Industries / Construction and property / Commercial and residential construction / Fencing
Farm and wildlife fencing
Fences that hold or keep out animals across open land: stock and electric fences, long barrier fences against pests, fences that let wildlife pass, and fences replaced by collars.
7
Milestones achieved, from 1901 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1901
Rabbits outrun the fence
Rabbits spread west across Australia and had reached Eucla, near Western Australia's border on the south coast, by 1896. Western Australia began its Number 1 Rabbit Proof Fence in 1901, 1,834 km from the south coast to the northwest coast, but by 1902 rabbits had already been found west of the line. A Number 2 fence, 1,166 km long, followed in 1905.
LessonA barrier must be closed ahead of the animals it is meant to stop; the State Library of Western Australia calls the fence a "unique, if inadequate, response".
Credit
Government of Western Australia
1938
Fences that shock
An electric fence energizer is the box that sends short pulses of high voltage down a wire. Bill Gallagher and Herbert Christie, working separately, built energizers in New Zealand about 1938, by the account of Te Ara, the encyclopedia of New Zealand, which notes the electric fence itself was not a New Zealand invention. Gallagher began by electrifying the family car to stop a horse rubbing on it. Light electric wire let farmers fence off part of a paddock and ration grass; trials of permanent electric fencing at Massey Agricultural College in 1958 brought "a radical change in fencing methods", and a 1961 change in New Zealand law let energizers run off mains power, strong enough to burn off plants touching the wire.
Credit
Bill Gallagher (Gallagher Group), Herbert Christie, and William Riddet at Massey
1946
A fence paid by rates
South Australia's Dog Fence Act 1946 set up the Dog Fence, now 2,110 km from the Great Australian Bight to the New South Wales border, to protect sheep from wild dogs and dingoes. Under the Act the Dog Fence Board charges rates on every landholding larger than 10 km² in the rateable area, whatever the land is used for, to pay for upkeep.
Credit
Parliament of South Australia; the Dog Fence Board
2005
Wildlife deaths on fences counted
Justin Harrington, a graduate student of Michael Conover at Utah State University, checked fences along roads in Colorado and Utah from June 2004 to June 2005. He found about 0.25 pronghorn, mule deer and elk caught dead in fences per km each year, and carcasses lying within 10 m of fences outnumbered those caught in them about two to one. Young animals were 8 times as likely to die as adults, most were caught between the top two wires, and woven wire topped by a single barbed strand was the deadliest fence.
Credit
Justin L. Harrington and Michael R. Conover, Utah State University (thesis 2005; Wildlife Society Bulletin, 2006)
2018
Predator-proof havens surveyed
Sarah Legge, John Woinarski and 25 co-authors counted Australia's havens free of foxes and cats: 17 mainland fenced areas covering 346 km², the largest 123 km², and 101 islands. The fenced havens held 27 threatened mammal taxa, and 43 of their 49 populations had been moved there by people.
LessonKatherine Moseby and John Read, who co-founded four havens, wrote in 2023 that populations inside rose by an average of 680% over 17 years while those outside fell 63%, but animals raised behind fences are often killed by cats and foxes when released, and havens that lost funding and fell into disrepair let predators back in, as cats did to fenced Shark Bay bandicoots.
Credit
Legge and colleagues, Wildlife Research (2018)
2022
Collars test fence designs
Jesse DeVoe, Kelly Proffitt and Joshua Millspaugh tracked 265 female pronghorn with hourly GPS collars in six Montana herds through 5,581 encounters with 979 km of mapped fence. Woven wire fences sharply cut crossings. Strand fences were fairly easy to cross at first, whether their lowest wire sat below or above the 41 cm that guidelines recommend; animals that balked took about 14 hours on average to get across. The authors advised removing woven wire or replacing it with strand fence.
Credit
DeVoe, Proffitt and Millspaugh (University of Montana and Montana Fish, Wildlife and Parks), Ecosphere (2022)
2025
Virtual fences legal in Australia
A virtual fence is a GPS collar that warns an animal with a sound, then a mild electric pulse, as it nears a boundary set in an app. In December 2025 Victoria and New South Wales made virtual fencing legal for cattle, and South Australia amended its Animal Welfare Regulations 2012 the same month, joining Queensland, the Northern Territory, Tasmania and Western Australia. Victoria requires approved systems, training from the maker, regular collar checks and a physical boundary fence; collars on other livestock need a scientific licence.
