The earthquakes are terrible because after the first shock, aftershocks often occur and cause considerable damage. These are very difficult to anticipate.

An earthquake , no matter how powerful, can have terrible consequences on the region it strikes. Replicas can also be very dangerous, sometimes even more than the first shock. These often come to complete structures already weakened by the previous shake …

Artificial intelligence to predict aftershocks of earthquakes

Scientists now have digital models helping to “predict” the arrival of replicas but in a new paper published in Nature, Google and Harvard unveil the result of their joint work. Using artificial intelligence, they hope to better predict these replicas. To do this, the researchers trained a network of neurons to detect particular patterns in a database of more than 131,000 earthquakes.

Then they made this neural network work on 30,000 similar events to see if the algorithm was able to identify the replicas. According to their first tests, this model is more efficient than the existing model using in particular the Mohr-Coulomb theory. The score obtained varies between 0 and 1, with 1 indicating an exact model.

A promising algorithm signed Google and Harvard

The artificial intelligence of Google and Harvard got a score of 0.849 versus “only” 0.583 for the Coulomb-based model. That being said, this AI-based model is still far from ready to officiate in the field. Apparently, it only works on replicas caused by permanent changes on the ground, what experts call static stress. For now, it is still too slow to work in real time but the results already obtained are promising. Let’s hope it can be improved enough to be very useful in all circumstances.