Using local, global, and simulated earthquakes to inform earthquake resilience efforts in the Pacific Northwest

Thesis (Ph.D.)--University of Washington, 2022In this dissertation, we investigate how the geometry and rock composition of the Seattle and Tacoma basins influences strong ground motions during local earthquakes by surveying and interpreting strong-motion seismic records and generating 3D ground-motion simulations. We also evaluate the performance of an earthquake early warning system for the West Coast of the United States using historical records of local and global intraslab earthquakes and ground-motion simulations of hypothetical magnitude 9 megathrust earthquake scenarios on the Cascadia subduction zone (CSZ).Chapter 2 is a characterization of sedimentary basin effects within the Seattle and Tacoma basins using Pacific Northwest Seismic Network and U.S. Geological Survey strong-motion recordings of five local earthquakes (M 3.9–6.8), including the 2001 Nisqually earthquake. We observe basin-edge generated surface waves at sites within the Seattle basin for most ray paths that cross the Seattle fault zone. We also note previously undocumented basin-edge surface waves in the Tacoma basin during one of the local earthquakes. To place quantitative constraints on basin amplification, we determine amplification factors by computing the spectral ratios of inside-basin sites to outside-basin sites at 1, 2, 3, and 5 s periods. Ground shaking is amplified in the Seattle basin for all the earthquakes analyzed and for a subset of events in the Tacoma basin. We find that the largest amplification factors in the Seattle basin are produced by a shallow crustal earthquake located to the southwest of the basin. Our observation suggests that future shallow crustal and megathrust earthquakes rupturing west of the Puget Lowland will produce greater amplification within the Seattle basin than has been seen for intraslab events. We also perform ground-motion simulations using a finite-difference method to validate a 3D Cascadia velocity model (CVM) by comparing properties of observed and synthetic waveforms up to a frequency of 1 Hz. Basin-edge effects are well reproduced in the Seattle basin, but are less well resolved in the Tacoma basin. Continued study of basin effects in the Tacoma basin would improve the CVM. In Chapter 3, we investigate whether assuming a fixed shallow depth in the ShakeAlert network-based earthquake early warning system is sufficient to produce accurate ground-motion based alerts for intraslab earthquakes. ShakeAlert currently uses a fixed focal depth of 8 km to estimate earthquake location and magnitude. This is an appropriate way to reduce computational costs without compromising alert accuracy in California, where earthquakes typically occur on shallow crustal faults. In the Pacific Northwest (PNW), however, the most common moderate-magnitude events occur within the subducting Juan de Fuca slab at depths between ~35 and 65 km. Using a dataset of seismic recordings from 37 Mw 4.5+ intraslab earthquakes from the PNW and Chile, we replay events through the Earthquake Point-Source Integrated Code and eqInfo2GM algorithms to estimate source parameters and compute modified Mercalli intensity (MMI) alert threshold contours. Each event is replayed twice—once using a fixed 8 km depth and a second time using the actual catalog earthquake depth. For each depth scenario, we analyze MMI III and IV contours using various performance metrics to determine the number of correctly alerted sites and measure warning times. We determine that shallow depth replays are more likely to produce errors in location estimates of greater than 50 km if the event is located outside of a seismic network. When located within a seismic network, shallow and catalog depth replays have similar epicenter estimates. Results show that applying catalog earthquake depth does not improve the accuracy of magnitude estimates or MMI alert threshold contours, or increase warning times. We conclude that using a fixed shallow earthquake depth for intraslab earthquakes will not significantly impact alert accuracy in the PNW. Chapter 4 is an evaluation of ShakeAlert performance for M 9 megathrust earthquakes in the PNW. Since there are no recordings of large magnitude earthquakes on the CSZ, we use synthetic seismograms from a suite of 30 simulated M 9 earthquake scenarios on the Cascadia megathrust with varying hypocenters, down-dip rupture extents, slip distributions, and locations of high-stress drop subevents to test the performance of ShakeAlert algorithms. We implement new features not currently set up in the operational ShakeAlert system (version 2.1.5), such as an upgraded version of the FinDer algorithm capable of utilizing generic and fault specific templates, a set of generic crustal templates that increase the maximum allowed rupture length from 300 km to 1362 km, a new version of the eqInfo2GM algorithm that uses precomputed distance tables to determine the spatial extent of ShakeAlert MMI alert threshold contours, and contour distance tables generated with the Next Generation Attenuation – West 2 ground motion models. We measure the timeliness and accuracy of source estimates and evaluate the performance of ShakeAlert alert contours using a station-based alert classification scheme. We also a develop a population-based alert classification method by aligning a 30 arc-second resolution population grid with Voronoi diagrams computed from the classified sites for each scenario. Using raster statistics, we estimate the approximate population in the PNW that would receive timely accurate alerts during an offshore M 9 earthquake. We also observe the range of expected warning times with respect to the spatial distribution of the population. Our results, disaggregated by MMI alert threshold, show that most of the population could receive alerts with positive warning times for an alert threshold of MMI III, but that the number of late and missed alerts increases as the MMI alert threshold is increased. For MMI V, an average of just under 60% of the population would be alerted prior to the arrival of threshold level shaking. Large regions of late and missed alerts for alert thresholds MMI IV and V are caused by delays in alert updates, inaccurate FinDer source estimates, and undersized alert contours. We also evaluate whether some end-users in the MMI V (moderate shaking) late alert zones could receive an alert prior to experiencing MMI VI (strong) or MMI VII (very strong) level shaking. Correct timely alerts increase by about 10% for MMI V using this warning time definition. Finally, we investigate an alerting strategy where ShakeAlert sends out an alert to the entire PNW region when the system detects at least an M 8 earthquake on the coast. This strategy eliminates all missed alerts and all late alerts except at sites close to the epicenter. The mean percentage of timely correct alerts is similar to using an alert threshold of MMI III, but the range of warning times is significantly greater and there is less risk of over-alerting in California