Men-Andrin Meier

Evidence for Universal Earthquake Rupture Initiation Behavior

Earthquake onsets provide a unique opportunity to study physical rupture processes because they are more easily observable than later rupture stages. Despite this relative simplicity, the observational basis for rupture onsets is unclear. Numerous reports of evidence for magnitude-dependent rupture onsets (which imply deterministic rupture behavior, e.g. Colombelli et al., 2014) stand in contradiction to a large body of physics-based rupture modeling efforts, which are mostly based on inherently non-deterministic principles (e.g. Rice, 1993). Here we make use of the abundance of short-distance recordings available today; a magnitude-dependency of onsets should appear most prominently in such recordings. We use a simple method to demonstrate that all ruptures in the studied magnitude range (4 < M < 8) share a universal initial rupture behavior and discuss ensuing implications for physical rupture processes and earthquake early warning.

How “good” are real‐time ground motion predictions from Earthquake Early Warning systems?

Real-time ground motion alerts, as can be provided by Earthquake Early Warning (EEW) systems, need to be both timely and sufficiently accurate to be useful. Yet how timely and how accurate the alerts of existing EEW algorithms are is often poorly understood. In part, this is because EEW algorithm performance is usually evaluated not in terms of ground motion prediction accuracy and timeliness but in terms of other metrics (e.g., magnitude and location estimation errors), which do not directly reflect the usefulness of the alerts from an end user perspective. Here we attempt to identify a suite of metrics for EEW algorithm performance evaluation that directly quantify an algorithms ability to identify target sites that will experience ground motion above a critical (user-defined) ground motion threshold. We process 15,553 recordings from 238 earthquakes with M > 5 (mostly from Japan and southern California) in a pseudo-real-time environment and investigate two end-member EEW methods. We use the metrics to highlight both the potential and limitations of the two algorithms and to show under which circumstances useful alerts can be provided. Such metrics could be used by EEW algorithm developers to convincingly demonstrate the added value of new algorithms or algorithm components. They can complement existing performance metrics that quantify other relevant aspects of EEW algorithms (e.g., false event detection rates) for a comprehensive and meaningful EEW performance analysis.

Evolution of seismicity near the southernmost terminus of the San Andreas Fault: Implications of recent earthquake clusters for earthquake risk in southern California

Three earthquake clusters that occurred in the direct vicinity of the southern terminus of the San Andreas Fault (SAF) in 2001, 2009, and 2016 raised significant concern regarding possible triggering of a major earthquake on the southern SAF, which has not ruptured in more than 320 years. These clusters of small and moderate earthquakes with M ≤ 4.8 added to an increase in seismicity rate in the northern Brawley seismic zone that began after the 1979 M_w 6.5 Imperial Valley earthquake, in contrast to the quiet from 1932 to 1979. The clusters so far triggered neither small nor large events on the SAF. The mostly negative Coulomb stress changes they imparted on the SAF may have reduced the likelihood that the events would initiate rupture on the SAF, although large magnitude earthquake triggering is poorly understood. The relatively rapid spatial and temporal migration rates within the clusters imply aseismic creep as a possible driver rather than fluid migration.

Earthquake Early Warning ShakeAlert System: Testing and Certification Platform

Earthquake early warning systems provide warnings to end users of incoming moderate to strong ground shaking from earthquakes. An earthquake early warning system, ShakeAlert, is providing alerts to beta end users in the western United States, specifically California, Oregon, and Washington. An essential aspect of the earthquake early warning system is the development of a framework to test modifications to code to ensure functionality and assess performance. In 2016, a Testing and Certification Platform (TCP) was included in the development of the Production Prototype version of ShakeAlert. The purpose of the TCP is to evaluate the robustness of candidate code that is proposed for deployment on ShakeAlert Production Prototype servers. TCP consists of two main components: a real‐time in situ test that replicates the real‐time production system and an offline playback system to replay test suites. The real‐time tests of system performance assess code optimization and stability. The offline tests comprise a stress test of candidate code to assess if the code is production ready. The test suite includes over 120 events including local, regional, and teleseismic historic earthquakes, recentering and calibration events, and other anomalous and potentially problematic signals. Two assessments of alert performance are conducted. First, point‐source assessments are undertaken to compare magnitude, epicentral location, and origin time with the Advanced National Seismic System Comprehensive Catalog, as well as to evaluate alert latency. Second, we describe assessment of the quality of ground‐motion predictions at end‐user sites by comparing predicted shaking intensities to ShakeMaps for historic events and implement a threshold‐based approach that assesses how often end users initiate the appropriate action, based on their ground‐shaking threshold. TCP has been developed to be a convenient streamlined procedure for objectively testing algorithms, and it has been designed with flexibility to accommodate significant changes in development of new or modified system code. It is expected that the TCP will continue to evolve along with the ShakeAlert system, and the framework we describe here provides one example of how earthquake early warning systems can be evaluated.

P-wave arrival picking and first-motion polarity determination with deep learning

Determining earthquake hypocenters and focal mechanisms requires precisely measured P-wave arrival times and first-motion polarities. Automated algorithms for estimating these quantities have been less accurate than estimates by human experts, which is problematic for processing large data volumes. Here, we train convolutional neural networks to measure both quantities, which learn directly from seismograms without the need for feature extraction. The networks are trained on 18.2 million manually picked seismograms for the southern California region. Through cross-validation on 1.2 million independent seismograms, the differences between the automated and manual picks have a standard deviation of 0.023 seconds. The polarities determined by the classifier have a precision of 95% when compared with analyst-determined polarities. We show that the classifier picks more polarities overall than the analysts, without sacrificing quality, resulting in almost double the number of focal mechanisms. The remarkable precision of the trained networks indicates that they can perform as well, or better, than expert seismologists.

