USA: New eyes on wildfires
By Jon Kelvey
Shortly before dawn on 8 November 2018, a fire erupted in the dry mountain grass in Butte County, Calif. By sunset, driven by dry winds, the fire had consumed the town of Paradise, its 20,000 citizens struggling to flee before the flames as official evacuation calls were outpaced by the speed of the conflagration. The Camp Fire would ultimately leave more than 80 people dead and $16.5 billion in damage, the most destructive wildfire in California history.
With climate change expected to increase the risk of such fires, better advanced warning systems could make the difference between life and death. That’s the idea behind the work of James MacKinnon, a software engineer at NASA Goddard Space Flight Center in Greenbelt, Md.
Inspired by the Camp Fire and a similar fire that sprung upon Fort McMurray, Alberta, Canada, in May 2016, MacKinnon wondered if satellite-based remote sensing could help.
Satellites that can sense fires are nothing new, but there are limitations to current options. Carrying the Moderate Resolution Imaging Spectroradiometer (MODIS), NASA’s Earth-imaging satellites Terra and Aqua can detect wildfires, MacKinnon says. But with two satellites flying in polar orbit, it can be up to 48 hours between visits to any spot on the ground, and the data must then be transmitted and analyzed—too long a delay, MacKinnon says, for reliable fire management.
Testing solutions for “a problem we can work around”
If he could do the data analysis on board the satellite, MacKinnon realized, he could cut out the middleman, beaming down only the practical information about fires.
Researchers trained their neural network on 1 million examples split 50-50 between “fire” and “nonfire” until it detected known fires with 99.96% accuracy.
“But there is a problem with this, a big problem, and that’s the type of computers that are on satellites are extremely weak,” he says. “But it is a problem we can work around.”
MacKinnon first developed a relatively simple neural network that can run on the low-powered processors found on small satellites. Then, he trained it on years of MODIS data—1 million examples split 50-50 between “fire” and “nonfire”—until it detected known fires in that data with 99.96% accuracy.
The project will soon get a test run on the International Space Station.
“We have an instrument called the Compact Thermal Imager that we just sent up there with the most recent SpaceX resupply” in December, MacKinnon says, and the plan is to upload his algorithm to the instrument to test operability by late 2019.
But Terra and Aqua are only two satellites. MacKinnon believes the optimal solution would be a large fleet of CubeSats, flying close to Earth to provide global coverage and even do so on a budget.
“You put up something like Terra or Aqua and it costs billions of dollars, but if you put up a CubeSat, it might cost you a few hundred thousand dollars,” he says.
CubeSats, however, introduce another constraint: The MODIS instrument draws something on the order of 100 watts, according to MacKinnon, far more power than an individual CubeSat can provide.
That’s where MacKinnon’s collaborator Charles Ichoku comes in. A fire scientist, Ichoku had worked with MacKinnon at NASA before taking up an Earth and environmental professorship at Howard University in September, but he’s still working on a compact thermal imaging device capable of flying on a CubeSat.
“It would be great to have the opportunity to continue the work we started, in order to synergistically build up the necessary hardware and software,” Ichoku says. “Especially in the face of increasing threats from wildfires at urban/wildland interfaces due to changing climate.”
Fully on board with MacKinnon’s vision of detecting fires earlier, Ichoku wants to go further, developing a high-resolution instrument that can map intensity within fires, soot, and carbon emissions.
To do so, Ichoku’s instrument must have far greater resolution than MODIS, which can resolve details down to about only 1 square kilometer. Ichoku’s initial calculations show his instrument will resolve down to 50 square meters from a height of 500 kilometers, a resolution that can “provide maps of the configuration of the fire” (which parts are burning most intensely), so “it will be easier to direct the fire intervention crews to go to places that are more urgent to address.”
Practical information for firefighters
That type of detail would be really important, according to Curtis Brown, safety, EMS and research and development chief with the California Department of Forestry and Fire Protection (Cal Fire), because in the Golden State at least, initial detection isn’t the largest problem anymore.
“I think we kind of know when the fire starts,” Brown says. “Everyone has a cell phone and is on high alert after recent fires, but how hot the fire is, where it is going, that’s really needed.”
California is looking for just such a solution, according to Brown. On 8 January, Governor Gavin Newsom signed Executive Order 4-19, a general reform of state procurement policies that also highlights Cal Fire and the need “to identify innovative solutions to the State’s wildfire crisis, with a goal of working-solutions deployments in 2019 and final awarded deployments by Spring 2020.”
The ultimate deployment of a dedicated CubeSat fleet would likely take longer than that—Ichoku said his instrument is probably 4 to 5 years away—but California’s search for answers is just the kind of interest that could ensure MacKinnon’s project is green-lighted for further funding and development.
“NASA is always open to partnership,” MacKinnon wrote in an email. “Having buy-in from the government of California would definitely go a long way in getting it approved.”