AI

AI Fuses Infrasound Waves and Satellite Thermal Imaging to Forecast Explosive Volcanic Eruptions

By Reporting from New York

NEW YORK — What happened

Geophysicists have published results on a multimodal early-warning system named PyroCast that combines ground-based infrasound sensor arrays with orbital infrared imaging to monitor active volcanoes. During blind back-testing and live continuous monitoring across active volcanic systems, the model successfully predicted explosive paroxysmal eruptions an average of 4.2 hours prior to magma breaching the surface, providing significantly more advance notice than classic ground-seismometer spikes alone.

Why it matters

Explosive volcanic eruptions pose catastrophic threats to nearby communities and global aviation. While seismic stations measure ground trembling, seismic noise from hydrothermal boiling or tectonic settling frequently triggers false alarms. Conversely, explosive transitions can occur rapidly with minimal seismic warning if magma chambers are already open. By combining atmospheric infrasound—sound waves oscillating below 20 Hz, which travel thousands of kilometers through the atmosphere—with rapid-scan geostationary thermal sensors, AI can spot the unique atmospheric-thermal 'fingerprint' of ascending gas slugs before violent decompression occurs.

Deep dive

PyroCast uses a multimodal spatiotemporal transformer. The acoustic arm processes continuous waveforms from micro-barometer arrays, searching for minute harmonic tremors between 0.1 and 5 Hz caused by deep gas expansion inside volcanic conduits. The orbital arm tracks pixel-level temperature shifts from meteorological satellites scanning the crater summit every five minutes. The cross-attention mechanism correlates subterranean gas acoustics with surface radiative heat output. When subterranean acoustic resonance accelerates without an immediate surface thermal release, the model identifies acute pressure build-up and issues a probability score for imminent explosive venting.

Report check

Peer-reviewed retrospective validation on historical datasets from Mount Etna, Stromboli, and Popocatépetl confirmed an 84% detection rate for major paroxysms with an average warning window of 4.2 hours. Unverified claims in regional press suggested the tool could forecast quiet effusive lava flows months in advance; the research team clarified that PyroCast is strictly designed for rapid explosive events driven by gas overpressure, not long-term magmatic accumulation cycles.

Open questions

Will atmospheric weather disturbances, such as severe cyclonic storms or high-altitude jet stream shearing, introduce acoustic distortions that degrade infrasound reliability? Furthermore, establishing reliable operational pipelines between space agencies and local civil defense agencies remains an administrative bottleneck.