AI Forecasting Wildfire Behavior with Physics

Researchers in the US have made significant strides in the field of wildfire prediction by adapting a generative AI framework to analyze satellite data for forecasting purposes rather than just wildfire detection. The model, developed at the University of Southern California (USC), utilizes satellite data to monitor the progression of wildfires in real time. This information is then input into the AI framework, enabling accurate predictions regarding the fire's potential path, intensity, and growth rate.

With multiple wildfires currently devastating California due to a dangerous mix of wind, drought, and extreme heat, innovative technologies are crucial for monitoring and tracking these natural disasters. The largest wildfire in the state this year, the Lake Fire, has already burned over 38,000 acres in Santa Barbara County. Various AI-based tools have been created to detect, monitor, and track wildfires, showcasing the importance of technological advancements in combating these destructive events.

The approach taken by researchers involves using a physics-informed method to infer the history of a wildfire from satellite observations. By analyzing the fire arrival time, which indicates when the fire reaches a specific location, valuable insights into the wildfire's behavior can be obtained. A conditional Wasserstein Generative Adversarial Network (cWGAN) is employed to predict fire arrival times based on satellite data, allowing for a deeper understanding of how wildfires evolve.

The cWGAN model was put to the test on four California wildfires spanning from 2020 to 2022, with predictions for fire extent compared against high-resolution airborne infrared measurements. The results showed a high level of accuracy, with an average difference of only 32 minutes between predicted and reported ignition times. This demonstrates the effectiveness of the AI framework in providing precise and timely information to aid firefighting efforts.

“This model represents an important advancement in our wildfire management capabilities,” stated Bryan Shaddy, a doctoral student at USC Viterbi School of Engineering and the study's corresponding author. “By offering more accurate data, our tool enhances the work of firefighters and evacuation teams on the front lines of wildfire response.” Assad Oberai, Hughes Professor at USC Viterbi, and co-author of the study emphasized the significance of studying past fire behavior to develop predictive models for future wildfires.