One of the great debates in seismology today revolves around the question of whether earthquake ruptures are self-similar, cascading failures, or whether their size is somehow predetermined at the start of the rupture. If earthquakes are self-similar there is theoretically no way to determine the magnitude of an event until the rupture has completely terminated, while if it is deterministic the magnitude should be immediately discernible. Recent advances in Earthquake Early Warning methodologies provide new insight into the fundamental physics of earthquake rupture and highlight the importance of understanding the answer to this question.

Observations of the amplitude and frequency content of early P-wave arrivals suggest that some information about the final size of an earthquake is already present within a few seconds of the initiation of rupture, in agreement with a host of other observations that show a degree of scaling between large and small earthquakes. While this suggests that earthquakes are deterministic, there is likewise a large body of work, both observational and model-based, that indicates that this is not true and earthquakes are self-similar.

This work documents the process of calibrating and testing the ElarmS Earthquake Early Warning methodology in northern California on the Northern California and Berkeley Digital Seismic Networks. In the process the work adds to the body of observations which show a dependency on event magnitude of P-wave frequency content and amplitude. These observations are corroborated with a new set of independent observations of kinematic slip distributions. These new observations indicate that the early slip on a fault also scales with magnitude and suggest again that earthquakes are not entirely self-similar cascading events.

In an effort to assign a physical mechanism to the observations of scaling, both in P-waves and in kinematic slip inversions, a hypothetical model is tested wherein the intensity of the early rupture imparts more or less energy to the rupture front and affects the likelihood of the rupture continuing or dying out in the face of unfavorable conditions further along the fault plane. The results of testing this hypothesis are somewhat equivocal, but they are suggestive of the likely truth, that earthquakes exhibit aspects of both deterministic and cascading rupture to some degree. Understanding the details of the interplay between these two aspects is crucial to the successful application of Earthquake Early Warning systems, especially in rare large earthquakes for which there is little empirical data on the performance of these systems.

Earthquake Alarm Systems, or ElarmS, is a network-based earthquake early warning (EEW) system developed in California. This dissertation concerns the transition of ElarmS from research prototype to production-grade code. We test ElarmS' performance for large earthquakes with an extensive dataset of large events from Japan. Using the Japanese dataset and results, we develop a statistical error model for the magnitude, location, and ground motion prediction algorithms. We adapt ElarmS to run continuously throughout the state and analyze system latencies due to various seismic networks and instruments. We then rewrite ElarmS completely in newer, more efficient code, while redesigning the association and alert algorithms for improved performance. Finally, we apply an Artificial Neural Network (ANN) at the end of system processing, to detect and block false alerts. The ANN allows ElarmS to send earlier, faster alert messages, when fewer stations have contributed data to an event estimate. ElarmS now sends realtime alert messages to the California Integrated Seismic Network's ShakeAlert EEW system.

The Farallon plate system has been subducting off the western United States since at least the middle Mesozoic. This plate has undergone virtually every subduction process during this time including a long episode of flat-slab subduction, generation of microplates, and formation of oceanic plateaus. The shallow remains of this plate are two small microplates, the Gorda and Juan de Fuca, in the Pacific Northwest. The anomalous nature of these two small plates and the missing deeper evidence of subduction has motivated this study.

The USArray seismic experiment has provided unprecedented spatial sampling of the seismic wavefield in the continuous United States. Utilizing this dataset, new imaging methods have been implemented and older imaging methods have been revitalized. This study first uses ambient seismic noise in the Pacific Northwest to extract short period Rayleigh waves which are sensitive to lithospheric scale structure. Phase velocities from this model are then combined with teleseismic delay times of body waves and surface waves to image the structure of the continuous United States from the surface through the mantle transition zone. The resolving power of this model allows tracing of the Farallon plate from the trench to the lower mantle.

The seismic velocity structure of the continuous United States is broadly composed of a slow western half and fast eastern half separated by the Rocky Mountain Front. The low velocity of the western U.S. contains several high velocity anomalies. While previous work has focused on individual anomalies and suggested they represent lithospheric instabilities, a larger regional view indicates that these are the western remnants of the Farallon plate. Below the thick cratonic lithosphere of the eastern U.S., the Farallon plate contains significant topography due to a subducted heterogeneity of the oceanic plate and a viscosity contrast through the mantle transition zone. The velocity models presented herein provide a cohesive picture of Farallon subduction for the past 150 Ma.

This study has two major focuses, which are united by the theme of applying data processing and imaging techniques to extract valuable information from large volumes of seismic data.

The first focus is on the use of seismic imaging to improve our understanding of the tectonic settings of two of North-America's subduction zones; namely Cascadia and Alaska. Both exhibit enigmatic features and provide insight into fundamental seismotectonic processes relating to the interplay between subduction, mantle flow, volcanism and continental crust growth. Cascadia, for example, represents the final stages of subduction of the once giant Farallon plate. Using data collected from an ambitious deployment of ocean bottom seismometers across the entirety of the Juan-de-Fuca and Gorda plates offshore Oregon and Washington, this study maps the geometry of upper mantle flow in the region, which has important tectonic implications.

