Pssst… over there, a predator amongst us
Until recently, wildlife ecologists have had to rely on global positioning systems (GPS) that are limited in their ability to capture collective animal behaviours for field studies. With the support of the Marie Skłodowska-Curie programme, the UNGULATE project took video-equipped drones, computer vision and machine learning to Kenya to shed new light on the dynamics of collective vigilance and information transfer in hooved animals in natural herds.
The whole is more than the sum of its parts
With GPS, an animal raising its head or sitting on its haunches can be analysed in the context of things like group size, sex and reproductive status. However, information flow through the group has remained enigmatic. Blair Costelloe, project fellow, notes: “We set out to quantify the vigilance behaviour of individual ungulates (hooved animals that include zebra, buffalo and impala) in natural herds, to understand how these individual strategies generate collective vigilance patterns, and how information transfers between individuals.” Costelloe turned to modern technology to get a better handle on group dynamics. Fixed-wing aircraft-like drones covered large areas and captured overlapping photos used to reconstruct a 3D map via photogrammetry. Helicopter-like drones hovered over herds, taking high-resolution video. Deep convolutional neural networks, depth referring to each ‘layer’ of processing for different aspects, e.g. edges, were then used to train software to identify animals accurately and differentiate them from the environment. The team’s software toolkit, DeepPoseKit, has been described in a recent eLife publication.
A natural (somewhat) controlled experiment
While observational data would be ideal, the ‘sample size’ would be small or the experiment extremely long if Costelloe had to wait to see the animals reacting to natural predators. Since the animals at her field site are not accustomed to people, she and her team played the role of ‘predators’. Costelloe recorded baseline activity levels in an undisturbed state before approaching the animals. Then, as she explains, “we watched the information flow through the group when team members approached by noting when each animal lifted its head and looked towards us. We will compare this to predictive networks based on visual contact, the distance between neighbours or the behavioural state of individuals.” The video capability has opened a whole new world on wildlife ecology. Aside from real-time behaviour, Costelloe has captured intricate networks of trails made by animals moving through the landscape over time. These could constitute a sort of collective ‘memory’ of the animal community engraved on the landscape.
The hunt is on
The team is now working with wildlife managers and conservation practitioners in Kenya, as well as computer scientists and roboticists, to develop practical tools to survey and monitor wildlife. Costelloe plans to study vigilance behaviour in mixed-species herds to evaluate how within-species ‘rules’ might change when multiple prey with differing vulnerabilities are considered. She is also seeking partners to study prey responses to actual predators. Project coordinator Iain Couzin concludes: “There is an urgent need to develop new multiscale imaging technologies for ecological systems. We face increasingly pressing needs to understand how we are impacting life on our planet and to develop quantitative tools to conserve biodiversity.” UNGULATE has made an excellent start.
Keywords
UNGULATE, animal, vigilance, behaviour, collective, herds, drones, video, wildlife, information flow, hooved, predators, Kenya, prey, ecology, conservation, biodiversity