The limits of earthquake early warning: Timeliness of ground motion estimates

In only rare cases will earthquake early warning systems be able to provide useful warnings for high levels of ground motion. The basic physics of earthquakes is such that strong ground motion cannot be expected from an earthquake unless the earthquake itself is very close or has grown to be very large. We use simple seismological relationships to calculate the minimum time that must elapse before such ground motion can be expected at a distance from the earthquake, assuming that the earthquake magnitude is not predictable. Earthquake early warning (EEW) systems are in operation or development for many regions around the world, with the goal of providing enough warning of incoming ground shaking to allow people and automated systems to take protective actions to mitigate losses. However, the question of how much warning time is physically possible for specified levels of ground motion has not been addressed. We consider a zero-latency EEW system to determine possible warning times a user could receive in an ideal case. In this case, the only limitation on warning time is the time required for the earthquake to evolve and the time for strong ground motion to arrive at a user’s location. We find that users who wish to be alerted at lower ground motion thresholds will receive more robust warnings with longer average warning times than users who receive warnings for higher ground motion thresholds. EEW systems have the greatest potential benefit for users willing to take action at relatively low ground motion thresholds, whereas users who set relatively high thresholds for taking action are less likely to receive timely and actionable information.

Applying Depth Distribution of Seismicity to Determine Thermo-Mechanical Properties of the Seismogenic Crust in Southern California: Comparing Lithotectonic Blocks

We analyze waveform-relocated seismicity (1981–2016) and other geophysical and geological datasets from 16 lithotectonic crustal blocks in southern California. We explore how earthquake depth histograms (EDH) are related to crustal strength, lithology, and temperature of the crust. First, we calculate relative EDHs to quantify the depth distribution of seismicity for each lithotectonic block. Second, we calculate depth profiles of maximum differential stress (“yield strength envelopes”, YSEs) using Byerlee’s law and a non-linear dislocation creep law. We use observed average heat flow values, strain rates, and states of stress to parameterize YSEs for five different crustal candidate lithologies in each lithotectonic block. We assume that seismicity ceases where the mechanical rock strength falls below a critical threshold level, and identify the YSE that best predicts the depth extent of seismicity in each block. The lithologies of the best matching YSEs are found to agree well with expectations from past tectonics: they are mostly quartz-dominated except for the feldspar-rich diorite lithologies in the Great Valley, the southernmost western Sierra Nevada, Inner Continental Borderland, and Rifted crust in the Salton Trough. Similarly, the inferred thermo-mechanical properties, including differential stress, lithology, and geotherms reflect the previously mapped tectonic variability between the 16 lithotectonic blocks. On average, the differential yield stress is smaller and peaks at a shallower depth in hotter and more quartz rich crust but is larger and peaks at greater depths for colder and predominantly diorite crust. The good agreement between the modeled YSEs, the EDHs and tectonic considerations suggests that EDHs indeed reflect long-term geophysical properties of the crust and can be used to infer thermo-mechanical properties at depth. In contrast, shallow seismicity may be more likely to reflect short-term strain transients from fluid flow or recent anthropogenic disturbances.

Machine Learning Seismic Wave Discrimination: Application to Earthquake Early Warning

Performance of earthquake early warning systems suffers from false alerts caused by local impulsive noise from natural or anthropogenic sources. To mitigate this problem, we train a generative adversarial network (GAN) to learn the characteristics of first‐arrival earthquake P waves, using 300,000 waveforms recorded in southern California and Japan. We apply the GAN critic as an automatic feature extractor and train a Random Forest classifier with about 700,000 earthquake and noise waveforms. We show that the discriminator can recognize 99.2% of the earthquake P waves and 98.4% of the noise signals. This state‐of‐the‐art performance is expected to reduce significantly the number of false triggers from local impulsive noise. Our study demonstrates that GANs can discover a compact and effective representation of seismic waves, which has the potential for wide applications in seismology.

Generalized Seismic Phase Detection with Deep LearningShort Note

To optimally monitor earthquake‐generating processes, seismologists have sought to lower detection sensitivities ever since instrumental seismic networks were started about a century ago. Recently, it has become possible to search continuous waveform archives for replicas of previously recorded events (i.e., template matching), which has led to at least an order of magnitude increase in the number of detected earthquakes and greatly sharpened our view of geological structures. Earthquake catalogs produced in this fashion, however, are heavily biased in that they are completely blind to events for which no templates are available, such as in previously quiet regions or for very large‐magnitude events. Here, we show that with deep learning, we can overcome such biases without sacrificing detection sensitivity. We trained a convolutional neural network (ConvNet) on the vast hand‐labeled data archives of the Southern California Seismic Network to detect seismic body‐wave phases. We show that the ConvNet is extremely sensitive and robust in detecting phases even when masked by high background noise and when the ConvNet is applied to new data that are not represented in the training set (in particular, very large‐magnitude events). This generalized phase detection framework will significantly improve earthquake monitoring and catalogs, which form the underlying basis for a wide range of basic and applied seismological research.