When an earthquake takes place, naturally the size of the earthquake itself is something to be concerned over, where it could be something small, or it could result in a massive natural disaster. However the aftershocks are also something that need to be considered, where it could result in structures weakened by the initial earthquake to collapse.
Now there are existing models that can help to predict where these aftershocks might occur, but in a paper published in Nature (via The Verge), Google and Harvard are working together and employing the use of AI which they hope will be better at predicting these aftershocks. In order to do this, the researchers trained a neural network to look for patterns in a database of more than 131,000 events.
They then took that neural network and applied it to a database of 30,000 similar events to see if it was effective at identifying aftershocks. Based on their tests, they found that their model is better compared to the existing model known as the Coulomb failure stress change. The score is a scale from 0 to 1, with 1 indicating a perfectly accurate model.
Google and Harvard’s AI managed to score a 0.849, while the Coulomb model managed a 0.583. That being said, this AI model isn’t quite ready to be put into the field just yet. Apparently it only works on aftershocks caused by permanent changes to the ground known as static stress, and is still too slow to work in real-time, but it does show promise and we wouldn’t be surprised if it (or some variation of it) were to be used in the future.