Detecting natural resources like oil reserves, geothermal sources and so on are done by tracking seismic waves that are produced naturally by earthquakes or artificially via explosives or underwater air guns. The narrow range of low-frequency seismic waves around one hertz can give a more clear picture of what is inside earth's crust. That said the currently available seismic wave detectors often miss out on the lower frequency waves because of Earth’s loud seismic hum. There might be a solution to this problem. Thanks to machine learning.

In a paper appeared in the journal Geophysics, the researchers at MIT described how they trained a neural network (a set of algorithms modelled after the neural workings of the human brain) on hundreds of different simulated earthquakes. The team presented the trained network with a single high-frequency seismic wave produced from a new simulated earthquake. The algorithms can recognize patterns in data that are transmitted into the network and can cluster these data into categories, or labels.  

The network was able to imitate the physics of wave generation. It correctly estimated the quake’s missing low-frequency waves. “The new method could allow researchers to artificially synthesize the low-frequency waves that are hidden in seismic data, which can then be used to more accurately map the Earth’s internal structures,” says co-author Laurent Demanet, professor of applied mathematics at MIT.

Hongyu Sun (Co-author of the paper) and Demanet adapted a neural network for signal processing, specifically, to identify designs in seismic data. If a neural network was served enough examples of earthquakes, and the ways in which the resulting high- and low-frequency seismic waves travel through a particular composition of the Earth, the network will be able to dig the unknown associations among different frequency elements and extrapolate any missing frequencies if the network were only given an earthquake’s partial seismic profile.