A wide whistle and a high-pitched trill against the low hum of an insect: this is the music of the forest that all scientists have ears for to get an idea of ​​biodiversity.

A sound recording from an Ecuadorian forest is part of new research to determine how artificial intelligence (AI) could study animal life in regenerating habitats.

When scientists want to measure forest regeneration, they can look at large areas using tools like satellites. But determining how quickly wildlife is returning to an area is a more difficult challenge, and sometimes requires an expert to sift through sound recordings and distinguish between animal calls.

University of Würzburg professor and ornithologist Jorg Muller wondered if there was another method.

“I saw a gap that we still need to fill, especially in the tropics, and the best methods to measure the enormous diversity,” he explains to AFP.

He focused on bioacoustics, which uses sound to learn about the lives of animals and their habitats.

It is not a recent research tool, but has recently been associated with computer learning to process large amounts of data faster.

Jorg Muller and his team made audio recordings at locations in Ecuador’s Choco region, ranging from recently abandoned cacao plantations and pastures to agricultural lands being restored after exploitation.

First, they asked experts to listen to the recordings and select birds, mammals and amphibians.

They then performed an acoustic index analysis, which provides a dimension of biodiversity based on measures such as noise volume and frequency.

Finally, they made two weeks of recordings using an AI-assisted computer program designed to distinguish the calls of 75 birds.

More entries needed

The program was able to consistently recognize bird calls, but was it able to correctly identify the biodiversity of each site?

To test this, the team used two basic data points: one from experts who listened to audio recordings, and another based on samples of sounds made by insects from each location.

Although the pool of sounds available to train the AI ​​model meant it could only recognize a quarter of the bird calls experts could identify, the method was able to correctly estimate the level of biodiversity at each site. confirms a study published Tuesday in the journal Nature Communications.

“Our results show that soundscape analysis is a powerful tool for monitoring the recovery of wildlife communities in highly diverse tropical forests.”

“The diversity of the soundscape can be measured effectively, economically and sustainably,” both in agricultural areas and in old and regenerating forests, the same source adds.

There are still gaps, notably the lack of animal sounds to train AI models.

And this approach allows us to consider only species that announce their presence.

“Of course there is no information about silent plants or animals. But birds and amphibians are very sensitive to ecological integrity, they are very good proxies,” Muller told AFP.

He believes the tool could become increasingly useful given the current push for “biodiversity credits,” a way to monetize the protection of animals in their natural habitats.

“The ability to directly measure biodiversity, rather than relying on indicators such as tree growth, facilitates and enables external evaluation of conservation efforts and promotes transparency,” the study states.

sah/cn/bds