Virtual reconstruction of modern and fossil hominoid crania: consequences of reference sample choice

Most hominin cranial fossils are incomplete and require reconstruction prior to subsequent analyses. Missing data can be estimated by geometric morphometrics using information from complete specimens, for example, by using thin-plate splines. In this study, we estimate missing data in several virtually fragmented models of hominoid crania (Homo, Pan, Pongo) and fossil hominins (e.g., Australopithecus africanus, Homo heidelbergensis). The aim is to investigate in which way different references influence estimations of cranial shape and how this information can be employed in the reconstruction of fossils. We used a sample of 64 three-dimensional digital models of complete human, chimpanzee, and orangutan crania and a set of 758 landmarks and semilandmarks. The virtually knocked out neurocranial and facial areas that were reconstructed corresponded to those of a real case found in A.L. 444-2 (A. afarensis) cranium. Accuracy of multiple intraspecies and interspecies reconstructions was computed as the maximum square root of the mean squared difference between the original and the reconstruction (root mean square). The results show that the uncertainty in reconstructions is a function of both the geometry of the knockout area and the dissimilarity between the reference sample and the specimen(s) undergoing reconstruction. We suggest that it is possible to estimate large missing cranial areas if the shape of the reference is similar enough to the shape of the specimen reconstructed, though caution must be exercised when employing these reconstructions in subsequent analyses. We provide a potential guide for the choice of the reference by means of bending energy.