Researchers at the University of Leon have created an artificial intelligence platform called CROWNFIRE-AI that predicts dangerous wildfires up to 16 days in advance.
The system analyzes environmental conditions to forecast crown fires, which spread rapidly through the tops of tree canopies and develop extreme intensity.
By combining orbital satellite observations from NASA and the European Space Agency with local weather forecasts and terrain data, the tool achieves a 90 percent prediction accuracy.
Satellite data and machine learning analysis
The CROWNFIRE-AI platform relies on a Random Forest machine learning algorithm trained on more than 18,000 recorded wildfire incidents across Spain.
Researchers designed the system to evaluate vegetation moisture and structure alongside physical terrain factors such as slope and elevation.
Atmospheric factors including dryness, high temperatures, and low humidity are also integrated to extend the evaluation window to 16 days.
Data inputs draw from NASA's Moderate Resolution Imaging Spectroradiometer instrument aboard the Terra and Aqua satellites, as well as the European Space Agency's Sentinel-2 satellite mission under the Copernicus Earth observation programme.
Meteorological forecasts from the ERA5-Land reanalysis dataset and the Global Forecast System complement the satellite streams to calculate localized risk levels.
Emergency planning and risk mapping
The University of Leon development team processes all atmospheric and geospatial measurements using Google Earth Engine, synthesizing complex technical data into regional risk maps.
Scientists said the primary objective is to generate clear visual maps that emergency managers and forestry officials can use without requiring complex computer skills.
The platform does not pinpoint exact individual fire events, but calculates the statistical probability that specific areas will meet the environmental risk factors required for crown fires to form.
According to the development team, early warnings allow civil protection crews to position resources strategically, design evacuation plans, and clear combustible plant material before fires start.
The US space agency noted that MODIS satellite products are already used globally to track large fires, but researchers said CROWNFIRE-AI aims to surpass existing monitoring standards by anticipating extreme fire behavior beforehand.
Platform access and regional calibration
Tested against real Sentinel-2 satellite records, the predictive model achieved a 90 percent accuracy rate and an F1-score of 0.88.
The university highlighted that training the model on real historical fire observations provides stronger scientific backing than relying strictly on simulated scenarios.
The CROWNFIRE-AI platform is free to access online, though users must register for a Google Earth Engine account to view interactive geospatial scenarios.
Because researchers built the system using specific historical records from Spain, deploying the tool in other regions requires further calibration using local environmental data.
