top of page
  • Writer's pictureSciEnvy

How can AI help bats in Borneo? | Natalie Yoh

Updated: Aug 11, 2020

How can machine learning be used to help conserve forest bat species in Borneo?

Photo by Ryan Harvey on Unsplash

Why bats in Borneo?

What do you think of when you think of Borneo’s wildlife? Maybe Orangutans or Birds of paradise? For most people, bats aren’t the first group that spring to mind but they are an integral part of Borneo’s tropical rainforests. There are over 1300 species of bats worldwide (that’s four times as many bats as there are primates) and they make up 40% of Borneo’s land mammal species. As well as dispersing the seeds of approx. 300 tree species on the island they are also vital in regulating insect populations by consuming over half their body weight in insects each night. Sadly, just as with many more well recognised species, bats in Borneo are severely threatened by habitat loss and deforestation from logging and palm oil plantations.

Difficulties in monitoring bats

The problem is - we don’t know enough about these bats to find ways to protect them from these threats. Bats are notoriously difficult to see and many have developed sophisticated echolocation calls which enables them to navigate the dense forest in the dark but this also enables them to detect (and avoid) traps. If we are unable to monitor these species we are unable to study how their populations are changing in response to forest loss or whether conservation efforts are being effective. Approximately 80% of the bats in Borneo use echolocation to hunt for insect prey which provides the potential to use an alternative monitoring technique. Acoustic monitoring of bats is used successfully in a number of countries including; the UK, North America, France, and increasingly also mega-diverse countries such as Mexico. This involved deploying ultrasonic detectors which record the calls of any passing bats. This is an effective way of monitoring bat activity in numerous habitats but it generates huge datasets of recordings. Researchers are then required to process these recordings to determine which calls are from which bat species. To do so, call characteristics must be compared to pre-existing reference libraries (and this often means sifting through numerous insect recordings first to find the bats). This is a time-consuming exercise which requires specialist training and introduced the risk of human error when assigning species identification. For this reasons, the technique is not considered feasible for many projects with limited resources or time. But what if the process was autonomised?

Machine learning for bat identification

As part of my Ph.D. I hope to reduce these limitations by designing a classifier that uses machine learning, an application of artificial intelligence, to automatically recognise bat species from their echolocation calls. This approach has been used to establish projects such as Nature-Smart Cities where bat activity in London’s parks can be monitored in real-time. But in order to be able to do this, a classifier must first be trained using a collection of bat calls from known species. I’m aiming to collate reference calls from numerous researchers working across Borneo to produce this reference library as well as undertaking additional fieldwork to collect more calls. Once up-and-running, I plan to compare the effectiveness of acoustic monitoring against traditional live-capture techniques to work out what is the best way to monitor how bats in Borneo are being affected by deforestation. Most importantly, I am optimistic we can use this information to help conserve these species in human-modified landscapes.

To find out more about bat conservation in Southeast Asia please click here.



bottom of page