Stanford University

Due to COVID-19, all classes have moved to an online format and many events at Stanford have been canceled or moved to a virtual format. This calendar may not reflect all changes. Please confirm event status with the event contact before attending.

x
north america from space

Paul Johnson, Los Alamos, Machine Learning Tools Reveal Fault Physical Characteristics

When:
Thursday, Feb 13, 2020 12:00 PM
Where:
Mitchell Building, room 350/372
Sponsor:
Geophysics Department

GEOPHYSICS DEPARTMENT SEMINAR

Paul Johnson

Geophysics Group, Los Alamos National Laboratory 

Machine Learning Tools Reveal Fault Physical Characteristics 

Date: Thursday, February 13, 2020

Location:  Mitchell 350/372

Time:  12:00 pm - 1:15 pm

Host: Greg Beroza

                                          

                                                            

We analyze continuous seismic data with machine learning (ML), with the goal of identifying hidden signals connected to the earthquake cycle. In the laboratory, we find that continuous seismic waves originating in the fault zone are imprinted with fundamental information regarding the physics of the fault. Statistics of these low-amplitude, noise-like signals identified with supervised ML approaches can be used to estimate fault friction, fault displacement, and forecast upcoming failure with great accuracy. These results hold true for both stick-slip and slow-slip frictional regimes. Similarly, when we scale the approach to study slow-slip events in the Cascadia subduction zone, we find that continuous seismic waves contain information about the instantaneous fault displacement at all times. Because we cannot easily locate this continuous seismic signal and therefore cannot be certain it arises at the plate interface, we develop a supervised ML approach based on convolutional neural networks (CNN’s) to train on known tremor events listed in independent catalogs and located approximately at the subducting plate interface. By training on cataloged tremor events, the CNN is able to identify far more, previously undetected tremor, and we find that it takes place at nearly all times as well. The noise-like signal identified in the lab and in Cascadia and the known tremor have characteristics in common but differences as well, and the relation to crustal motions that will be discussed. Remarkably, the CNN trained at a single station in Cascadia also identifies tremor in Chile, New Zealand, Japan and also along the Parkfield section of the San Andreas Fault, suggesting universal fault friction characteristics.

IconsList of icons used on the sitemaillinkedindouble carrot leftarrow leftdouble carrotplayerinstagramclosecarrotquotefacebooktwitterplusminussearchmenuarrowcloudclock