Skip to main content
31.03.2025 | ב ניסן התשפה

Lightning Strike: The AI Model That Predicts Wildfires

Scientists from Bar-Ilan University’s Department of Computer Science help crack the code behind lightning-induced wildfires—offering a powerful new weapon in the fight against climate change

Image
Lightning Strike: The AI Model That Predicts Wildfires Before They Start

Wildfires are no longer rare catastrophes. They’re becoming seasonal,—almost expected. The damage is staggering: in just one recent case, a wildfire that tore through Los Angeles left dozens dead and caused an estimated $200 billion in damage.

While many fires are sparked by human activity, lightning is behind some of the most devastating—and unpredictable—blazes. These fires ignite in remote regions, go unnoticed for days, and grow uncontrollably before first responders even know they’ve begun.

Now, a team of Israeli researchers has developed a revolutionary AI model that can predict where and when lightning will spark the next fire—with over 90% accuracy.

And it all started in the Computer Science Department at Bar-Ilan University.

A Burning Need

Climate change is making extreme weather more extreme. More frequent lightning storms, hotter and drier conditions, and shifting ecosystems mean more fuel—and more ignition points. In some regions, like Canada and the western U.S., lightning-caused wildfires now account for the majority of burned land. And the numbers are only going up.

What makes lightning fires so dangerous is how sneaky they are. A strike can hit, smolder underground for days, and suddenly erupt into a full-blown inferno. And because these strikes often occur far from civilization, firefighting crews have no chance to contain them early.

The Breakthrough

In a study published in Scientific Reports (Nature Publishing Group), researchers—including Dr. Oren Glickman and Dr. Assaf Shmuel from BIU’s Department of Computer Science, alongside colleagues from Ariel and Tel Aviv Universities—unveiled a machine learning model that maps and predicts lightning-induced wildfire risk on a global scale.

Unlike older models, which relied on regional patterns and limited data, this one was trained on high-resolution, global satellite data spanning seven years (2014–2020). The AI takes into account not just lightning strikes, but vegetation, weather conditions, and topography to assess the risk of fire ignition.

It was then tested using wildfire data from 2021—and nailed the results with over 90% accuracy. That level of precision has never been seen before.

Why It Matters

This isn't just an academic achievement. With better understanding of where and when lightning fires are likely to occur, meteorological services, fire departments, and emergency planners can act sooner, respond smarter, and potentially save lives and ecosystems.

One of the study’s most important findings is that lightning-induced wildfires behave very differently from human-caused ones. In fact, predictive models designed for human activity simply don’t work when applied to lightning fires—highlighting the urgent need for specialized approaches tailored to each fire type.

The research also underscores just how rapidly climate change is accelerating the wildfire threat. The model’s forecasts show a consistent rise in the risk of lightning fires as the planet warms—driven by both increased lightning activity and more frequent "fire weather" conditions.

Crucially, the study shows that using broad, high-quality global data sets dramatically improves prediction accuracy, demonstrating how Big Data and AI can serve as powerful tools in climate resilience.

What Does the Future Hold?

This AI model is not yet running in real-time forecasting systems—but it lays critical groundwork. It shows that with the right tools, we can get ahead of nature’s most destructive forces.

With the growing implications of climate change, new modeling tools are required to better understand and predict its impacts; machine learning holds significant potential to enhance these efforts,” says Dr. Shmuel.