As the number of seismic sensors grows, it is becoming increasingly difficult for analysts to pick phases manually and comprehensively, yet such efforts are fundamental earthquake monitoring. Despite years improvements in automatic phase picking, match performance experienced analysts. A more subtle issue that different may differently, which can introduce bias into locations. We present a deep-neural-network-based arrival-time picking method called "PhaseNet" picks arrival times both P S...

Abstract Earthquake signal detection and seismic phase picking are challenging tasks in the processing of noisy data monitoring microearthquakes. Here we present a global deep-learning model for simultaneous earthquake picking. Performing these two related tandem improves performance each individual task by combining information phases full waveform signals using hierarchical attention mechanism. We show that our outperforms previous traditional phase-picking algorithms. Applying to 5 weeks...

Seismology is a data rich and data-driven science. Application of machine learning for gaining new insights from seismic rapidly evolving sub-field seismology. The availability large amount computational resources, together with the development advanced techniques can foster more robust models algorithms to process analyze signals. Known examples or labeled sets, are essential requisite building supervised models. has data, but reliability those labels highly variable, lack high-quality sets...

Abstract Earthquake signal detection is at the core of observational seismology. A good algorithm should be sensitive to small and weak events with a variety waveform shapes, robust background noise non-earthquake signals, efficient for processing large data volumes. Here, we introduce Cnn-Rnn Detector (CRED), detector based on deep neural networks. CRED uses combination convolutional layers bi-directional long-short-term memory units in residual structure. It learns time-frequency...

In this letter, we use deep neural networks for unsupervised clustering of seismic data. We perform the in a feature space that is simultaneously optimized with assignment, resulting learned representations are effective specific task. To demonstrate application method signal processing, design two different consisting primarily full convolutional and pooling layers apply them to: 1) discriminate waveforms recorded at hypocentral distances 2) first-motion polarities. Our results precisions...

Abstract The 2016–2017 central Italy seismic sequence occurred on an 80 km long normal-fault system. initiated with the Mw 6.0 Amatrice event 24 August 2016, followed by 5.9 Visso 26 October and 6.5 Norcia 30 October. We analyze continuous data from a dense network of 139 stations to build high-precision catalog ∼900,000 earthquakes spanning 1 yr period, based arrival times derived using deep-neural-network-based picker. Our contains order magnitude more events than routinely produced local...

Abstract The ever-increasing networks and quantity of seismic data drive the need for seamless automatic workflows rapid accurate earthquake detection location. In recent years, machine learning (ML)-based pickers have achieved remarkable accuracy efficiency with generalization, thus can significantly improve location previously developed sequential methods. However, inconsistent input or output (I/O) formats between multiple packages often limit their cross application. To reduce format...

Earthquake phase association algorithms aggregate picked seismic phases from a network of seismometers into individual earthquakes and play an important role in earthquake monitoring. Dense networks improved picking methods produce massive data sets, particularly for swarms aftershocks occurring closely time space, making challenging problem. We present new method, the Gaussian Mixture Model Association (GaMMA), that combines mixture model measurements (both amplitude), with location, origin...

Abstract The two principle earthquakes of the July 2019 Ridgecrest, California, earthquake sequence, M W 6.4 and 7.1, their immediate foreshocks thousands aftershocks present a challenging environment for rapid analysis characterization this sequence as it unfolded. In study, we analyze first 6 days using continuous data from available seismic networks to detect locate associated with sequence. We build high‐precision catalog deep‐neural‐network‐based picker—PhaseNet sequential association...

Abstract The important task of tracking seismic activity requires both sensitive detection and accurate earthquake location. Approximate locations can be estimated promptly automatically; however, depend on precise phase picking, which is a laborious time‐consuming task. We adapted deep neural network (DNN) picker trained local data to mesoscale hydraulic fracturing experiments. designed novel workflow, transfer learning‐aided double‐difference tomography, overcome the 3 orders magnitude...

Abstract Fault-zone fluids control effective normal stress and fault strength. While most earthquake models assume a fixed pore fluid pressure distribution, geologists have documented valving behavior, that is, cyclic changes in unsteady migration along faults. Here we quantify through 2-D antiplane shear simulations of sequences on strike-slip with rate-and-state friction, upward Darcy flow permeable zone, permeability evolution. Fluid overpressure develops during the interseismic period,...

