Considering a moderate seismic intensity threshold, the system showed that it could provide 10 to 60 seconds of advance warning in areas 20 to 300 kilometers from the epicenter, and the false alarm rate was quite low.
Based on a higher intensity threshold, it was found that the system could provide a warning of 20 to 40 seconds in advance, but this was only valid for points more than 100 kilometers from the epicenter. The test results were published in the journal Communications Earth & Environment.
The first earthquake examined in the study occurred at 04:00 in the morning near Pazarcık and its magnitude was measured as 7.8. Nine hours later, a second earthquake of magnitude 7.6 occurred near Elbistan, approximately 100 kilometers further north. The two tremors affected a large area of approximately 350 thousand square kilometers; 14 million people lived here. The confirmed death toll in Türkiye and Syria was 59 thousand. The United Nations estimated that approximately 1.5 million people were homeless.
Early earthquake warning systems
Early earthquake warning systems aim to predict the arrival of a dangerous earthquake several seconds to several tens of seconds in advance. This way, people can take protective measures, and facility and infrastructure managers can put security measures in place.
This time advantage arises from the fact that earthquakes produce two different seismic waves caused by fractures in the earth’s crust. Longitudinal waves (P waves) compress and stretch the crust in the direction of propagation. Transverse waves (S waves) create up-down oscillations and are the waves that cause the greatest damage. P waves propagate faster than S waves; While it travels at a speed of 6-7 kilometers per second, S waves propagate at a speed of 3-4 kilometers.
As the waves spread from the epicenter, seismic stations in more distant areas detect P waves first. This signal is analyzed to predict whether a destructive S wave will occur and if necessary, an alarm is given. However, as the signal duration increases, the reliability of the prediction increases. Therefore, it is very difficult to design an effective early warning system. The faster you want to be, the greater the risk of not being able to raise an alarm at some points where intense shaking will occur.
Predicting violence in the first seconds
The researchers used data obtained from 110 seismic stations operated by Türkiye’s Disaster and Emergency Management Authority (AFAD) under the Ministry of Internal Affairs. The stations were located at a distance of 20 to 300 kilometers from the epicenter.
“We reproduced the real-time transmission of the signals recorded from the stations, as if we were watching the earthquake live,” said Aldo Zollo, who coordinated the study.
Researchers attempted to estimate the maximum ground acceleration that would occur at the end of the earthquake by analyzing seismic signals as they were recorded. This parameter can be converted into macroseismic intensity scales that express the effects that people and buildings will be exposed to. One of the most well-known of these scales is the Mercalli scale, developed by Italian volcanologist Giuseppe Mercalli in 1902.
The relationship between ground acceleration and damage varies from region to region; because this relationship largely depends on the quality and durability of the building stock.
Whether or not an alarm is issued is based on the predicted macroseismic intensity in the area covered by the early warning system. The selected severity threshold significantly affects system performance.
The innovation tested in the research is in the way P waves are processed. The model developed in 2023 combines two approaches. At points with seismic stations, maximum ground acceleration is estimated directly from the P wave amplitude. In regions where there are no stations, the model estimates the hypocenter (breaking point in the earth’s crust) and magnitude of the earthquake before the P wave amplitude; It then uses these parameters to calculate the maximum ground acceleration through relationships based on local geological features.
This second approach is less reliable, but essential for creating maps for a large region. Because in the first moments of the break, points far from the epicenter or regions without stations have not yet been exposed to P waves.
Threshold and balance
Two scenarios were evaluated. In the first one, the alarm is triggered if the predicted ground acceleration exceeds macroseismic intensity IV (moderate); In the second scenario, it is given if the intensity exceeds VI (strong).
In both scenarios, no alarm is given in the first 9 seconds of the break. This time is necessary for P waves to reach the 10 closest stations and for researchers to estimate the magnitude and hypocenter.
After 9 seconds, the system’s correct prediction rate immediately reached 85% in the first scenario, and increased to 100% after 60 seconds. In the second scenario, there were more missed alarms in the first 35 seconds; After 60 seconds, the rate of correct alarms exceeded false and missed alarms.
For correctly predicted stations, the warning period increases with distance from the epicenter and is generally longer at lower thresholds. In the high threshold scenario, an alarm can be given 20-40 seconds in advance for stations 100-300 kilometers away from the epicenter.
“The magnitude of the earthquake depends on the size of the area where the rupture occurs and the relative displacement of the blocks,” Zollo says. At the same time, the duration of the P signal depends on the size of the fracture area and the fracture speed. Therefore, the magnitude estimate increases over time and reaches saturation as larger portions of the signal are analyzed. The lower threshold is crossed earlier and saves more time; A higher threshold requires a longer wait.
Therefore, there is a technical balance. The intervention to be applied and the target to be protected (for example, high-speed trains or students in schools) are decisive in choosing this threshold.
Based on experience in the USA and Japan, a false alarm is more acceptable than a missed alarm. However, the most appropriate threshold is the one that minimizes the risk by considering the costs and probabilities of false and missed alarms together.
