The fact that the eastern woodpecker has not been seen or heard in south-east Queensland since its home in the Gondwana rainforest was destroyed in the 2019/20 summer blackfires is in some ways unsurprising.
First, it is believed that there were fewer than 40 individual birds in its northern population.
Add to this the fact that he is a “nondescript brown bird”, shy and secretive, which flies along the ground between the bushes doing the coolest thing he can to not be seen.
This makes calling the rough bird the most effective way to track it down.
Usually, this would involve a person entering the woods and playing a recording of a call in an attempt to coax a wild bird into a response.
“But you have to be in the right place at the right time and the bird has to want to respond,” says Susan Fuller of Queensland University of Technology.
So the Queensland University of Technology researchers teamed up with BirdLife Australia and Healthy Land and Water to place five acoustic monitors in the northern range of the rough-hewn bird mid-last year, returning only to replace the batteries and weeks later for the recordings.
The results were encouraging, confirming the presence of an elusive bird feared to be lost in south-east Queensland.
The potential for this type of monitoring, called passive acoustic monitoring, has intrigued scientists for more than a decade. Fuller says that recent advances in computer science and artificial intelligence have helped make this possibility a reality.
“We’ve always come back to the same stumbling block of someone having to sit down and take recordings minute by minute, manually select calls,” says the assistant professor at the QUT Center for the Environment.
For a large conservation project, this can amount to terabytes of data—a trove that is impossible for a human to review exhaustively.
In this case, Queensland University of Technology computer scientist Dr Lance de Vine has developed an AI model that can be trained to recognize bristle bird calls between hours upon hours of field recordings.
“Without AI we can’t do it,” says Fuller. “This is a game changer for us.”
The breakthrough was based on ecological understanding and human experience—threatened species BirdLife project officer and QUT PhD candidate Callan Alexander was the first to select the raspy bird call from the recordings.
Using Alexander’s trained ear, de Vine was able to gradually train the AI program to accurately identify an endangered bird call from other, similar sounds, and then let it lose it over the rest of the recordings, of which he detected 350 calls of eastern woodpeckers on the two. month period.
After that initial breakthrough, the researchers now have 20 screens on a larger scale of habitats.
Fuller says AI offers other great potential for conservation, including identifying individual animal calls from recordings, not just species.
The scientist says Soundscapes can provide unique insights into the overall health of an ecosystem.
Viewed as a spectrograph (a visual representation of a spectrum of frequencies), the acoustic 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 from a poorer ecosystem,” says Fuller. “And we can calculate, from that, the acoustic diversity index, which just tells us, say, this site has more species than that.”
This kind of information can prove invaluable in monitoring the restoration of degraded habitats, for example.
“We can use acoustics almost as a fingerprint of the environment,” says Fuller.