The study emerges at a time when California and most of the western United States is experiencing a more frequen and severe fire season. Several fires driven by even more perilous weather conditions of wind, drought, and heat are spreading across the state. Among them, the Lake Fire in the largest one in the state this year and has already burnt through more than 38,000 acres in Santa Barbara County.
”This model is a good advancement towards containing wildfire,” said Bryan Shaddy, a doctorate student in Aerospace and Mechanical Engineering at the USC Viterbi School of Engineering and the author of the paper. Through providing data more accurate and timely, our tool supports the firefighters and the evacuation teams in the fight against wildfires on the front line.
Reverse-engineering wildfire behavior with AI
Before the actual analysis, the researchers needed to accumulate the actual historical wildfire data from high spatial resolution satellite imagery. Thus, based on the analysis of the previous wildfires the researchers received the conception of how the fire of each type began, how it evolved to be a fire and how it was put off. Their examination made it possible to outline the patterns that influence them with the help of factors including weather, fuel which is considered to be trees, brush, etc. & topography.
They then described a generative AI machine learning model also known as a conditional Wasserstein Generative Adversarial Network (cWGAN) as to how these factors influence the manner in which the wildfires will unfold in the future. and in the area related to the match with the satellite image, their shown model of how the fire spread they trained shape that can be seen in the satellite image.
They then check how well the cWGAN model perform for the real was that happen in California in the period of 2020 to 2022 to predict the fire spread zone.
From such observations we create a model of how the next fire is likely to behave, said Assad Oberai, Hughes Professor and Professor of Aerospace and Mechanical Engineering at USC Viterbi and co-author of the study.
Using AI to predict wildfires: A feeling that is achieving or could achieve an impressive model.
Using AI to predict wildfires: Impressive model
Thus, Oberai and Shaddy were glad to see that even if the cWGAN has been trained on basic simulation data assuming ideal conditions such as flat-topography and a single-direction wind, it passed its tests sufficiently well on the actual wildfires of California. However, they blamed this success on the fact that the cWGAN was applied with actual wildfire data from satellite imagery, not independently of it.
As Oberai whose research interest is using computers to simulate fluid dynamics in order to gain physical insight of various phenomena, has simulated virtually all things, from the turbulence over aircraft wings, infection diseases to study the proliferation of cells inside tumors and their behavior. Speaking of everything that he has modeled, he observes that wildfires are possibly among the hardest.
“Wildfires involve intricate processes: Fuel such as grass, shrubs or trees burn and start a chemical reaction that produces heat and wind – the behavor of the fire is influenced by the topography and and or weather – in dampness fire do not spread easily but when dry they spread very fast. ”These are, in fact, highly complex, and often chaotic, and nonlinear processes, and you need to factor in all these variables To the extent that you are going to do it faithfully, you need high-end computing.”
Other co-authors are an undergraduate student Valentina Calaza in the Department of Aerospace and Mechanical Engineering at USC Viterbi; Deep Ray who is a postdoctoral scholar of the University of Maryland, College Park was previously in USC Viterbi; Angel Farguell and Adam Kochanski from San Jose State University, Jan Mandel of University of Colorado, Denver; James Haley and Kyle Hilburn from Colorado State University, Fort Collins;
It should be noted that the research was supported by Army Research Office, NASA and Viterbi CURVE program.
