NZ Herald were in the air in the hours after the dramatic eruption of White Island and captured stunning footage of NZ's most active volcano. Video / Auckland Rescue Helicopter Trust / George Novak
A University of Canterbury-led team has developed an AI model that could help scientists forecast future eruptions.
Their just-published study drew on data from earlier events — including the Whakaari White Island disaster — to identify early seismic warning signals
Patterns were found to be transferrable across multiple volcanoes — offering a potential forecasting tool for less-studied volcanoes around the world.
Could scientists one day forecast volcanic eruptions?
It’s a tantalising possibility a team of Kiwi scientists has brought closer to reality with a model developed with data from dozens of eruptions — including 2019’s Whakaari White Island disaster.
Their pioneering approach, described in a just-published study, used artificial intelligence to pick out hidden seismic patterns recorded in the lead-up to 41 earlier eruptions across 24 volcanoes and more than 70 years.
Along with the shock Whakaari eruption that claimed the lives of 22 people, the team also drew on observations from Ruapehu’s 2007 eruption and Tongariro’s first-in-a-century blow in 2012.
As with earthquakes, uncertainty and variability have made it essentially impossible for scientists to predict the likelihood and timing of such events.
But the specific pre-eruption cues identified by the team’s machine-learning model may ultimately help change that.
Crucially, the warning signals were shown to follow repeatable patterns across multiple volcanoes — challenging the current paradigm that each volcano has its own unique patterns.
That meant the model could be applied to less-studied volcanoes elsewhere in the world, where some 29 million people live in areas at risk of eruption.
“This finding could be a breakthrough for eruption forecasting,” the study’s lead author, Dr Alberto Ardid said.
“Timely and accurate eruption forecasting can save lives, reduce economic losses, and minimise losses due to disruptions to air travel, agriculture, and global supply chains.”
The Dome Shelter near Mt Ruapehu's crater lake was almost submerged in 2007's phreatic eruption. Photo / Alan Gibson
Study co-author Professor David Dempsey said the team worked closely with international volcano observatories to ensure their models offered “actionable” data — and they now planned to share their codes with an open-access policy.
“The modelling tool we’ve come up with is relatively simple and it’s complementary to existing practices of volcanic observations, but it provides an extra layer of information.
“It means we can start to think about forecasting eruptions at volcanoes that have never had instrumentally recorded eruptions — such as Mt Taranaki.”
GNS Science senior volcano geophysicist Dr Craig Miller described the new findings as a “promising result” that could complement the methods GNS uses to look for signs of future eruptions.
“Volcano forecasts are critical to minimising the social and economic impacts of any volcanic unrest.”
The study, published in major journal Scientific Reports overnight, comes after Ardid and colleagues used data-crunching approaches to discover several “volcanic burps” at Mt Ruapehu that went unnoticed at the time.
Elsewhere, Ardid has helped pioneer a new wildfire forecasting system that updates using real-time weather data — and could help firefighters protect lives and property from a growing global threat.
Sign up to The Daily H, a free newsletter curated by our editors and delivered straight to your inbox every weekday.
