Morgan, Julia2020-12-022020-12-022020-122020-11-30December 2Blank, David G. "Investigating Earthquake Rupture Using Particle Dynamics Simulations." (2020) Diss., Rice University. <a href="https://hdl.handle.net/1911/109603">https://hdl.handle.net/1911/109603</a>.https://hdl.handle.net/1911/109603The three chapters of this thesis explore the physics that control earthquakes, particularly their precursors, using numerical models. In the first chapter, we use the discrete element method to create numerical analogs to subduction megathrusts with natural roughness and heterogeneous fault friction. Boundary conditions simulate tectonic loading, inducing fault slip. Intermittently, slip develops into complex rupture events that include foreshocks, mainshocks, and aftershocks. We probe the kinematics and stress evolution of the fault zone to gain insight into the physical processes that govern these phenomena. Prolonged, localized differential stress drops precede dynamic failure, a phenomenon we attribute to the gradual unlocking of contacts as the fault dilates prior to rupture. Slip stability in our system appears to be governed primarily by geometrical phenomena, which allow both slow and fast slip to take place at the same areas along the fault. Similarities in slip behavior between simulated faults and real subduction zones affirm that modeled physical processes are also at work in nature. In the second chapter, we develop numerical simulations of dynamic perturbations passing along a heterogeneous pre-stressed fault, to better understand the physical mechanisms that control delayed dynamic triggering of earthquakes. Our results demonstrate that weak portions of the fault that host ongoing slow slip can transfer stress in response to the perturbations, loading asperities poised for failure. We find that the magnitude of perturbation, the state of the asperity, as well as deformation of the surrounding material, jointly control the delay time between perturbation and triggered event. The slow-slip modulated delayed triggering model that we propose can account for the wide range of observed delay times in nature, including the two end-member cases of no delay and no triggering. Triggered slow slip events in nature might provide warning signs of impending earthquakes, underscoring the importance of high-resolution monitoring of active fault zones. In the third chapter, we present a new approach to characterizing precursory earthquake slip using machine learning. We use a binary classification model rooted in state-of-the-art deep learning techniques to predict whether or not complete-interface rupture is imminent along a simulated megathrust fault. The deep learning models are trained on labeled 2D space-time input features taken from the synthetic fault system. We contrast the performance of two different neural networks trained on three different types of data, in order to determine the relative predictive power of each. The neural networks are able to discriminate imminent complete rupture precursors from everything else, thus providing a relative size and time forecast. Vertical displacements along the fault demonstrate particularly good predictive power and improved performance is seen when multiple model predictions are used in tandem. The results confirm previous observations that precursory deformation scales with upcoming event size, consistent with the preslip model for earthquake nucleation. The methods we propose are adaptable and can be modified to use 3D data in the future.application/pdfengCopyright is held by the author, unless otherwise indicated. Permission to reuse, publish, or reproduce the work beyond the bounds of fair use or other exemptions to copyright law must be obtained from the copyright holder.slow slipearthquake precursorsearthquake triggeringmachine learningThesis2020-12-02