Credit
the state governments; Gallagher's eShepherd system grew from animal behavior research by CSIRO, Australia's national science agency
Watch for
Dog Fence rebuild runs late
In 2018 more than two-thirds of South Australia's Dog Fence, 1,600 km, was over 100 years old and damaged by kangaroos, emus, feral camels, wild dogs, weather and drifting sand. The Australian and South Australian governments and the livestock industry put A$29 million into rebuilding it, with completion first planned for June 2024. The Dog Fence Board says almost 1,140 km had been rebuilt by 30 April 2026 and another 340 km was under way, and that rain, closures of the Woomera defence range and contractor shortages made a June 2026 finish unlikely, so crews are working toward December 2026. In June 2025 the board and the NSW Government also agreed to build up to 290 km of new fence on the border between the two states.
Achieved whenthe department or an outside auditor reports all 1,600 km finished and in service, with wild dog incursions counted before and after.
Claimed by
Dog Fence Board, May 2026
Open
01
Fence crossings scored by AI
A model that watches trail-camera video at fences and labels each event by species and outcome: jumped over, crawled under, went through, turned back or was caught. Release it with an open set of labeled clips, so studies of wildlife-friendly fence designs no longer need people to sort footage by hand.
The briefCredit · unclaimed02
Measured fence savings from collars
A farm-by-farm record of the physical fence actually removed, left unbuilt or still maintained after a farm adopts virtual fencing, with what was spent on fencing before and after, so the claim that collars replace fences can be checked.
The briefCredit · unclaimed
Open 01 · the brief
Fence crossings scored by AI
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A model that watches trail-camera video at fences and labels each event by species and outcome: jumped over, crawled under, went through, turned back or was caught. Release it with an open set of labeled clips, so studies of wildlife-friendly fence designs no longer need people to sort footage by hand.
Why it is still open
Fence-crossing studies still score images by eye. Emily Burkholder and colleagues (Wildlife Society Bulletin, 2018) used remote cameras from 2010 to 2016 to study how deer crossed fences before and after modification. A 2015 program by Kaitlyn Visser, Michael Janzen and colleagues at The King's University, Alberta, used background subtraction rather than learning on about a million images, cut the images left for people to a third, and identified crossings correctly 60% to 92% of the time. Web and GitHub searches in November 2026 found no deep-learning crossing classifier and no labeled fence-crossing set.
Built with what exists
MegaDetector (open, MIT license) to find animals in frames; Google's SpeciesNet camera-trap models (cameratrapai, Apache 2.0) to name species; a video classifier fine-tuned on labeled clips; CVAT or Label Studio for labeling; trail cameras with a video mode.
First step
Ask a wildlife agency or a researcher who has run a fence-camera study for existing footage, and label 200 events.
Done looks like
At least 2,000 labeled crossing events from at least 10 fence sites, released openly; on sites held out of training, the model's outcome labels agree with two human labelers at least 90% of the time; and the hours needed to process one season's footage reported against hand sorting.
Open 02 · the brief
Measured fence savings from collars
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A farm-by-farm record of the physical fence actually removed, left unbuilt or still maintained after a farm adopts virtual fencing, with what was spent on fencing before and after, so the claim that collars replace fences can be checked.
Why it is still open
The University of Arizona's March 2025 analysis (Dari Duval and colleagues) modeled avoided fence at an assumed $20,000 per mile and found that, under its baseline assumptions, none of the systems examined would pay for themselves over the hardware's life; it is built on hypothetical ranches. Victoria's rules keep a physical boundary fence. Virginia Cooperative Extension began a year-long study in 2026 of whether virtual fences cut maintenance costs. No survey measuring fence removed or avoided on working farms was found.
Built with what exists
A one-page interview form; farmers' own fence sketches redrawn in QGIS or Google Earth; receipts for fencing materials and labor; a spreadsheet; Zenodo to publish the anonymized records.
First step
Reach five farms that use virtual fences through an extension office, and agree on the form with them.
Done looks like
20 farms in at least 2 countries, each with kilometers of fence before adoption and 12 months after, kilometers removed, kilometers planned but not built, and fencing spending, set beside Arizona's modeled break-even of roughly 2 to 9 miles of fence avoided in the first year.