Artificial intelligence is suitable for monitoring volcanoes

Helmholtz Center Potsdam GFZ German Research Center for Geosciences

Read out

More than half of Earth's active volcanoes are not instrumentally monitored. For example, outbreaks can occur that could at least theoretically warn people without triggering an alarm. Researchers from the Technical University Berlin and the German Research Center for Geosciences GFZ in Potsdam have now created the MOUNTS volcano monitoring platform, which brings together various measurement data and analyzes satellite imagery using, among other things, 'machine learning'.

More than half of Earth's active volcanoes are not instrumentally monitored. For example, outbreaks may occur that could at least theoretically warn people without triggering an alarm. In a first and still early step on the way to a volcano warning system, a new volcano monitoring platform was developed in a research project led by Sébastien Valade from the Technical University of Berlin (TU Berlin) and the GFZ German Research Center for Geosciences in Potsdam analyzed using "artificial intelligence". By testing recent events, Valade and his colleagues demonstrated that their MONUNTS (Monitoring Unrest from Space) platform can combine multiple datasets with different types of data for full volcanic monitoring. The results of the team were published in the journal Remote Sensing.

Of the 1500 active volcanoes worldwide, up to 85 break out each year. Because of the cost and uncertainties of maintaining volumetric instrumentation, less than half of active volcanoes are monitored with ground-based sensors, and even fewer are considered well-controlled. Volcanoes that are considered dormant or extinct are usually not observed instrumentally. But they can erupt unexpectedly and massively, as happened in 2008 at the Chaitén volcano in Chile, which awoke after 8, 000 years of inactivity.

Eruptions are often accompanied by precursor signals

Satellites can provide crucial data when ground-based monitoring is limited or absent. Continuous long-term observations from space are the key to better recognize signs of geological unrest. Eruptions are often - though not always - accompanied by precursor signals that can take several hours to a few years. These signals may include changes in seismic behavior, soil deformations, gas emissions, rising temperatures, or a combination thereof.

"With the exception of seismicity, all of these phenomena can be monitored from space by using different wavelengths in the electromagnetic spectrum, " says Sébastien Valade, head of the MOUNTS project. It is funded by GEO.X, a research network for geosciences founded in 2010 in Berlin and Potsdam, and carried out at the TU Berlin and the GFZ. "In the MOUNTS monitoring system, we use different satellite sensors to detect and measure changes in volcanoes, " adds Valade. "And we also included seismic data from GFZ's global GEOFON network and United States Geological Survey USGS data." Display

Part of the project was to test whether "Artificial Intelligence" (AI) can be successfully integrated into the data analysis process. The AI ​​algorithms were mainly developed by Andreas Ley of the TU Berlin. For the automatic detection of large deformation events he used so-called artificial neural networks. The researchers trained them with computer-generated images that were modeled on real satellite imagery. From this large number of synthetic examples, the software learned to detect larger deformation events in real, previously unknown satellite data. This area of ​​data science is referred to as "machine learning."

"It was an important 'test balloon' for us to see how we can integrate machine learning into the system, " says Ley. "At the moment, our deformation detector solves only one task. Our vision is to integrate several AI tools for different tasks. Because these tools typically benefit from learning large amounts of data, we want them to continuously learn from all the data the system collects globally

MOUNTS monitors 17 volcanoes around the world

The main challenges facing S bastien Valade and his co-authors were the handling of large amounts of data and software development issues. "But these problems can be solved, " says Valade. I believe that automated monitoring systems using AI and data from multiple sources such as remote sensing and Earthbound sensors in the not-too-distant future will help to warn people more timely and more reliable.

The analysis currently provided by the MOUNTS monitoring platform already allows a comprehensive understanding of different processes in different climatic and volcanic environments around the world: the spread of magma below the surface From the distribution of volcanic material during the eruption to the morphological changes in the affected areas and the emission of gases into the atmosphere. The researchers successfully tested MOUNTS for recent events, such as the Krakatoa outbreak in Indonesia in 2018, or outbreaks in Hawaii and Guatemala.

The system is currently monitoring 17 volcanoes worldwide, including Popocat petl in Mexico and tna in Italy. The platform's website is freely accessible on the Internet and designed so that new data can be easily integrated thanks to global coverage and free access to data. (Remote Sensing; doi: 10.3390 / rs11131528)

Source: Helmholtz Center Potsdam GFZ German Research Center for Geosciences

- Kay Sanders