Caltech Logo

IST Lunch Bunch

Tuesday, October 8, 2019
12:00pm to 1:00pm
Add to Cal
Annenberg 105
How deep learning is transforming earthquake seismology
Zach Ross, Assistant Professor, Geophysics, Caltech,

The volume of seismic data recorded around the world is exploding. At the same time, standard techniques for earthquake detection still routinely miss the smallest earthquakes, which represent the vast majority of seismic activity that occurs naturally. These hidden events, however, are essential to advancing our understanding of earthquakes and faults because they fill in the gaps between larger events and provide a more complete picture of how earthquake sequences evolve in space and time. Reliable measurements of time and amplitude properties of seismic waves also enable tomographic images of Earth's interior, and delineate the earthquake rupture process below the surface. Earthquake early warning requires rapid identification that a large earthquake has occurred with only a tiny fraction of the total data available at that moment.

Seismology is a field that is rich in labeled data and has many difficult data-driven problems with some unique challenges. These aspects have led to a surge in recent applications of deep learning to seismology, resulting in state-of-the-art performance on numerous tasks. In this talk, I will discuss several important problems in seismology related to earthquake detection, localization, and earthquake early warning, and the development of deep learning algorithms to address them. Machine learning will play a prominent role in the future of seismology by improving real-time earthquake monitoring, as well as advancing earthquake science to the next level from analysis of large high-dimensional datasets.

For more information, please contact Diane Goodfellow by email at [email protected].