Earlier volcano warnings could mean more false alarms—but they could also save lives

New Zealand's Whakaari/White Island volcanic disaster in December 2019 claimed 22 lives and injured 25 others, marking it as the country's deadliest volcanic event in recent history. However, this was not an isolated occurrence; over the past two decades, the nation has witnessed several sudden volcanic explosions with the potential to cause harm, though many were fortunate enough to escape without incident.

Notably, Japan's Mount Ontake erupted in 2014 without any prior warning, resulting in the death of 63 hikers. Researchers have now published a study in Nature Communications proposing that machine-learning techniques could provide more timely warnings of impending eruptions. By analyzing subtle changes in continuous vibrations around a volcano, these algorithms could potentially identify signs of an eruption hours before it occurs.

While implementing such a system could lead to more false alarms, the potential benefits in terms of saving lives could be significant. Conventional warning systems rely on experts interpreting complex signals and assessing the level of risk, which can be challenging when changes occur rapidly over minutes or hours. Automated warnings, drawing on real-time monitoring data, could complement expert judgment by providing quick alerts when volcanic unrest suddenly changes.

However, there is always a trade-off between the number of false alarms and the potential to save lives. The study reanalyzed the performance of a machine-learning eruption forecaster at five volcanoes worldwide, including five in New Zealand, and found that the forecaster correctly predicted four out of five eruptions at Whakaari within a 48-hour window.

The downside, however, was around 15 days each year when a warning would have been issued without an actual eruption. While 15 days may seem substantial, the decision of whether this threshold is too high depends on the stakes involved. Implementing precautionary measures could reduce preventable losses by 30-90%, potentially saving lives and reducing economic disruption.

The balance of costs and benefits will vary depending on the specific volcano and its unique circumstances. For instance, closing Mount Ruapehu during the busy ski season could result in significant financial losses and disruption, but these costs must be weighed against the potential loss of life from a timely evacuation. Other warning systems, such as tsunami warnings, already operate with a higher rate of false alarms, understanding that some level of uncertainty is necessary for public safety.

Introducing automated warnings does not mean replacing human volcano experts; rather, it could serve as a valuable complement to their expertise. These systems would rely on monitoring data collected by organizations like New Zealand's GeoNet, while human volcanologists would still play a crucial role in interpreting volcanic behavior and managing crises.

The proposed automated warnings could be particularly useful during sudden eruptions, providing an initial warning while experts assess the situation and determine the appropriate response. Ultimately, earlier warnings could help protect not only lives but also critical infrastructure like ski areas, walking tracks, nearby communities, roads, power lines, and other essential facilities.

The study aligns with the United Nations' Early Warnings for All initiative, which aims to provide universal access to multi-hazard warning systems by 2027. By accepting a certain number of false alarms, communities may gain more time to evacuate and avoid the catastrophic consequences of a sudden volcanic eruption.