Deep neural networks will register anomalies in the magnetic field. This will help predict earthquakes and tsunamis faster than ever before.
The destruction caused by earthquakes and tsunamis directly indicate the need for an effective forecasting method. There are already systems in the world that alert people to disasters just before seismic waves occur. However, the second wave is often caught off guard faster than a warning is given. That is why faster and more accurate means are needed, which will ultimately save many lives. An article about the study was published in IEICE Communications Express.
It is known that earthquakes and tsunamis accompany local changes in the geomagnetic field. During earthquakes, the piezomagnetic effect first occurs, as a result of which changes in the geomagnetic field occur. In the case of a tsunami, there is a sudden movement of gigantic volumes of water into the sea, which entails changes in atmospheric pressure. This, in turn, affects the ionosphere, ultimately changing the geomagnetic field. In both cases, this can be recorded by a network of observation posts at different locations. The main advantage of this approach to the problem is speed. Considering that electromagnetic waves propagate at the speed of light, an event can be instantly registered by observing changes in the geomagnetic field.
But how do you know if a change is abnormal or not? In different areas, the geomagnetic field is an oscillatory signal, so this method should be based on an understanding of what is the “norm” of the field for a particular area.
Yuta Katori and Kan Okubo from the Tokyo Metropolitan University took up the development of a method for measuring at different points in Japan and the designation of the permissible values of the geomagnetic field at different observation points. Specifically, they applied a sophisticated machine learning algorithm known as deep neural network (DNN). By loading a large amount of historical measurement data into the algorithm, they allowed it to create and optimize an incredibly complex multi-level set of operations that maximize the data of direct measurements. Using more than half a million data points, scientists were able to create a network capable of measuring the magnetic field at an observation point with unprecedented accuracy.
Given the relatively low computational cost of DNN, the system can be connected to a network of highly sensitive sensors to achieve ultra-fast recording of earthquakes and tsunamis, which in turn will provide an effective warning system capable of minimizing losses and saving thousands of lives.