The fact the eastern bristlebird had not been seen nor heard in south-east Queensland since its Gondwana rainforest home was ravaged in the black summer bushfires of 2019/20 was, in some ways, unsurprising.
For one, there are thought to have been fewer than 40 individual birds in its northern population.
Couple that to the fact it is a “nondescript, brown bird”, shy and secretive, that flits along the ground between shrubs doing its darnedest not to be seen.
That makes the bristlebird’s call the most efficient way of tracking it down.
Normally, that would involve a person going into the forest and playing a recording of a call in an effort to coax a response from a wild bird.
“But you’ve gotta be at the right place at the right time and the bird’s got to want to respond,” says Queensland University of Technology’s Susan Fuller.
So QUT researchers teamed up with BirdLife Australia and Healthy Land and Water to place five acoustic monitors in the bristlebird’s northern range mid last year, returning only to replace batteries and weeks later for the recordings.
The results were heartening, confirming the existence of the elusive bird feared lost to south-east Queensland.
The potential of this kind of monitoring, called passive acoustic monitoring, has excited scientists for more than a decade. But it is recent advances in computer science and artificial intelligence that have helped make that potential a reality, Fuller says.
“We’ve always come back to the same stumbling block of someone having to sit down and go through the recordings minute by minute, manually identifying the calls,” the associate professor at QUT’s Centre for the Environment says.
For a large conservation project, that could amount to terabytes of data – a trove physically impossible for a human to comprehensively review.
In this case, QUT computer scientist Dr Lance De Vine developed an AI model that could be trained to recognise bristlebird calls among the hours and hours of field recordings.
“Without AI we can’t do this,” Fuller says. “This is a gamechanger for us.”
The breakthrough was still grounded in ecological understanding and human expertise – it was BirdLife threatened species project officer and QUT PhD candidate Callan Alexander who first picked a bristlebird call from the recordings.
Using Alexander’s trained ear, De Vine was able to gradually coach the AI program to accurately identify the endangered bird call from other similar noises, and then let it loose upon the rest of the recordings, from which it discovered 350 eastern bristlebird calls over the two-month period.
After this preliminary breakthrough, the researchers now have 20 monitors over a broader range of habitat.
AI offers considerable further potential for conservation, Fuller says, including to identify the calls of individual animals from recordings, not just species.
The scientist says soundscapes can provide unique insights into the overall health of an ecosystem.
When displayed as a spectrogram (a visual representation of the spectrum of frequencies), an audio recording provides a measurable snapshot of the number of species making calls in a patch of habitat.
“You can see a healthy ecosystem and it’s very different to a poorer one,” Fuller says. “And we can calculate, from that, an acoustic diversity index, that just tells us, say, that this site has more species than that does.”
This kind of information could prove invaluable in monitoring the restoration of degraded habitat, for example.
“We can use acoustics as almost a fingerprint of the environment,” Fuller says.
