Wildfire modelling project links fire spread with smoke impact
Iain Hoey
Share this content
Wildfire modelling and evacuation planning
George Mason University is developing a wildfire digital twin designed to simulate fire behaviour in real time and forecast air pollution impacts during fast-moving incidents in Southern California.
The work is being led by Chaowei “Phil” Yang, professor in the Geography and Geoinformation Science Department in the College of Science, in partnership with researchers from California State University—Los Angeles, NASA Jet Propulsion Laboratory and the City of Los Angeles.
The project was framed against the Palisades and Eaton Fires, which killed 28 people, destroyed more than 16,000 structures and displaced tens of thousands of residents.
Wildfires are difficult to predict because many factors affect their growth, spread and speed.
That uncertainty can slow evacuation decisions and make it harder for response teams to judge who should be moved away from flames or hazardous air pollution.
Yang said: “The goal is to develop an artificial intelligence (AI)-based system that can provide real-time, high-resolution simulation and forecasting of wildfire behavior and model the resulting air pollution and air quality impacts for better informed public health responses.”
Yang added: “Inhaling wildfire smoke can cause serious and long-lasting damage to the breathing system.
“We need to get those people impacted to safety as well as those in direct line of the spreading flames.”
Data sources and system development
George Mason University said the digital twin will combine data from satellites, unmanned aerial vehicles (UAVs), ground observations and citizen reports to simulate wildfire progress and the effect of potential interventions.
The system draws in information on fuel sources, moisture, load and consumption, along with wind speeds, temperatures and real-time fire sensors.
George Mason’s high-performance computing (HPC) cluster is used to automate data interpretation, and machine learning modelling is used to calibrate data sets to improve accuracy.
Model accuracy is expected to improve as more data sets become available.
Yang explained: “When we eventually provide information to firefighters and local agencies, we will give them a range of possibilities and a confidence level in those possibilities, such as ‘the fire will move in this direction with about 90% confidence, or 20% confidence,’
“That’s important for them when they’re trying to make these quick decisions about where to put fire fighters.
“It makes the information actionable instead of just data sets.”
The project also includes student participation from high school level through to post-doctoral research, with work focused on cloud computing and digital transformation.
Anusha Srirenganathan said: “Working closely with researchers from different disciplines helped me grow as a collaborator, and my work on the project strengthened my abilities in large-scale satellite data processing, spatial cloud computing, and AI/ML modeling for environmental applications,”
Yang said the team is also working on a Chesapeake Bay digital twin for flood forecasting and a conflict resolution digital twin with an alert system.