Imaging the Earth’s Near Surface with Dense Seismic Observation

Citation

Yang, Yan (2025) Imaging the Earth’s Near Surface with Dense Seismic Observation. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/s4rx-6n15. https://resolver.caltech.edu/CaltechThesis:07182024-175725823

Abstract

Understanding the Earth's near surface is critical for assessing seismic hazards and ensuring environmental sustainability. In this thesis, I explore the use of advanced observation and analysis techniques for near-surface imaging with big seismic data.

Chapters 2-6 focus on the applications of Distributed Acoustic Sensing (DAS). DAS is an emerging sensing technology that transforms fiber-optic cables into dense seismic arrays. In Chapter 2, I introduce a high-performance Python tool for computing seismic ambient noise cross-correlation with large volumes of DAS data. In Chapter 3, I perform ambient noise tomography using a DAS array in Ridgecrest, California, to resolve spatial variation of the near-surface structure, revealing its correlation with earthquake ground shaking amplification. In Chapter 4, I use surface wave scattering observed in the DAS noise cross-correlation for fault zone detection and characterization. In Chapter 5, I analyze three years of DAS noise cross-correlation to monitor seismic velocity changes, providing insights into vadose zone soil moisture dynamics and water resource management in the context of climate change. In Chapter 6, I use a DAS array at the South Pole to characterize firn structure for a better understanding of cryosphere mass balance.

Chapters 7 and 8 focus on imaging geological structures in the urban Los Angeles region using dense arrays of geophones. Chapter 7 uses converted S-to-p phases recorded by a dense network of low-cost accelerometers to map the basin depth. Chapter 8 investigates shallow seismicity in the Long Beach area to illuminate complex fault structures.

In Chapters 9 and 10, I apply a state-of-the-art machine learning framework known as a neural operator for solving seismic wave equations. The trained neural operator enables full seismic waveform modeling with substantial advancements over conventional numerical techniques including its fast speed, generalizability, and convenient differentiability for full waveform inversion.

Thesis Files