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

Mousavi, S. Mostafa; Ellsworth, William L.; Zhu, Weiqiang; Chuang, Lindsay; Beroza, Gregory C.

doi:10.1038/s41467-020-17591-w

Cited by 617 publications

(444 citation statements)

References 42 publications

Supporting

Mentioning

437

Contrasting

Unclassified

Order By: Relevance

“…The major difference between the proposed algorithm and the relatively shallow deep learning architectures, for example, transformer‐deep learning network (Mousavi et al., 2020) is that the proposed algorithm is a different implementation of deep learning for a similar end goal. A drawback coming along with our proposed strategy is the increased training and model convergence times caused by the high complexity of the deep architecture (

\approx

24 million parameters).…”

Section: Discussionmentioning

confidence: 99%

SCALODEEP: A Highly Generalized Deep Learning Framework for Real‐Time Earthquake Detection

et al. 2021

View full text Add to dashboard Cite

The detection of earthquake signals is a fundamental yet challenging task in observational seismology. A robust automatic earthquake detection algorithm is strongly demanded in view of the ever‐growing global seismic dataset. Here, we develop an automatic earthquake detection framework based on a deep learning approach (SCALODEEP). It extracts high‐order features embedded in three‐component seismograms by encoding a time‐frequency representation of the data (scalogram) into a deep network with skip connections. The SCALODEEP is trained and validated on an open‐source dataset from North California, and then employed to seismicity detection in four areas, including Arkansas, Japan, Texas, and Egypt. Despite vastly varying characteristics of regional earthquakes (e.g., focal mechanism, duration, and noise level), SCALODEEP successfully detects seismic signals over a broad range of local magnitudes (as low as −1.3normalML) and outperforms conventional algorithms such as STA/LTA, FAST, and template matching. Compared to recently proposed deep learning based frameworks (e.g., CRED and Earthquake transformer), SCALODEEP achieves a superior generalization ability via a sophisticated network architecture. In summary, our study offers a promising new tool to improve existing earthquake detection systems and, as importantly, sheds light on designing an effective deep learning network for generalized earthquake detection.

show abstract

\approx

24 million parameters).…”

Section: Discussionmentioning

confidence: 99%

SCALODEEP: A Highly Generalized Deep Learning Framework for Real‐Time Earthquake Detection

et al. 2021

View full text Add to dashboard Cite

show abstract

“…In the recent years, detectors based on machine learning / deep learning (ML/DL) have been proposed [ 24 , 25 ]. The method of [ 24 ] delivers probabilities associated with the existence of an earthquake event and two different seismic phases for each time point by using encoders that intrinsically capture the temporal dependencies in seismic data. The decoders consist of carefully designed set of deep learning network models comprising tens of layers and about 372000 tunable parameters for detecting and picking the seismic phase arrivals.…”

Section: Literature Reviewmentioning

confidence: 99%

“…Finally, despite their high levels of complexity, the construct of these models may result in an unacceptable level of false alarm rates. A remedy suggested by [ 24 ] is to explicitly incorporate the spectral features of seismic signals.…”

Section: Literature Reviewmentioning

confidence: 99%

A prediction framework with time-frequency localization feature for detecting the onset of seismic events

2021

View full text Add to dashboard Cite

Onset detection of P-wave in seismic signals is of vital importance to seismologists because it is not only crucial to the development of early warning systems but it also aids in estimating the seismic source parameters. All the existing P-wave onset detection methods are based on a combination of statistical signal processing and time-series modeling ideas. However, these methods do not adequately accommodate some advanced ideas that exist in fault detection literature, especially those based on predictive analytics. When combined with a time-frequency (t-f) / temporal-spectral localization method, the effectiveness of such methods is enhanced significantly. This work proposes a novel real-time automatic P-wave detector and picker in the prediction framework with a time-frequency localization feature. The proposed approach brings a diverse set of capabilities in accurately detecting the P-wave onset, especially in low signal-to-noise ratio (SNR) conditions that all the existing methods fail to attain. The core idea is to monitor the difference in squared magnitudes of one-step-ahead predictions and measurements in the time-frequency bands with a statistically determined threshold. The proposed framework essentially accommodates any suitable prediction methodology and time-frequency transformation. We demonstrate the proposed framework by deploying auto-regressive integrated moving average (ARIMA) models for predictions and the well-known maximal overlap discrete wavelet packet transform (MODWPT) for the t-f projection of measurements. The ability and efficacy of the proposed method, especially in detecting P-waves embedded in low SNR measurements, is illustrated on a synthetic data set and 200 real-time data sets spanning four different geographical regions. A comparison with three prominently used detectors, namely, STA/LTA, AIC, and DWT-AIC, shows improved detection rate for low SNR events, better accuracy of detection and picking, decreased false alarm rate, and robustness to outliers in data. Specifically, the proposed method yields a detection rate of 89% and a false alarm rate of 11.11%, which are significantly better than those of existing methods.

show abstract

“…Recently, deep learning approaches have been also utilized for detection purposes in passive seismic recordings. However, they are restricted only to the extraction of surface waves from DAS data [23][24][25] or earthquake-induced seismicity [26][27][28][29][30][31][32][33][34][35]. The other demonstrated solutions for data selection are mostly semi-automatic and/or performed on preprocessed and cross-correlated data, which in turn requires extra operator workload and computational cost.…”

Section: Introductionmentioning

confidence: 99%

Automatic Image-Based Event Detection for Large-N Seismic Arrays Using a Convolutional Neural Network

2021

View full text Add to dashboard Cite

Passive seismic experiments have been proposed as a cost-effective and non-invasive alternative to controlled-source seismology, allowing body–wave reflections based on seismic interferometry principles to be retrieved. However, from the huge volume of the recorded ambient noise, only selected time periods (noise panels) are contributing constructively to the retrieval of reflections. We address the issue of automatic scanning of ambient noise data recorded by a large-N array in search of body–wave energy (body–wave events) utilizing a convolutional neural network (CNN). It consists of computing first both amplitude and frequency attribute values at each receiver station for all divided portions of the recorded signal (noise panels). The created 2-D attribute maps are then converted to images and used to extract spatial and temporal patterns associated with the body–wave energy present in the data to build binary CNN-based classifiers. The ensemble of two multi-headed CNN models trained separately on the frequency and amplitude attribute maps demonstrates better generalization ability than each of its participating networks. We also compare the prediction performance of our deep learning (DL) framework with a conventional machine learning (ML) algorithm called XGBoost. The DL-based solution applied to 240 h of ambient seismic noise data recorded by the Kylylahti array in Finland demonstrates high detection accuracy and the superiority over the ML-based one. The ensemble of CNN-based models managed to find almost three times more verified body–wave events in the full unlabelled dataset than it was provided at the training stage. Moreover, the high-level abstraction features extracted at the deeper convolution layers can be used to perform unsupervised clustering of the classified panels with respect to their visual characteristics.

show abstract

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

Cited by 617 publications

References 42 publications

SCALODEEP: A Highly Generalized Deep Learning Framework for Real‐Time Earthquake Detection

SCALODEEP: A Highly Generalized Deep Learning Framework for Real‐Time Earthquake Detection

A prediction framework with time-frequency localization feature for detecting the onset of seismic events

Automatic Image-Based Event Detection for Large-N Seismic Arrays Using a Convolutional Neural Network

Contact Info

Product

Resources

About