“This is a completely new way of [identifying] large earthquakes,” said Richard Allen, a seismologist at the University of California, Berkeley, who was not involved in the study. “If we implement this algorithm, we’ll have a lot more confidence that this is a really big earthquake, and we’ll be able to issue alerts more quickly over a larger area.”
Scientists typically detect earthquakes by monitoring ground vibrations, or seismic waves, with devices called seismographs. The amount of advance warning they are able to provide depends on the distance between the earthquake and the seismograph, as well as the speed of the seismic waves, which can travel below 6 kilometers per second. Networks in Japan, Mexico and California provide seconds or even minutes of advance warning, and the method works well for relatively small earthquakes. But above magnitude 7, seismic waves can saturate seismographs. That makes the most damaging earthquakes, such as the Great East Japan Earthquake, the hardest to identify, Allen said.
Recently, researchers involved in the search for gravitational waves realized that these gravitational signals, traveling at the speed of light, might also be used to monitor earthquakes. “The idea is that whenever mass moves anywhere, the gravitational field changes, and … everything feels it,” said Bernard Whiting, a physicist at the University of Florida who works on the Laser Interferometer Gravitational-Wave Observatory. Surprisingly, this signal even shows up in seismographs.”
In 2016, Whiting and his colleagues reported that ordinary seismometers could detect these gravity signals. Earthquakes cause massive changes in mass; these changes send out gravitational effects that deform both the existing gravitational field and the ground beneath the seismograph. By measuring the difference between the two, scientists think they can create a new kind of earthquake early warning system. Gravity signals appear on seismographs before the first seismic wave arrives, in a section of seismograms that are traditionally ignored. By combining the signals from dozens of seismometers, scientists can identify patterns that explain the size and location of large events, Whiting said.
Now, Andrea Licciardi, a postdoc at the University of the Côte d’Azur, and his colleagues have built a machine learning algorithm to do this pattern recognition. They trained the model on hundreds of thousands of simulated earthquakes and then tested it on a real dataset in the Northeast. The model accurately predicted the magnitude of the earthquake in about 50 seconds, researchers reported Thursday in the journal Nature, faster than other state-of-the-art early warning systems.
“It’s not just a seed of an idea — they’ve shown it can be done,” Whiting said. “What we’re showing is a proof-of-principle. What they’re showing is a proof-of-implementation.”
Gravity signals are too weak to detect earthquakes smaller than magnitude 8.3 with current technology, and the system is unlikely to provide additional early warning in earthquake zones already covered by seismographs. But it can provide more reliable estimates of the size of large earthquakes, which is critical, especially for predicting tsunamis, which often take an extra 10 or 15 minutes to arrive, Allen said. With this technique, seismologists in Japan can accurately determine the magnitude of the Tohoku region and issue an appropriate alert “within 1 or 2 minutes of the onset of an earthquake,” said seismologists, also at the University of the Côte d’Azur, in the paper. said co-author Jean-Paul Ampuero. “In 2011, it would take a few hours. It would be very good.”
But the technology is not yet in use. It’s not yet processing data in real time. The model will be deployed in Japan — but only for earthquakes generated by specific fault areas that could produce “big earthquakes.” The algorithm needs to be trained individually for use in different regions, and the researchers are currently training it for seismic networks in Peru and Chile, Licciardi said. Still, “we have a first-generation algorithm … that’s a huge improvement,” Allen said. “Now let’s find out if it actually works.”
