Zooming into earthquake ruptures: from kinematics to dynamic

Abstract

Earthquakes, especially the large ones, cause huge hazards and threaten people living nearby. It is important to improve our understanding of the earthquake source process. I do so by combining seismic data, observations, and simulation of dynamic ruptures. I first made the striking observation that the megathrust earthquakes present a ubiquitous pattern of coseismic rupture with updip low-frequency radiation and downdip high-frequency radiation, based on the backprojection and spectral analysis of teleseismic P waves. I tied this observation to the unique kinematics and dynamics of megathrust earthquakes. To relate the backprojection images, I used synthetic seismograms from theoretical and kinematic sources to explain that the backprojection images are proportional to the slip history, albeit a spatial smoothing operator that I derived. To further illustrate the observations of depth-frequency relation during megathrust earthquakes, I build a suite of 2D dynamic rupture models of megathrust earthquakes in realistic Earth structures. I find that systematic variations in the slip rate functions along-dip of the fault exhibit a systematic variation of frequency content.

Given the variety in model settings, I attribute the ubiquitous depth-frequency radiation observation to the interaction of the earthquake with the Earth's free surface, which is my preferred first-order explanation. Diving more into the observations of earthquake source time functions, we also find that heterogeneous representation of fault properties is necessary to explain the complexity in the source time function, and earthquake rupture is a dynamical process that is self-organized. These systematic studies on the seismic signature of earthquakes have shown the power of combining observations and simulations to decipher source physics from observations.

Many observations rely on the quality of these signals at a broad range of frequencies, which leads me to my last chapter. I develop a machine-learning multi-task model to separate the earthquake signals from the ambient seismic vibrations in complex seismic data. I have tested this method on an island station in Hawaii contaminated by tectonic, volcanic earthquake signals, and a strong microseismic source around the island. I show the promises of such an approach to improve earthquake and ambient noise seismology.