Can We Predict Earthquakes in the Future? A Controversial Debate in the Scientific Community

[Un article de The Conversation écrit par Quentin Bletery – Géophysicien, Directeur de Recherche IRD à l”Observatoire de la Côte d’Azur., Institut de recherche pour le développement (IRD)]

The earthquakes and the tsunamis they generate caused the death of nearly a million people in the past twenty-five years. Alert systems exist, but they are based on the first signals issued by the earthquake, and therefore provide only a few alert before the first tremors, a few tens of minutes before the possible tsunami. If significant progress is made to improve these systems, these are intrinsically limited in terms of alert time, because they only use signals issued by an earthquake already initiated.

To alert more than a few seconds in advance, it will therefore be able to be able to predict the earthquakes before they are triggered. Can we one day achieve this goal?

Humans have sought to predict earthquakes for a long time. In the 1970s, many scientists thought that this goal was at hand. In California, an earthquake seemed to occur every twenty-two years, which pushed seismologists to predict that an earthquake would occur, according to this recurrence scheme, in 1988. But the announced earthquake did not take place until 2004, and before the failure of this prediction, the scientific community became, in the 1990s, more and more skeptical to the idea of ​​the predictability of the earthquakes. This event made the idea of ​​predicting the “taboo” earthquakes for many years, especially in the United States, where any mention of the term “prediction” has been prohibited. Today it remains considered impossible by a large part of the community.

However, experimental and theoretical work suggests the existence of an earthquake preparation phase. The seismological data (which record the soil vibrations) and geodesic (which record the movements of the soil) are growing strongly and gives us more and more thin information on what is going on in depth. Under these conditions, will the prediction of earthquakes remain impossible forever?

What is an earthquake?

An earthquake is a rapid shift between two “earth blocks” along the interface that separates them: a fault. The greatest “earth blocks” are known as tectonic plates. These move slowly (a few millimeters a year) compared to each other, but their relative movement is largely blocked along the faults that separate them. The faults being blocked, and the plates that do not stop moving, a “slip deficit” accumulates. This deficit is compensated, in a few seconds, during rare but violent events: earthquakes.

But then, could it be that the rapid shift occurring during earthquakes begins with a slow sliding accelerating until the seismic rupture? This is shown by the experiences carried out in the laboratory and what physical models predict.

Experiments and models suggest the existence of a preparatory phase of earthquakes during which the slow shift accelerates and gradually stretches until reaching a critical size from which an earthquake is triggered. This critical size is very low in most of the experiments made in the laboratory, because these experiences are made on “flaws” very small perfectly flat and homogeneous sizes.

Theoretical work has shown that this critical size can be much greater and the duration of the much longer (and potentially detectable) preparatory phase) for heterogeneous flaws which are much more realistic than the flat homogeneous flaws in laboratory experiences.

Do we observe signals before the earthquakes?

If we observe many pre-seismic signals in the laboratory, this is much more difficult on natural flaws. The reason is simple: the sensors are much further. Researchers mainly use two types of sensors: seismological sensors and geodetic sensors. Seismological sensors (seismometers) record soil vibrations, while geodetic sensors (such as GPS) measure the movements of the soil.

To search for slow sliding clues along the faults, it seems natural to look at GPS data. Unfortunately, when we look at the movements measured by GPS in the hours, or days, preceding the big earthquakes, we see essentially what is called noise, that is to say signals which do not have a tectonic origin, but which are due to correction errors in the processing of GPS data. This noise masks any potential pre-seismic signal which would necessarily be low amplitude. How could we thus extract a signal of low amplitude “drowned” in rough data?

We proposed an approach, in an article published in 2023 in the newspaper Scienceto increase the signal ratio on noise and bring out of potential low -amplitude signals. The approach does not aim to predict earthquakes, but to explore the existence of weak signals in ” stackant (That is, simply, by adding) all the GPS data recorded before all the major earthquakes.

