Shoebox-sized saviors: New AI CubeSats could detect forest fires 500 times faster

Improving fire detection times with AI CubeSats

The new method could help tackle forest fires before they generate large amounts of heat, making them easier for ground crews to control.

The project, funded by the SmartSat CRC and led by the University of South Australia (UniSA), has overcome some hurdles holding similar methods back. It uses a cutting-edge artificial intelligence (AI) algorithm to process and compress vast hyperspectral imagery aboard CubeSat satellites orbiting Earth in real time. 

The result is a system that uses AI to quickly scan images of vast swathes of our planet, quickly detecting early signs of forest fires from outer space.

“It is not only the algorithm itself, but the combination of a CubeSat constellation, AI onboard imagery processing, and smart data downlinking,” Dr Stefan Peters explained to Interesting Engineering (IE) in an interview over email.

Typically, up to six hours can pass between a satellite’s capture of an image of wildfires and the information being passed on to ground teams. 

Peters told IE that the new method uses CubeSats—small satellites roughly the size of a shoebox—with complex sensors and onboard processing capabilities that greatly speed up the process.

Though these CubeSats provide great benefits, they “have limited energy and limited storage,” Peters continued. “Therefore, efficient onboard processing is important. Smart onboard processing and rapid downlinking of detected fire locations to Earth will help to get fires under control earlier before they expand into devastating disasters. Ultimately, [it will] save lives.”

Detecting fire 500 times faster

Peters and his team are developing their system for South Australia’s first cube satellite, called Kanyini. The South Australian Government has provided $6.5 million in funding for the Kanyini CubeSat project. Its goal is to help the region’s space sector grow while supporting the nation’s wider space strategy.

The team’s system detects the very earliest signs of forest fires by continuously capturing hyperspectral images of Australia. The system’s AI model quickly analyzes the images. It rapidly highlights any smoke, allowing for a fast, real-time detection method. Traditionally, Earth observation satellites have not been able to analyze complex images of Earth in real-time.

The researchers claim that their onboard AI model reduced the volume of data downlinked to 16 percent of its original size. This is when compared to on-ground-based processing of hyperspectral satellite imagery.

Their model essentially allows the AI to choose which images to beam to Earth, allowing for more efficient operations. The team also said this resulted in a 69 percent reduction in energy consumption. Most importantly, it detected fire smoke 500 times faster than traditional on-ground processing.

The team tested and demonstrated their AI model using satellite imagery of recent Australian bushfires. Using machine learning, they had previously trained the AI to detect smoke in images. The team published their findings in a new paper in the IEEE Journal.

From cartography to satellite fire detection

Peters started his academic career as a research associate at the Department of Cartography of the Technical University of Munich. After a stint as a Senior Lecturer in GIS and Cartography at the University of Technology Malaysia, he moved to Australia in 2016. 

There, he started working as a lecturer at the University of South Australia, where he worked in geospatial and environmental science ever since.

Since Peters moved to Australia, the country has experienced some of the most devastating wildfires in recorded history.

During the 2019-2020 bushfires, up to 19 million hectares were burned, and almost three billion animals were impacted, according to the World Wildlife Fund (WWF). Given his expertise in environmental science, it might seem natural that Peters would gravitate toward wildfire detection. Interestingly, though, his passion for cartography first led him down his current path.

“It all started with my passion for maps and map analytics to understand the dynamics of our planet,” Peters explained. “This brought me to Remote Sensing and Earth Observation – working with satellite imagery to detect fires, landslides, floods, and other information.”

“A few years back, I was involved in a forestry-sponsored research review on wildfire detection and suppression,” he continued. “Our review identified the great potential of CubeSat constellations and AI onboard for fire detection. This is where our research journey with SmartSat-CRC began.”

The benefit of CubeSats

The cost-effective nature of CubeSats, as well as the improved accessibility to space thanks to companies like SpaceX, has helped to open up a whole host of new possibilities.

One of the great benefits of using CubeSats is that they are far cheaper than more traditional satellites. This is largely down to their size and weight. A CubeSat can be as small as a shoebox or a toaster and weigh between three and 30 Kg.

“CubeSats are [far] cheaper and faster to develop and launch than larger LEO Earth Observation missions,” Peters said. “The real challenges related to CubeSat-based early disaster applications are onboard storage, onboard processing, and efficient downlink of captured imagery.”

Another benefit CubeSats provides is that the revisiting time can be vastly improved if launched as a constellation. “In other words, instead of a six-hour revisiting time, a fire can be revisited by CubeSats every hour, or even less,” Peters explained, adding that it would take 20-plus CubeSats to achieve those times.

The economic implications of faster fire detection

While CubeSats are cheaper to launch and operate, faster fire detection could have massive economic implications. 

The 2019-2020 Australia bushfires were the costliest natural disaster in the country’s history. Scientists estimated it may have cost roughly A$88 billion in property damage and economic losses.

Peters said his team had made some internal calculations about the economic benefits of their fast fire detection method. However, they are not sharing the data at this time.

While the primary focus is on forest fire detection, the AI technology could also be applied to rapidly detect other disasters such as floods, earthquake damage, and landslides.

The team works with several companies, including Esper, to build its system. Esper is an Australian startup that enables clients to capture hyperspectral imagery from space at a fraction of the cost. In March, the company launched its first demonstration satellite aboard SpaceX’s Transporter-10 mission.

Kanyini, meanwhile, is set to fly aboard SpaceX’s Transporter-11 mission, which was supposed to launch on July 11. Unfortunately, that mission has been delayed due to a rare engine failure during a recent SpaceX Falcon 9 mission.

Once it does take to the skies, though, Kanyini could usher in a new era of rapid fire detection, enabling ground teams to tackle potentially disastrous fires before they escalate, ultimately saving lives and protecting vast areas of forest.