AI provides early warning for debris avalanches

Detection of an upcoming debris flow as early as possible is key to protecting human lives, landscapes and property. Several hundred mudslides and debris avalanches happen every year in Switzerland alone. The main reason for the occurrences of this natural phenomenon is climate change: permafrost is becoming increasingly unstable and extreme weather events are on the rise.

To date, seismologists around the world have based their seismic monitoring of endangered regions primarily on methods such as radar, laser devices, geophones, and video cameras. The disadvantage of such systems is that they have to be installed in accessible, low-elevation valley sections. Consequentially, as good as all these systems are in combination, they can only identify the debris flow a few minutes before it starts descending towards the valley.

Researchers from ETH Zurich and WSL have been operating a station measuring debris flow in the Illgraben, a huge rocky basin set high above the Rhone valley in the canton of Valais, for over 20 years. It is one of the most active debris flow channels in the Swiss Alps – the flows can be observed several times a year. In a new study published in 2020, the scientists realised a tremendous improvement to the conventional monitoring of such steep alpine terrain by introducing seismic sensors that are normally used to measure earthquakes.

Małgorzata Chmiel, lead author of the paper and postdoctoral researcher at the Laboratory of Hydraulics, Hydrology and Glaciology at ETH Zurich, explains: 

The result of their work is a detector capable of differentiating the debris flow from other ground vibrations such as trembling of the earth, road traffic and even herds of cows. These seismic sensors can be installed at distances of several kilometres, making it possible to track movement at elevations higher than ever before. For their research purposes, a network of seismometers was installed around Illgraben.

But the best technology in this field is not worth much if it cannot be automated. With the help of machine learning and historical data, the researchers developed an automatic detection system that is able to successfully detect new events without any human monitoring. The result after one year: all of the 13 debris flows that occurred at Illgraben were detected properly by the new system. The research data show that, on average, the new technology detects debris flows 20 minutes earlier than existing systems.

While the study provides promising evidence that debris flows can be detected much earlier thanks to seismic data and machine learning, it is still unclear to what extent the new detection system trained at Illgraben can also detect debris avalanches or mudslides in other areas. For this reason, the scientists want to further develop the algorithm so that it can also function successfully with less or, ideally, even with no location-specific historical data.