Staff writers
Wildfire detection is set to reach new heights – literally – thanks to a special CubeSat in low Earth orbit.
This shoebox-sized satellite, Kanyini, was developed at the University of South Australia as part of the federally-funded SmartSat Cooperative Research Centre.
The SA Government is the key backer of the Kanyini mission, with space-based data collection for use by environmental agencies to monitor water quality, or for emergency services to mitigate fire activity as its priority.
The mission is due for launch later this year on a SpaceX Transporter-11.
Its bushfire monitoring functions will combine data collection from a hyperspectral imager that captures reflected light wavelengths from the Earth to create detailed maps of Australia’s surface with a ‘lightweight’ AI system.
This AI system takes what the imager sees and analyses it to detect possible fire sites.
According to data published in the IEEE Journal of Selected Topics in Applied Earth and Remote Sensing, the on-board AI model reduced data beamed back to Earth by 16% and used almost 70% less energy to perform the analysis while detecting fire smoke 500 times faster than conventional processing at ground facilities.
That’s because the AI model on Kanyini can distinguish between fire smoke and cloud cover.
“Smoke is usually the first thing you can see from space before the fire gets hot and big enough for sensors to identify it, so early detection is crucial,” says Stefan Peters, a geospatial scientist at UniSA.
“For most sensor systems, only a fraction of the data collected contains critical information related to the purpose of a mission.
“Because the data can’t be processed on board large satellites, all of it is downlinked to the ground where it is analysed, taking up a lot of space and energy. We have overcome this by training the model to differentiate smoke from cloud, which makes it much faster and more efficient.”
A test simulation of the AI model was used to ‘detect’ a previous fire event in the Coorong region of South Australia. According to the research group, this process took place in 14 minutes.
