Neural network has learned to notice earthquakes by changes in the Earth’s gravity

(ORDO NEWS) — French scientists have developed an artificial intelligence capable of registering earthquakes faster than the most sensitive seismographs – by weak changes in the planet’s gravitational field.

Strong earthquakes are manifested not only by shocks of the earth’s crust, but also by changes in the Earth’s gravitational field.

Unlike seismic vibrations propagating in the interior of the planet, its gravitational field changes almost instantly, at the speed of light, allowing you to recognize the threat much faster.

To determine this signal almost as quickly, scientists from France used artificial intelligence. During tests, their model detected an earthquake in just 50 seconds.

The authors of the paper believe that their approach can complement existing seismic monitoring systems with completely new capabilities for early warning of such threats.

So far, such installations rely on the readings of sensitive seismometers. They are extremely useful and regularly save human lives, but do not allow you to quickly recognize the degree of danger.

So, to determine the magnitude of an earthquake, geophysicists rely on P-waves, which propagate from the epicenter faster than others and arrive at the detector very first.

And if the earthquake is strong enough, this signal is often too strong and does not immediately accurately estimate the magnitude.

However, any earthquake is associated with a redistribution of mass in the bowels of the planet, and hence with some changes in its gravitational field.

Previously, such a signal was considered too weak to distinguish it from the general background. It was only in 2017 that this was done for the first time .

Then similar work was done on archival data for many powerful earthquakes that have occurred in recent decades.

She showed that, in principle, a gravitational signal makes it possible to register an earthquake much faster than the arrival of tremors and take the necessary safety measures. It remains to find the right tool.

The PEGSNet machine model developed by Andrea Licardi and his co-authors became such a tool. She was trained on data from real earthquakes, as well as 500 thousand simulated events – without them, there would not be enough information to prepare the neural network.

We are talking about really powerful shocks with a magnitude of at least 8.3: for weaker ones, even AI is not yet able to isolate the desired signal.

For verification, PEGSNet received gravity monitoring data collected during the famous 2011 Japan earthquake.

It was one of the most powerful earthquakes in modern history: the magnitude reached 9, and the epicenter was 70 kilometers from the east coast of Japan, so seismic tremors reached the nearest island in a couple of minutes, and tsunami waves in a couple of tens of minutes.

The new method made it possible to notice shocks along the gravitational field earlier than seismographs did – just 50 seconds after the start.

Thanks to AI, such high-speed recording of earthquakes has become not just an idea, but a working method that can now be fully implemented in a variety of seismic monitoring stations.

Perhaps, in some places they will be supplemented by new means of predicting an approaching tsunami wave from fluctuations in the Earth’s magnetic field, which we have already talked about .

Online:

Contact us: [email protected]

Our Standards, Terms of Use: Standard Terms And Conditions.