A giant precursor to a giant earthquake may have been detected by gravity satellites

By Ross Stein, Ph.D. and David Jacobson, M.Sc., Temblor

Click here to protect yourself from natural disasters

Spanish Version

French Version

The devastating Tohoku earthquake in 2011 and the subsequent tsunami caused damage valued at billions of dollars and the death of thousands of people. A new study, published this month finds a large precursory gravity change over the Japanese archipelago starting a few months before the M=9.0 Tohoku earthquake. (Photo from SFDEM)


A study by Isabelle Panet and her French colleagues published this month in the high-profile journal, Nature Geoscience, finds a large precursory gravity change over the Japanese archipelago starting a few months before the M=9.0 Tohoku earthquake. The signals were processed from the GRACE (Gravity Recovery and Climate Experiment) satellite duo, launched in 2002. The authors found nothing comparable to the precursory signal in the preceding seven years. They also found a large gravity signal associated with the Tohoku earthquake itself, and with the post-seismic period, as others have previously published, using the same data – exceeding in size, in March 2011, the previously known GRACE signals. Because of its uniqueness and importance, the Panet et al. analysis was subjected to substantial, protracted, and skeptical review, and so was exhaustively tested, and modeled. Its publication does not mean that it’s right, but that it has not been proven wrong.

This map shows the gravitational gradient around the location of the Tohoku earthquake prior to, during, and after the quake. mEötvös are the gravitational gradient of the Earth, that is, the change in the gravitational acceleration vector from one point on the Earth’s surface to another. ‘Pac’ is the Pacific Plate, ‘PHS’ the Philippine Sea plate, and ‘Eur’ the European plate. In the first two frames, the megathrust slab depth contours are every 200 km, while in the third, they are every 100 km. (Figure modified from Panet et al., 2018)


Now come four critical ‘If’s

If the result is reproducible by others; if a similar precursory signal is found for at least some other megathrusts events, such as the 2004 M=9.2 Sumatra, 2010 M=8.8 Maule, Chile, or 2005 M=8.6 Nias–Simeulue, Indonesia, earthquakes; if no similar signal is seen where large quakes do not follow; and if the GRACE data could be processed in near-realtime to monitor and search for such anomalies before great quakes strike—then this could truly be a breakthrough. That is an exceedingly high bar, one that virtually no other earthquake prediction scheme has ever been able to hurdle. But it is not impossible, and when one considers the death toll tusnamis, and to a lesser extent, shaking, from these goliaths extract, it would be a breakthrough.


What is being measured here?

A megathrust earthquake slips a gently inclined fault by 20-60 m (70-200 feet) over a very wide zone that can be hundreds to almost a thousand kilometers long, and hundreds of kilometers wide. This occurs where an oceanic plate is thrust under a continent or island arc in a process geologists call ‘subduction.’ In the earthquake, the crust of the earth is suddenly thickened or ‘stacked’ offshore, and so locally the mass increases. If you were to weigh yourself offshore Tohoku before and after the M=9, you’d weigh a bit more after the quake, because your weight is the product of your mass and the mass of the sliver of earth beneath you, which is now greater. Onshore, the crust is thinned and stretched, and so you would weigh less. The GRACE satellites detect mass redistributions within the Earth and its fluid envelopes by sensing the associated changes in the gravitational field.

The original twin GRACE Earth science spacecraft, launched in 2002 on a five-year mission, were retired in November 2017, more than 15 years after they were launched. It’s successor will be launched on May 19, 2018. These satellites speed up and slow down relative to each other due to changes in mass in the Earth beneath them. These tiny accelerations, as they come together and pull apart, change their relative distance, which is measured by laser or microwave ranging systems and then converted to gravity field variations of the Earth below. The GRACE satellites make repeated passes over the same spot, so changes in gravity can be detected in time, every few weeks. (Image from: NASA)


What could go wrong with these observations?

Apart from this study, the best evidence suggests that great earthquakes nucleate and rupture just like small ones, but for unknown reasons, the big ones just keep rupturing. The speed at which the rupture grows is almost the same for a M=7 and M=9 quake, the only difference being that the M=7 starts to peter out after about 10 seconds, and for a M=9, event this may not begin until 60 seconds or more. For Panet et al. to be right, this view of rupture process must be wrong, because there would be preconditions that promote the occurrence of a much larger event; some events would know they are going to be great.

