Listening to the bush: How AI can help NZ rid its wilderness of 'hold‑out' pest possums

How AI is learning to hear the last possums

Possums are among New Zealand's most destructive introduced mammals. They damage native forests, prey on wildlife and remain a major target of the country's Predator Free 2050 program, as reported by both NZ Herald and Phys.org.

But removing most possums from an area through predator control is only part of the challenge. Finding the last few survivors can be much harder. Even a handful of remaining animals can eventually rebuild a population, making it essential that conservation workers have effective tools to detect them.

One promising solution lies in microphones that record sounds overnight, combined with artificial intelligence (AI) software capable of scanning thousands of hours of audio for possum calls.

The problem of false alarms

A new study, published in the journal Acoustics Australia, shows the approach can work well—but only if the AI learns not to mistake other animals for possums. The research, written by Akbar Ghobakhlou and colleagues, was reported by both NZ Herald and Phys.org, with the latter providing the study's DOI.

AI is becoming increasingly useful for analysing wildlife recordings, allowing vast audio datasets to be processed automatically rather than requiring volunteers or researchers to listen to every recording. This is particularly valuable during the final "mop-up" stage of eradication, when only a handful of animals remain.

Many AI models can also run directly on small, battery-powered recording devices in remote forests, avoiding the need to upload huge amounts of audio for processing. But AI models designed to run on low-power devices often generate more false alarms, wrongly attributing the calls of other animals to possums. For conservation teams, a false alarm can mean traveling to remote locations in search of an animal that isn't there.

The 'cross-model confusion mapping' approach

To tackle this problem, the researchers developed a new training approach called "cross-model confusion mapping." Rather than asking the AI directly which sounds were possums, they first used BirdNET—a widely used AI system trained to identify more than 6,000 bird species.

Although BirdNET has never been trained to recognize possums, that proved to be an advantage. Because it could only classify sounds as bird species, every possum call was forced into the bird species it most closely resembled. That revealed which bird species were most likely to be confused with possums.

To human ears, these calls sound different, but AI doesn't "hear" sound the way we do. It analyzes visual representations of sound frequencies over time, known as spectrograms, where the patterns are more alike than they seem to us.

The researchers then added recordings of those birds to their training data as what they call "hard negatives"—examples the AI found difficult to distinguish from possums but needed to learn were not possums. The method is conceptually similar to teaching the model to recognize not only what a possum sounds like, but also which other sounds it might be mistaken for.

The approach was tested on recordings from native New Zealand forests, according to both outlets. Models trained with this method produced far fewer false alarms while maintaining high detection accuracy. The approach also worked on forest recordings containing bird species not encountered during training, suggesting it can generalize to new environments. Notably, the ruru (morepork) was not identified by BirdNET as a species likely to be confused with possums.

Toward broader applications

The researchers note that further field testing is needed before the system can be widely deployed. They also suggest the approach could be adapted to detect other invasive pest species such as stoats and rats, as reported by both outlets.

The findings offer an optimistic view of AI's role in conservation. As described in both articles, the results were encouraging and suggested that such AI systems could become an increasingly valuable tool for conservationists in New Zealand, potentially helping to achieve the Predator Free 2050 vision.

As the research continues, conservation managers will likely watch for field trials that confirm these promising results in real-world settings. The new training technique could become a cornerstone for using low-power devices for pest detection in remote, forested territory.