AI enters the brumby battle

Gone Feral is a special project created by senior students from the Australian Centre for Advanced Journalism. This article about brumbies is written by Harshita Roy.

As arguments continue about ways to measure the number of brumbies in the Kosciusko National Park, there’s been a call for scientists on both sides of the debate to collaborate more.

The management and culling of feral horses throughout high country national parks has driven fierce divisions for decades, with tensions escalating in recent years into what’s widely described as a culture war conflict between brumby defenders and people concerned about their documented impacts on the fragile alpine environment and threatened native species.

The latest legal skirmish played out in the New South Wales Supreme Court this week when a judge dismissed an attempt to stop the aerial shooting of brumbies in Kosciuszko National Park, rejecting arguments brought by pro-brumby advocates that the NSW environment minister did not have adequate information when she authorised the method in 2023.

Both sides have enlisted scientific arguments to back their cases over the years, including on the central question of the population of wild horses. Earlier this year, an aerial survey commissioned by pro-brumby advocates enlisting artificial intelligence tools was released, which they claimed provided evidence that officials have significantly overestimated feral horse numbers in Kosciuszko National Park.

Herd of brumbies in snow
Brumbies in Kosciuszko National Park. Credit: Brook Mitchell / Stringer / Getty Images

A survey commissioned by pro-brumby campaigners which enlists artificial intelligence tools claims to show that officials have significantly overestimated feral horse numbers in Kosciuszko National Park.

The survey found just 569 horses in a 21,200-hectare section in the north of the park – about three per cent of its total area. According to survey organiser, Rocky Harvey, the  count would likely be over 5,000, based on his estimates applying data from the NSW National Parks and Wildlife Service 2023 survey.

But retired senior ACT Government ecologist Dr Don Fletcher Fletcher and Dr David Berman of the University of Southern Queensland, say the  “independent horse count” aerial survey is seriously flawed.

They say, for example, it did not include a formula to estimate the number of horses likely to be present but not seen or counted. Nor did it include an estimate of the proportion of the horse population in the survey.

Berman added that he believed the AI-assisted methodology had promising potential, urging further testing and development to better track horse numbers in the future. He recommended incorporating thermal imaging. By trialling it in a contained area with a known horse population, it could provide a formula to estimate the number of horses the model doesn’t detect.

Conducted by Airborne Logic, a remote sensing agriculture and environment consultancy based in Adelaide, the survey used high-resolution aerial images taken from a low-flying aircraft to stitch together an orthophoto map, smoothing out terrain and camera distortions. An object detection tool, based on machine learning, was then trained to scan the image and identify horses.

The private survey was commissioned and crowdfunded as “an independent count of wild horses in the Kosciusko National Park” by Harvey – a builder and butcher from Jindabyne – and Macquarie University biostatistician Claire Galea. On their Gofundme page, they argue that better, more accurate techniques are required to resolve widespread “dispute on actual wild horse numbers in the park”. 

Galea declined to respond to questions, other than to say she was no longer involved in anything to do with the wild horses due to “threats and harassment”. 

While the survey report includes a peer review by an independent AI expert, this is confined to the technology and its application and doesn’t explore the ecological component.

“The plan to use this technology and the interpretation of its result were done by cull protestors. None of that was peer-reviewed,” says Fletcher, who is a threatened species management expert who remains actively involved in the NSW Senate inquiry on feral horses, including making several written submissions.  In July, he published a detailed critique of the aerial survey online.

The AI method was developed to challenge the established, widely-used distance sampling methodology used by the NSW Government to estimate wild horse numbers. This survey, conducted by ecologist Dr Stuart Cairns of the University of New England, collects observations from a pair of specialist observers flying at low altitude over the area, seated each side of the aircraft and recording sightings of horses and related data, such as the distance of the animal or cluster from the observer, habitat cover and cloud cover.  No horses are photographed or described.

These are then analysed “using advanced software to reliably estimate the wild horse population in the entire survey area”, says the NSW National Parks and Wildlife Service, and peer-reviewed by independent experts from CSIRO and the Queensland Department of Agriculture and Fisheries.

The distance sampling method applies a derived mathematical formula to estimate the number of horses that are present but not seen, which the Airborne Logic method lacks, says Berman, who also works for Queensland Forestry with contractor HQPlantations and The Brumby Project to manage and rehome feral horses.

Co-organiser Rocky Harvey argues that the number of horses not seen are “absurd” and vastly overestimate the true numbers.

“What we set out to do was to demonstrate to government that there are better ways to do it,” says Harvey, echoing the view of many in the pro-brumby community that the parks service is not using the best available science to count wildlife.

Last year, Harvey dumped a dead horse’s head outside the national parks office in Jindabyne to bring attention to the mass culling of feral horses and their improper carcass disposal.

Berman urges more cooperation to bridge the trust deficit between the two sides. “National Parks, they get sick of being attacked and do not provide enough information, and the other side then gets upset and attacks more.”

He suggests engaging the concerned community groups in research. “You have got to gain trust with the community … You’ve got to get them as part of the team working on this problem.

 “So Rocky Harvey needs to be in there working with the rangers to solve this problem.”

Recent data released by the NSW government has revealed that 8,944 feral horses have been removed from Kosciuszko since 2021. Of these 5,963 were aerially culled.

The NSW government plans to bring down the feral horse population in Kosciuszko from 17,393 estimated in November 2023 to 3,000 by 2027, using various lethal and non-lethal control methods.

This story is co-published with The Citizen, a publication of the Centre for Advancing Journalism at the University of Melbourne.

Sign up to our weekly newsletter

Please login to favourite this article.