Ambient noise differential adjoint tomography reveals fluid-bearing rocks near active faults in Los Angeles.

Water scarcity is a pressing issue in California. We develop ambient noise differential adjoint tomography that improves the sensitivity to fluid-bearing rocks by canceling bias caused by noise sources. Here we image the shallow S-wave velocity structure using this method beneath a linear seismic array (LASSIE) in Los Angeles Basin, which shows significant velocity reduction marking a major regional water producer, the Silverado aquifer, along with other fluid-bearing structures. Based on the S-wave tomography and previous P-wave studies, we derive the porosity in Long Beach and discover that the rock from 1-2 km depth surrounding the Newport-Inglewood Fault contains abundant fluids with pore-fluid fraction ~0.33. The high-porosity rock around the fault coincides with previously observed week-long shallow seismicity south of LASSIE array in Long Beach. The imaged S-wave velocity in the top layer shows a similar trend in the geotechnical layer Vs 30, suggesting additional applications to ground motion prediction.

Slow earthquake scaling reconsidered as a boundary between distinct modes of rupture propagation.

The scaling law for slow earthquakes, which is a linear relationship between seismic moment and duration, was proposed 15 y ago and initiated a debate on the difference in physical processes governing slow vs. fast (ordinary) earthquakes. Based on new observations across a wide period range, we show that linear scaling of slow earthquakes remains valid, but as a well-defined upper bound on moment rate of ~1013 Nm/s. The large gap in moment-rate between the scaling of slow and fast earthquakes remains unfilled. Slow earthquakes occur near the detectability threshold, such that we are unable to detect deformation events with lower moment rates. Observed trends within slow earthquake categories support the idea that this unobservable field is populated with events of lower moment rate. This suggests a change in perspective - that the proposed scaling should be considered as a bound, or speed limit, on slow earthquakes. We propose that slow earthquakes represent diffusional propagation, and that the bound on moment rate reflects an upper limit on the speed of those diffusional processes. Ordinary earthquakes, in contrast, occur as a coupled process between seismic wave propagation and fracture. Thus, even though both phenomena occur as shear slip, the difference of scaling reflects a difference in the physical process governing propagation.

A comprehensive suite of earthquake catalogues for the 2016-2017 Central Italy seismic sequence.

The protracted nature of the 2016-2017 central Italy seismic sequence, with multiple damaging earthquakes spaced over months, presented serious challenges for the duty seismologists and emergency managers as they assimilated the growing sequence to advise the local population. Uncertainty concerning where and when it was safe to occupy vulnerable structures highlighted the need for timely delivery of scientifically based understanding of the evolving hazard and risk. Seismic hazard assessment during complex sequences depends critically on up-to-date earthquake catalogues-i.e., data on locations, magnitudes, and activity of earthquakes-to characterize the ongoing seismicity and fuel earthquake forecasting models. Here we document six earthquake catalogues of this sequence that were developed using a variety of methods. The catalogues possess different levels of resolution and completeness resulting from progressive enhancements in the data availability, detection sensitivity, and hypocentral location accuracy. The catalogues range from real-time to advanced machine-learning procedures and highlight both the promises as well as the challenges of implementing advanced workflows in an operational environment.

Deep-learning seismology.

Seismic waves from earthquakes and other sources are used to infer the structure and properties of Earth's interior. The availability of large-scale seismic datasets and the suitability of deep-learning techniques for seismic data processing have pushed deep learning to the forefront of fundamental, long-standing research investigations in seismology. However, some aspects of applying deep learning to seismology are likely to prove instructive for the geosciences, and perhaps other research areas more broadly. Deep learning is a powerful approach, but there are subtleties and nuances in its application. We present a systematic overview of trends, challenges, and opportunities in applications of deep-learning methods in seismology.

Toward improved urban earthquake monitoring through deep-learning-based noise suppression.

Earthquake monitoring in urban settings is essential but challenging, due to the strong anthropogenic noise inherent to urban seismic recordings. Here, we develop a deep-learning-based denoising algorithm, UrbanDenoiser, to filter out urban seismological noise. UrbanDenoiser strongly suppresses noise relative to the signals, because it was trained using waveform datasets containing rich noise sources from the urban Long Beach dense array and high signal-to-noise ratio (SNR) earthquake signals from the rural San Jacinto dense array. Application to the dense array data and an earthquake sequence in an urban area shows that UrbanDenoiser can increase signal quality and recover signals at an SNR level down to ~0 dB. Earthquake location using our denoised Long Beach data does not support the presence of mantle seismicity beneath Los Angeles but suggests a fault model featuring shallow creep, intermediate locking, and localized stress concentration at the base of the seismogenic zone.

Statistical bounds on how induced seismicity stops.

