How Robots Are Helping Scientists Predict Earthquake Aftershocks

Researchers analyzed hundreds of thousands of earthquakes to predict the location of aftershocks using machine learning to explore changes in ground stress.

By Russ Banham, Contributor

Recent advances in computing power, deep learning, and physics simulation software present the possibility of mitigating the impact of an earthquake. This is good news in light of research that predicts a cataclysmic earthquake will strike the Western United States in the near future.

Scientists at Harvard University and the Massachusetts Institute of Technology (MIT) are approaching the urgent need to minimize earthquake damage from separate fronts. Harvard researchers are focused on the use of deep learning algorithms to posit where earthquake aftershocks are likely to occur, while MIT researchers have developed a seismic wave diversion structure on a supercomputer that draws the destructive waves away from protected areas like neighborhoods and downtown business centers.

The ingenuity of the technologies is equaled by their timing. Scientists at the United States Geological Survey (USGS) predict a 99.7 percent chance of a 6.7-magnitude earthquake striking Los Angeles within the next 30 years. That’s the same magnitude of the 1994 Northridge Earthquake that killed 72 people, injured more than 10,000 others, and destroyed thousands of homes, buildings, and cars in the surrounding region, costing more than $40 billion in property damage.

The prognosis is even worse for residents of the Pacific Northwest coastal region, home to the 620-mile long Cascadia Subduction Zone, where the Juan de Fuca ocean plate dips under the North American continental plate. Seattle and Portland, both inside the zone, confront an eight to 20 percent chance of experiencing the “Big One,” what seismologists call a full-margin rupture resulting in a magnitude 8.7 to 9.2 earthquake.

“We’re hoping to do our part in reducing the devastating impact of these seismic events.”—Robert Haupt, senior technical staff scientist, MIT’s Lincoln Laboratory

Most earthquakes are nowhere near as catastrophic. Of the half a billion or so detectable earthquakes that occur across the world each year (of which about 100,000 of them are felt), roughly 100 cause significant property damage and potential loss of life, noted Robert Haupt, senior technical staff scientist at MIT’s Lincoln Laboratory. “We’re hoping to do our part in reducing the devastating impact of these seismic events,” he says.

A Seismic Muffler

Haupt is a key architect of what Lincoln Lab is calling a “seismic muffler.” The concept calls for drilling a V-shaped array of sloping boreholes hundreds of feet deep on both sides of the structures to be protected, such as power plants, airport runways, office buildings, and other protected assets.

The one- to three-feet diameter boreholes—which are cased in steel and look like a set of trench walls from above—divert hazardous surface waves generated by an earthquake away from the protected asset. By the time this destructive wave force reaches ground surface, it dissipates—much like the acoustic energy coming from a combustion engine of an automobile is softened by the car’s muffler.

Haupt and his scientific colleagues at the lab have successfully tested a variety of borehole spacing models to dissipate hazardous seismic waves. In this work, they used 3D high-performance supercomputers and physics-based simulation software, such as SPECFEM3D seismic wave propagation software and COMSOL Multiphysics Acoustic software. The team uses artificial intelligence (AI) to sift through incredibly large data volumes involving earthquake detection probability. “There’s no way you can plot up all the multiple dimensions using spreadsheets or pen and paper,” Haupt says.

The findings were impressive. “We performed a series of tabletop exercises on 3D supercomputers that indicated a V-shaped array of mufflers can decrease the ground shaking effects of a 7.0-magnitude earthquake to a 5.5-magnitude earthquake and lower,” Haupt explains. “That’s a pretty significant reduction in ground motion.”

He’s not kidding. According to the USGS, a 7.0-magnitude temblor is 177.8 times stronger (energy release) than a 5.5-magnitude quake. Lincoln Lab recently received a patent for its innovative technology and is in licensing talks with several interested parties, whose names Haupt declined to disclose. Field testing is expected to be underway this summer.

Regarding the expense, Haupt estimates the cost would be less than what real estate developers currently pay to secure a skyscraper with base isolation systems, in which spring-like pads are inserted between a building’s foundation and a building to absorb ground motions. “Based on extensive 3D supercomputer computations, we believe it would protect a lot more buildings at about the same cost,” he says.

Location, Location, Location

Harvard University’s scientists have focused their research on the location of earthquake aftershocks, which follow the main shock and can occur for weeks, months, and even years after the primary event. Although smaller in magnitude than the initial temblor, some aftershocks pack a wallop—the case with a 2015 magnitude 6.7 aftershock recorded in Nepal.

While scientists are able to calculate the magnitude of aftershocks with some degree of precision, they’ve struggled with predicting their location. To improve the odds, Harvard research scientists Brendan Meade, a professor of earth and planetary sciences, and Phoebe DeVries, a postdoctoral fellow working in Meade’s lab, collected a massive volume of data from more than 130,000 earthquakes worldwide. Using AI technology, they analyzed this database to discern where the aftershocks had occurred, mapping them across a series of 5-kilometer square grids.

The researchers then compared the findings with a physics-based computer model calculating the stresses and strains of the Earth during the main shock. The model incorporated deep learning algorithms to ferret out specific correlations between the strains and stressors and the aftershock locations.

The next stage of the research, published in the scientific journal Nature, called for testing the outcomes against 30,000 mainshock and aftershock events. The results were promising, encouraging the value of pursuing additional research. Thanks to AI, as Meade told The Harvard Gazette, “Problems that are dauntingly hard are extremely accessible these days.”

That’s exceedingly good news for anyone living in an earthquake-prone region. And hopefully just in time, too.

Russ Banham is a Pulitzer-nominated financial journalist and best-selling author.