Researchers at the University of Texas have developed an artificial intelligence that can predict earthquakes with 70% accuracy and a week in advance. The AI used seismic data from earthquakes in China, which hosted a competition for this type of technology. According to the project leaders, the technology can evolve with access to data from regions with more robust seismic activity monitoring.
The AI’s accuracy rate is linked to the region where the earthquake will occur. The algorithm analyzed various data from a location and marked the future epicentre on the map. Even if it is off by kilometres, this would help governments to remove the population from risk areas or from inappropriate housing within a certain radius of the predicted epicentre.
AI has “simple operation”, say researchers
In the University of Texas statement, the researchers explain that, given the right proportions, the functioning of the AI is a simple case of machine learning. They created the artificial intelligence based on earthquake statistics and physics.
The AI was then trained on a database of five years of seismic records. After this training, the system evaluated the current records and provided a prediction of where the epicentre of the earthquake would be. In terms of date, the AI aims to predict tremors a week in advance.
The mean absolute error, a metric used in machine learning, was 381k m. However, as we can see in the previous image, the predictions for weeks 10, 11 and 13 were very close to the real epicentre.
It is worth remembering that the earthquake in Turkey and Syria had its epicentre in the province of Gaziantep, but was felt in Beirut, 384 km away.
“Predicting earthquakes is the Holy Grail”
For Sergey Fomel, a professor in the Department of Economic Geology and a member of the research team, earthquake prediction is the Holy Grail. “We are not yet close to making predictions anywhere in the world, but what we have achieved shows that what we thought was an impossible problem is, in principle, solvable,” says Fomel.
According to the researchers, AI will perform best in areas with effective seismic monitoring. California, Greece, Italy, Japan, Turkey and Texas are examples of places that meet this requirement. These states and countries are in high-risk regions and experience frequent earthquakes.
Yangkang Chen, the lead researcher on the research, says the next step is to train the AI with more physics and data to create a more generalized technology. The idea, he says, is to have a ChatGPT for earthquakes that could be used anywhere in the world.
With information: SciTechDaily