A fossil ankle has a practical advantage over a magnificent horn: researchers may have many more ankles to work with. Their shapes also record anatomy relevant to movement. Extracting that information demands comparisons, because a recognisable difference between bones does not automatically identify the species that left them.[2]
A 2026 study of As-Sahabi in Libya applies machine learning to 217 bovid bones from a roughly 6.8-million-year-old rock unit: 159 ankle bones and 58 proximal toe bones. Models trained on modern animals first assign a subfamily, then infer habitat using measurements and that assignment.[1]
The resulting collection suggests a mixed landscape. Its three largest habitat groups are open environments at 48%, light woodland at 29% and forest at 20%; a smaller heavy-cover group is also present. These are proportions of classified fossils. They are not a survey measuring how much ground each vegetation type occupied.[1]
The method assumes the relationship between anatomy, lineage and habitat transfers from living animals. Its habitat stage also treats the preceding subfamily prediction as correct.[1]
That dependency has a useful precedent. A 2012 study compared ankle bones from living pronghorn and fossil collections. Variation could help reject the proposition that a fossil assemblage contained only one species, while still failing to identify every individual bone reliably. The authors also found that a previously published African bovid classification did not transfer effectively to modern pronghorn. A successful anatomical comparison has a biological domain.[2]
A simple hypothetical shows why the unit of measurement matters. Imagine a collection with 50 bones assigned to open country and 50 to woodland. Each category contributes half the bones. Doubling the number of bones from the open-country animals changes that share to two-thirds, even if the imagined landscape itself stays exactly the same. Converting counts into land area would therefore require an additional relationship between animals, recoverable bones and hectares. The arithmetic alone cannot supply it.
The older study’s distinction remains productive: an assemblage can reveal a pattern even when each specimen resists an exact name. Abundant, ordinary bones can contribute evidence that conspicuous diagnostic fossils leave too sparse to assemble.[2]
Sources & further reading
- As-Sahabi bovid postcranial fossils Palaeobiodiversity and Palaeoenvironments · 1 May 2026 · Accessed 20 Oct 2026
- Artiodactyl postcrania and species identification Palaeontologia Electronica · Accessed 20 Oct 2026