Learning earthquake sources using symmetric autoencoders
Authors:
Pawan Bharadwaj,
Madhusudan Sharma,
Isha Lohan,
Abhinav Pratap Singh
Abstract:
We introduce Symmetric Autoencoder (SymAE), a neural-network architecture designed to automatically isolate earthquake information from far-field seismic waves. SymAE represents the measured displacement field using a code that is partitioned into two interpretable components: source and path-scattering information. We achieve this source-path representation using the scale separation principle an…
▽ More
We introduce Symmetric Autoencoder (SymAE), a neural-network architecture designed to automatically isolate earthquake information from far-field seismic waves. SymAE represents the measured displacement field using a code that is partitioned into two interpretable components: source and path-scattering information. We achieve this source-path representation using the scale separation principle and stochastic regularization, which traditional autoencoding methods lack. According to the scale separation principle, the variations in far-field band-limited seismic measurements resulting from finite faulting occur across two spatial scales: a slower scale associated with the source processes and a faster scale corresponding to path effects. The primary challenge in leveraging SymAE to enhance our understanding of earthquake sources lies in the lack of direct interpretability of the latent code it generates. However, despite this hurdle, we present two strategies. Firstly, we employ clustering analysis of the source codes to reveal connections between the characteristics of various earthquakes. This method is less susceptible to variations in path effects, allowing us to assess the correlation in source properties between observed earthquakes and their corresponding finite-fault models. Secondly, we introduce the generation of virtual seismograms with desired source and path information. Through this approach, we engineer seismograms to incorporate the source characteristics of a specific earthquake from different azimuths while maintaining consistent subsurface scattering effects. We propose that these virtual seismograms may facilitate a comprehensive and precise examination of directivity effects for the earthquake in question. The paper conducts an analysis of nearly forty complex earthquake events along with synthetic earthquakes.
△ Less
Submitted 12 March, 2024; v1 submitted 5 April, 2023;
originally announced April 2023.
