Computer program quickly detects buildings damaged by forest fires


By McKenzie Prillaman, Correspondent

A new computer program powered by artificial intelligence only takes a few minutes to determine if homes and buildings have been destroyed by wildfires after the smoke clears.

Developed by scientists at Cal Poly in San Luis Obispo and Stanford University, DamageMap assesses destruction by scanning post-fire aerial and satellite images.

Residents of the Santa Cruz Mountains who were evacuated in the CZU Lightning complex fire in the summer of 2020 say such a program would have relieved them of a lot of stress and angst.

“I spent days thinking, ‘My house might be on fire right now,'” said Lisa Smith Beasley, a Boulder Creek resident who was ordered to leave her home during the devastating event.

Beasley’s house survived the flames. But she and other residents of the Greater Bay Area often waited weeks to hear from authorities if their homes were successful.

Andrew Fricker, a space ecologist at Cal Poly who co-developed DamageMap, said the program is expected to significantly reduce this waiting period. Once fully developed, it would be available free of charge to the public and emergency responders.

“There are so many people in California who are affected by every fire season,” Fricker said. “And it will only get worse.”

Fricker and his colleagues at Cal Poly and Stanford published their peer-reviewed work on the program in the November issue of the International Journal of Disaster Risk Reduction.

Computer programs that detect damage from natural disasters from aerial and satellite photos have been in development for a quarter of a century. But most of them require before / after photos to be taken with similar angles, lighting, and photo quality, an imperfect system that requires an expensive and continually updated catalog of images.

To determine which structures have burned down, DamageMap relies solely on post-fire images and a digital database showing the location of houses and buildings.

Over the past four decades, the number of acres and homes burned in the West has increased dramatically, fueled in part by climate change.

California’s deadliest and most destructive wildfire – the 2018 Butte County Campfire – inspired the creation of DamageMap.

Hell severely damaged Fricker’s childhood home in Chico, where his parents still lived, but luckily the wildfire did not bring the house to the ground.

During the evacuation, Fricker struggled to know if the house was still standing. “I was frantically trying to download all the satellite images I could get, trying to get information for myself and our neighbors,” he recalls.

Wanting to spare others the same distress, Fricker gathered aerial footage of the destruction of Camp Fire and assessments of damage to door-to-door structures in Cal Fire. With this data, he and a team of Cal Poly undergraduates created a rudimentary prototype of DamageMap.

He took the prototype to Google’s 2019 Geo for Good summit, where he met Stanford graduate student Krishna Rao. At the event, the two built an improved version of the program. And in the years that followed, they continued to collaborate and recruited more scientists to work on the project.

Last year’s CZU Lightning Complex fire in Santa Cruz and San Mateo counties was California’s ninth most destructive wildfire. The event burned more than 80,000 acres and destroyed nearly 1,500 structures, including 911 homes in Santa Cruz County.

Despite evacuation orders, many of Beasley’s neighbors remained behind. If a program like DamageMap had been available to continuously inform evacuees of the condition of their homes, she said, more people would likely have fled for safety.

“It’s not knowing who made it so bad because you couldn’t look to the future at all,” said M’Liss Jarvis Bounds, another evacuee from Boulder Creek. She waited three weeks to learn that her house had survived the blaze.

DamageMap works by first creating a database of the locations of houses and buildings before the fire using satellite images or aerial photos. Then he examines the photos after the fire and decides which structures are damaged based on features such as collapsed or blackened roofs.

The application uses “machine learning”, a form of artificial intelligence, or AI, to identify buildings that have burned down.

Typically, computer programmers feed tens of thousands of images into a program so that it learns to identify specific patterns. Facebook, for example, uses machine learning to recognize faces and suggest people “tag” photos.

By developing DamageMap, researchers introduced nearly 50,000 images of burnt and intact structures into the program, including photos of the Tubbs fire in 2017 in Santa Rosa, the Southern California wildfires of 2017, and Woolsey Fire in 2018 in Los Angeles and Ventura counties. Afterwards, the programmers tested how well DamageMap learned what a fire-damaged structure looked like by showing the app another 18,000 images of Camp Fire and Carr Fire 2018 in Shasta and Trinity counties.

The program correctly identified the charred structures in the second set of photos at least 92% of the time in about 18 minutes, according to the published article. But he made mistakes when trees or other objects blocked the view of buildings and when roofs blended into the surroundings.

While not intended to replace human post-fire assessments, technology that can quickly and accurately assess damage does call for emergency responders.

“As the technology and machine learning technology develops, we will certainly use it in the unfortunate event of another campfire or another Tubbs fire, where it mows many structures. both, “said Will Brewer, GIS analyst and developer at Cal Fire.

For now, Fricker and his team are improving the program by providing DamageMap with more data to learn. The more post-fire images he sees, the better he can identify the damage.

The developers say a lack of funding prevents the program from being available for wider use. So far, a grant of $ 18,000 from Cal Poly has been the main source of funding, but Fricker estimates that it will take an additional $ 80,000 for the request to be operational to the public.

Fricker said he has to pay dedicated computer programmers to continue training him and has to find a suitable online platform to host the program, which could be costly.

“The code is working and we have a lot of data,” Fricker said. “If people were motivated to make this known to the public for the next fire season, it could be done. “


Gordon K. Morehouse