Eavesdropping on wildlife to reduce human-tiger conflict

A recent paper led by Dr Arik Kershenbaum, a College Associate Professor here in Zoology, describes how both wildlife and humans could benefit from networks of acoustic detectors to monitor tiger activity in Nepal. 

Here Dr Kershenbaum writes about his group’s research, and how a faux tiger skin came in handy.

Endangered Bengal tiger populations are increasing across India and Nepal – a conservation success story, but one that also creates tragic problems for local communities. The Tharu people of southern Nepal live a traditional lifestyle, gathering subsistence supplies from the forest on a daily basis, and are strongly supportive of conservation in general, and the conservation of tigers in particular. The forest is part of their traditional way of life, and that includes all the creatures in it. 

However, the toll on livestock and human life is terrifying: there were 32 confirmed human fatalities from tiger attacks between 2007 and 2014 alone. Our goal is to make the forest a safer place for local villagers, while maintaining the forest’s natural fauna – including tigers.

Our research group, established in 2018, uses new technologies to monitor animal population through the sounds that they make. Our first project, in Wisconsin USA, brought together researchers from Cambridge, Europe, and the USA to study the interactions of wolves, dogs, and coyotes by analysing their howls and barks. In 2020, two former Cambridge students travelled to Romania to study the spread of golden jackals there, and their interactions with people and other fauna. And since 2022, we have been running a long-term study in Vietnam, monitoring the 70 remaining individuals of the critically endangered cao vit gibbon via their loud and distinctive songs.

Although tigers rarely make any sound at all – and particularly not while hunting – other forest animals are always on the lookout for predators and give loud alarm calls when they spot a tiger. Spotted deer (Axis axis) are abundant and very loud. Our idea was to listen for those alarm calls and to create a risk map, showing in real time which parts of the forest are experiencing a lot of alarm calls, and therefore have a higher risk of tiger presence. We work with local government forest rangers, who routinely patrol the community forests. By sending alerts directly to a ranger’s cellphone, along with a risk map, they can head to the high-risk areas of the forest and advise villagers to move to safer areas for foraging.

Apart from the innovative use of prey alarm calls to track the tigers, we are also pushing a promising paradigm in applying AI to real world problems. Rather than using powerful and expensive computers to process large amounts of data, we deploy small, low-power devices across the forest, each one performing a lightweight AI process – just detecting whether they are hearing a deer alarm call. Each device is also equipped with a low-power radio and communicates with nearby devices, sharing information and passing messages to a gateway device that then forwards these messages to the internet. As a result, we can monitor tiger presence in a forest in Nepal, from an office in Cambridge.

Before even developing such a system, we had to make sure that spotted deer really do make distinctive alarm calls to tigers, and not just to every disturbance (e.g. humans moving through the forest). To do that, one of our team members draped themselves in a faux tiger skin and crawled through the forest towards the deer, while another researcher recorded the sounds that were produced. Although it sounds unconvincing, this kind of fake predator presentation is remarkably effective – and even scared more than one villager who happened to pass by!

Read the paper: Kershenbaum, A., Markham, A., Root-Gutteridge, H., Smith, B., Anderson, C., McClaughry, R., Chaudhary, R., Vishwakarma, A., Cummins, S. and Dassow, A. (2026), An autonomous network of acoustic detectors to map tiger risk by eavesdropping on prey alarm calls. Remote Sens Ecol Conserv. https://doi.org/10.1002/rse2.70061