Researchers have been working on a number of new technologies, including satellites, underwater seismic sensors and software to detect earthquakes and even predict them. One new tool unveiled in the journal Science Advances last week looks to be very promising.
Drawing on speech recognition technology, researchers developed a convoluted neural network called ConvNetQuake that detects even the tiniest tremor against geological background noise, much in the same way the human ear can detect someone speaking in a moving car by tuning out vehicle background noise.
Preparing for the study
The study was led by Thibaut Perol, a senior deep learning scientist at Gram Labs, an AI startup based in Washington, DC. Along with his collaborators, Massachusetts Institute of Technology graduate student Michaël Gharbi and Harvard assistant professor Marine Denolle, Perol trained the network with real seismic data from 2014 to 2016.
As the authors point out in their study, most earthquake detection methods are designed for moderate and large earthquakes, with only nine earthquakes of magnitude greater than 5.0 in Oklahoma that might have been triggered by nearby disposal wells between 2008 and 2017.
This means that most low-magnitude earthquakes are being missed. And small earthquakes occur much more frequently than most people realize. In Southern California, roughly 10,000 earthquakes occur each year, most of them barely perceptible.
And what makes ConvNetQuake particularly useful is that it does not require triangulation, like traditional seismic monitors. “What people used to do is they would use a lot of seismic stations. You don’t need three stations anymore. You can do it with one,” said Perol.
Perol also notes there is an important distinction between actual earthquakes and seismic noise. Noise can be caused by anything from a tractor rolling by to ocean waves. “If you try to record someone talking inside a car, the noise made by the car is quite constant,” Perol explains. And this background noise is what covers up small earthquakes.
The ConvNetQuake algorithm builds on recent advances in deep learning and was trained on a large dataset of labeled raw seismic waveforms that helped it to learn a compact representation that can discriminate seismic noise from earthquake signals.
According to the authors, “It is more accurate than state-of-the-art algorithms and runs orders of magnitude faster. In addition, ConvNetQuake outputs a probabilistic location of an earthquake’s source from a single station.”
Study findings
Because seismic activity in Oklahoma has been correlated with increased fracking activity, ConvNetQuake was used in a particularly active area near Guthrie, OK. In this region, the Oklahoma Geological Survey (OGS) cataloged 2021 seismic events from 15 February 2014 to 16 November 2016.
GS.OK029 and GS.OK027 are the two stations that record continuously the ground motion velocity. The colored circles are the events in the training data set. Each event is labeled with its corresponding area.
The ConvNetQuake system detected 17 times more earthquakes than were recorded in the Oklahoma Geological Survey earthquake catalog.
“The reason we’re interested in detecting and locating the small ones is we can get an idea of how seismic activity evolves,” Perol says. “The software will detect tons of earthquakes. Basically, all of this will be done in real time, and all of this information could be given to scientists so they could assess the risk faster.”
While ConvNetQuake cannot predict when the next “big one” will hit, it is superior to conventional methods being used today. ConvNetQuake can potentially provide very rapid earthquake detection and location, which is useful for earthquake early warning.
And as the study points out, it could easily be used to monitor geothermal systems, natural resource reservoirs, volcanoes and seismically active and well-instrumented plate boundaries such as the subduction zones in Japan or the San Andreas Fault system in California.
