Do earthquakes produce signals before they rupture? Maybe, scientists say

A new study suggests a warning sign may be available two hours before an earthquake strikes. An independent analysis suggests otherwise, and underscores the importance of rapid, open exchange of data and code among researchers.

By Rebecca Owen, Science Writer (@beccapox)

Citation: Owen, R., 2023, Do earthquakes produce signals before they rupture? Maybe, scientists say, Temblor,

Living in seismically active regions means coexisting with the possibility of earthquakes and the damage that can ensue. Early warning systems may give few seconds’ notice via a text alert or push notification — and this crucial bit of time is enough to help residents know what’s happening and take protective actions. However, these notifications aren’t meant to predict earthquakes. The earthquake is already in progress when alerts are delivered. An earthquake prediction would need to get three things right: time, location, and magnitude. As of now, there is no way to predict an earthquake.

For half a century, researchers have been on the hunt for precursory signals that foretell where an earthquake will occur in the minutes, hours, or days before the ground begins to shake — just one of the three necessary ingredients.

A new study published in Science suggests that a precursory signal might exist, and it could be tracked through movements recorded by GPS data in the two hours before a large quake. However, improved early warning systems — and earthquake prediction — are still a long way from reality.

Reflected and direct signals reaching a GPS antenna. Credit: Lucia Bellino & Zach Courter via EarthScope
Reflected and direct signals reaching a GPS antenna. Credit: Lucia Bellino & Zach Courter via EarthScope

A signal in the noise?

The study’s authors started by assuming that earthquakes begin with a precursory event — some seemingly imperceptible slow slippage of the fault that can’t be felt on the surface, but whose motion could be detected by GPS.

“How do you go from no slip to a seismic slip? Is it something that happens instantaneously or does the acceleration process take some time that you can maybe observe? Our study was a way to investigate these questions,” says Quentin Bletery, co-author of the study and a geophysicist at Cote D’Azur University in France.

GPS stations around the world record their exact locations every five minutes. Bletery and his colleague, Jean-Mathieu Nocquet, compiled data from the Nevada Geodetic Laboratory, selecting GPS stations located near 90 recorded earthquakes of magnitude 7.0 and above that occurred within the past twenty years. They compiled 3,026 snapshots of each time series in the 48 hours before an earthquake struck and stacked the five-minute increments of data together to look for trends that would otherwise be lost in the noise for individual events.

In the 46 hours before an earthquake, no discernible patterns emerged. “It is pretty chaotic and it looks random because it is dominated by the noise of the GPS data,” says Bletery. “But in the last two hours, it looks like there is something going on that suggests that the fault starts slipping, accelerating just before the event.”

The GPS sensors seemed to be indicating increasing fault slip in the two hours before an earthquake. As a control case, the researchers also looked at 100,000 random 48-hour time series when no known earthquakes occurred to verify that their findings were correlated with earthquake activity. They found similar precursory patterns in the data only 0.03% of the time in the control, suggesting that their findings may indicate an elusive precursory signal after all.

U.S. Geological Survey GPS Station called “The Husband” located in Three Sisters, Oregon. Credit: Dave Sherrod, USGS, Public domain, via Wikimedia Commons
U.S. Geological Survey GPS Station called “The Husband” located in Three Sisters, Oregon. Credit: Dave Sherrod, USGS, Public domain, via Wikimedia Commons


“This study really was different in that it tried to grab all the data to find coherent and consistent features,” says Roland Burgmann, a geophysicist and geodesist at UC Berkeley, who wrote a corresponding perspective on the study. “If every earthquake is somewhat different in size, duration, and maybe even a different process — that doesn’t help if you want to predict earthquakes. They looked for a common theme.”

Studying the past to predict the future

Without accurate means of prediction, earthquake research relies on studying the past. “The authors could not have done this study if they hadn’t known when, where, and how big the earthquakes were,” Burgmann says. In other words, their study was retrospective; we still cannot make any precursory announcements before an actual earthquake.

Patterns in activity — like foreshocks, a relatively rare precursory feature of earthquakes which can occur minutes, hours, days, or months ahead of an event — can be gleaned only after the fault has ruptured, buildings have been destroyed, or lives have been lost. That is because no diagnostic feature of foreshocks distinguishes them from run-of-the-mill earthquakes. And so, despite some remarkable foreshock-mainshock sequences, no one has learned how to use them as precursors. For instance, the 2011 Tohōku Earthquake in Japan was one of the earthquakes in the researchers’ dataset, and foreshocks shook that region in the weeks to days before the main shock. Yet, the magnitude and timing of the devastating mainshock was a surprise.

“We already know that many large earthquakes have precursory behavior. So it isn’t surprising that we can look back retrospectively and find signals that seemed to predict the upcoming earthquakes,” says Kyle Bradley, a geologist who was not part of the study published in Science. ”What has proved more elusive is finding ways to actually tell when precursory behavior really heralds an upcoming large quake.”


Earthquake precursors? Not so fast

Bradley co-authored an analysis of the research that, in the end, offered a differing perspective on how to interpret the GPS data. (This analysis has not been peer reviewed.) Along with co-author Judith Hubbard, Bradley used the information provided in the Science publication to replicate — and build on — the authors’ findings by removing what’s called “common mode noise,” which is noise that’s common to most GPS stations in a given network.

Their efforts showed no two-hour precursory slip and no clear signal indicating that an earthquake was imminent. Instead, they found that the supposed signal was simply GPS noise — not an indicator of an impending earthquake.


“This high level of transparency makes it easy for people to understand what really goes into the calculations and conclusions,” says Bradley, referring to the Science paper’s complete methods description and data repository. Even though the results differed, access to the code and data in Bletery’s article means that the conversations and findings surrounding earthquake precursory signals will continue openly as researchers dig into the data.

Paul Segall, a geophysicist at Stanford University, who was not involved in the original study or subsequent analysis, says that with an idea so provocative as a precursory earthquake signal, scientists “would jump on it, try to test it, and that’s exactly what happened. That’s how we want science to work.”

Monitoring subtle movements in real time

Earthquake prediction is still an impossibility. But, researchers can use better GPS data, more accurate measurements, and more substantial monitoring to gather information and continue to identify whether subtle movements of the earth do, indeed, precede an earthquake. Different instrumentation, like strainmeters, could be used to verify information relayed by GPS, suggests Burgmann. Such equipment would also be useful to monitor the ocean floor, where many subduction zones are located and continuously operating GPS stations are not.

“Without high precision instrumentation very near the focal region of a future earthquake, it will be hard, if not impossible, to detect precursors,” says Segall. “Bletery [and Nocquet’s] paper tried another approach, using regional GPS stations from a large number of quakes to overcome this problem.” The data, he says, are still useful for studying what happens during and after an earthquake, “even if they don’t reveal precursors.”


Bletery, Q., & Nocquet, J. M. (2023). The precursory phase of large earthquakes. Science, 381(6655), 297-301.


Text © 2023 Temblor. CC BY-NC-ND 4.0

We publish our work — articles and maps made by Temblor — under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license.

For more information, please see our Republishing Guidelines or reach out to with any questions.