This figure, from Meier et al. (2017), suggests that no little earthquake knows it’s marked for future greatness. They all start growing at about the same rate, but the small ones fade sooner. The ‘Moment’ is a measure of quake size or energy.


There is an enormous amount of signal processing to reach these results, some of which is tuned to the orientation and scale of the subduction zone. This is why reproduction of their results by independent investigators is so important. The Asian monsoon rains add mass to the Earth’s surface, as does the shrinking and growing of snow and ice. Storms and currents alter the mass of the ocean offshore. All of these signals can last months, and so must be fully removed. But while there is a remaining monsoon signal in eastern China in the Panet et al. results in their August 2010 and February 2011 maps, these are smaller and more isolated than the precursory anomaly, which persists over time. When Panet et al. stacked all of the signals over 2006-2009, no anomaly as large as the precursory feature is seen. And when they look at a huge area centered on the anomaly during 2011, this one feature stands out by its size and intensity.

The left globe in this figure shows cumulated July–February anomalies stacked over 2006–2009. The right globe shows the zoomed out area from the first panel in the first figure, which is the time just prior to the M=9.0 Tohoku earthquake. (Figure from Panet et al., 2018)


If real, what is the cause of the anomaly?

The authors argue that the descending slab extended or stretched in the earth’s mantle. If the source of the precursory mass change is hundreds of kilometers deeper than the eventual earthquake rupture, and if it is not too localized, it would explain why the anomaly is even broader than the co-seismic signal. Further, the source would have to be very deep for the deformation to have evaded detection by Japan’s unmatched GPS monitoring network. There is a subtle GPS excursion before the Tohoku quake (Mavrommatis et al. 2014), but it took about a decade to occur and so would seem to be unrelated.


What next?

In short order—perhaps a year—we will know if this fizzles, or is the beginning in a new chapter in our understanding of great quakes, and possibly, a new seismic sentinel to protect the public. The chances of that are small, but that is what science is for, and that is what makes it exciting.

Isabelle Panet from the Institut National de l’Information Géographique et Forestière and Université Paris Diderot, Sorbonne, and coauthor Clément Narteau, from the Institut de Physique du Globe de Paris, and Université Paris Diderot, Sorbonne, photographed on the rooftop of the IPGP building by Ross Stein last week.



Mavrommatis, A. P., P. Segall, and K. M. Johnson (2014), A decadal-scale deformation transient prior to the 2011 Mw9.0 Tohoku-oki earthquake, Geophys. Res. Lett., 41, 4486–4494, doi:10.1002/2014GL060139.

M.-A. Meier, J. P. Ampuero, T. H. Heaton, The hidden simplicity of subduction megathrust earthquakes, Science 357, 1277–1281 (2017) 22 September 2017

Isabelle Panet, Sylvain Bonvalot, Clément Narteau, Dominique Remy and Jean-Michel Lemoine, Migrating pattern of deformation prior to the Tohoku-Oki earthquake revealed by GRACE data, Nature Geoscience | VOL 11 | MAY 2018 | 367–373

  • Tiegan Hobbs

    A great article presenting the “what if’s” of earthquake precursors! Really neat result.

  • Suppose it pans out. What about the timing ? Authorities could say there’s a 90% chance of a 8.0 earthquake here within X days/hours. What then? Mass evacuation? Advisory? Thoughts, anybody?

  • Ross Stein

    A several-month period would be sufficient to temporarily close collapse-risk buildings, to undertake some engineering retrofits, to institute enhanced safety measures, rehearse emergency response, strengthen hospital preparations, and cause people to change travel plans, and consider evacuations. But we are getting ahead of ourselves.

  • Ross Stein

    Thank you, Tiegan.

    To make predictions, one must establish three metrics: the successful quake prediction rate (prediction then quake), the missed quake rate (quake but no prediction), and the false alarm rate (prediction but no quake). Predictions are an optimization between the missed event rate and the false alarm rate.

    Once those metrics are established retrospectively from GRACE, then prospective monitoring with the new satellite becomes key. If the warning is two months long but it takes three months to process the data, it still won’t work. So ultimately, routine processing will be needed.

    This is a noble experiment that needs to be undertaken.