Can AI Help with Earthquake Aftershocks?

Predicting the location or timing of an earthquake itself might be an impossible task, but for the last 25 years, scientists have done their best to understand how a quake might precipitate aftershocks in the surrounding area. Previous methods of investigation have focused on fault line orientation and have shown varying degrees of success.

However, a helping hand from our computer counterparts might be able to change all that. A new study published in the journal Nature has used artificial intelligence (AI) to assess information from more than 130,000 earthquakes and shown far greater accuracy in predicting the whereabouts of aftershocks than previously existing methods, giving its researchers hope that it can provide a model for the future.

Second-guessing Mother Nature

As well as increasing the risk of other disasters like landslides and flooding, earthquakes also often instigate a series of aftershocks in the neighbouring area. Ever since a notable cluster of tremors devastated parts of California and Nevada in 1992 - striking Landers, Big Bear and Yucca Mountain in quick succession - scientists have worked to try and better understand how and when these shocks are likely to happen.

Traditional methods of interpretation have analysed different criteria surrounding the initial earthquake, with most researchers prioritising the Coulomb failure stress change, which affects fault orientations in the vicinity. However, this is a very complicated subject, as the fault lines are disturbed by both shear stress (that moving them horizontally) and normal stress (that affecting them vertically). This makes it very difficult to predict aftershocks with any accuracy.

A new approach

Recent advances in technology have meant that it has been able to benefit science in a whole range of different areas. Earlier this year, scientists announced they were combining forensic science and AI to combat plastic pollution, and the team of researchers behind the aftershock study were keen to leverage the power of machine learning to similar effect.

By inputting data from 130,000 earthquake and aftershock pairs into an AI modelling system, the team were able to gain a better understanding of the relationship between the two. The approach differed in that it included a wider array of information, including location, magnitude, different measures of stress and maximum capacities of stress. In this way, the model was able to predict aftershock locations in 30,000 other incidents with far greater accuracy than humans can alone.

Slow to react

While the study does provide encouraging evidence that AI can help scientists and meteorologists to better understand and prepare for aftershocks, it has one major drawback – timing. This model was able to estimate where and when an aftershock would occur precisely because it had access to such a large amount of data, which simply won’t be available as earthquakes occur in real-time.

In order to offer useful predictions in the future, the AI system must be able to react more rapidly. At present, it can’t offer actionable insight until days after the initial earthquake, by which time the majority of aftershocks may have already taken place. Therefore, it’s certainly a positive study and a sign of how technology might be able to help science further down the line, but it has some way to go before becoming a useful tool in the field.