machine learning model is able to identify signs of an imminent earthquake from weak changes in the magnetic field associated with deep energy accumulation.
2014 California magnitude 6.0 earthquake - One of 19 earthquakes used to train neural networks
earthquakes usually look completely abrupt, although potential energy is gradually accumulated, far before it is released with a powerful impact. Geologists are looking for ways to detect threats early and take action to save people and infrastructure. Monitoring local magnetic fields can be one of such tools.
In fact, the increase in depth pressure can change the characteristics of underground substances, including conductivity , and currents will appear in the accumulated gas. Such a process should also be reflected in the characteristics of the magnetic field, which is generated at the most center of the earth and recorded near the surface. Evidence in this area has existed before, but it cannot reliably confirm this hypothesis.
With the help of Artificial Intelligence , it is now possible to notice changes in the magnetic field. Scientists wrote this in an article published by in the Journal of Geophysical Research: Solid Earth. The study was conducted by scientists at QuakeFinder, a nonprofit science foundation, along with Stellar Solutions and the Google Accelerated Science team.
They used seismic monitoring and magnetic field observations related to earthquakes that occurred in California from 2005 to 2019, and magnitude reached at least 4.5. Information received from 125 magnetometers located in the state. It is very noisy because solar activity and even road traffic can lead to weak anomalies of the magnetic field. However, even in this noisy context, machine learning models are able to identify signals.
After testing the test samples, scientists were convinced that AI could consider signs of an imminent earthquake based on the magnetometer. “We do not claim that such signals existed before any earthquake,” emphasized the author of the work, . However, the new approach may be a valuable tool for predicting threat proximity in advance. In some cases, he may have noticed signs of an earthquake as early as 24-72 hours before the first aftershock.
