Automated Bayesian layer counting of ice cores

• Main Author: Wheatley, Joseph
• Other Authors: Blackwell, P. ; Wolff, E. ; Abram, N. ; Mulvaney, R.
• Published: University of Sheffield 2015
• Subjects: 510
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.647036

The polar ice sheets hold a continuous record of climatic and environmental information, in the composition and concentrations of various chemicals, particles and gasses, extending back over hundreds of thousands of years. In order to interpret these data we must first learn about the underlying relationship between depth and age. Ice cores are vertical samples of the ice sheets. Some signals measured from them have annual cycles which show as quasi-periodic seasonality; layer counting uses this periodicity to obtain a chronology for the core. This is currently achieved manually, which is time-consuming and open to inconsistency and human error. We present a method to standardise an ice core signal, isolating its seasonality, and to split it into sections with well-defined cycle counts and those with uncertain cycle counts. We show how the uncertain sections can be presented for manual assessment, and describe how the possible reconstructions can be identified and assigned probabilities based on their implied cycle lengths. We also develop a multivariate fully Bayesian approach, which models the signals as phase-shifted sine waves with continuously varying mean and amplitude. We use Markov chain Monte Carlo algorithms to enable inference about the age-depth relationship, and specifically the number of years covered by a particular section of ice core, including quantitative assessment of the uncertainty involved. We provide examples, applying our methods to several chemistry signals measured from ice cores from Greenland and Antarctica.