Frequency filtering is widely used in routine processing of seismic data to improve the signal-to-noise ratio (SNR) recorded signals and by doing so subsequent analyses. In this paper, we develop a new denoising/decomposition method, DeepDenoiser, based on deep neural network. This network able simultaneously learn sparse representation time-frequency domain non-linear function that maps into masks decompose input signal interest noise (defined as any non-seismic signal). We show...

Full-waveform inversion (FWI) is an accurate imaging approach for modeling the velocity structure by minimizing misfit between recorded and predicted seismic waveforms. However, strong nonlinearity of FWI resulting from fitting oscillatory waveforms can trap optimization in local minima. We have adopted a neural-network-based full-waveform (NNFWI) method that integrates deep neural networks with representing model generative network. Neural naturally introduce spatial correlations as...

The deep magmatic architecture of the Hawaiian volcanic system is central to understanding transport magma from upper mantle individual volcanoes. We leverage advances in earthquake monitoring with learning algorithms image structures underlying a major swarm nearly 200,000 events that rapidly accelerated after 2018 Kīlauea caldera collapse. At depths 36 43 kilometers, we resolve 15-kilometers-long collection near-horizontal sheeted identify as sill complex. These sills connect lower...

Earthquake monitoring workflows are designed to detect earthquake signals and determine source characteristics from continuous waveform data. Recent developments in deep learning seismology have been used improve tasks within that allow the fast accurate detection of up orders magnitude more small events than present conventional catalogs. To facilitate application machine-learning algorithms large-volume seismic records, we developed a cloud-based workflow, QuakeFlow, applies multiple...

Earthquake monitoring by seismic networks typically involves a workflow consisting of phase detection/picking, association, and location tasks. In recent years, the accuracy these individual stages has been improved through use machine learning techniques. this study, we introduce new, end-to-end approach that improves overall earthquake detection jointly optimizing each stage pipeline. We propose neural network architecture for task multi-station processing waveforms recorded over network....

Earthquake monitoring in urban settings is essential but challenging, due to the strong anthropogenic noise inherent seismic recordings. Here, we develop a deep-learning-based denoising algorithm, UrbanDenoiser, filter out seismological noise. UrbanDenoiser strongly suppresses relative signals, because it was trained using waveform datasets containing rich sources from Long Beach dense array and high signal-to-noise ratio (SNR) earthquake signals rural San Jacinto array. Application data an...

Abstract Distributed Acoustic Sensing (DAS) is an emerging technology for earthquake monitoring and subsurface imaging. However, its distinct characteristics, such as unknown ground coupling high noise level, pose challenges to signal processing. Existing machine learning models optimized conventional seismic data struggle with DAS due ultra-dense spatial sampling limited manual labels. We introduce a semi-supervised approach address the phase-picking task of data. use pre-trained PhaseNet...

Earthquake focal mechanisms provide critical in-situ insights about the subsurface faulting geometry and stress state. For frequent small earthquakes (magnitude< 3.5), their are routinely determined using first-arrival polarities picked on vertical component of seismometers. Nevertheless, quality is usually limited by azimuthal coverage local seismic network. The emerging distributed acoustic sensing (DAS) technology, which can convert pre-existing telecommunication cables into arrays...

Abstract Distributed Acoustic Sensing (DAS) is a promising technique to improve the rapid detection and characterization of earthquakes. Previous DAS studies mainly focus on phase information but less amplitude information. In this study, we compile earthquake data from two arrays in California, USA, one submarine array Sanriku, Japan. We develop data‐driven method obtain first scaling relation between magnitude. Our results reveal that amplitudes recorded by different regions follow similar...

ABSTRACT Detecting offshore earthquakes in real time is challenging for traditional land-based seismic networks due to insufficient station coverage. Application of distributed acoustic sensing (DAS) submarine cables has the potential extend reach and thereby improve real-time earthquake detection early warning (EEW). We present a complete workflow modified point-source EEW algorithm, which includes machine-learning-based model P- S-wave phase picking, grid-search location method, locally...

Abstract We revisited the June 2010 to October 2011 Guy‐Greenbrier earthquake sequence in central Arkansas using PhaseNet, a deep neural network trained pick P and S arrival times. applied PhaseNet continuous waveform data used phase association hypocenter relocation locate nearly 90,000 events. Our catalog suggests that consists of two adjacent sequences on same fault second may be associated with wastewater disposal well west Fault, rather than wells north east were previously implicated....