The application of artificial intelligence (AI) with the use of big data in the study, detection and advanced warning of seismic movements has achieved one of its first experimental successes thanks to an algorithm developed by a team of experts from the United States, China and Egypt.
The first results of the trial carried out with a new AI algorithm developed at the University of Texas in Austin (United States) in seismic risk regions in southwestern China for seven months indicate that 70% could be correctly predicted. of earthquakes, up to a week in advance. The conclusions of this research and its experimental testing have been presented in an article published in the Bulletin of the Seismological Society of America (September 5, 2023)
The AI algorithm was prepared to detect statistical changes in seismic data in real time, based on information accumulated in large databases about previous earthquakes, highlights the University of Texas.
The result of the trial was a weekly forecast in which the AI successfully predicted 14 earthquakes recorded within approximately 321 km (200 miles) of where it estimated they would occur and with almost exactly the calculated strength. The system failed (that is, did not give the expected warning) in detecting an earthquake and reported eight false warnings.
The University of Texas, to which three of the eleven signatories of the study belong, has highlighted that "it is not yet known whether the same approach will work in other places, but the effort is a milestone in research for earthquake forecasting driven by AI."
"Predicting earthquakes is the holy grail," said Sergey Fomel, a professor at the University of Texas Bureau of Economic Geology and co-author of the study. "We are far from being able to make accurate predictions anywhere in the world, but our work indicates that what until now was believed to be an unsolvable problem [earthquake prediction] may have a solution."
The completed and presented essay is the result of an international call or competition held in China in which the AI developed by the University of Texas came in first place among 600 other designs. The initiative was led by office seismologist and lead AI developer Yangkang Chen of the University of Texas.
Earthquake prediction involves hundreds of teams around the world but has so far yielded very few results, recalled Alexandros Savvaidis, senior research scientist who directs the Texas Seismological Network (TexNet) Program. If its effectiveness is confirmed in other areas, an AI prediction system like the one now, "even without exceeding 70% success, could help minimize enormous economic and human losses, and has the potential to drastically improve preparation for earthquakes around the world,” says Savvaidis.
The authors of the new study explained that their method had been successful by following a relatively simple machine learning approach. The AI was given a set of statistical characteristics based on the team's knowledge of earthquake physics, then given access to a five-year database of seismic records in the analysis area; perfecting your calculations through self-learning. Finally, once launched, the new AI algorithm analyzes the real data at each moment and issues predictions about the date, place and intensity of the earthquakes.
Researchers are confident that in places with seismic monitoring networks and large databases, such as California, Italy, Japan, Greece, Turkey and Texas, AI could improve its success rate and limit its predictions to a few tens of kilometers.
One of the next steps is to test AI in Texas, as the state experiences a high rate of minor and some moderate magnitude earthquakes.
One of the improvements planned for the new system is to incorporate more parameters and data on subsurface physics in which it is intended to predict the occurrence of earthquakes. .
"Our future goal is to combine physics and data-driven methods to give us something generalized, like chatGPT, that we can apply anywhere in the world," said Yangkang Chen.
