Insights into Lake Baikal Radiocarbon Age Offsets from a Database of AMS Radiocarbon Age Estimates

Abstract. Radiocarbon dates are an essential tool for dating non-varved lake sediments, however their interpretation is hindered by issues such as reservoir age or contamination which culminate in age estimates that can be thousands of years younger or older than the true depositional age of that sediment (we call this an age offset). Often, precise estimators of the radiocarbon age offset are not available, as in the case of Lake Baikal. Linear regression of uncalibrated radiocarbon dates has been used to estimate the age offset, with answers ranging from 0 to 1500 ¹⁴C yr BP. These have been interpreted to suggest that different regions of Lake Baikal have different age offsets, although some dispute this. Other estimators have returned estimates of approximately 2000 ¹⁴C yr BP. Despite this, most previous studies have not included any estimates of uncertainties for these age offsets in their proxy analysis, or have included uncertainty of, at most, ± 90 ¹⁴C yr. Here, we present a complete database of published AMS radiocarbon dates from Lake Baikal sediment cores up to 2023 and, using this, review the use of linear regression on uncalibrated radiocarbon ages as a method for estimating age offsets from the sediments of Lake Baikal. We apply a standardised linear regression age offset method to all cores in our database to better quantify the age offset of Total Organic Carbon (TOC) in the lake’s sediments. We conclude that there is no statistically significant evidence from linear regression methods for a large difference in age offset in different regions of Lake Baikal. Our results return a lake-wide age offset estimate of TOC of 1.56 ± 0.75 ¹⁴C kyr BP, suggesting previous studies in Lake Baikal have significantly underestimated the temporal uncertainty of radiocarbon ages. Finally, our results are a caution that linear regression-based age offset estimates in lake sediments have a large uncertainty that might only be observable with multiple datasets.