Unsupervised Wildfire Detection Using Multispectral MTG-FCI Data: A Feasibility Study.

Area of Science:

• Earth and Space Science
• Environmental Monitoring
• Remote Sensing Technology

Background:

• The Meteosat Third Generation (MTG) satellite features the Flexible Combined Imager (FCI) sensor, offering enhanced temporal and spatial resolution for environmental monitoring.
• Traditional fire detection methods often rely on lower-frequency polar-orbiting sensors, which may miss transient fire events.

Purpose of the Study:

• To assess the feasibility of near-real-time fire detection using data from the MTG-FCI sensor.
• To compare a conventional threshold method with an experimental Lightweight U-Net autoencoder for unsupervised fire detection.

Main Methods:

• Two unsupervised methods were applied to MTG-FCI data over Italy: a radiometric threshold technique and a Lightweight U-Net autoencoder for anomaly detection.
• The autoencoder was trained on fire-free data, identifying fires based on reconstruction error anomalies and z-score analysis.
• Validation utilized Sentinel-2 imagery, Fire Radiative Power (FRP), and European Forest Fire Information System (EFFIS) data.

Main Results:

• MTG-FCI successfully triggered active fire alerts before polar overpasses in 67.32% of synchronized cases.
• A median early detection lead time of 21 minutes was achieved, with some instances showing an advance of up to 6 hours.
• The high temporal resolution of MTG-FCI enables robust near-real-time alerting, capturing transient fires missed by polar-orbiting sensors.

Conclusions:

• The MTG-FCI sensor is highly effective for near-real-time fire detection, significantly improving early warning systems.
• Despite limitations in spatial resolution for detailed mapping, the temporal frequency provides a critical advantage for detecting dynamic fire events.
• This technology enhances geostationary environmental monitoring and disaster response capabilities.