A new machine learning tool developed by researchers from Columbia University and New York University can forecast the risk of forest fires across the western United States months in advance.
As reported by the Society for Industrial and Applied Mathematics, the tool uses climate data and mathematical models to provide seasonal fire forecasts within minutes.
This marks the first time machine learning has been applied to make monthly fire risk predictions on a seasonal basis.
According to the researchers, the new approach was driven by the 2023 fire season in Canada, which saw larger burn areas and longer-lasting fires.
The fires significantly affected air quality across large parts of North America.
Jatan Buch, a postdoctoral research scientist at Columbia University, explained the limitations of current wildfire forecasts, which primarily rely on data from the National Interagency Fire Center.
These forecasts offer low spatial resolution and provide basic fire danger indicators without accounting for uncertainty in climate factors.
“Fires are getting more intensive and destructive as a result of our warming world,” Buch said.
“If we can identify risk areas with a high level of certainty early on – even months in advance – we can ensure better fire control, evacuation, and public awareness planning.”
The tool, known as SEASFire, analyzes data using dozens of sources, including the European Centre for Medium-Range Weather Forecasts’ seasonal system SEAS5.
It predicts fire risk based on long-term trends, including climate history, vegetation, topography, lightning, and human activity.
The SEASFire model forecasts elevated fire risks for regions in California and the Pacific Northwest in the coming months.
Gabriel Provencher Langlois, Visiting Assistant Professor at NYU, emphasized the improved capabilities of their tool over existing models: “Our model can analyze the volumes of data that exist and provides significantly more information than what’s currently available with other forecasts.”
The SEASFire tool allows for more accurate predictions of fire-prone areas by integrating climate predictions and fire risk factors on a long-term scale.
This helps address one of the major challenges in wildfire prediction: identifying climate conditions conducive to extreme fires months in advance.
The team plans to release the SEASFire tool to the public free of charge.
They are currently developing a website where users can input a zip code to receive forecasts for wildfire risk in their specific area.
The researchers hope this will increase public awareness and improve preparedness in fire-prone regions.
“Fires cause significant personal and ecological damage and involve huge costs,” said Buch.
He stressed the importance of using tools like SEASFire to better predict fire risks and prepare accordingly.
A new machine learning tool, developed by researchers from Columbia University and New York University, can predict forest fire risk across the western United States months in advance.
The tool, known as SEASFire, provides detailed predictions based on climate data, including inputs from the SEAS5 seasonal forecasting system.
SEASFire aims to improve fire risk prediction for regions such as California and the Pacific Northwest, which are expected to experience elevated fire risks in the coming months.
The researchers are preparing to make the tool publicly accessible, allowing individuals to receive long-term fire forecasts by inputting their zip code.
