Using AI to find dinosaur fossils

Ever looked at imprints in coastal rocks and wondered if they were left by a dinosaur?

It’s unlikely, of course, and even if they had survived for 100m years or so, it could be extremely tricky to pin them down to a certain animal.

Paleontologists are frequently uncertain of the originator of a print and often disagree amongst themselves – so could AI help? And will it help us find more prints, more quickly and easily?

“Personally I think that machine learning is useful here, because deciding if a footprint is a theropod (like, a T. rex) or an ornithopod (like, an iguanodon) involves subjectivity. Especially in cases where there is disagreement,” says Peter Falkingham, professor of paleobiology at LJMU.

Peter, who along with Masters student Michael Jones, postdoc Dr Jens Lallensack and the mathematicians Dr Ian Jarman and Dr Ivo Siekmann trained six machine learning models to classify footprints either as theropod or ornithopod. Both are tridactyl (three-toed) with the former meat-eating and the latter herbivorous.

Of all models, one called multi-layer perceptron (MLP) performed best, correctly classifying footprints of both groups of dinosaurs 90% of the time.

The idea is to classify footprints that are not classified, yet, or that palaeontologists are uncertain about. Machine learning methods do this by processing examples of footprints that have already been classified by experts – the method then “learns” the patterns that distinguish one class from another.

"Trackways preserve fossilised behaviour, something we are unable to get from the bones of an animal alone" - Professor Peter Falkingham

Although we don’t have the remains of the animals that made the tracks, scientists can learn a lot about the lives of dinosaurs from the footprints they left behind.

They can tell us when certain dinosaurs roamed, how they moved also give clues on things like how different species interacted. The shape of the footprints and their arrangement within a trackway can reveal even more information, such as how fast a dinosaur was moving.

Trackways are important, says Professor Falkingham, because they preserve fossilised behaviour, something that we are unable to get from the bones of an animal alone.

Individual footprints are quite commonly found in the UK, especially on the coasts around Sussex, the Isle of Wight and Yorkshire, and recently a whole trackway was discovered in Oxfordshire.

The scientists also say that the development of a non-human ‘spotter’ opens up future possibilities, such as spotting fossilised tracks from drones, identifying new tracksites we didn’t know about.

The research is published in the Journal of Vertebrate Paleontology. Classification of dinosaur footprints using machine learning