Improving Earthquake Rapid Response and Early Warning Performance with Geodesy and Machine Learning
Loading...
Date
2025-02-24
Authors
Dybing, Sydney
Journal Title
Journal ISSN
Volume Title
Publisher
University of Oregon
Abstract
This dissertation focuses on the work I have undertaken to investigate how quickly and accurately we can characterize earthquakes while and immediately after they occur, and potential ways to improve the systems we use for this. I focus on how we can use the modern technology we have for Earth monitoring as well as modern data processing and analysis techniques such as machine learning to improve the capabilities of the systems we rely on for earthquake disaster response. In this dissertation I present an exploration into the question of how early earthquakes are distinguishable by magnitude using borehole strainmeters, where we found that earthquakes do not appear to be strongly deterministic (i.e., large earthquakes are not inherently different from small earthquakes in their beginning stages). This has implications for how long it takes to accurately determine the magnitude of an earthquake, particularly for large events which rupture over longer periods of time. I then discuss our development of a machine learning algorithm for improving the performance of earthquake early warning systems for large earthquakes. This algorithm allows for discrimination between noisy GNSS waveforms which do actually contain seismic waves from earthquakes and those which do not, which enables us to reduce the amount of high-noise/low quality data that enters algorithms which determine the magnitude of such earthquakes. While earthquake early warning systems tend to operate only over specific regions such as the U.S. West Coast, organizations such as the USGS’s National Earthquake Information Center also must rapidly publish information such as magnitude about worldwide earthquakes to aid in response efforts. However, magnitude estimation across a large range of earthquake sizes and tectonic settings is technically difficult. To help streamline this process, we developed another machine learning model which allows for the estimation of earthquake magnitudes uniformly for all locations and tectonic settings worldwide, which is also presented in this dissertation. Six multi-frame animations with more figures from this chapter are included as supplemental video files. This dissertation includes previously published and unpublished co-authored material.
Description
165 page PDF and 6 MP4 data files.
Keywords
earthquake, geodesy, machine learning, seismology