Of course, brutally adding all the data would not make sense, because depending on the type of earthquake and the source-station configuration the movements potentially generated by a hypothetical pioneer shift will not all be in the same direction. We therefore calculated all the displacements expected on each station before each large earthquake, then we calculated the scalar product between the expected trips and the movements observed.

The scalar product is a measure of consistency between the expected and observed displacements. If the displacements observed are more or less in the same direction as the expected displacements, their scalar product will be positive. Otherwise, it will be negative. Thus, if GPS measures contain only noise unrelated to a pre-seismic signal potential, scalar products have an equal probability of being positive or negative, and summoning a large number of these should give a result close to 0. On the contrary, if GPS measures contain a low pre-seismic signal, we expect, summarizing a large number of scalar products, to obtain rather positive values.

Thus, we have summoned the scalar products of the expected displacements and observed every five minutes during the 48 hours preceding all the events of magnitude greater than 7. This represents a total of 3,026 GPS time series recorded before 90 earthquakes (Figure 1).

The result is a temporal series describing consistency between the expected displacements and the displacements observed according to the time before the major earthquakes (Figure 2).

This temporal series shows an increase in consistency between expected displacements and observed within two hours preceding the earthquakes, which could be the trace of an acceleration of the slow pre-seismic shift leading to the seismic rupture. The signal is subtle, but we have reproduced the exercise on data recorded at 100,000 different dates (not preceding earthquakes) and obtained a similar signal in only 0.3 % of cases, making it a very significant statistically signal.

Pre-seismic signal or noise?

Given the potential implications, we have published with the article all our codes and all our data so that the scientific community can verify and work on alternative approaches, from which we would not have thought.

Four days after the publication of the article, two American scientists, converted into popularization, published, on their blog, an article in doubt concerning the tectonic origin of the signal. This article had a strong media echo, due to the activity of popularizer of the authors and the force of the advanced main argument: after having corrected GPS data of what the authors consider noise (fluctuations in the data they consider not linked to tectonic activity), the signal disappears from stack. The conclusion of the authors of this blog is that the signal is the result of an unlucky coincidence of improbable factors showing GPS noise as a tectonic signal.

We recently published an article in the newspaper Seismica showing that the probability of such a coincidence is extremely low, and advanced the hypothesis that the proposed noise correction can alter the detection of a real signal.

To date, the scientific community is shared and the debate around this question is more alive than ever. An online debate was even organized between the authors of the blog and ourselves before an audience of scientists (a first in the scientific community working on earthquakes). The terms of the debate are extremely technical and the outcome is uncertain.

The multiplication of GPS stations (and other geophysical instruments) provides an increasing number of observations as new earthquakes occur and augur the fact that, if the experts fail to determine the origin of the signal, the time will do so – the signal becoming more and more and less, clear as new data will be added to stack.

When will the prediction?

If the signal turns out to be the result of a unfortunate combination of correlated noise, then the prospect of the prediction will move away a little more. If the signal turns out to be proof of a phase of preparatory slip of the earthquakes, then this perspective will come closer. A little only, because, even in this hypothesis, the approach that we have proposed unfortunately will not be able to predict the earthquakes.

Indeed, this uses all the data recorded before all the past events by hypothesizing that the epicenter and the mechanisms of the earthquakes are known. She cannot thus have any predictive ambition.

We generally consider that a stack Amplifies the signal ratio at noise of a factor equal to the square root of the number of observations. In our case, that would mean that the signal identified in Figure 2 has been amplified by a factor 55 thanks to stack. It would therefore be necessary to increase the sensitivity (or reduce noise) of the GPS records of a factor (at least) 55 to be able to identify a signal at a single station.

Such progress represent major technological advances and are improbable in the years to come. However, if the existence of a preparatory phase (potentially observable) of earthquakes is confirmed, this would certainly motivate the development of new technologies and the deployment of dense stations networks which could make the prediction of earthquakes a not so distant perspective.