On the planet, about 800 million people live within 100 kilometers of volcanoes. Only a small part of these fire-breathing mountains is monitored.
It is not always possible to predict volcanic activity using existing methods. Because eruptions sometimes entail human sacrifice and destruction, the scale of which is truly impressive.
Fortunately, science does not stand still. Recently, a team of volcanologists led by Juliet Biggs from the University of Bristol in the UK suggested using radar observations from the Sentinel-1 satellite. He collects information about all volcanoes, including potentially dangerous.
Depending on the location in orbit of the Earth, the device updates information about the volcanoes of the world every 6, 12 or 24 days. Moving over previously studied places, he re-measures the distance to the surface of the Earth. This method allows to detect new swelling and subsidence of the soil due to underground movement of magma near volcanoes.
However, space observations are limited by meteorological conditions and atmospheric anomalies in seismically active areas. So, emissions of a large amount of water vapor can significantly distort the picture and mislead specialists.
For the first time, scientists encountered these problems in November 2017, when they watched the Agung volcano on the island of Bali in Indonesia. Then the forerunners of the eruption were hundreds of small shocks that occurred from August to November.
Observation of this volcano, according to Biggs, taught her team a lot. Experts have found a way to “circumvent” the data distortion, which helped in advance to evacuate and in time to remove more than 140 thousand people from the danger zone.
The researchers were able to notice the rise of the soil by only 10 centimeters in the direction of the neighboring volcano. Due to this, the existence of an underground connection between them was discovered.
Now a geophysicist from the University of Bristol, Fabien Albino (Fabien Albino) – a member of the Biggs team – is developing ways that could help overcome atmospheric disturbances.
The methodology, which the research team reports in an article published in the journal Advancing Earth and Space Science, provides not only satellite observation, but also subsequent analysis of data by artificial intelligence. This will help track the slightest changes, find patterns in the behavior of volcanoes and predict the level of danger and nature of the eruption.
Scientists have developed a neural network, which has already analyzed 30 thousand images of more than 900 volcanoes and selected 100 images for closer study. At 39 of them, researchers did find evidence of a near eruption.
Now, to improve the quality of the algorithm, scientists train the network using computer models of volcanic eruptions. This has already helped to increase the recognition accuracy by half.
An alternative approach is offered by geophysicist Andrew Hooper (Andrew Hooper) from the University of Leeds. His method is also based on accurate satellite measurements of soil deformation near a volcano, but in this algorithm it is important how this deformation changes over time (it can remain constant, accelerate, slow down or behave in some other way). The program allows you to track even small changes over long periods of time.
Both teams plan to test their approaches on a global database of terrain changes near volcanoes.
Meanwhile, another technique was presented by Matt Pritchard, a volcanologist from Cornell University in the United States. He develops an algorithm that uses satellite data such as surface temperature, gas and ash emissions for analysis. The scientist expects that machine learning and data accumulated by NASA in recent years of observation will help to significantly improve the ability to predict eruptions.
It is curious that this technique is used to search for places of use of slave labor.
The authors will present their research at an upcoming volcanology conference in Santa Fe, New Mexico.