Opportunities and challenges in the reconstruction of environmental processes using segmented micro-CT scans of coral skeletons
Year: 2025
Type: conference-abstract
Abstract: Over the past decades coral skeletons have become established as excellent archives of environmental changes, with seasonal banding in either skeletal density or trace element geochemistry providing a method of reconstructing annual to sub-annual processes. Density banding was first observed using radiography to produce two-dimensional images of a skeletal slab. More recently, micro-tomography has been used to develop three-dimensional reconstructions of a skeletal sample, which provides opportunities to understand microstructural changes and associated variations in bulk density in different compartments of coral skeletons. Here, we use micro-tomography to track changes in density in two species of Caribbean coral, Siderastrea siderea and Pseudodiploria strigosa. Specifically, we develop a method that involves unsupervised learning for instance segmentation of microstructrual features such as corallites. We then combine this with semantic segmentation to separate individual voxels as either open space or coral skeleton, from which we can calculate and then track changes in porosity and thus bulk density. The results show that there are clear changes in the rate of extension throughout an annual cycle, and by tracking these variations we observe clear changes in porosity associated with environmental disturbances such as volcanic eruptions. Additionally, we show that these changes manifest differently in different compartments of the coral skeleton. Finally, we develop panoptic segmentation methods as a tool to overcome non-linearities in coral extension across the surface of the coral, by following the extension of individual features throughout the growth of the skeletal sample.
Authors: Tom Sheldrake, James Vincent
Institution: University of Geneva
Language: English
Subfield: Ecology
Field: Environmental Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4408433567