Earthquakes caused by human activities receive scrutiny due to the risks and hazards they pose. Seismicity that occurs after the causative anthropogenic operation stops has been particularly problematic-both because of high-profile cases of damage caused by this trailing seismicity and due to the loss of control for risk management. With this motivation, we undertake a statistical examination of how induced seismicity stops. We borrow the concept of Båth's law from tectonic aftershock sequences. Båth's law anticipates the difference between magnitudes in two subsets of seismicity as dependent on their population count ratio. We test this concept for its applicability to induced seismicity, including ~ 80 cases of earthquakes caused by hydraulic fracturing, enhanced geothermal systems, and other fluid-injections with clear operational end points. We find that induced seismicity obeys Båth's law: both in terms of the magnitude-count-ratio relationship and the power law distribution of residuals. Furthermore, the distribution of count ratios is skewed and heavy-tailed, with most earthquakes occurring during stimulation/injection. We discuss potential models to improve the characterization of these count ratios and propose a Seismogenic Fault Injection Test to measure their parameters in situ. We conclude that Båth's law quantifies the occurrence of earthquake magnitudes trailing anthropogenic operations.

Ambient noise Love wave attenuation tomography for the LASSIE array across the Los Angeles basin.

The Los Angeles basin is located within the North America-Pacific plate boundary and contains multiple earthquake faults that threaten greater Los Angeles. Seismic attenuation tomography has the potential to provide important constraints on wave propagation in the basin and to provide supplementary information on structure in the form of the distribution of anelastic properties. On the basis of the amplitude information from seismic interferometry from the linear LASSIE array in the Los Angeles basin, we apply station-triplet attenuation tomography to obtain a 2D depth profile for the attenuation structure of the uppermost 0.6 km. The array crosses four Quaternary faults, three of which are blind. The attenuation tomography resolves strong attenuation (shear attenuation Qs ~ 20) for the fault zones and is consistent with sharp boundaries across them.

A risk-based approach for managing hydraulic fracturing-induced seismicity.

Risks from induced earthquakes are a growing concern that needs effective management. For hydraulic fracturing of the Eagle Ford shale in southern Texas, we developed a risk-informed strategy for choosing red-light thresholds that require immediate well shut-in. We used a combination of datasets to simulate spatially heterogeneous nuisance and damage impacts. Simulated impacts are greater in the northeast of the play and smaller in the southwest. This heterogeneity is driven by concentrations of population density. Spatially varying red-light thresholds normalized on these impacts [moment magnitude (M w) 2.0 to 5.0] are fairer and safer than a single threshold applied over a broad area. Sensitivity tests indicate that the forecast maximum magnitude is the most influential parameter. Our method provides a guideline for traffic light protocols and managing induced seismicity risks.

Revisiting evidence for widespread seismicity in the upper mantle under Los Angeles.

We revisit the finding of widespread deep seismicity in the upper mantle imaged with a dense, temporary nodal seismic array in Long Beach, California using back-projection to detect candidate events and trace randomization to develop a reliable imaging threshold for candidate detections. We find that nearly all detections of small events at depths greater than 20 kilometers in the upper mantle fall below the reliability threshold. We find a modest number of small, shallower events in the crust that appear to align with the active Newport-Inglewood Fault. These events occur primarily at 15- to 20-kilometer depth near the base of the seismogenic zone. Localized seismicity under fault zones suggests that the deep extensions of active faults are localized and deforming, with stress concentration leading to a concentration of small events, near the seismic-aseismic transition.

Laboratory earthquake forecasting: A machine learning competition.

Earthquake prediction, the long-sought holy grail of earthquake science, continues to confound Earth scientists. Could we make advances by crowdsourcing, drawing from the vast knowledge and creativity of the machine learning (ML) community? We used Google's ML competition platform, Kaggle, to engage the worldwide ML community with a competition to develop and improve data analysis approaches on a forecasting problem that uses laboratory earthquake data. The competitors were tasked with predicting the time remaining before the next earthquake of successive laboratory quake events, based on only a small portion of the laboratory seismic data. The more than 4,500 participating teams created and shared more than 400 computer programs in openly accessible notebooks. Complementing the now well-known features of seismic data that map to fault criticality in the laboratory, the winning teams employed unexpected strategies based on rescaling failure times as a fraction of the seismic cycle and comparing input distribution of training and testing data. In addition to yielding scientific insights into fault processes in the laboratory and their relation with the evolution of the statistical properties of the associated seismic data, the competition serves as a pedagogical tool for teaching ML in geophysics. The approach may provide a model for other competitions in geosciences or other domains of study to help engage the ML community on problems of significance.

Earthquake transformer-an attentive deep-learning model for simultaneous earthquake detection and phase picking.

