It seems that if. This is what Google says, and a study it has done with Harvard University. Phoebe DeVries is the researcher in the Astronomy department of the aforementioned University who has coordinated the initiative to predict the location of earthquake aftershocks using artificial intelligence (AI).
And how have they done it? According to Harvard University, “We have collaborated with machine learning (ML) experts at Google to see if we could apply deep learning and determine where replicas could occur. We started with a database with information on more than 118 major earthquakes around the world ”.
From these data, they point out at the University, a neural network was applied to analyze the relationship between static stress changes caused by the locations of the initial tremor and the aftershocks. The end result was a significantly improved model for predicting aftershock locations, which could one day help deploy emergency services and evacuate areas at risk of aftershock. Although this system is still imprecise, it represents a huge advance over classical techniques.
According to DeVries, “With the study we also got an unexpected result that can help us better understand the cause of earthquakes, and not just aftershocks. By applying neural networks to the dataset, we were able to take a closer look at the specific factors that you consider important or useful for that prediction, rather than just taking the prediction results as is. This opens up new possibilities for physical theories that allow us to better understand this phenomenon of nature ”.
This study, published at the end of August in the journal Nature, is not the only one that tries to decipher if AI is capable of helping seismologists to the secrets of earthquakes. In February of this year another project, led by deep learning scientist Thibaut Perol, currently at the startup Gram Labs and also a disciple of Harvard University, was published by Science Advances.
It recounted how Perol had used artificial intelligence to detect earthquakes through a system they called ConvNetQuake, which is, according to the aforementioned publication, the first neural network designed to detect and locate earthquakes. The solution uses a specialized algorithm that can observe measurements of ground movement known as seismograms, and determine whether the seismic activity is just “noise” or an earthquake.
Thibault Perol’s research team used seismic activity in Oklahoma to train and test ConvNetQuake, finding that their system detected 17 times more earthquakes than were recorded in the Oklahoma Geological Survey catalog of earthquakes. It seems, then, that AI can help predict and forecast earthquakes, which is undoubtedly a breakthrough that will help save many lives.
