 53 Atlantic cores. The Bayesian routine Bacon (Blaauw & Christen, 2011) calculates probabilities of sample age models with user-specified priors and likelihoods for accumulation rates and radiocarbon age misfits. The Bacon algorithm was recently implemented in the software toolbox PaleoDataView (Langner & Mulitza, 2019). The deterministic routine Undatable (Lougheed & Obrachta, 2019) emulates the results of probabilistic models with bootstrapping to identify outliers and scaled Gaussian distributions that affect confidence interval widths. Undatable was recently applied to generate age models for a compilation of 92 Atlantic cores (Waelbroeck et al., 2019). Both Bacon and Undatable require subjective parameter settings that determine age model uncertainties. A new Bayesian age model algorithm, BIGMACS (Lee et al., in review; https://arxiv.org/abs/1907.08738), is designed to run without subjective parameter settings using a prior model of sedimentation rate variability based on a compilation of 37 radiocarbon-dated cores (Lin et al., 2014). Here we compare the results of Bacon, Undatable and BIGMACS to evaluate the extent to which the choice of software package may impact the final age model. Specifically, we construct three age models (one with each routine) for 53 of the Atlantic cores compiled by Waelbroeck et al., 2019. For consistency, all methods are run using radiocarbon ages calibrated with the Intcal20 curve and time-dependent modeled reservoir ages for each cores location (Heaton et al., 2020). The results of the three routines are compared for each core based on median age, confidence interval widths, and simulated changes in accumulation rates. An evaluation of these three routines may help future studies decide which method and associated parameter settings are most appropriate for the cores involved in the study, and the extent to which these choices may affect the interpretation of paleoclimate data.