We hear it from earthquake experts all the time: there’s no way to predict when the next big earthquake will happen.
But could 21st Century technological advances change that?’
Some researchers, including those at Los Alamos National Lab in New Mexico, are hoping tools like artificial intelligence and machine learning could lead to a seismic breakthrough.
dr. Paul Johnson, a staff scientist in the geophysics group at Los Alamos, says researchers aren’t trying to do real earthquake prediction, which deals with magnitude location and time. Rather, they’re just looking to the timing of a quake, which he says is “hard enough.”
Using computers and a simulated earthquake machine at Penn State, researchers applied machine learning to document millions of data points created by seismic waves from its lab-created fault. They then performed experiments to see if some of the patterns that the computers found could help predict the next rupture.
“It worked almost immediately,” Johnson said. “And we were just stunned at how quickly it worked and how easily it worked on a problem that we had struggled with for so long.”
Researchers then tested the theory on a somewhat newly discovered type of real-world earthquake seen in the Pacific Northwest.
It’s an earthquake that nobody even feels. They’re called “slow slip earthquakes,” where one tectonic plate, slowly, over the course of a month or so, slips underneath another.
Johnson said the theory worked on those hard to detect earthquakes, too. He said the computers can warn of the next slow slip earthquake “immediately after” the previous one.
But Johnson warns scientists are still a long way from proving the models can predict other types of earthquakes, like the Northridge quake of 1994.
“For seismogenic earthquakes, the ones we really care about as citizens, we’re not there yet,” Johnson said.
Lucy Jones, founder of the Dr. Lucy Jones Center for Science and Society, says she’s heard every possible theory related to predicting earthquakes.
“Among the general public, we have a lot of thought about connecting to the lunar cycle and tides, hot weather, early morning, all of those,” Jones said. “We remember the earthquakes that fit the pattern and forget the ones that don’t because we want the pattern to be true.”
So far no theories have proven true, Jones said, and she believes the research will in the end – like past attempts – only reinforce the theory that big earthquakes are random and cannot be predicted.
“It’s important to do this study because we have this conundrum of, ‘is it random?’” Jones said. “Is it really random or is it random because it’s too complicated? And machine learning has a chance of helping us separate that out.”
Jones is not part of the research team at Los Alamos, but says she’s interested in seeing how the research plays out and expects there will be much debate on the subject for a long, long time.
“At this point, we still don’t know. We still argue about it,” Jones said. “It’s that point in science which makes it interesting.”
Suggest a correction