AI firefighters: how tech is revolutionising bushfire detection and saving Australia

An Australian-founded start-up has developed a new system that can spot a plume of smoke from 20km away and stop bush fires exploding into ‘mega’ infernos.

Pano AI’s technology aims to prevent ‘mega fires’ from igniting.

Artificial intelligence has become the latest tool in a firefighter’s arsenal, with start-ups raising tens of millions of dollars from investors to use the technology to spot blazes — even at night — within minutes.

Pano AI is the latest company to secure capital to accelerate uptake of its fire detection platform, which can spot a plume of smoke from 20km away, completing a $US44m ($68m) raising.

It dwarfs a $2m raise Tasmanian-based Fire Foresight is in the early stages of planning, but underscores demand for such technology as fires become more frequent, intense and outside traditional seasons like Los Angeles’s wildfires earlier this year.

Arvind Satyam — originally a North Sydneysider who moved to Silicon Valley in 2006 — co-founded Pano AI, soon after Australia’s devastating bushfires in 2020. At the time, The New York Times featured a picture of a kangaroo hopping across a landscape engulfed in flames on its front page with the headline ‘nowhere else to run’. The Times in London published a similar front page.

The image taken by Sydney based photographer Matthew Abbott while covering the NSW bushfires for the New York Times, has made headlines around the world, and prompted Arvind Satyam to create Pano AI.

“That is something that I looked at and it emotionally moved me, and I’m thinking ‘how can you do something different here’,” Mr Satyam said.

Five years later, Pano AI has booked more than $US100m in contracts to deploy its system, which has been launched across Australia in five states and also in the US in 11 states.

In its latest raise — which brings its total funding to $US89m — it has secured backing from Giant Investors, Liberty Mutual Strategic Ventures and Tokio Marine Future Fund. Other investors include Congruent Ventures, Initialized Capital and Salesforce Ventures.

Pano’s AI model has been trained on more than two billion images and can detect smoke across a variety of different landscapes and environmental conditions, including haze, fog and even snow.

Pano AI co-founders Sonia Kastner and Arvind Satyam.

“Detecting smoke on a natural landscape, which varies, turns out to be a really hard problem. What you see in northern parts of Tasmania versus what we may be seeing in the Riverina region or the US … we’re seeing dramatic changes,” Mr Satyam said.

“In Colorado, in our first season, the foliage started turning from green to red to orange and brown, and the AI wasn’t able to distinguish some of those colours. We had to train it.

“We have to detect smoke in all those environments, and we also need to detect smoke at night.”

At night, the system uses heat signatures, but even then it’s not fail-proof.

“You need to understand is that a fire or is that fireworks or a flare. We’ve got incredibly high resolution imagery – even at night-time – and we can zoom in to see what’s happening. That provides a huge advantage.

“The AI has just gotten so good but we don’t rely on the artificial intelligence because the last thing we want to do is send notifications to first responders that could be false positives that we can quickly check off.”

This is where Mr Satyam says Pano AI is different from other fire detection providers because it also has a team of human analysts to verify what the AI system detects.

“For instance, if that AI has made a detection and our analysts in dedicated regions see that it’s in a farm and it’s a tractor going down a hill, we don’t notify that.

“Making sure it’s actually smoke and not dust is really important.”

In Australia, Pano protects more than 2 million hectares of forestry plantations and agricultural land as well as surrounding communities, state forests and national parks.

Its system uses two six megapixel cameras with 30x optical zoom lenses rotating constantly to get a “full 360-degree picture of what’s going on”. Mr Satyam said this differed to other providers who mainly use fixed cameras that can cost firefighting agencies precious time.

“A camera might be looking west and there is a fire in the east, so someone has got to manually move it, so you don’t see the early moments of a fire.

“And we need eyes on that incident as quickly as we can get it. Early detection followed by rapid initial attack is the absolute best way to stop a mega fire. Just like any other analogy we could use — whether it’s cancers — by being able to better detect it early, you’re able to better allocate resources.”

Pano AI's system installed at Oak Grove Butte in Ripplebrook, Oregon.

While the six megapixel resolution might seem low by consumer camera standards — even compared to an iPhone, which has a 48MP main camera sensor — Mr Satyam said it was enough to provide sharp images and detect smoke from 20 kilometres away.

Crucially, he said it allowed images to be uploaded quickly in remote areas.

“We needed to make sure that we get the highest amount of spec that allows us to get that imagery in remote environments, and then apple AI in real time.”

Mr Satyam said utilities were also becoming big customers, along with firefighting agencies. In 2019, California’s Pacific Gas and Electric filed for bankruptcy, primarily from liabilities from wildfires in 2017 and 2018.

“We’re also providing intel to utilities, forestry, private landowners, so they’re able to know where that fire is relative to the asset. So if you have, if you’re a utility that’s running the grid, and you know that that fire is 100 meters from energised transmission lines, you can de-energise that part of the grid, and you can stop it from becoming a much larger incident.”

Mr Satyam declined to disclose the company’s financials, including its profitability. “We’re really focused on making sure that we bring private capital to accelerate investment on this. We’ve also booked over $US100m, in contracts — that’s over $150m Aussie — in less than five years,” he said.

“So for us, the goal is to really scale the tech and continue to be a company that’s investable, so we can really go build a lot of the capability that’s needed.”