A thin pipe and a shrill note harmonize with the underlying buzz of insects in the background. This symphony represents the forest’s unique sound and is observed by researchers to measure its biodiversity.

New research is using artificial intelligence to monitor animal populations in recovering habitats, such as the recording from a forest in Ecuador.

To assess reforestation, scientists can utilize tools like satellites and lidar to survey extensive areas of land.

However, accurately assessing the speed and volume at which wildlife is repopulating an area can prove to be a complex task. This may involve the expertise of a professional to analyze audio recordings and identify animal vocalizations.

Jorg Muller, a field ornithologist and professor at the University of Wurzburg Biocenter, pondered the possibility of an alternative approach.

In an interview with AFP, he stated that there is a need for improved quantification methods to address the vast diversity in the tropics and enhance conservation efforts.

He delved into the field of bioacoustics, which utilizes sound to gain a deeper understanding of animal behavior and environments.

This research tool has been used for a long time, but now it is being combined with computer learning in order to efficiently handle large amounts of data.

Muller and his colleagues collected sound recordings at various locations within Ecuador’s Choco region. These sites included recently deserted cocoa farms, fields that were once used for agriculture but are now being restored, and mature forests.

Initially, specialists were tasked with listening to the recordings and identifying birds, mammals, and amphibians.

Next, the researchers conducted a study using an acoustic index analysis to determine the level of biodiversity in a given area. This analysis takes into account various factors from the surrounding soundscape, such as the intensity and frequency of noises.

Ultimately, they processed a fortnight’s worth of audio recordings using a computer program aided by artificial intelligence, which had been taught to identify 75 distinct bird vocalizations.

More recordings needed

Was the program able to accurately identify the relative biodiversity of each location by consistently recognizing the trained calls?

To verify this, the group utilized two reference points: one from the professionals who listened to the audio recordings, and another from insect samples taken at each site, which act as a substitute for measuring biodiversity.

According to the study, the AI model was limited in its ability to recognize bird calls, only able to identify 25% of the calls that experts could. However, it still accurately measured levels of biodiversity in different areas using the available library of sounds.

According to the study published in the journal Nature Communications on Tuesday, analyzing soundscapes can effectively track the restoration of animal populations in highly diverse tropical forests.

The measurement of soundscape variety can be done efficiently and reliably over a range of environments, from actively farmed land to regenerating and mature forests,” the statement said.

Some limitations remain, such as a lack of animal vocalizations available for AI training.

This method is limited to detecting species that make known their existence.

Muller, speaking to AFP, stated that there is no data available on plants or quiet creatures. However, birds and amphibians are highly attuned to the health of their environment and can serve as effective indicators.

He thinks that the tool has the potential to be more valuable as there is currently a trend towards the implementation of “biodiversity credits,” which involves making money from conserving animals in their natural environment.

The study stated that being able to accurately measure biodiversity, instead of using indirect measures like tree growth, supports and enables outside evaluation of conservation efforts and encourages transparency.