StalAge – An algorithm designed for construction of speleothem age models

Abstract

Here we present a new algorithm ( StalAge), which is designed to construct speleothem age models. The algorithm uses U-series ages and their corresponding age uncertainty for modelling and also includes stratigraphic information in order to further constrain and improve the age model. StalAge is applicable to problematic datasets that include outliers, age inversions, hiatuses and large changes in growth rate. Manual selection of potentially inaccurate ages prior to application is not required. StalAge can be applied by the general, non-expert user and has no adjustable free parameters. This offers the highest degree of reproducibility and comparability of speleothem records from different studies. StalAge consists of three major steps. Firstly, major outliers are identified. Secondly, age data are screened for minor outliers and age inversions, and the uncertainty of potential outliers is increased using an iterative procedure. Finally, the age model and corresponding 95%-confidence limits are calculated by a Monte-Carlo simulation fitting ensembles of straight lines to sub-sets of the age data. We apply StalAge to a synthetic stalagmite ’sample’ including several problematic features in order to test its performance and robustness. The true age is mostly within the 95%-confidence age limits of StalAge showing that the calculated age models are accurate even for very difficult samples. We also apply StalAge to three published speleothem datasets. One of those is annually laminated, and the lamina counting chronology agrees with the age model calculated by StalAge. For the other two speleothems the resulting age models are similar to the published age models, which are both based on smoothing splines. Calculated uncertainties are in the range of those calculated by combined application of Bayesian chronological ordering and a spline, showing that StalAge is efficient in using stratigraphic information in order to reduce age model uncertainty. The algorithm is written in the open source statistical software R and available from the authors or as an electronic supplement of this paper.