Using Deep Learning to View the Rich Data of Seismology
Seismology is a data-rich science. Vast arrays employing thousands of sensors continuously sample and monitor for seismic waves and motion. Sampling rates can involve 100 points per second. Plus, seismologists need to be able to weed out extraneous “noise” such as Fourth of July reworks or the rumbling of big-rig trucks. us, seismologists have turned to folks at places like Google to help enlist so-called “deep-learning” techniques of analysis.
Deep learning involves machine learning algorithms able to take raw input and extract higher-level features. For instance, in image recognition, deep learning can take the raw input of tiny pixels and progressively, layer-by-layer, assemble the image of a face.
In a review summary in the journal Science, S. Mostafa Mousavi (Stanford University) and Gregory C. Beroza (Google, Mountain View, CA) note that deep-learning techniques are becoming vital in seismology, particularly with ever more reams of data generated by new sensing technologies, from ber optic cables to new features in smart devices. ey conclude that to fully maximize deep-learning seismology, there’s a need to better educate geoscientists on data science literacy and to partner with others possessing advanced data science expertise.