Abstract
Long-distance vocalisation is a characteristic of African lion (Panthera leo) behaviour and is important for maintaining territorial boundaries as well as locating distant group members. Vocal signalling is, however, a flexible behaviour that involves varying costs and benefits depending on environmental, social and spatial factors. Motivated by previous data collection limitations, we developed a novel approach to investigate the influence of atmospheric conditions and animal home range geography on lion vocal behaviour using acoustic and accelerometer biologgers. To compensate for the short lifetime of the acoustic biologger, we trained a machine learning model to detect lion roars from long term acceleration signals which yielded over 500 nights of data from 7 individual lions. Analysis of detected roar events revealed that vocalisations occurred mainly at night with a peak just before dawn. The relative likelihood of vocalisation was negatively related to wind speed and temperature and positively related to absolute humidity suggesting that lions preferred to roar under conditions that reduce sound attenuation and thereby maximise calling area. Roar occurrence was found to be dependent on an animal's location relative to its home range with lions demonstrating an apparent avoidance for vocalising beyond the home range boundary. Lions were also more likely to roar repetitively while closer to rivers and water points within their home range. This study is the first of its kind and not only improves the understanding of lion vocal behaviour but can inform new approaches for recording animal vocalisations remotely.