A new technique developed by the National Center for Atmospheric Research (NCAR) uses artificial intelligence to effectively update vegetation maps relied on by computer models of forest fires to accurately predict the behavior and spread of fires.
In a recent study, scientists demonstrated the method using the Eastern Problem Fire in 2020 in Colorado, which burned land that was mischaracterized in fuel inventories as a healthy forest. In fact, the blaze, which spread explosively, scorched a landscape that had recently been devastated by pine beetles and storms, leaving significant areas of dead and fallen timber.
The research team compares fire simulations generated by a state-of-the-art forest fire behavior model developed at NCAR using both standard fuel inventory for the area and one that has been updated with artificial intelligence (AI). The simulations, which used AI-updated fuels, did a much better job of predicting the area burned by the fire, which eventually grew to more than 190,000 acres of land on both sides of the continental border.
“One of our main challenges in modeling forest fires is to get accurate data, including fuel data,” said NCAR scientist and lead author Amy Decastro. “In this study, we show that the combined use of machine learning and satellite imagery provides a viable solution.”
The study was funded by the National Science Foundation of the United States, which is a sponsor of NCAR. The simulation simulations were performed at the NCAR-Wyoming supercomputer center of the Cheyenne system.
Use of satellites to report damage from pine beetles
In order for a model to accurately simulate a forest fire, it requires detailed information about current conditions. This includes local time and terrain, as well as the characteristics of the plant matter that provides fuel for the flames – what is actually available for combustion and in what condition. Is it alive or dead? Is it wet or dry? What kind of vegetation is it? how many? How deep is the fuel on the ground?
The gold standard for fuel datasets is produced by LANDFIRE, a federal program that produces a number of geospatial datasets, including information on forest fire fuels. The process of creating these fuel arrays for forest fires is extensive and includes satellite imagery, landscape simulation, and information gathered personally during research. However, the amount of resources needed to produce them means that in practice they cannot be updated frequently, and forest disturbances – including forest fires, insect infestations and development – can radically change available fuels in the meantime.
In the case of the eastern fire problem that started in Grand County, Colorado and burned east in Rocky Mountain National Park, the latest LANDFIRE fuel data set was released in 2016. Over the next four years, pine beetles had caused widespread tree death In the area.
To update the fuel data set, the researchers turned to the Sentinel satellites, which are part of the European Space Agency’s Copernicus program. Sentinel-1 provides surface texture information that can be used to identify vegetation type. (Grass has a very different texture from trees, for example.) And Sentinel-2 provides information that can be used to infer the health of a plant from its greenery. Scientists are feeding the satellite data into a machine learning model known as the “random forest,” which they trained in the U.S. Forest Service’s Insect and Disease Detection Study. The survey is conducted annually by trained staff who assess the mortality of trees from the air.
The result was that the machine learning model was able to accurately update the LANDFIRE fuel data, turning most fuels categorized as “wood litter” or “wood litter” into “sloping vents”, the designation used for heavy forest areas. trees mortality.
“LANDFIRE data is super valuable and provides a reliable upgrade platform,” said Decastro. “Artificial intelligence has proven to be an effective tool for updating data in a less resource-intensive way.”
Positioned for positive impact
To test the effect that the updated fuel inventory will have on forest fire simulation, the researchers used a version of the NCAR model and weather forecasting model known as WRF-Fire, which is specifically designed to simulate the behavior of forest fires. .
When WRF-Fire was used to simulate the eastern fire problem using the uncorrected LANDFIRE fuel data set, it significantly underestimated the size of the area that the fire would burn. When the model was restarted with the adjusted data set, it was able to predict the burned area with a much greater degree of accuracy, which shows that dead and felled timber helped spread the fire much more than if the trees were still they were alive.
For now, the machine learning model is designed to update an existing fuel map and can get the job done quickly (in minutes). But the project’s success also shows the promise of using a machine learning system to start regularly producing and updating zero-fuel fuel maps in large regions at risk of forest fires.
The new NCAR study is part of a broader trend to explore possible applications of AI for forest fires, including efforts to use AI to more quickly assess fire perimeters. NCAR researchers also hope that machine learning can help address other ongoing challenges in modeling behavior in forest fires. For example, machine learning may be able to improve our ability to predict the properties of fire-generated embers (how large, how hot, and how dense), and the likelihood that these embers will cause spot fires.
“We have so much technology and so much computing power and so many resources at our fingertips to solve these problems and protect people,” said NCAR scientist Timothy Giuliano, co-author of the study. “We are in a good position to make a positive impact; we just need to keep working on it.”