December 5, 2022


Everything You Value

New forecasting system should help predict wildfires accurately

The East Troublesome Fire, which burned almost 200,000 acres in Northern Colorado in 2020, could have been predicted significantly more accurately by newly developed artificial intelligence, according to news release from the University Corporation for Atmospheric Research.

The new wildfire forecasting system, which was created by researchers at the University of Colorado Boulder and the University Corporation for Atmospheric Research, uses satellite imagery to more accurately measure and update the state of vegetation in wildfire-prone areas and predict the spread of fires.

Currently, the most advanced wildfire imaging system is LANDFIRE, a program developed by federal agencies, including the Department of Interior and the U.S. Forest Service, which produces datasets on available fuels — such as downed trees — to forecast wildfire spread.

Given the large scope of the LANDFIRE program, however, regularly updating the changing landscapes in wildfire-prone areas has posed a challenge. The East Troublesome Fire imaging used data collected in 2016 and did not incorporate the high rate of pine beetle tree kills that helped fuel the fire.

“One of our main challenges in wildfire modeling has been to get accurate input, including fuel data,” Amy DeCastro, the lead author of the new study and a scientist at the National Center for Atmospheric Research, wrote in a statement.

The new artificial intelligence system uses Sentinel satellites, which are produced by the European Space Agency, to collect more detailed information about vegetation, including its moisture, depth, and whether it is alive.

DeCastro described the new data collection method a “viable solution” to the previous issue of low information. In tests of the new machine learning system, the artificial intelligence program was able to classify areas with high tree mortality as high-risk more accurately than with the use of LANDFIRE data alone.

“The LANDFIRE data is super valuable and provides a reliable platform to build on,” DeCastro wrote. “Artificial intelligence proved to be an effective tool for updating the data in a less resource-intensive manner.”

The new machine learning will be able to update LANDFIRE data in a matter of minutes to predict wildfire spread, which can help advance intelligence on the sort of wildfires that have rampaged the area in past months, such as the Marshall Fire, which caused damage to more than 1,000 homes.

Timothy Juliano, a co-author of the study and a specialist in weather prediction at the National Center for Atmospheric Research, said this breakthrough may encourage scientists who are working on technology to improve wildfire prediction, imaging and prevention.

“We have so much technology and so much computing power and so many resources at our fingertips to solve these issues and keep people safe,” he wrote in a statement. “We’re well-positioned to make a positive impact; we just need to keep working on it.”