An earthquake early warning system cannot predict when and where an earthquake will occur; however, it can give people a few seconds to take cover before the most destructive waves arrive. According to Quentin Bletery, a researcher at the French National Research Institute for Sustainable Development (IRD) and the Géoazur laboratory, even a few seconds in such a situation can be of practical importance.

"If you are given 10 seconds before it happens, there are several things you can do," the researcher noted, describing the standard recommendation during an earthquake: drop to the floor, take cover under a table, and stay there until the shaking ends.

According to Bletery, the problem is not only the speed at which seismic waves propagate. Once a strong earthquake begins, it can be physically difficult for a person to do anything at all: the intense movement can literally cause them to freeze. Therefore, a warning received a few seconds before the strong phase of an earthquake may make it possible to get into a safe position in advance.

Why the system can issue a warning in advance

The operating principle of early warning is relatively simple. If seismic stations are located closer to the earthquake source than the city, they can detect its onset before the most destructive waves reach the city.

P-waves arrive first. They are weaker and usually do not cause the main damage, but it is precisely their detection that allows the system to begin estimating the earthquake’s parameters. The more destructive waves arrive later.

Thus, the system tries to use the time difference between detecting the first waves and the arrival of the strongest shaking. "We are constantly trying to gain a few more seconds," Bletery noted.

One of the developments presented by the researcher is particularly interesting because it does not need to wait for data from several stations at once in order to assess an earthquake.

AI identifies an earthquake using data from a single station

The system was developed for Peru after the country’s government set the task of creating an early warning system for earthquakes of magnitude 6 and above.

Along Peru’s coastline, 111 seismic stations were installed. However, the researchers deliberately chose a path that differs from the traditional approach: the system was supposed to work using data from a single station rather than waiting for confirmation of an earthquake from several stations at once.

The developed system includes seven machine learning models. The first continuously analyzes seismic data and determines the probability of a P-wave appearing. After it is detected, the other models estimate the wave’s arrival time, magnitude, the distance from the station to the epicenter, and the depth of the earthquake, after which the system determines its location.

To do this, the algorithm analyzes the characteristics of the seismic signal, including in the time, spectral, and other domains. The model is based on about 140 parameters, and training was carried out on approximately 100,000 observations from a global database, including data from Japan.

According to Bletery, despite the fact that the system uses only about three seconds of recording from a single station, the magnitude and distance estimates proved to be sufficiently accurate.

The system has already been tested in Peru for three years. During this time, it worked correctly for three earthquakes of magnitude 6 and above and, according to the researcher, did not trigger without a corresponding event.

The system is already operating online, although sirens necessary for directly alerting the population have not been installed in all cities of Peru.

How much time can be gained

According to the researchers’ calculations, the time gained depends on the location of seismic stations relative to the epicenter and the city. For example, simulations of the consequences of major earthquakes showed that in some places the system could theoretically provide about 5 seconds, and in others 10 seconds.

At the same time, the task is not simply to warn as many people as possible. It is important to ensure that the warning reaches precisely those who will be in the zone of noticeable and potentially dangerous impact, rather than residents of areas where there will be no strong shaking.

Another advantage of the developed approach is that it is relatively easy to transfer to other countries. The algorithm is lightweight enough and can be integrated directly into a seismic sensor.

What this means for Armenia

Bletery also separately considered the situation with Armenia’s seismic network.

According to him, the internationally accessible part of the network currently includes eight seismic stations. If we imagine that the 1988 earthquake were to occur today and the system had to wait for the event to be registered at five stations, the seismic waves would have to spread across almost the entire territory of Armenia. For an early warning system, that is too long.

However, using the single-station approach fundamentally changes the situation. If an earthquake occurs near one of the stations, that station can be the first to detect the event and transmit information to the system before the seismic waves reach Yerevan or another населенный пункт.

Thus, even with a relatively small seismic network, the transition from multi-station to single-station detection can provide several additional seconds to warn the population.

For an earthquake, these few seconds can become the critical interval between the moment when a person can still calmly take cover and the moment when strong shaking already makes this much more difficult.