Applying morphometrics and machine learning to Australian zooarchaeology: case studies from Boodie Cave, Barrow Island

Abstract

Zooarchaeological research in Australia faces important methodological challenges relating to the taxonomic identification of marsupial postcrania. Marsupials are highly diverse and isolated postcranial specimens can be difficult to differentiate to meaningful taxonomic levels. Compared to craniodental morphology, comparatively little research has been undertaken that systematically tests how to differentiate postcrania of closely related marsupials. The colonial disruption of Indigenous ecologies has also significantly altered marsupial biogeography and resulted in many extinctions which further complicates our understanding of the past. A focus on craniodental remains reduces the sample of identified specimens which has implications for understanding taxonomic diversity and abundance in the past. In addition, postcranial data are essential for understanding butchery, body part transport and bone processing practices in the past.

This paper presents the results of methodological experiments using traditional and geometric morphometrics to identify how closely related marsupials vary morphologically. Skeletal elements from two economically important but speciose groups, the Macropodoidea (kangaroos and wallabies) and Peramelemorphia (bandicoots and bilbies) were examined. These data are then combined with machine learning to statistically classify unknown specimens from Boodie Cave, Barrow Island. Taxonomic identification of macropod and peramelemorphian postcrania can be achieved with a high degree of accuracy using machine learning, even where bone morphology visually overlaps. The application of these methods to Boodie Cave has revealed previously unidentified rare and extinct taxa and provides independent evidence of biogeographical shifts associated with changing climatic and environmental conditions over the last 50,000 years.

There is substantial scope for machine learning, using morphometric predictors, to be used as a powerful complementary tool by zooarchaeologists working with biodiverse wild fauna where visual differentiation of specimens is challenging. Statistical classification also provides a solution to long standing issues regarding the replicability and transparency of zooarchaeological data. Drawing on case studies from Boodie Cave this presentation will highlight how these methods can be practically applied to increase sample sizes and enhance our understanding of Indigenous foraging strategies in Australia.

About the presenter

Erin Mein is a PhD candidate at the University of Queensland. Her current work focuses on developing novel quantitative methods to help taxonomically identify marsupial remains from Australian archaeological sites. She has a particular interest in the archaeology of the Australian arid zone and palaeoecological implications of zooarchaeological data. Erin received her BA (Honours) from the University of Western Australia and spent seven years working as a Heritage Consultant in NSW before commencing her postgraduate studies. She is currently employed at Flinders University on the ARC Centre of Excellence for Biodiversity and Heritage (CABAH) funded Ozboneviz project. This project aims to elevate our capacity for research on palaeozoological assemblages by providing a high quality, open access, 3D digital reference collection and to develop teaching resources for tertiary and secondary school contexts.