Overview
- The University of León published CROWNFIRE‑AI in mid‑August and hosts the cloud platform on its servers with free access after registration.
- The model combines Sentinel, MODIS and other satellite data with topography, forest inventories and GFS forecasts and was trained on about 18,874 reference points from 33 large Spanish fires between 2017 and 2024.
- Developers report strong validation in Spain, citing roughly 90% accuracy and an F1 score near 0.88 on Sentinel‑2 data, while stressing the tool estimates probability of canopy fire rather than predicting exact ignitions.
- In Mexico separate projects are at work: IPN students built a prototype to map pre‑fire risk in Parque Nacional El Tepozteco, and UNAM with Conafor is developing near‑real‑time smoke detection using GOES‑R processed at LANOT with roughly 10‑minute scanning but limited daytime and spatial resolution.
- Researchers emphasize that crown fires are growing more frequent because of heatwaves, drought and fuel build‑up, that tools need local calibration to work outside Spain, and that AI is meant to augment prevention and patrol decisions rather than replace firefighting brigades.