A Sampling-Based Approach to Estimate Tree-Ring Width Chronologies

Sammanfattning

In the field of dendrochronology, scientists work with the process of ordering different trees in time by using their annual tree ring width (TRW). The TRW data are logged in time series and cross-dated with each other to find out in which period of time they both were alive. The two TRW series are then conjoined by taking the average of their ring widths in each overlapping year, and this can be repeated adding more and more trees to create a chronology. This is done adding one tree after the other, analysing where each individual tree ’fits in’, adding the ’best fitting’ series, then next best, third best and so forth. This is a time-consuming process, matching TRW series and determining which series fit in and which do not. In this project a workflow was developed with the aim of automatically cross- dating a given number of TRW series in a probabilistic manner, utilising principles from different statistical methods including Markov Chain Monte Carlo (MCMC) and Artificial Bee Colony (ABC). The workflow was developed to sample from the multivariate distribu- tion of the TRW series’ datings, and gradually converge towards the distribution showing the TRW series’ most probable datings. The developed method proved to be somewhat successful in correctly cross-dating both artificially generated and real tree ring data. It furthermore managed to provide insight into the uncertainty of individual trees’ ring width data. Although this workflow is an early-stage approach and possible improvements were identified, it could potentially help scientists develop chronologies faster and provide them with uncertainty information, whilst merely acting as a tool and not a replacement of the scientist.