Researchers and scientists are resorting to artificial intelligence to predict, or at least detect in time, volcano eruptions. Considering that there are more than 800 million people living close to active volcanoes, such research takes on a certain importance.
Using satellite images, the MOUNTS project (Monitoring Unrest from Space) currently monitors 18 active volcanoes, including Etna, and does so by analyzing satellite images. However, the data sets provided by the satellites are so large and full-bodied, and are updated almost daily, that manual checks are not possible and computer algorithms must be used.
The images also reflect the smallest changes related to the deformation of the terrain near volcanoes. Just to make the “control” phase more efficient, the researchers developed artificial neural networks, a type of artificial intelligence, to automatically detect important ground deformation events, signals that suggest that the magma itself is moving underneath the terrain.
In essence, researchers, as reported by Andreas Ley, a researcher at the Technical University of Berlin involved in the MOUNTS project, do not want to continuously monitor volcanoes, they want computers to report when something interesting is happening.
The system has already been able to detect early signals of different eruptions. Last month, for example, he detected a deformation of the land connected to an evolution which then took place on the island of Reunion concerning the volcano Piton de la Fournaise.
On this occasion, the system itself sent automatic emails not only to the researchers but also to the users who had registered with the appropriate website to get updates.
The deformations of the ground detectable by the satellites do not cover all situations and for this reason, the researchers decided to integrate these important data with other equally important data such as those related to gas emissions near the cone and to volcano temperature increases or of the area around it. This data is naturally collected via ground sensors. Now, with all this data, it is possible to work on algorithms based on automatic learning to predict eruptions more and more efficiently.