Using shape analysis and deep learning to classify trace fossils

Trace fossils capture the presence and behaviours of ancient organisms in the form of trackways, trails, and burrows preserved in the rock record. They are important sources of evidence for investigating changes in biodiversity through deep time, such as during evolutionary radiations and mass extinctions. In the past decade, there has been a surge in studies using trace fossils in meta-analyses of major evolutionary events; however, all this work relies on a robust taxonomic framework for trace fossils (i.e., ichnotaxonomy), which is currently lacking in many cases. To make ichnotaxonomy less qualitative and subjective, we need to: (1) find ways to quantify the complex morphological variation in trace fossils that results from interactions among the anatomy of the trace maker, its behaviour, and the underlying substrate conditions; and (2) explore the potential of deep learning as a tool for being able to automatically identify ichnotaxa.

This project will focus on trackways made by arthropods, drawing on an existing dataset comprising digital images of thousands of specimens. The project has two sequential objectives: (1) employ morphometrics to develop an objective system that we can use to review and accurately classify ichnotaxa; and (2) use this to generate a training dataset to teach a deep learning model to automatically classify trace fossils, thereby making ichnotaxonomy more robust and accessible. Achieving these objectives will enable broader scientific engagement with trace fossil data, allowing their full potential to be unlocked for exploring the ecology and evolution of past species and environments throughout Earth’s history.