Title: FLAME: Physics-Guided Neural Operators for Onboard Satellite Methane Detection in Hyperspectral Imagery

Abstract: Methane is identified as a significant factor in near-term climate change. Rapid identification of methane emission sources is essential for climate intervention strategies. Spaceborne hyperspectral imagery serves as the primary data source for this purpose. Due to the high volume of data generated by these sensors, ground-based detection is considered impractical, requiring detection capabilities onboard satellites. Classical detection methods present high computational costs for onboard hardware, while deep learning models offer speed but exhibit limitations in detection quality.

The authors propose FLAME, a physics-guided neural operator that integrates the physics of methane absorption directly into its architecture. Evaluation on a methane detection benchmark indicates that FLAME achieves the highest detection accuracy among all evaluated methods. The model reduces the pixel-level false positive rate by nearly three times compared to the strongest neural baseline. Additionally, FLAME utilizes fewer parameters than other learned baselines and operates within the latency constraints of onboard satellite hardware.

Source: arXiv:2606.01577v1

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC