Google and Harvard are teaming up to AI to predict aftershock locations more accurately. Source: Money Control

An earthquake is one of the most devastating natural disasters. It hits without warning and can cause huge damages. In addition to this, earthquake aftershocks too can pose serious risks to people and their safety. The aftershocks can rumble over the affected area months after the initial earthquake hits. However, there are ways that can be used by scientists to predict the size and timing of these aftershocks. But the technology is not as accurate as it should be. Pinpointing the origin of the aftershocks is always the biggest challenge. New research by Harvard and Google suggests that this might change soon.

In one research article published in the Journal of Nature, researchers try to demonstrate how deep learning can be used to predict aftershock locations more accurately compared to existing technologies. The scientists started by training a neural network to look for patterns in a database of over 131,000 aftershock events. They then used the network to predict 30,000 shocks from the same 131,000 used for the training. The network was able to reliably predict the aftershocks compared to current models. According to the scientists involved in the study, on a scale of 0-1 in which 1 represents 100% accuracy, traditional aftershock detection models were only able to hit 0.583 while the AI was able to hit 0.849. This shows that the new deep learning technology is almost perfectly accurate.

With the new deep learning AI, scientists are able to predict the location where aftershocks will happen. Source: Mashable

The researchers believe they can still improve this and bring the accuracy level even higher. The scientists also note that traditional prediction models used to only predict how big the aftershocks would be and when they would happen. However, with the new deep learning AI, they are able to predict the location where they will happen. This can make a huge difference in evacuating people and reducing the overall damages.

Despite what looks like very promising research, the new technology is still far away from deployment into the real world. There are a number of reasons for this. For example, the AI model focuses on aftershocks that have been caused by permanent changes to the ground. These changes are also known as statics stress. There are other factors that may contribute to aftershocks other than static stress too. For instance, follow up earthquakes can also be caused by the rumblings felt on the ground soon after the earthquake hits. This is called dynamic stress.

As of right now, the AI models take too long to predict the shocks in real time. Source: First Post

In addition to this, the AI is also too slow to work on a real-time basis. This is because serious aftershocks usually occur within 24 hours after an earthquake hits. In some other cases, the shocks might be felt later than this but the frequency tends to decrease as the days go by. The AI models take too long to predict the shocks that come at a much higher frequency. Nonetheless, this is a significant breakthrough. The scientists who are working on the deep learning AI say that this is not a perfect solution but it does open a new door to exciting research that could make a difference in the near future.