HighFM: Towards a Foundation Model for Learning Representations from High-Frequency Earth Observation Data

Researchers have developed HighFM, a foundation model for analyzing high-frequency Earth observation data using over 2TB of satellite imagery to enable real-time disaster monitoring. The model adapts masked autoencoding frameworks with temporal encodings to capture short-term environmental changes and demonstrates superior performance in cloud masking and fire detection tasks.

• →HighFM is the first foundation model specifically designed for high temporal resolution, multispectral Earth observation data.
• →The model leverages over 2TB of SEVIRI satellite imagery from the Meteosat Second Generation platform for training.
• →Enhanced architecture includes fine-grained temporal encodings to capture short-term environmental variability for real-time monitoring.
• →Benchmarking shows consistent performance gains over traditional baselines in cloud masking and active fire detection tasks.
• →The approach offers a scalable path toward AI-powered disaster detection and tracking systems.