Sciencehabit shares a report with Science.org: Two minutes after the world’s largest tectonic plate shook off the coast of Japan, the country’s meteorological agency issued a final warning to about 50 million people: a magnitude 8.1 earthquake triggered a tsunami that hit the coast. But just hours after the waves arrived, experts determined the true magnitude of the March 11, 2011 earthquake in Tohoku. Ultimately, it reached a magnitude of 9 – releasing more than 22 times more than energy experts had predicted, and leaving at least 18,000 dead, some in areas that never received warnings. Now scientists have found a way to quickly get more accurate estimates of size, using computer algorithms to identify traces of gravitational waves that break out of the fault at the speed of light.
Recently, researchers involved in the hunt for gravitational waves – ripples in space-time created by the movement of massive objects – realized that these gravitational signals moving at the speed of light can also be used to observe earthquakes. “The idea is that as soon as mass moves anywhere, the gravitational field changes, and … it all feels that way,” says Bernard Whiteing, a physicist at the University of Florida who worked on the gravitational wave observatory laser interferometer. “What was amazing was that the signal would be present even in seismometers.” Of course, in 2016, Whiteing and colleagues reported that conventional seismometers can detect these gravitational signals. Earthquakes lead to large mass shifts; these shifts cause gravitational effects that deform both existing gravitational fields and the ground beneath seismometers. By measuring the difference between the two, scientists have concluded that they may create a new kind of early earthquake warning system. Gravitational signals appear on seismometers before the arrival of the first seismic waves, in the part of the seismogram that is traditionally ignored. By combining signals from dozens of seismometers on top of each other, scientists can identify patterns to interpret the size and location of major events, Whiting says.
Now Andrea Lichardi, a postdoc at the University of the Cote 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 before testing it on a real-world data set from Tohoku. The model accurately predicted the magnitude of the earthquake in about 50 seconds – faster than other modern early warning systems, researchers said today in Nature. “It’s more than the germ of an idea – they’ve shown that it can be done,” White says. “What we showed was proof of principle. What they show is proof of realization.”