Scaling Up Data-Driven Earthquake Monitoring

Abstract

Modern seismology is undergoing a data revolution driven by advances in machine learning, fiber-optic sensing, and cloud computing. This dissertation develops scalable, data-driven approaches to earthquake monitoring that span the full data lifecycle from acquisition and storage to analysis. I compile a curated benchmark seismic dataset from 21 years of Pacific Northwest recordings that improves machine learning phase picker performance in the region. I develop a cloud-native storage framework for large-scale fiber-optic sensing data and demonstrate its effectiveness on an ambient noise cross-correlation workflow using data from the SeaDAS-North deployment. I apply a machine learning wavefield reconstruction method to seafloor fiber-optic data in Cook Inlet, Alaska to enhance earthquake detection and phase picking in an edge computing setting. I present a cloud-native workflow that processes approximately 1.3 petabytes of continuous seismic recordings spanning two decades, producing a publicly accessible global phase pick database. I develop a multi-stage machine learning framework that reconstructs earthquake ground motion fields and estimates probabilistic hypocenter locations from sparse sensor observations, with applications to earthquake early warning. Finally, I provide a comprehensive review of cloud computing and storage in seismology. Together, these contributions advance the scalability of earthquake monitoring and will contribute to the development of next-generation seismic monitoring systems.