A nearly 20-year hiatus in major seismic activity in southern California ended on 4 July 2019 with a sequence of intersecting earthquakes near the city of Ridgecrest, California. This sequence included a foreshock with a moment magnitude (Mw) of 6.4 followed by a Mw 7.1 mainshock nearly 34 hours later. Geodetic, seismic, and seismicity data provided an integrative view of this sequence, which ruptured an unmapped multiscale network of interlaced orthogonal faults. This complex fault geometry persists over the entire seismogenic depth range. The rupture of the mainshock terminated only a few kilometers from the major regional Garlock fault, triggering shallow creep and a substantial earthquake swarm. The repeated occurrence of multifault ruptures, as revealed by modern instrumentation and analysis techniques, poses a formidable challenge in quantifying regional seismic hazards.

Earthquakes follow a well-known power-law size relation, with smaller events occurring much more often than larger events. Earthquake catalogs are thus dominated by small earthquakes yet are still missing a much larger number of even smaller events because of signal fidelity issues. To overcome these limitations, we applied a template-matching detection technique to the entire waveform archive of the regional seismic network in Southern California. This effort resulted in a catalog with 1.81 million earthquakes, a 10-fold increase, which provides important insights into the geometry of fault zones at depth, foreshock behavior and nucleation processes, and earthquake-triggering mechanisms. The rich detail resolved in this type of catalog will facilitate the next generation of analyses of earthquakes and faults.

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 (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 is not represented in the training set (in particular, very large magnitude events). This generalized phase detection (GPD) framework will significantly improve earthquake monitoring and catalogs, which form the underlying basis for a wide range of basic and applied seismological research.

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 are 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 s. 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.

This article provides an overview of current applications of machine learning (ML) in seismology. ML techniques are becoming increasingly widespread in seismology, with applications ranging from identifying unseen signals and patterns to extracting features that might improve our physical understanding. The survey of the applications in seismology presented here serves as a catalyst for further use of ML. Five research areas in seismology are surveyed in which ML classification, regression, clustering algorithms show promise: earthquake detection and phase picking, earthquake early warning (EEW), ground-motion prediction, seismic tomography, and earthquake geodesy. We conclude by discussing the need for a hybrid approach combining data-driven ML with traditional physical modeling.

The vibrant evolutionary patterns made by earthquake swarms are incompatible with standard, effectively two-dimensional (2D) models for general fault architecture. We leverage advances in earthquake monitoring with a deep-learning algorithm to image a fault zone hosting a 4-year-long swarm in southern California. We infer that fluids are naturally injected into the fault zone from below and diffuse through strike-parallel channels while triggering earthquakes. A permeability barrier initially limits up-dip swarm migration but ultimately is circumvented. This enables fluid migration within a shallower section of the fault with fundamentally different mechanical properties. Our observations provide high-resolution constraints on the processes by which swarms initiate, grow, and arrest. These findings illustrate how swarm evolution is strongly controlled by 3D variations in fault architecture.

We discuss several outstanding aspects of seismograms recorded during >4 weeks by a spatially dense Nodal array, straddling the damage zone of the San Jacinto fault in southern California, and some example results. The waveforms contain numerous spikes and bursts of high-frequency waves (up to the recorded 200 Hz) produced in part by minute failure events in the shallow crust. The high spatial density of the array facilitates the detection of 120 small local earthquakes in a single day, most of which not detected by the surrounding ANZA and regional southern California networks. Beamforming results identify likely ongoing cultural noise sources dominant in the frequency range 1-10 Hz and likely ongoing earthquake sources dominant in the frequency range 20-40 Hz. Matched-field processing and back-projection of seismograms provide alternate event location. The median noise levels during the experiment at different stations, waves generated by Betsy gunshots, and wavefields from nearby earthquakes point consistently to several structural units across the fault. Seismic trapping structure and local sedimentary basin produce localized motion amplification and stronger attenuation than adjacent regions. Cross correlations of high-frequency noise recorded at closely spaced stations provide a structural image of the subsurface material across the fault zone. The high spatial density and broad frequency range of the data can be used for additional high resolution studies of structure and source properties in the shallow crust.

Distinct on-fault and off-fault seismicity in the trifurcation area of the San Jacinto fault zone.