Machine learning helped demystify a California earthquake swarm

Circulating groundwater triggered a four-year-long swarm of tiny earthquakes that rumbled beneath the Southern California town of Cahuilla, researchers report in the June 19 Science. By training computers to recognize such faint rumbles, the scientists were able not only to identify the probable culprit behind the quakes, but also to track how such mysterious swarms can spread through complex fault networks in space and time.

Seismic signals are constantly being recorded in tectonically active Southern California, says seismologist Zachary Ross of Caltech. Using that rich database, Ross and colleagues have been training computers to distinguish the telltale ground movements of minute earthquakes from other things that gently shake the ground, such as construction reverberations or distant rumbles of the ocean (SN: 4/18/19). The millions of tiny quakes revealed by this machine learning technique, he says, can be used to create high-resolution, 3-D images of what lies beneath the ground’s surface in a particular region.

In 2017, the researchers noted an uptick in tiny quake activity in the Cahuilla region that had, at that point, been going on for about a year. Most of the quakes were far too small to be felt but were detectable by the sensors. Over the next few years, the team used their computer algorithm to identify 22,000 such quakes from early 2016 to late 2019, ranging in magnitude from 0.7 to 4.4.

Such a cluster of small quakes, with no standout, large mainshock, is called a swarm. “Swarms are different from a standard mainshock-aftershock sequence,” which are typically linked to the transfer of stress from fault to fault in the subsurface, Ross says. The leading candidates for swarm triggering come down to groundwater circulation or a kind of slow slippage on an active fault, known as fault creep.

“Swarms have been somewhat enigmatic for quite a while,” says David Shelly, a U.S. Geological Survey geophysicist based in Golden, Colo., who was not connected with the study. They are particularly common in volcanic and hydrothermal areas, he says, “and so sometimes, it’s a bit harder to interpret the ones that aren’t in those types of areas,” like the Cahuilla swarm (SN: 5/14/20).

“This one is particularly cool, because it’s [a] rare, slow-motion swarm,” Shelly adds. “Most might last a few days, weeks or months. This one lasted four years. Having it spread out in time like that gives a little more opportunity to examine some of the nuances of what’s going on.”

Data from the Cahuilla swarm, which is winding down but “not quite over,” Ross says, revealed not only the complex network of faults beneath the surface, but also the evolution of the fault zone over time. “You can see that the sequence [of earthquakes] originated from a region that’s only on the order of tens of meters wide,” Ross says. But over the next four years, he adds, that region grew, creating an expanding front of earthquake epicenters that spread out at a rate of about 5 meters per day, until it became about 30 times the size of the original zone.

That diffusive spread, Ross says, suggests that moving groundwater is triggering the swarm. Although the team didn’t directly observe fluids moving underground, the scientists speculate that beneath the fault zone lies a reservoir of groundwater that previously had been sealed off from the zone. At some point, that seal broke, and the groundwater was able to seep into one of the faults, triggering the first quakes. From there, it moved through the fault system over the next few years, triggering more quakes in its wake. Eventually, the seeping groundwater probably ran up against an impermeable barrier, which is bringing the swarm to a gradual halt. 

Being able to identify what causes such mysterious events is extremely important when it comes to communicating with people about earthquake hazards, Ross says. “Typically, we have very limited explanations that we can provide to the public on what’s happening,” he says. “It gives us something that we can explain in concrete terms.”

And this discovery, he adds, “gives me a lot of confidence” to continue to apply this technique, such as on the last 40 years of amassed seismic data in Southern California, which likely contains many more previously undetected swarms.

The study highlights how seismologists are increasingly acknowledging the importance of fluids in the crust, Shelly says. And, he adds, it emphasizes how having so many tiny quakes can illuminate the hidden world of the subsurface. “It’s kind of like having a special telescope to look down into the crust,” he adds. Combining this wealth of seismic data with machine learning is “the future of earthquake analysis.”