Using AI to Predict Volcanic Eruptions

Of Earth's 1,500 active volcanoes, only a fraction are monitored. This is primarily due to the extreme conditions that volcanoes produce, and as a result sensors and other monitoring equipment tend to have extremely short lifespans.

Volcanoes that are considered non-threatening, or dormant, rarely have any form of monitoring what-so-ever, even though the Chilean volcano Chaitén erupted in 2008 after nearly 8,000 years of dormancy.

Today, volcanologists are using satellite imagery and artificial intelligence to monitor most volcanoes with the hopes of possibly predicting eruptions.

These predicted eruptions could prove vital considering some 800 million people live within 62 miles of an active volcano. Humans have been living in close proximity to volcanoes for centuries, mostly due to the fact that the soil near volcanoes is so fertile.

With the help of MOUNTS, or Monitoring Unrest from Space, researchers are currently tracking 18 volcanoes from all over the world. A recent paper in Remote Sensing explains how MOUNTS searches for signals through satellite images captured throughout the electromagnetic spectrum. These signals include changes in the surrounding landmass and temperature spikes, as well as gas emissions. An inactive volcano that has a sudden change in behavior could mean an eruption is nearing.

MOUNTS has the ability to gather substantial information on the monitored volcanoes, more information than what can be sorted through manually. Researchers have therefore employed artificial intelligence. With the development of an artificial neural network, landmass deformation-a telltale sign of magma moving under the surface-can be detected. The algorithm would then compare images from different times to verify any changes.

"We don't want to monitor them all the time," says Andreas Ley, a TU Berlin researcher who worked on MOUNTS. "We want the system to tell us when something interesting is going on."

MOUNTS has already proven itself useful by detecting signals in some of the most recent eruptions and then sending automated emails to registered users. All of the MOUNTS data is accessible online for free.

However, MOUNTS still has a few obstacles to overcome before it becomes a reliable and trustworthy system. As for now MOUNTS has trouble with seasonal pictures that show the same area but with varying foliage, as well as images that may have been distorted due to precipitation or vapors in the atmosphere.

Although the technology has potential, it still isn't capable of replacing ground-based sensors. While satellite imagery can help predict a volcanic eruption up to two years in advance, some volcanoes erupt abruptly, only minutes after showing signs of possible activity.

Researchers are currently working on incorporating the images of temperature changes and gas emissions into the algorithm in order to better predict volcanic eruptions.

Join the Discussion

Recommended Stories

Real Time Analytics