Scientists could use AI to detect looming volcanic eruptions
The 2019 Whakaari/White Island volcanic eruption killed 22 people and has resulted in ongoing legal action in an attempt to hold to account those who allowed tourists and employees on the island that fateful day.
But a team of University of Canterbury scientists have attempted to salvage something positive from the disaster - knowledge that could give advance warnings to protect life and limb in future eruptions.
Canterbury Civil and Natural Resources Engineering postdoctoral fellow Dr Alberto Ardid used GeoNet seismometer data from Whakaari and a machine-learning algorithm to detect a "seismic frequency pattern" in the lead up to the fatal eruption.
Ardid used data from instruments recording ground noise and shaking prior to 18 earthquakes around the world, including Ruapehu, Tongariro, and Whakaari in the Bay of Plenty. The training data helped finetune an artificial intelligence system that detected the frequency patterns that occurred most regularly before large eruptions.
Whakaari/White Island source: Wikipedia
The research, published today in the journal Nature Communications, raises the prospect of an AI-based early warning system being developed to warn of impending volcanic eruptions. The issue isn't just topical for Whakaari over two years on from the eruption.
Sensors around Mt Ruapehu have recorded heightened seismic activity over the last month, described as "the longest period of tremor recorded over the past 20 years". The volcanic Alert Level for the active volcano remains at Level 2 and scientists think it unlikely that an eruption will take place.
The key discovery of the Canterbury team was a peak in the displacement seismic amplitude ratio (DSAR). This points to a blockage in the shallow part of a volcano which can trap hot gas, build up pressure and even trigger an explosion.
"We found the same DSAR signal would build and peak in the days before the last five eruptions at Whakaari, and many of the eruptions at Ruapehu and Tongariro," Ardid says.
He says the patterns identified in one volcano can apply to other volcanos as well.
"The next step is to see how well this signal works as a warning system for volcanoes around the world," adds Ardid.
"It's really exciting that we can potentially contribute to a warning system that could help save lives."
Dr David Dempsey, a Senior Lecturer in Civil and Natural Resources Engineering at the University of Canterbury who worked closely with Dr Ardid on the research and is a co-author on the journal article, says the DSAR signal could become part of a precautionary evacuation alarm system, particularly in tourist areas. The volcanoes would need to have active seismic monitoring in place to gather the data needed to inform the AI system.
"Active volcanoes, including Whakaari, Ruapehu, Tongariro, and others around the world where visitors and skiers are likely to be nearby, are unpredictable and sometimes hazardous. Early warning systems could save lives and avoid debilitating injuries.
You must be logged in in order to post comments. Log In