Artificial intelligence helps scientists predict volcanic eruptions with up to 95% accuracy Science and health

Scientists have developed a new technique to predict volcanic eruptions using machine learning tools to predict when the most dangerous volcano in the United States — Mount St. Helens located in Washington — will erupt.

According to the Daily MailThe new system, developed at the University of Granada (Spain), is able to predict an accident days before it occurs, using the new tool to identify local seismic activity patterns and improve emergency plans. The use of an artificial intelligence strategy made it possible to detect moments of unrest, eruption, and pre-eruption with an accuracy of up to 95%.

The study was announced 10 days after the Pacific Northwest Systematic Network detected about 350 earthquakes in the area, showing signs that the volcano is active. This month alone, there have been 38 earthquakes.

With the help of machine learning, Spanish scientists want to avoid future problems. In 1980, the eruption of Mount St. Helens killed 57 people and changed the local biome.

The tool they developed analyzed all the signals emitted to detect patterns of progression from one stage to another, until the peak of the eruption was reached.

According to the data, the pre-eruption signals match movements seen in 2024 — tremors, magma accumulation and pressure. The study used mathematical formulas to analyze the signals and calculate the four main fundamental monitoring properties: power, predictability, sharpness of activity peaks, and change in signal frequencies.

Based on these results, scientists can determine the state of the volcano: unstable, pre-eruptive and eruptive, with magma erupting.

“This is a reliable numerical value to confirm the probability of a volcanic eruption in the near term and would improve monitoring of the volcanic system and the ability to predict the occurrence of an eruption,” the researchers said in the journal Frontiers in Earth Science. “The probability of being present in an eruption typically reaches 80% when an eruption is about to start, demonstrating that the methodology has potential as a global monitoring tool.”