Earthquakes are more complicated than you think. It’s not just the big quake that’s dangerous. Smaller, follow-up quakes triggered by the initial shock can rumble around an affected area for months. This might not seem much but it can topple structures weakened by the parent quake. With current technology, Scientists have been able to somewhat predict the size and timing of the aftershocks. But it’s quite difficult to predict them accurately. New research from scientists at Harvard and Google suggests AI might be able to predict earthquake aftershocks.

A paper was published in the journal Nature this week, which explains how researchers are trying to use deep learning to increase the accuracy of these predictions. Scientists trained a neural network to look for patterns in a database of more than 131,000 “mainshock-aftershock” events, before testing its predictions on a database of 30,000 similar pairs.

It might not come as a shock (pun intended) but using a neural network is much more accurate compared to the current “Coulomb failure stress change” model. The new method is up to 1.5x more effective.


Brendan Meade is a professor at Harvard who helped author the paper. He believes that the results are quite promising. “There are three things you want to know about earthquakes,” said Meade. “When they are going to occur, how big they’re going to be and where they’re going to be. Prior to this work, we had empirical laws for when they would occur and how big they were going to be, and now we’re working the third leg, where they might occur.”

AI can do a much better job than humans in this use case because they’re much better at finding patterns in complex datasets. Seismology is no different, it can be very difficult for us humans to find patterns in the data we collect. Seismic events involve a ton of variables. Ranging from the makeup of the ground in different areas to the types of interactions between seismic plates to the ways energy propagates in waves through the Earth. Even the most intelligent of scientists have trouble understanding these factors and how they correlate within the dataset.

As with every project under research, this will take years to be even close to being practical. There are areas where even the AI still struggles. There are some factors that the AI cannot work with. For example, it can analyse “static stress” but cannot do the same with “dynamic stress”. But there is hope, maybe within a few years, we will be able to accurately predict earthquakes and their aftershocks?

As Phoebe DeVries, a Harvard postdoc who helped lead the research said, “We’re still a long way from actually being able to forecast [aftershocks] but I think machine learning has huge potential here.”

Read more interesting articles here.