Applications of deep learning in seismology
Abstract/Contents
- Abstract
- Seismic waveforms contain valuable information about earthquakes and earth structure. Dense seismic monitoring networks are deployed across the world and collect massive amounts of observational data; however, the sheer amount of this data poses a challenge for seismic data processing and analysis. Developing effective and efficient algorithms and models for seismic data analysis is thus important for studying earthquake physics, improving earthquake forecasting, and mitigating earthquake hazards. In the work presented in this thesis, I have explored a promising approach to advancing earthquake detection and inversion using deep learning. Deep learning has in recent years achieved super-human performance in solving many challenging problems, such as image recognition, protein folding, and playing Go or Atari games. In contrast to conventional algorithms that rely on expert-designed features and decision rules, deep neural networks can automatically learn characteristic features and statistical criteria from large training data sets accompanied by manual labels. The huge amount of archived seismic data collected in the past few decades provides excellent training resources for deep learning, making it a very promising approach to studying seismic signals and addressing research challenges, such as detecting hidden small earthquakes whose numbers dominate earthquake catalogs. However, at the time I started my PhD research, there was limited work on deep learning applications in seismology. To explore the potential of deep learning in seismology, I focused on two directions: First, I developed modular deep learning algorithms to improve earthquake monitoring including signal denoising, phase picking, phase association, and earthquake detection. The results of my work show that these deep learning algorithms significantly improve earthquake monitoring by detecting up to orders of magnitude more small earthquakes than are detected in standard catalogs, and by doing so reveal a far more detailed picture of earthquake sequences and fault structures. In addition, I utilized cloud computing to scale-up our detection workflow to solve the big data challenge in mining large archived data sets. Second, I studied the connection between deep learning optimization and conventional seismic inversion, such as full-waveform inversion. I developed a new inversion approach to solving seismic inverse problems using automatic differentiation and proposed a new regularization method by parameterizing inversion targets using neural networks. The results show that the rapid development of deep learning frameworks and neural network architectures can improve seismic inversion to constrain physical parameters of interest from detected seismic waveforms. In all, these applications of deep learning to both earthquake monitoring and inverting underlying parameters demonstrate that deep learning is an effective tool to improve the extraction of useful information from seismic data and that it holds great promise for future developments in seismology.
Description
Type of resource | text |
---|---|
Form | electronic resource; remote; computer; online resource |
Extent | 1 online resource. |
Place | California |
Place | [Stanford, California] |
Publisher | [Stanford University] |
Copyright date | 2021; ©2021 |
Publication date | 2021; 2021 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Zhu, Weiqiang, (Researcher in geophysics) |
---|---|
Degree supervisor | Beroza, Gregory C. (Gregory Christian) |
Thesis advisor | Beroza, Gregory C. (Gregory Christian) |
Thesis advisor | Dunham, Eric |
Thesis advisor | Ellsworth, William L |
Degree committee member | Dunham, Eric |
Degree committee member | Ellsworth, William L |
Associated with | Stanford University, Department of Geophysics |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Weiqiang Zhu. |
---|---|
Note | Submitted to the Department of Geophysics. |
Thesis | Thesis Ph.D. Stanford University 2021. |
Location | https://purl.stanford.edu/hf729kf3190 |
Access conditions
- Copyright
- © 2021 by Weiqiang Zhu
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...