More than 29 million people around the world live within 10 km of a volcano, and understanding the behavior of volcanoes, and being able to predict when they will erupt or spew ash into the air, is vital to safeguarding people’s well-being. However, predicting the behavior of volcanoes is difficult, especially if they have been dormant, and monitoring them can be challenging, as taking samples or deploying equipment presents physical hazards. And although theoretical models can approximate how a particular volcano behaves given its location, geological composition, and the behavior of the Earth’s magma beneath it (among other things), there are still many unknown variables, and each volcano is unique.
Now a researcher at the University of Granada in Spain, Zuccarello aims to automatically analyze volcanic activities to develop early warning models that could save the lives of people living near volcanoes. In the last decade, data collection methods have improved significantly, with new and more sensitive equipment, and researchers now have access to an unprecedented flood of data. For example, they can access real-time information about how the Earth shakes in the vicinity of the volcano (seismic activity), the propagation of sound waves from deep within the Earth, and the chemicals present within the volcano and how they are changing. Volcano observatories need to analyze large amounts of data in a short period of time. “There is a need for faster and error-free techniques to collect this data,” said the expert.
His VOLCANOWAVES project, which includes researchers based in Spain, the United Kingdom, Italy, Mexico and Argentina, uses machine learning to identify patterns in seismic activity around a volcano in an effort to predict when or if it will erupt. In particular, Zuccarello is analyzing low-frequency events, such as volcanic tremors, which are generally related to the movement of magma within the pipes of a volcano. Speech recognition In the last decade, the application of machine learning to pattern identification has been integral to speech recognition, but researchers are now using it to forecast the behavior of volcanoes. “Although these fields vary significantly in terms of context and source, the object of the analysis is the same: studying their harmonics over time in search of patterns,” he said.
The main output of the project will be a set of algorithms, to be completed when the project ends later this year, and it hopes that they will be widely used in the scientific community to monitor volcanoes on a day-to-day basis. “Speech and seismic signals share important properties,” says Guillermo Cortés, a specialist in signal processing and machine learning at the University of Udine in Italy. He ran a project called VULCAN.ears, which also used speech recognition technology to understand what volcanoes say.
Cortés and his colleagues developed a real-time volcano monitoring system that automatically detects and labels volcanic “events” in data streams from monitoring stations that detect seismic signals. This system then creates activity catalogs to find patterns of behavior. Roberto Carniel, geophysicist at the University of Udine and scientific supervisor of the project, says: “The arrival of machine learning and applied deep learning techniques is discovering new solutions to old problems. (Now) it is easier to mix results from various monitoring areas involving the study of seismic signals, infrasonic signals, magnetic signals, geochemical analysis of gases and fluids, deformation, thermal and video cameras, to produce more robust and reliable predictions ” .
The team developed a supervised machine learning-based volcanic seismic reconnaissance system, in which they analyzed data that had already been tagged by other experts, teaching the software to identify volcanic events such as volcanic tremors, ashfall or explosions within the volcano. This approach is similar to finding words in conversation, labeling their parts of speech, and finding unique language patterns for each volcano. It is a break with the classical methods for building catalogs of the behavior of volcanoes, explains Cortés. These methods involve automatic event detection and manual classification by experts. “They typically perform this task on a daily basis, which could be too slow in a situation involving a population at risk from an unexpected eruption,” he says.
Time can be of the essence when it comes to volcanoes, particularly in the event of ashfalls, landslides and landslides, he says. In those cases, “detection and classification in real-time operation is essential” to reduce decision-making time if nearby communities need to be evacuated.
Cortés’s ultimate goal was to develop a system that is universal and independent of volcanoes that could easily be integrated into any volcano observatory. To build this, the researchers have created a universal database from dozens of volcanoes around the world and used their machine learning techniques to build universal models. A preliminary version is available online.
However, for Carniel, the important thing now is that volcanic observatories around the world move the work forward. “They are the real key to promoting the independent idea of volcanoes, installing the volcanic seismic reconnaissance system in their own observatories, sharing resources and giving valuable feedback,” he said. These observatories are, after all, the front line of countries’ efforts to protect their citizens from volcanoes within their borders, and scientists must be able to hear the whispers of volcanoes to predict when they will start screaming.
Original article
This article was originally published in Horizon, the EU Research and Innovation Magazine