The Alaskan subduction zone contrasts with Cascadia because the downgoing oceanic lithosphere here is older and more seismically active. This study uses data from an unprecedented seismic survey of Alaska to construct high resolution, three dimensional models of the velocity structure from the surface down to about 400km depth. These elucidate many fascinating features of this region, including the geometry of the downgoing material and its relationship to several enigmatic volcanic provinces.

The second focus of this study is concerned with earthquake early warning. This is a technology that aims to provide seconds to minutes of warning to people before the onset of severe shaking during an earthquake, potentially allowing them to take life-saving actions. MyShake is a smartphone application developed by the U.C Berkeley Seismological Laboratory that allows mobile devices to detect earthquakes. MyShake is available for public download and is providing a rich dataset for investigation. This study describes an algorithm designed to use networks of MyShake devices to quickly locate and issue warnings about earthquakes, and a simulation workflow designed to test this approach. The results indicate that MyShake networks have exciting potential to issue useful earthquake early warnings worldwide.

Understanding of Earth’s evolution has been hindered, in part, by the technical challenges associated with placing seismic instruments on the seafloor. As technology improves, more arrays of ocean bottom seismometers are being deployed around the ocean basins. Perhaps the most ambitious such array, the Cascadia Initiative, covered the entirety of a small oceanic plate, the Juan de Fuca plate, with ocean bottom seismometers in a four-year experiment. That array is the primary motivation for this study.

The two models presented in this work, CASC16-P and CASC19-S, are teleseismic tomographic studies. They are each the first of their kind to use offshore and onshore data to simultaneously image the P- and S-wave seismic velocity structure of the mantle beneath the Juan de Fuca and North American plates. These models have provided insight into the tectonics of oceanic plates: how they are created, how they might interact with the mantle over the course of their life, how they might be influenced by an overriding continental plate, how they behave in the mantle after subduction, and how subduction could cease.

This work focuses on the on- and offshore regions of the Pacific Northwest of the United States—in particular, on the interaction between the Juan de Fuca plate, the North American plate, and the upper mantle. But it also demonstrates the value in oceanic arrays. The offshore instruments used in this study have allowed the identification of new features offshore, and better resolution of features near the coastlines. Ocean bottom seismology represents a rapidly growing new frontier that has the potential to investigate earth evolution in detail that was not previously possible. The work we present here has both helped illuminate the tectonics of the Pacific Northwest, and helped generate detailed hypotheses about the nature of global tectonics that will be tested elsewhere.

MyShake is a free citizen science smartphone application that has been publicly available since 2016 and has been downloaded globally 2.6 million times. Since becoming an earthquake early warning delivery service for the US West Coast, MyShake has attracted its greatest concentration of users in the states of California, Oregon, and Washington. As a citizen science project, MyShake collects various types of data, which can be broadly categorized as long-form waveforms and real-time snippets. When a phone operating MyShake is still for 30 min, MyShake will begin monitoring the accelerometer, which is hardware built into all smartphones. A machine learning-trained pair of algorithms scan for earthquake-like motion and will record five minutes of waveform data when an earthquake is suspected to be occurring. Waveform recording can also be triggered by the delivery of an early warning message, in anticipation of imminent earthquake shaking. These data are uploaded to secure servers when phones are connected to wi-fi and power. Real-time data collected by MyShake includes state of health information sent in data snippets called heartbeats, alert receipts messages, short-term-average over long-term-average (STA/LTA) triggers whenever a monitoring phone suddenly moves, and aggregate peak acceleration values.

The data collected by MyShake represents an at-scale opportunity to explore the capabilities of a smartphone seismic network. In this work, I investigate four use-cases of smartphone data: 1) assessing how smartphones perform as early warning delivery tools; 2) using smartphone waveform data to measure ground shaking intensities; 3) incorporating real-time smartphone earthquake detections into a traditional earthquake early warning system; and 4) measuring the fundamental frequency of buildings using smartphone waveforms. The first and third objectives form a circle, generating and then delivering early warning to private citizens. We find that phones are both efficient delivery mechanisms, with warnings typically arriving on phones within 5 seconds of alert generation, and efficient earthquake detectors, with real-time triggers arriving at a backend server in the same amount of time as traditional data. Measuring acceleration and converting to shaking intensity helps evaluate the accuracy of the shaking extent produced by early warning. Using waveform data to measure fundamental frequency ventures into the realm of structural health monitoring, a critical field for earthquake preparedness and rapid response. All together, MyShake represents a novel and high-potential opportunity to make seismology accessible and applicable within communities while also supporting the creation of seismic knowledge that can be used to make the same communities safer.