Koalas are an iconic Australian species now classified as nationally endangered due to habitat loss, predation, disease, and vehicle collisions. Yet, because of their nocturnal and arboreal nature, koala movement on the ground between trees remains poorly described. Ground visits are fraught with danger, with two-thirds of koala deaths occurring due to vehicle strikes and dog attacks. Quantifying these behaviours can provide insights for conservation efforts targeted to their time on the ground. To quantify detailed movement patterns, we collared nine koalas in a highly fragmented agricultural landscape with tri-axial accelerometers for 8.50 ± 1.17 days. Using an annotated dataset of observed behaviours, we trained Random Forest models to classify four main behavioural states: Motionless in Tree, Feeding & Grooming in Tree, Other Movement in Tree, and Walking. Koalas spent 57.5% ± 14.2% of their time moving in trees, 26.8% ± 13.5% motionless in trees, 15.6% ± 2.7% feeding and grooming, and only 0.2% ± 0.1% walking on the ground (around 3 min daily). Koalas walked most frequently between 2 am and 5 am. Our findings highlight a major discrepancy between mortality rates on the ground and frequency of ground time, indicating that focused mitigation during these infrequent events could yield disproportionately high conservation benefits.
Keywords: accelerometry; behavioural ecology; koala conservation; machine learning; movement ecology.