Aug 30 2018
Following a major earthquake, for many weeks or months, the surrounding area is usually prone to powerful aftershocks that can largely hinder the recovery efforts and leave an already damaged community reeling.
Although empirical laws, such as Ohmori’s Law and Bäth’s Law, have been developed by researchers to detail the probable size and timing of those aftershocks, it has been more challenging to grasp techniques for forecasting their location.
However, intrigued by a suggestion from scientists at Google, Brendan Meade, a Professor of Earth and Planetary Sciences, and Phoebe DeVries, a post-doctoral fellow working in his lab, have been employing artificial intelligence technology to make efforts to find a solution to the problem.
Both the researchers used deep learning algorithms to analyze a database of earthquakes from across the globe to make attempts to predict at which places the aftershocks could occur. As a result, they developed a system that, although being imprecise, could forecast aftershocks considerably better compared to random assignment. The study has been reported in a paper published in the journal Nature on August 30, 2018.
“There are three things you want to know about earthquakes - you want to know when they are going to occur, how big they’re going to be and where they’re going to be,” stated Meade. “Prior to this work we had empirical laws for when they would occur and how big they were going to be, and now we’re working the third leg, where they might occur.”
“I’m very excited for the potential for machine learning going forward with these kind of problems—it’s a very important problem to go after,” stated DeVries. “Aftershock forecasting in particular is a challenge that’s well-suited to machine learning because there are so many physical phenomena that could influence aftershock behavior and machine learning is extremely good at teasing out those relationships. I think we’ve really just scratched the surface of what could be done with aftershock forecasting ... and that’s really exciting.”
The concept of employing artificial intelligent neural networks to attempt to predict aftershocks first emerged a number of years ago, during the first of Meade’s two sabbaticals at Google in Cambridge.
Meade stated that while working on a related problem with a group of scientists, a colleague suggested that the then-emerging “deep learning” algorithms might render the problem more tractable. Meade later partnered with DeVries, who had been employing neural networks to transform high-performance computing code into algorithms with the ability to run on a laptop to focus on aftershocks.
“The goal is to complete the picture and we hope we’ve contributed to that,” stated Meade.
In order to achieve this, Meade and DeVries started by accessing a database of observations made after over 199 major earthquakes.
“After earthquakes of magnitude 5 or larger, people spend a great deal of time mapping which part of the fault slipped and how much it moved,” stated Meade. “Many studies might use observations from one or two earthquakes, but we used the whole database ... and we combined it with a physics-based model of how the Earth will be stressed and strained after the earthquake, with the idea being that the stresses and strains caused by the main shock may be what trigger the aftershocks.”
Using this information, they then separated an area found into 5-km square grids. In each grid, the system analyzes whether there was an aftershock and instructs the neural network to search for correlations between locations at which aftershocks occurred and the stresses produced by the main earthquake.
“The question is what combination of factors might be predictive,” stated Meade. “There are many theories, but one thing this paper does is clearly upend the most dominant theory—it shows it has negligible predictive power, and it instead comes up with one that has significantly better predictive power.”
According to Meade, the system pointed to a quantity called the second invariant of the deviatoric stress tensor—simply known as J2.
“This is a quantity that occurs in metallurgy and other theories, but has never been popular in earthquake science,” stated Meade. “But what that means is the neural network didn’t come up with something crazy, it came up with something that was highly interpretable. It was able to identify what physics we should be looking at, which is pretty cool.”
According to DeVries, this interpretability is highly significant since for a long time, artificial intelligence systems have been considered by many researchers to be black boxes—with the ability to produce an answer based on some data.
“This was one of the most important steps in our process,” she stated. “When we first trained the neural network, we noticed it did pretty well at predicting the locations of aftershocks, but we thought it would be important if we could interpret what factors it was finding were important or useful for that forecast.”
However, taking on such a challenge with largely complex real-world data would be a difficult task. Hence the researchers instead instructed the system to develop forecasts for synthetic, highly idealized earthquakes and then investigated the predictions.
“We looked at the output of the neural network and then we looked at what we would expect if different quantities controlled aftershock forecasting,” she stated. “By comparing them spatially, we were able to show that J2 seems to be important in forecasting.”
Meade stated that since the network was trained using earthquakes and aftershocks from across the world, the resulting system held good for various different types of faults.
“Faults in different parts of the world have different geometry,” stated Meade. “In California, most are slip-faults, but in other places, like Japan, they have very shallow subduction zones. But what’s cool about this system is you can train it on one, and it will predict on the other, so it’s really generalizable.”
“We’re still a long way from actually being able to forecast them,” she stated. “We’re a very long way from doing it in any real-time sense, but I think machine learning has huge potential here.”
According to Meade, in the future, his attempts are to predict the magnitude of earthquakes themselves with the help of artificial intelligence technology with the aim of helping to avert the destructive impacts of the disasters one day.
“Orthodox seismologists are largely pathologists,” stated Meade. “They study what happens after the catastrophic event. I don’t want to do that - I want to be an epidemiologist. I want to understand the triggers, causing and transfers that lead to these events.”
According to Meade, eventually, the purpose of the research is to underscore the potential for deep learning algorithms to answer questions that, to date, researchers barely knew how to ask.
“I think there’s a quiet revolution in thinking about earthquake prediction,” he stated. “It’s not an idea that’s totally out there anymore. And while this result is interesting, I think this is part of a revolution in general about rebuilding all of science in the artificial intelligence era.”
“Problems that are dauntingly hard are extremely accessible these days,” he added. “That’s not just due to computing power - the scientific community is going to benefit tremendously from this because ... AI sounds extremely daunting, but it’s actually not. It’s an extraordinarily democratizing type of computing, and I think a lot of people are beginning to get that.”