A Bayesian approach to reconstructing quantitative climate using geochemical lake sediment data

Palaeoclimate reconstructions from proxy data provides the opportunity to extend the instrumental climate record over the last few millennia. This extension allows the identification of climate trends which are not observed in the short observational period, contextualise current changes in the climate system, and be used in climate model sensitivity tests to strengthen future projections. The skill in reconstructing accurate palaeoclimate on centennial to millennial timescales using proxy data has increased over the past decades, however reconstructions at high temporal resolutions (1 to 10 years) are limited. This is mostly due to the sampling resolution of available proxy records and/or uncertain chronologies. Micro X-ray Fluorescence (m-XRF) core scanning data provides multivariate information about the changing geochemical composition of a sediment sequence. When combined with a tightly resolved chronology, such as one from a varved sediment sequence, m-XRF data can be used as a proxy for past changing environmental and climate conditions at an annual to sub-annual timescale. However, despite this, the reconstructions are only semi-quantitative and therefore only used to assess relative changes in climate and environmental variability.

We have developed a novel approach which transforms XRF core scanning data into a quantified climate reconstruction using a Bayesian statistical approach. This method uses a calibration period measuring the relationship between the different geochemical elements and a climate parameter, as well as measuring the covariance and non-linear responses between the different geochemical elements. This learnt statistical relationship is then applied on the proxy data through time with quantified uncertainties calculated at each timestep.

We present the simplification of this approach in a user-friendly R package (SCUBIDO, Simulating Climate Using Bayesian Inference with Proxy Data Observations) which provides functions for the stages of model calibration, application, and validation. We demonstrate this approach using the annually laminated lake sediments of Diss Mere (UK) to reconstruct annual mean temperature for the past 10,000 years. The temperature reconstruction reveals good consistency from the proxy data to the TraCE-21k transient climate simulation, and other palaeoclimate reconstructions from the surrounding region.