Earthquake signal detection and seismic phase picking are challenging tasks in the processing of noisy data and the monitoring of microearthquakes. Here we present a global deep-learning model for simultaneous earthquake detection and phase picking. Performing these two related tasks in tandem improves model performance in each individual task by combining information in phases and in the full waveform of earthquake signals by using a hierarchical attention mechanism. We show that our model outperforms previous deep-learning and traditional phase-picking and detection algorithms. Applying our model to 5 weeks of continuous data recorded during 2000 Tottori earthquakes in Japan, we were able to detect and locate two times more earthquakes using only a portion (less than 1/3) of seismic stations. Our model picks P and S phases with precision close to manual picks by human analysts; however, its high efficiency and higher sensitivity can result in detecting and characterizing more and smaller events.

CRED: A Deep Residual Network of Convolutional and Recurrent Units for Earthquake Signal Detection.

Earthquake signal detection is at the core of observational seismology. A good detection algorithm should be sensitive to small and weak events with a variety of waveform shapes, robust to background noise and non-earthquake signals, and efficient for processing large data volumes. Here, we introduce the Cnn-Rnn Earthquake Detector (CRED), a detector based on deep neural networks. CRED uses a combination of convolutional layers and bi-directional long-short-term memory units in a residual structure. It learns the time-frequency characteristics of the dominant phases in an earthquake signal from three component data recorded on individual stations. We train the network using 500,000 seismograms (250k associated with tectonic earthquakes and 250k identified as noise) recorded in Northern California. The robustness of the trained model with respect to the noise level and non-earthquake signals is shown by applying it to a set of semi-synthetic signals. We also apply the model to one month of continuous data recorded at Central Arkansas to demonstrate its efficiency, generalization, and sensitivity. Our model is able to detect more than 800 microearthquakes as small as -1.3 ML induced during hydraulic fracturing far away than the training region. We compare the performance of the model with the STA/LTA, template matching, and FAST algorithms. Our results indicate an efficient and reliable performance of CRED. This framework holds great promise for lowering the detection threshold while minimizing false positive detection rates.

Machine learning for data-driven discovery in solid Earth geoscience.

Understanding the behavior of Earth through the diverse fields of the solid Earth geosciences is an increasingly important task. It is made challenging by the complex, interacting, and multiscale processes needed to understand Earth's behavior and by the inaccessibility of nearly all of Earth's subsurface to direct observation. Substantial increases in data availability and in the increasingly realistic character of computer simulations hold promise for accelerating progress, but developing a deeper understanding based on these capabilities is itself challenging. Machine learning will play a key role in this effort. We review the state of the field and make recommendations for how progress might be broadened and accelerated.

Stress drops of induced and tectonic earthquakes in the central United States are indistinguishable.

Induced earthquakes currently pose a significant hazard in the central United States, but there is considerable uncertainty about the severity of their ground motions. We measure stress drops of 39 moderate-magnitude induced and tectonic earthquakes in the central United States and eastern North America. Induced earthquakes, more than half of which are shallower than 5 km, show a comparable median stress drop to tectonic earthquakes in the central United States that are dominantly strike-slip but a lower median stress drop than that of tectonic earthquakes in the eastern North America that are dominantly reverse-faulting. This suggests that ground motion prediction equations developed for tectonic earthquakes can be applied to induced earthquakes if the effects of depth and faulting style are properly considered. Our observation leads to the notion that, similar to tectonic earthquakes, induced earthquakes are driven by tectonic stresses.

Earthquake detection through computationally efficient similarity search.

Seismology is experiencing rapid growth in the quantity of data, which has outpaced the development of processing algorithms. Earthquake detection-identification of seismic events in continuous data-is a fundamental operation for observational seismology. We developed an efficient method to detect earthquakes using waveform similarity that overcomes the disadvantages of existing detection methods. Our method, called Fingerprint And Similarity Thresholding (FAST), can analyze a week of continuous seismic waveform data in less than 2 hours, or 140 times faster than autocorrelation. FAST adapts a data mining algorithm, originally designed to identify similar audio clips within large databases; it first creates compact "fingerprints" of waveforms by extracting key discriminative features, then groups similar fingerprints together within a database to facilitate fast, scalable search for similar fingerprint pairs, and finally generates a list of earthquake detections. FAST detected most (21 of 24) cataloged earthquakes and 68 uncataloged earthquakes in 1 week of continuous data from a station located near the Calaveras Fault in central California, achieving detection performance comparable to that of autocorrelation, with some additional false detections. FAST is expected to realize its full potential when applied to extremely long duration data sets over a distributed network of seismic stations. The widespread application of FAST has the potential to aid in the discovery of unexpected seismic signals, improve seismic monitoring, and promote a greater understanding of a variety of earthquake processes.