Next-generation radiocarbon calibration: Incorporating information on calibration curve covariance

一个为期6个月的项目，调查IntCal 20放射性碳校准曲线内协方差信息对放射性碳校准的影响。放射性碳（14 C）是测定过去55，000年的最常用方法，并支持环境和考古科学的广度。由于过去14 C水平的波动，放射性碳测年需要根据国际公认的IntCal校准曲线进行校准。传统上，这些IntCal校准曲线提供了过去14 C水平的逐点估计，这些水平被视为独立于一年到下一年。这种独立性表明，真实的14 C水平可能会在一年内从不确定性包络线的底部变化到下一年的区间顶部，然后再次回到底部。这种极端的可变性是不现实的。实际上，曲线估计值具有协方差-我们期望它在其不确定性包络内随时间变化得更慢。当校准多个14 C测定时，这对许多环境建模至关重要，这种协方差很重要。针对最近的IntCal 20曲线实施的曲线构建的新方法能够存储单个校准曲线实现（针对已发布的逐点IntCal 20估计值进行了总结）。每个实现代表了55，000 - 0 cal yr BP的整个合理14 C历史，并对曲线协方差进行编码。该项目旨在探索使用这些实现（以及协方差信息）而不是逐点校准的效果。我们的发现有可能改写我们的校准方式。

A 6-month project to investigate the effect on radiocarbon calibration of covariance information within the IntCal20 radiocarbon calibration curve. Radiocarbon (14C) provides the most common method for dating the last 55,000 years and underpins the breadth of the environmental and archaeological sciences. Due to fluctuations in past levels of 14C, radiocarbon dating requires calibration against the internationally-agreed IntCal calibration curves. Traditionally, these IntCal calibration curves provide pointwise estimates of past 14C levels that are treated as independent from one year to the next. This independence suggests the true 14C level could change from the bottom of the uncertainty envelope in one year, to the top of the interval in the next, and then back to the bottom again. This extreme variability is unrealistic. In reality, the curve estimate has covariance - we expect it to vary more slowly over time within its uncertainty envelope. When calibrating multiple 14C determinations, essential for much environmental modelling, this covariance matters. The novel approach to curve construction implemented for the recent IntCal20 curves enabled storage of individual calibration curve realisations (which were summarised for the published pointwise IntCal20 estimates). Each realisation represents an entire plausible 14C history from 55,000 - 0 cal yr BP and encodes the curve covariance. This project aims to explore the effect of using these realisations (and hence covariance information) rather than pointwise calibration. Our findings have the potential to rewrite the way we calibrate.