Machine learning applications are already quite extensive. It has proven its potential in online fraud detection, video surveillance and recommendation systems. Even its applications in medicine are far-reaching.
So what would it be like if we could apply machine learning techniques to predict natural disasters such as earthquakes and tsunamis? The result is saving countless lives, while minimizing damages, too.
A team of geologists from Tokyo Metropolitan University are exactly up to that – and now they have developed a deep neural network machine learning model that studies earth’s geomagnetic field data for faster warnings before earthquakes and tsunamis.
Earthquakes send out accumulated stress along fault lines which causes local changes in geomagnetic field. Similarly, in the case of tsunami, vast movement of the sea cause variations in atmospheric pressure which in turn causes alterations in geomagnetic field. These localized changes in the geomagnetic field are then picked up by a network of observation points set up at various locations.
Systems that warn people just before the arrival of seismic waves already exist. But in terms of speed and accuracy, this AI based system outperforms the conventional warning systems.
One of the best approaches for this model is speed. Understanding electromagnetic waves travel at the speed of light, the system can instantly tell if earthquakes or tsunamis are about to happen by observing changes in geomagnetic field.
The signal sent out by geomagnetic field is usually sporadic. So how can the system tell if the detected field is anomalous or not? Well, it has its way to figure out what the “normal” field at a location is.
Yuta Katory and his team selected multiple locations around Japan. They then developed a technique to measure and estimate the geomagnetic field at different, specific observation points.
The technique is the state-of the art machine learning algorithms, known as a Deep Neural Network. Much like how humans learn from experiences, machine learns from previous computations and through series of tests and validations, it can generate top-notch predictions.
So the team fed vast amount of input from past measurements and let the algorithm generate and optimize complex set of operations that maps the data to the measurements. They used half a million data points taken from 2015 and they were able to successfully create a network that can detect the magnetic field with high accuracy.
In the future, the team hopes to pair DNNs with a network of high sensitivity detectors to achieve even better detection of earthquakes and tsunamis.