A comprehensive identification protocol for postcranial bones of Southeast Asian Ruminantia. Part 3: Morphological criteria for large-sized species

Abstract

Postcranial remains of large-bodied ruminants are abundant in the Late Quaternary record of continental Southeast Asia, yet their taxonomic attribution remains one of the least resolved problems in the zooarchaeology of this biodiversity-rich region. We address it here with an osteological identification protocol covering the six large ruminant species of the mainland: Bubalus bubalis, Bos javanicus, Bos sauveli, Bos gaurus, Budorcas taxicolor and Rusa unicolor, built from a comparative sample of 109 skeletons. A total of 148 discrete characters over 27 skeletal elements representing 442 states were defined on a restricted set of individuals, then scored across the entire sample. Most proved remarkably stable within taxa with 80.0% of the character/taxon combinations showing a dominant state at a frequency above 95%, and a substantial number of states occurred in a single family, subfamily, genus or species. Character combinations for taxonomic identification were then evaluated by the use of two supervised machine learning methods, a random forest model and a naive Bayes classifier at each of four taxonomic ranks. The two behaved almost identically, with mean accuracies of 95-94% for families, 93% for subfamilies, 89% for genera and 76% for species; only the three closely related and unevenly sampled species of Bos resisted consistent separation. The weighted naive Bayes models are distributed as a standalone web application in which any combination of states observed on a single element returns attribution probabilities for every rank. Together these results offer a transparent and statistically explicit alternative to expert judgement for the identification of Late Quaternary postcranial mammal remains, in Southeast Asia and elsewhere.