It is the largest earthquake in the history of Morocco. The center of the country was hit Friday evening, September 8, in the province of Al-Haouz, southwest of the city of Marrakech, by an earthquake measuring 7 on the Richter scale. The toll, still provisional and which continues to increase over the hours, lists at least 1,000 deaths and more than 1,200 injured, according to the authorities.
If the seismic risk is known in this region, particularly around the Atlas zone, the extremely rare intensity of this earthquake was not “predictable”, according to seismologists. Today, however, there are warning systems capable of detecting earthquakes and estimating their magnitude as quickly as possible. Some, using artificial intelligence (AI), would even be able to detect them even faster. Working on one of these warning systems, seismologist Quentin Bletery answers questions from L’Express.
L’Express: In an article for the university media The Conversation, you write that the prediction as such of earthquakes “is impossible”. For what ?
Quentin Bletery: Quite simply because science doesn’t yet know how to do it! Nothing special is observed before earthquakes occur. But two big fundamental questions arise: do we see nothing because there is nothing to see, or because we do not have sensors precise enough to detect waves? These questions remain quite open within the scientific community. Moreover, in one of my recent articles for the journal Science, I explain that by adding up all the data ever recorded before the earthquakes, we can detect a subtle signal. Which means that something may be happening before major earthquakes, but our tools are not yet sophisticated enough to detect it.
Despite everything, there are warning systems that can detect earthquakes and estimate their magnitude as quickly as possible. How do they work?
Warning systems exist in around fifteen countries around the world. But they don’t all work with the same algorithms. Currently, those used in real time do not work with artificial intelligence (AI). Their mode of operation is as follows: there is generally a network of seismological stations which record seismic waves on sensors to determine where the signal comes from. Depending on the amplitude of these first seismic waves, a magnitude is then estimated. Everything is very fast. This works because seismic waves propagate quickly, but not instantly. If towns are located further away than the first stations which record the earthquake, there is some time to alert the population. But it takes seconds.
However, these systems have some shortcomings. In the first seconds of recording, they are not able to determine whether it is a small or large earthquake. It can be seen that very large earthquakes are always underestimated. These alert systems therefore work better on small ones, which do not last long, because we can get the information quite quickly, unlike large ones. In a very large earthquake that lasts several minutes, the first recordings allow us to see what happens during the first seconds of the earthquake. However, the breakup is not necessarily over and the intensity can sometimes increase.
To remedy this you have developed other more sophisticated algorithms using artificial intelligence. What is their particularity?
We have in fact developed two algorithms based on AI: one based on classic seismic waves, which is operational, and the other on gravity waves, which is still at the prototype stage. For the second, we use a signal that we have only known since 2017. This is the disturbance of gravity due to an earthquake. The scientific community had missed this signal for a long time because it is very weak – about a million times smaller than seismic waves. But for very large earthquakes, we manage to detect it. What is interesting about this signal is that gravity propagates at the speed of light, faster than seismic waves. We can therefore hope to spot the signal earlier and make predictions in advance. Furthermore, gravity being an acceleration recorded by our seismometers, we thought about developing an AI algorithm which scans the data in real time. From these gravity signals, it is able to estimate the magnitude of an earthquake. It’s not quite instantaneous because you have to wait for the earthquake to be large enough to produce a sufficient signal.
Is this system already operational?
We have just published the prototype developed in Japan. For the moment, it is an algorithm that has not yet been integrated into a real-time alert system. On the other hand, we are beginning to install it in an alert system in Peru. This will be a supplement for very large magnitudes. Concretely, it should work as follows: depending on the location and magnitude, an alert radius will be automatically defined around the earthquake. Sirens will thus be triggered in all localities and town halls located in the area. In addition, an alert message will be displayed on the phones of people within the perimeter. In this case, either the exit from a building is close enough and therefore it is easy to leave it, or it is recommended to take shelter under a table or in secure rooms to protect yourself. The rest of the system uses another AI algorithm, based on classic seismic waves this time.
Could these AI-based detection algorithms have detected or prevented the Moroccan earthquake?
The algorithm on gravity waves could not have worked because it is starting to be effective in Japan on earthquakes of magnitude 8.3. The Moroccan earthquake was magnitude 6.9, so it would have been impossible to see it. However, the algorithm based on seismic waves – which will be implemented in Peru initially – is normally made for this. However, it should be noted that when an earthquake occurs near a large city, the time to alert the population is limited because seismic waves propagate very quickly.
Basically, the AI algorithm on seismic waves is rather effective for large cities located far from the epicenter. If the location is right above the epicenter, a warning system cannot do anything. The earthquake will be felt before reaching the nearest station. Thus, this system will work on large earthquakes whose epicenter is far from a large city and if there are seismometers in between. For the Moroccan earthquake, the epicenter is located 77 kilometers from Marrakech, which remains relatively close even if there are stations in between. We will test this system in the coming days to see if it could have been effective and limited the damage if it had been in place in Morocco.