Paris Soir - Can We Predict Volcanic Eruptions Like Weather? Scientists Seek Unified Physics

The summer of 1991 marked a pivotal moment in the history of volcanology. On June 12, Mount Pinatubo in the Philippines began convulsing, culminating three days later in a catastrophic explosion that obliterated its peak and replaced it with a 2.5-kilometer-wide chasm. The eruption triggered pyroclastic flows—incandescent avalanches of molten rock and gas—that tumbled down sterilized slopes, killing more than 800 people, primarily due to roofs collapsing under the weight of rain-saturated ash.

The disaster could have been far deadlier. Approximately 250,000 people lived in the volcano’s shadow across multiple cities and a sprawling U.S. Air Force base. When Pinatubo began belching steam in April, scientists from the United States and the Philippines deployed an array of instruments to track its inner tumult. Mike Poland, who was then part of that rapid geological assessment team and is now the scientist in charge at the U.S. Geological Survey’s Yellowstone Volcano Observatory, recalled the urgency of the moment.

“We didn’t know much about that volcano, and so there was this really rapid geological assessment,” Poland said. “And the assessment said, ‘Oh, crap, when this thing erupts, it only erupts big.’ And that became the basis for a forecast.”

By early June, ash and lava were escaping Pinatubo’s flanks, prompting an evacuation order just days before the cataclysmic event. While the scientists saved countless lives, their forecast was more of an educated guess than a precise prediction. They could not say with certainty that an explosive eruption would occur on June 12, nor could they predict its evolution. This imprecision characterizes forecasting at all well-monitored volcanoes today.

However, the field has made significant leaps since Pinatubo’s eruption. Instrumentation has become more advanced, machine learning has improved data interpretation, and scientists have gained a deeper understanding of the magmatic plumbing that drives volcanism. These advancements have prompted experts to ask a critical question: How close are we to forecasting volcano behavior with the same accuracy as weather forecasts?

Contemporary scientific prediction of weather is a relatively recent invention, grounded in mathematical equations derived at the start of the 20th century. Today, meteorologists can forecast atmospheric conditions up to two weeks into the future. In contrast, while some 800 million people live within 100 kilometers of an active volcano, and rare eruptions can affect the entire planet, volcanoes present unique forecasting challenges.

“The big difference between [volcanoes] and the weather forecasting is the weather is always happening,” said Jenni Barclay, a volcanologist at the University of Bristol in England. The atmosphere is perpetually visible and measurable. Magma, however, resides kilometers below Earth’s crust, and active volcanoes erupt only once every few decades on average.

Each volcano is also unique. The architecture of subterranean pathways, magma chemistry, eruption cadence, and styles differ from place to place. Eruptions are triggered by a complex interplay of factors: the temperature and pressure of the magma reservoir, the weakness of enclosing rock, gas and crystal content, magma depth, and regional tectonic plate motion.

“Geology is chaotic,” said Marius Isken, a geophysicist at the GFZ Helmholtz Center for Geosciences in Potsdam, Germany. “But there is order buried in the chaos. Can we find it?”

Volcanologists currently monitor volcanoes like orchestras composed of hundreds of instruments. Seismometers sense rock cracking as magma ascends; ground sensors and satellites track crustal shifts indicating magma flow; gas detectors reveal depressurization at shallow depths. Yet, knowing how this symphony will develop into a climax remains difficult. At the most comprehensively monitored volcanoes, scientists typically offer acute caution rather than prediction.

Alert systems, such as those used by the U.S. Geological Survey, notify the public of heightened unrest, but this does not guarantee an imminent eruption. “Only 50 percent of volcanic unrest that looks like it’s going to be an eruption ends up in an eruption,” said Jessica Johnson, a geophysicist at the University of East Anglia in England.

Some volcanoes can also ambush observers. Pockets of highly pressurized water trapped below the surface can be heated by adjacent magma. If these pockets rupture, they cause dangerous steam explosions that may unleash imprisoned magma, often with no discernible warning signs.

More predictive detail is possible for volcanoes studied over multiple eruption cycles. At Italy’s Stromboli and Etna volcanoes, which regularly spout lava fountains, scientists can confidently forecast outbursts hours in advance. “We have systems that can tell us that in a few hours, the volcano will erupt,” said Maurizio Ripepe, a geophysicist at the University of Florence.

Similarly, at Hawaii’s Kīlauea and Iceland’s Reykjanes Peninsula, scientists track migrating magma with such precision that they can predict where lava will emerge within an hour. However, Tom Winder, a volcano seismologist at the University of Iceland, noted that such precise forecasts are “relatively unusual” and apply to frequently active volcanoes unlikely to produce major explosive events. In most other cases, early warnings provide insufficient time for evacuation.

Forecasting is complicated because volcanoes cannot be reduced to simple models. They are complex geologic systems with hidden, labyrinthine plumbing. Twenty years ago, a lecturer told Poland that predicting major eruptions was a “pipe dream” due to the idiosyncratic nature of volcanoes. Poland disagreed, noting that all volcanoes are vessels of immense pressure and heat where molten rock eventually cracks, breaks, and explodes.

