Scientists at the National Center for Atmospheric Research (NCAR) have developed a new technique that uses artificial intelligence (AI) to improve wildfire forecasts. The technique helps efficiently update vegetation maps that are used by wildfire computer models to accurately predict fire behavior and spread.
The method was demonstrated using the 2020 East Troublesome Fire in Colorado. During this fire, the burned land was mischaracterized in fuel inventories as being healthy. But in reality, the area that burned was recently impacted by pine beetles and windstorms, which left large portions of dead and downed timber.
Comparing Wildfire Simulations
The team compared simulations of the fire generated by a wildfire behavior model that used the standard fuel inventory and another that was updated with AI. The AI simulations performed significantly better when predicting the area burned by the fire.
Amy DeCastro is an NCAR scientist and lead author of the study.
“One of our main challenges in wildfire modeling has been to get accurate input, including fuel data,” said DeCastro. “In this study, we show that the combined use of machine learning and satellite imagery provides a viable solution.”
The model simulations were run at the NCAR-Wyoming Supercomputing Center on the Cheyenne system.
Models meant to accurately simulate wildfires require a lot of detailed information about current conditions, such as the local weather and terrain. They also require information on the plant matter, which acts as fuel for the fire.
The LANDFIRE Dataset
The best fuel dataset is produced by LANDFIRE, which is a federal program that produces geospatial datasets that contain information on wildfire fuels. To create the wildfire fuel datasets, experts need a lot of satellite imagery, landscape simulation, and survey information. Because of the large amount of required data, it takes a long time to update the datasets. At the same time, the available fuels in an area can change quickly.
The team updated the fuel dataset by using the Sentinel satellites, which belong to the European Space Agency’s Copernicus program. Sentinel-1 provides data around surface texture, which can be used to identify vegetation type. Sentinel-2 provides data that can be used to infer the plant’s health from its greeness. This satellite data was fed into a machine learning model that was trained on the U.S. Forest Service’s Insect and Disease Detection Survey, which is conducted annually to estimate tree mortality from the air.
With these new additions, the machine learning model was able to accurately update the LANDFIRE fuel data.
“The LANDFIRE data is super valuable and provides a reliable platform to build on,” DeCastro said. “Artificial intelligence proved to be an effective tool for updating the data in a less resource-intensive manner.”
Testing the New System
The team then set out to test the effect the updated inventory would have on wildfire simulations, so they used WRF-Fire, which was developed by NCAR to simulate wildfire behavior.
They first used WRF-Fire to simulate the East Troublesome Fire with the unadjusted LANDFIRE fuel dataset, which resulted in it underpredicting the amount of area that would be burned. However, when the model was run with the adjusted dataset, it predicted this burned area with a far greater degree of accuracy. It did this by predicting that the dead and downed timber would help spread the fire.
This machine learning model is currently designed to update existing fuel maps, but it could eventually lead to the regular production and updating of fuel maps from scratch.
Researchers at NCAR also hope that machine learning will solve other major challenges in this area, such as improving our ability to predict the properties of the embers generated by a fire.
NCAR scientist Timothy Juliano is a co-author of the study.
“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,” said Juliano. “We’re well positioned to make a positive impact; we just need to keep working on it.”
- Neural Networks Learn Better by Mimicking Human Sleep Patterns
- 7 “Best” AI Translation Software & Tools (November 2022)
- Conversational AI Is Making Customers and Employees Happier. Here’s How.
- Electricity Helps Find Materials That Can “Learn”
- Lack of Trustworthy AI Can Stunt Innovation and Business Value