Science struggled to weigh up warning signs from New Zealand’s volcano

When cool water meets boiling magma, the result can be explosive. This is what is thought to have happened on Whakaari, the New Zealand volcano that erupted on Monday.

Six people are known to have died on Whakaari, also known as White Island, and eight are unaccounted for. Another 30 were taken to hospital with burns. Mobile phone footage shot by tourist Michael Schade, who had just left the island, captures a wall of thick grey smoke and ash billowing 3km into the sky.   

Volcanologists believe that it was a “phreatic” or hydrothermal eruption, a water-hits-magma outburst that can happen when there is surface water in the summits of large active volcanoes.

Water percolating downwards or on the surface encounters magma rising upwards, generating sudden flashes of steam. If this happens underground and the steam cannot escape, an explosion occurs. While steam-driven explosions are notoriously hard to forecast, Monday’s incident once again raises the question of whether eruptions can ever be predicted. While some volcanoes are monitored using ground instruments, another project is combining satellite imagery with artificial intelligence to identify imminent eruptions. Scientists have long wanted to read the volcanic runes — for good reason. There are about 1,500 potentially active volcanoes in 81 countries. An estimated 800m people live within 100km of one — close enough to be vulnerable to lethal hazards, such as superfast hot flows, projectiles and gases. Some 2,000 people have been killed by eruptions since 2000.

Locals have to weigh up the slim volcanic danger against the lure of fertile soils, abundant geothermal energy and lucrative tourism. Communities often trust that warning signs such as tremors will give them a chance to flee.

Indeed, ground instruments at the top of the uninhabited Whakaari recorded an uptick of activity in November. Still, a geohazard monitoring service concluded on December 3 there was no direct risk for visitors. There is now a dispute over how scientific information was used to inform tour operations.

Prediction should, in theory, be possible. Major eruptions are often preceded by signs that the magma chamber underneath a volcano is restless, including earthquakes, changes in ground topography, heat and gas emissions and gravitational and magnetic anomalies.

Those signs can be detected on the ground or from space. The Mounts (Monitoring Unrest from Space) project tracks 18 volcanoes. Parameters, such as gas emissions or ground disturbances, can be picked up by satellites. A paper published in Remote Sensing this year described how the Mounts neural network was able to combine enormous data sets to build a better picture of recent eruptions, including that of Hawaii’s Mount Kilauea in 2018. It is early days: AI can be confused by water vapour and seasonal variations in tree cover, but it does provide intelligence on areas too risky to measure directly.

The challenge, though, is to transform the data into a real-world risk assessment. Judgments about whether to evacuate are made with limited information, high uncertainty and a ticking clock. But false alarms also come with a cost. This happened with the 1976 evacuation of more than 70,000 people from La Soufrière, a volcano in Guadeloupe. Gas was released in the years that followed but no dramatic eruption materalised. That response is now regarded as a mistake: there were no deaths but the evacuation accelerated the region’s economic decline and alienated locals who lost homes and livelihoods.

Better science may provide a deeper understanding of the fire beneath our feet, but it is unlikely to make nuanced judgment calls — steering clear versus staying put — any less agonising.

The writer is a science commentator