AI helps link dinosaur footprints to their makers
Researchers are building an AI-based method to identify which dinosaurs made fossil footprints. The approach classifies tracks using eight measurable traits, offering a more objective comparison than visual guesswork alone.
German scientists recently reported an unusually long trackway from a 30-tonne dinosaur, underscoring how much information tracks can hold when analyzed at scale. Some of the most striking examples come from Texas, where 115-million-year-old prints have surfaced after droughts exposed old riverbeds.
Why footprints matter
Tracks are often more common than bones and show behavior in context. With enough prints in one place, you can infer habitat, crowding, directionality, and which animals likely crossed paths.
That said, matching a track to its maker remains a major challenge. Paleontologists have debated these assignments for generations.
The method in brief
The team analyzed 1,974 images and drawings of dinosaur tracks spanning roughly 150 million years. Their algorithm pulled out eight traits that explain most of the shape variation across prints and mapped those features to likely dinosaur groups.
- Overall footprint area and outline (foot-ground contact)
- Position of weight-bearing within the print
- Toe spread (splay)
- How toes connect to the foot
- Heel position
- Heel loading (relative pressure)
- Relative emphasis of toe vs. heel impressions
- Left-right asymmetry across the track
Lead researcher Gregor Hartmann notes the value here is an objective way to compare prints, reducing reliance on subjective interpretation. After identifying the key features, the team produced a reference chart that links patterns in those traits to suspected dinosaur groups, giving future studies a standard starting point.
What uncertainty looks like in tracks
Track shape isn't just about anatomy. The same animal can leave very different prints depending on gait, speed, and substrate conditions.
- Behavior: walking, running, turning, even swimming
- Substrate: moisture, grain size, firmness
- Post-deposition: how the print was buried and later eroded
Size varies widely too-from chicken-sized meat-eater tracks to bathtub-sized impressions from long-necked giants. Any classification needs to account for that spread without overfitting to size alone.
What this offers to researchers
- Standardized features: A shared set of traits enables consistent labeling across sites and studies.
- Comparability: New finds can be benchmarked against a broad historical dataset.
- Transparency: Feature-based assignments are easier to audit than purely subjective calls.
- Stronger inference: When combined with sedimentology and stratigraphy, feature profiles can narrow candidate trackmakers.
Practical next steps
- Integrate depth data from photogrammetry or laser scans to capture loading more directly.
- Quantify uncertainty for each prediction (confidence bands, posterior probabilities).
- Cross-validate against sites with associated body fossils where available.
- Publish trait extractors and reference charts for reproducibility and wider testing.
Context and sources
The study appears in the Proceedings of the National Academy of Sciences. For background on the journal and related research, see PNAS. For the institute leading the analysis, visit Helmholtz-Zentrum Berlin.
Build similar workflows
If you're setting up feature-based classification pipelines for morphology or other research images, structured training can speed up your workflow. Explore focused options here: AI certification for data analysis.
Your membership also unlocks:
