Artificial intelligence helps identify earthquakes

(by Giovanni Calcerano) Researchers at Harvard University and the Massachusetts Institute of Technology (MIT), led by Professor Thibault Perol, have developed an Artificial Intelligence (AI) system that can detect earthquakes faster than any other existing device. The results of their research were published in the February 14 in the journal Science Advances.

The system, called ConvNetQuake (Convolutional neural network for earthquake detection location - in Italian Convolutional neural network for the location of earthquakes) is able to detect large, medium and small earthquakes with an accuracy of 94,8%. It is also capable of predicting the location of these earthquakes with an accuracy of 74,5%.

The solution proposed by Perol and his colleagues is to use artificial intelligence to amplify the sensitivity of the seismographs in order to detect a number of earthquakes 17 times greater than previous methods and in a fraction of the time required so far.

The method is similar to that used by digital assistant voice detection software like Siri and Cortana, explains Perol. That is, it is a question of highlighting the hidden signal in the noise. For digital assistants, this means recognizing voice commands while ignoring the ambient background sound. For seismographs, this means instead eliminating the normal geological noises of the Earth (the so-called “environmental seismic noise”) to identify earthquakes even when they are very small or very far away.

To this end, Perol and his colleagues trained a convolutional neural network to recognize background noise using data taken from areas without telluric movements. The software has examined this input and has learned to recognize those models that allow it to define a standard geological noise. In this way, therefore, the neural network can remove background noise and distinguish the signals that conceal true earthquakes.

There is also the hypothesis that such a system could help predict earthquakes before they occur. This could be possible through the standardization of predictive models: that is, if it were possible to verify that a number of small earthquakes in rapid succession trigger a larger and potentially damaging earthquake, then an early warning could be issued whenever other small, similar ones were identified. , earthquakes in an equally short period of time.

Clearly, the idea of ​​using artificial intelligence to predict, and not just detect, earthquakes is very exciting, but it's not something the entire seismological community agrees on. But with the help of Perol and his colleagues, and the neural network they devised, this important possibility could actually become possible. We are now at a point where tremendous advances in instrumentation, machine learning, computer power, and the ability to handle huge data sets could lead to huge, and fundamental, advances in earthquake science.

Artificial intelligence helps identify earthquakes