AI predicts the spread of forest fires
Forest fires can destroy large areas. To fight them, fire departments need to know how the fire will behave. AI can help with this.
![Forest fire in Sicily](https://heise.cloudimg.io/width/610/q85.png-lossy-85.webp-lossy-85.foil1/_www-heise-de_/imgs/18/4/6/3/8/9/3/6/Waldbrand_Sizilien_20210804-c9c0cd01d9a5a3b2.jpeg)
(Image: Alessio Tricani / Shutterstock.com)
A team of scientists at the University of Southern California (USC) has developed a method based on artificial intelligence (AI) to accurately predict the spread of forest fires. This gives fire departments a better opportunity to respond to forest fires and fight them more effectively.
The USC AI model uses satellite images to track the spread of a forest fire in real time. This information is analyzed by an AI that predicts the probable course of the forest fire, its intensity and the growth rate of the fire. The researchers describe the underlying AI in the study "Generative Algorithms for Fusion of Physics-Based Wildfire Spread Models with Satellite Data for Initializing Wildfire Forecasts", which has been published in Artificial Intelligence for the Earth Systems.
The AI model is based on historical data on forest fires. The researchers collected high-resolution satellite images of forest fires. They examined these with regard to the behavior of the forest fires. This enabled the researchers to find out how and where the fires started, how they spread and how they could be contained. From the analysis, the scientists were able to extract patterns that depend on various factors such as weather, fuel (trees, undergrowth, grass and the like) and terrain.
"Complex processes take place during forest fires: Fuels such as grass, bushes or trees ignite, leading to complex chemical reactions that generate heat and wind currents. Factors such as topography and weather also influence fire behavior - under humid conditions, fires hardly spread at all, while under dry conditions they can spread quickly," says Hassad Oberai, professor of aerospace engineering at USC and co-author of the study. "These are highly complex, chaotic and non-linear processes. To model them accurately, you have to take all these different factors into account. You need advanced computers to do that."
Predicting the spread of forest fires
The USC researchers used the collected data to train a generative, AI-supported computer model, the Conditional Wasserstein Generative Adversarial Network (cWGAN). This enables the researchers to determine how the individual factors influence the development and spread of forest fires. They taught the model to recognize patterns in the satellite images that correspond to the spread of forest fires in their model.
The researchers tested the cWGAN model on real forest fires in California between 2020 and 2022. Although the model was initially trained with simulated data under ideal conditions, such as in flat terrain with a one-sided wind, the prediction performed well. The scientists attribute this to the fact that actual forest fire data from the satellite images was incorporated into the cWGAN model.
(olb)