Could deep-learning algorithms help us prepare for the Big One?
Apr 27, 2022

Could deep-learning algorithms help us prepare for the Big One?

Researchers have created tools to get us closer to predicting earthquakes, but Caltech's Egill Hauksson says that in this field, putting research into practice can be a challenge.

Contrary to what you might have seen in the Rock’s movie “San Andreas,” we still can’t predict earthquakes. But there have been recent advances in seismology assisted by artificial intelligence.

Researchers at Stanford used a deep-learning algorithm to detect more earthquakes in cities by filtering out the normal noise and vibrations of urban life. And a group at Pennsylvania State University used machine learning to analyze simulated fault movements in the lab and look for indicators that could help predict an impending quake.

I spoke with Egill Hauksson, a research professor of geophysics at the California Institute of Technology, who said these tools can give us a fuller understanding of earthquake patterns. The following is an edited transcript of our conversation.

A headshot of Egill Hauksson, research professor of geophysics at Caltech, wearing a pair of glasses and a blue shirt.
Egill Hauksson (Courtesy Caltech)

Egill Hauksson: The current algorithms detect very well the bigger events, but we are missing the smaller events. So you can think of the bigger events as being the bricks in a wall and the smaller events being the mortar in between the bricks. And understanding what’s happening in the mortar is very important for overall understanding of earthquakes.

Meghan McCarty Carino: What are some of the challenges for using research like Stanford and Penn State’s to actually apply to real-life earthquakes?

Hauksson: Often, taking this step of putting something into practice that was previously research is quite challenging. In part, for earthquakes we need to analyze all the data [in] real time, as opposed to in research you can take our current data and analyze it and come up with some very important and interesting findings. If you want to apply it real time, you have to have different specific algorithms and specific technology analyzing the data to ensure that you get the results as fast as possible.

McCarty Carino: What’s so hard about studying this kind of stuff? Like, what makes it so difficult to predict?

Hauksson: You can think of an earthquake starting at a depth of about 6 to 8 miles underground, and the only thing we really got is seismic waves. Some of the pieces that we are missing is we don’t know exactly what are the forces that are being applied to the fault and make it break. We don’t know the exact role of friction. So, those are some of the bigger challenges with understanding how earthquakes happen.

McCarty Carino: If you were going to bet on things, what do you think: Will we be able to predict earthquakes someday based on what you’re seeing now?

Hauksson: I don’t think so. You can think of it as a barber who has a barbershop. He knows there’s going to be 100 people walking in the door, say, this week. But he can’t tell you how many are going to walk in today or tomorrow. So it’s sort of one of these statistical problems that it’s, in a way, fairly randomized, but we know from our understanding of how the plates move on the surface of the Earth, the tectonic plates, that we will have earthquakes. But when and where things will break is, in my mind, impossible to predict.

A person checks the "MyShake" app on their smartphone in Hollywood on October 17, 2019.
In 2019, California launched the MyShake app, an earthquake detection and warning system, in the hope that residents will be alerted within seconds of a possible disaster and can “drop, cover and hold on.” (Chris Delmas/Getty Images)

Related links: More insight from Meghan McCarty Carino

Here’s more on the two research papers focused on deep learning, the one from Penn State on simulated earthquakes and the Stanford one on denoising cities.

Funny story: I once worked at a radio station right next to a light rail line, and for my first week or so every time a train went by I thought it was an earthquake for a split-second. But eventually my brain sort of blocked it out.

That’s basically what this deep-learning algorithm does, according to MIT Technology Review. It makes it possible to separate out what’s actually a small earthquake from a train going by.

And it just so happens that Tuesday was National Richter Scale Day. That, of course, is used to measure an earthquake’s magnitude, or at least that’s what this California kid thought. Apparently, seismologists don’t use it anymore.

They’ve replaced it with something called the moment magnitude scale, which, unlike the Richter scale, can measure quakes above magnitude 8. It’s another thing the movie “San Andreas” got wrong, in addition to scientists predicting earthquakes. But it’s still a great flick. I mean, come on. The Rock and earthquakes — it’s a perfect geological pairing.

The future of this podcast starts with you.

Every day, the “Marketplace Tech” team demystifies the digital economy with stories that explore more than just Big Tech. We’re committed to covering topics that matter to you and the world around us, diving deep into how technology intersects with climate change, inequity, and disinformation.

As part of a nonprofit newsroom, we’re counting on listeners like you to keep this public service paywall-free and available to all.

Support “Marketplace Tech” in any amount today and become a partner in our mission.

The team

Daniel Shin Producer
Jesús Alvarado Associate Producer