Despite this commonality, scientists agree that a vital piece of the forecasting puzzle remains unsolved. “We don’t even fully understand the underlying physics,” said Diana Roman, a volcanologist at Carnegie Science in Washington, D.C. Specifically, the transition from a stable magma reservoir to catastrophic failure is not fully understood.

“They have to have shared physics,” Roman said. If these underlying equations can be discovered, they could be applied to all volcanoes to output accurate predictions of eruption timing and style.

Scientists have identified some governing equations, but they apply only after eruptions begin. Using over a century of observations, researchers have derived the physics of volcanic hazards, such as lava and pyroclastic flows. The Navier-Stokes equations describe fluid movement, while the heat equation reveals cooling rates. These allow experts to predict where outpourings will emerge and how far they will reach once an eruption starts.

“This work saves lives, but it’s a fraction of the forecasting dilemma,” Poland said. “Using our weather analogy, this is like saying, ‘Once the rain starts to fall, we can forecast what watersheds might flood.’ Knowing when the storm will start requires getting at the subsurface physics of magma reservoirs.”

Current warnings rely on recognizing patterns in geophysical signals, such as escalating seismic activity. However, correlation is insufficient if patterns are inconsistent. “What we’re trying to do is looking at the causative relationships there … to understand the physics,” Johnson said. “If you understand what those patterns mean, [then] when those patterns change, we’re not that stuck.”

Johnson is part of Ex-X: Expecting the Unexpected, a multidisciplinary project led by the University of Bristol investigating dangerous volcanic escalations in the Eastern Caribbean. The region’s volcanoes can quickly transition from effusive lava flows to catastrophic explosions, as seen at La Soufrière in St. Vincent in 2020.

The Ex-X project will deploy hundreds of seismometers and fiber-optic cable networks to record tiny earthquakes during tranquility and unrest periods. Machine learning programs will be trained to identify minute shifts in seismic data, revealing hidden magmatic pathways and tracking magma movement in near real-time. The goal is to gain unprecedented detail on how small changes in magma behavior lead to eruptions, potentially uncovering shared fluid dynamics equations across diverse Caribbean volcanoes.

Seismology alone is insufficient. “We lack the physical understanding of what exactly is going on in a magma chamber,” Poland said. Key questions remain: What causes the nucleation of bubbles that propel magma? What combination of molten rock, crystals, and gas triggers an eruption? What drives the switch from lava effusion to ash blasting?

Geochemistry is also essential. Scientists analyze fresh or ancient lava and ash to identify subtle chemical changes. While numerical models simulate volcanic processes, laboratory experiments are needed to ground these models. In fall 2025, scientists successfully recreated conditions present at the birth of planets, including simulacra of magma and miniature hydrogen atmospheres.

“Ideally, volcanologists want to try something else truly ambitious,” Winder said: “Drill all the way down to where there is some magma sitting at depth, and really see these processes in situ.” This is a key objective of the Krafla Magma Testbed in Iceland, which aims to become the world’s first direct magma observatory.

Poland expressed optimism about the future. “There’s no reason we can’t think that, at some point in the future, we can have volcano forecasts that are like weather forecasts.” However, deriving a unified theory of volcanism will require a massive scientific effort akin to a geologic Manhattan Project.

This would involve monitoring a diverse constellation of volcanoes over multiple eruption cycles—decades of data collection. “You would like to think, ‘OK, volcanoes are pretty well monitored.’ But they’re not,” Roman said. Only a handful of “Cadillac volcanoes” have permanent networks. Even dangerous U.S. volcanoes in the Cascades, such as Mount St. Helens and Mount Rainier, are only partially covered by limited sensors.

With vast amounts of geophysical and geochemical data, scientists aided by machine learning could identify commonalities to derive foundational geophysical laws. This would allow the creation of an archetypal volcano model—a generic framework that can be layered onto any volcano globally.

For example, if concerns arise about Japan’s Mount Fuji, scientists could feed its seismicity, magmatic geochemistry, and deformation rates into the model. Software driven by governing equations could virtually fast-forward the volcano to predict its most probable eruption date, style, and duration. Zach Ross, a geophysicist and machine learning researcher at the California Institute of Technology, agreed this is “definitely the right way to be thinking about it.”

Skepticism remains, however. Winder stated he can only imagine accurate forecasting in exceptional circumstances, such as frequently erupting volcanoes in Hawaii or Iceland. Others are more sanguine, suggesting that while some volcanoes will always be troublesome, many eruptions should be forecastable.

“What we’re really missing is more data,” Isken said. “We have not really observed that many different systems going off. But I think that gap will fill over time.”

Roman is involved in the Subduction Zones in Four Dimensions (SZ4D) project, an international effort to monitor subduction zones in Chile, Alaska, and the Cascades. If funded, SZ4D would study triggers for landslides, earthquakes, and eruptions, hoping to uncover underlying physics common to these hazards.

Roman views SZ4D as a necessary colossal undertaking, similar to those required to understand weather and climate change. “It’s time for a big push,” she said. As volcanologists continue performing daily scientific work to protect millions, the prospect of days or weeks of warning before an eruption offers a thrilling vision for the future of the field.

B.Mercier--PS