In spring 2010 Iceland’s Eyjafjallajökull volcano erupted beneath an ice cap, mixing hot lava with a flood of meltwater, which blasted a plume of gas and ash over 10 kilometers into the sky. Hundreds of people were evacuated and the turmoil reached well beyond Iceland, with several European nations closing their airspace for days. Thankfully, Eyjafjallajökull killed no one—but it still caused its fair share of chaos.
In spite all the harm and havoc volcanic eruptions can wreak—even the nonfatal ones—scientists still cannot reliably forecast them. Although they have had success predicting dozens of eruptions, they lack a standard method. “The field of volcanology is quite a long way behind fields like meteorology, in terms of developing forecasts,” says David Pyle, a volcanologist at the University of Oxford. Volcanoes have complicated, unpredictable behavior—and, of course, much of their activity takes place underground, which makes them significantly harder to study and develop models for than, say, weather systems. “The real challenge at the moment is that for volcanoes where we have no observations of prior eruptions and where it’s not currently densely observed, it can be very difficult to anticipate what will happen next,” Pyle says. He adds that the methods scientists use for eruption forecasts today “are pretty qualitative.” But a team of researchers at the University of Savoy Mont Blanc is attempting to develop a more reliable, accurate and data-driven approach to anticipate eruptions like the one at Eyjafjallajökull—and potentially create a daily or even hourly volcano forecast—using satellites and a method called data assimilation.
Data assimilation is widely used in fields like meteorology—our weather forecasts depend on it. The method combines a model for systems such as weather or climate with real-world data points to develop predictions about the future. The strength of this technique is that the model is continuously fine-tuned—it compares its predictions against the real-world data and self-corrects in near-real time.
In a new study published Wednesday in Frontiers in Earth Science, the Savoy researchers applied data assimilation to a volcano model to see if the technique could accurately predict an important parameter for volcanic eruptions: magma overpressure. This is the excess pressure created by the volcano’s magma pushing outward, relative to the inward pressure created by the overlying rock. “For each volcano, there’s a critical overpressure value,” says Mary Grace Bato, lead author of the study and a PhD fellow at the Institute of Earth Science in France. “If this value is attained, then you would know that in a few days or months, there might be an eruption.” Being able to predict how this element of the system changes could help volcanologists make better forecasts.
For their study, the team created a simplified model based on the Grímsvötn volcano, also in Iceland. They then used synthetic satellite data, on how the volcano’s exterior ground deformed, to inform the model over time and make predictions. Bato offers a simple way to think about the relationship between ground deformation and magma overpressure: “Imagine that the volcano’s magma chamber is like a balloon. If you continuously fill this balloon with magma, the balloon inflates and causes the ground on top of it to deform. We can measure the deformation by using GPS or radar satellite data, and then we can infer the magma overpressure.” The team can also use that satellite data to fine-tune their model’s predictions for magma overpressure in the future. Current practices do not use this type of physics-based technique, explains Daniel Dzurisin, a research geologist at the U.S. Geological Survey’s Cascades Volcano Observatory. He says today’s eruption forecasting relies on combining monitoring data with information from global volcanic databases, local knowledge of a volcano's past behavior and scientific insight based on experience.
When the researchers compared their predictions to a simulation of the volcano, they found data assimilation was able to accurately forecast the shifts in magma overpressure. In addition to the technique’s forecasting strength, it also helped constrain the volcano’s underground features. “It shows that scientists can use data assimilation to better understand various components and behaviors occurring inside the volcano,” such as the geometry of its magma chamber and the rate of magma flow inside that reservoir, Bato explains. “These are the parameters that are very difficult to infer since they are buried at huge depths, greater than 10 kilometers.”
Bato points out that their study is just the start of testing data assimilation for volcanic forecasts. The team used a very simplified model, and their approach relied on synthetic data. In the real world, volcanoes are much more messy and complicated, and the method would need to employ genuine GPS and satellite radar data. Their study also did not incorporate other important predictors for volcanic eruptions such as earthquakes and gas output—although the researchers plan to include these measurements in future studies.
Scientists also still need to gain a better understanding of when the magma overpressure—the main focus of the study—will signal an eruption in the real world. “Conceptually, a critical overpressure value must exist. But that value is generally unknown, probably different for each volcano, and might change over time even [for] a single volcano,” Dzurisin explains. “Nonetheless, being able to estimate overpressure…would be an important step forward.” Dzurisin thinks data assimilation will become more widely used in the volcanology community. “This approach seems to have good promise for applying to the field of eruption prediction,” he says. “We have a very long way to go, though.” Oxford’s Pyle agrees that the method has potential. “It’s certainly an obvious way forward,” he says.
Next, Bato’s team plans to apply their approach to Grímsvötn using real satellite data. One day, Bato says, “I hope we can provide daily or even hourly forecasts for those cities or towns that are located near an active volcano, because that’s where this kind of research would be most valuable.”