RED-PAN: Real-Time Earthquake Detection and Phase-Picking With Multitask Attention Network

Liao, Wu-Yu; Lee, En‐Jui; Chen, Da‐Yi; Chen, Po; Mu, Dawei; Wu, Yih‐Min

doi:10.1109/tgrs.2022.3205558

Cited by 7 publications

(8 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although the specific problem of "prediction inconsistency" has been reported (Park et al 2023), DL-based pickers outperform traditional algorithms and achieve picking accuracies similar to those of skilled analysts (Mousavi and Beroza 2023), and the published models have been used in many studies. Similar to event detection/classification problems, stateof-the-art modules and architectures, such as RNN (Zhou et al 2019), attention (Liao et al 2021(Liao et al , 2022aLi et al 2022a), transformer , and edge convolutional module (Feng et al 2022b), were continuously incorporated into the models to improve their performance. In this approach, a model takes seismic waveforms as inputs and outputs the arrival times as scalar values (Ross et al 2018a) or a time series of probability values with a peak at a picked arrival time (Zhu and Beroza 2019;.…”

Section: Event Detection/classification and Arrival Time Pickingmentioning

confidence: 99%

“…While some studies have used single-station waveforms (Ross et al 2018a;Zhu and Beroza 2019;Woollam et al 2019;Wang et al 2019a;Zheng et al 2020;Liao et al 2021;Tokuda and Nagao 2023), others have applied multiple station records (Zhu et al 2022c;Li et al 2022c;Chen and Li 2022;Feng et al 2022b;Sun et al 2023). As both event detection and phase picking use seismic waveform as input data, they can be processed simultaneously in a single model using multitask learning, which has multiple outputs, including event and arrival time probabilities (Liao et al 2022a;.…”

Section: Event Detection/classification and Arrival Time Pickingmentioning

confidence: 99%

See 1 more Smart Citation

Recent advances in earthquake seismology using machine learning

Kubo,

Naoi,

Kano

2024

Earth Planets Space

View full text Add to dashboard Cite

Given the recent developments in machine-learning technology, its application has rapidly progressed in various fields of earthquake seismology, achieving great success. Here, we review the recent advances, focusing on catalog development, seismicity analysis, ground-motion prediction, and crustal deformation analysis. First, we explore studies on the development of earthquake catalogs, including their elemental processes such as event detection/classification, arrival time picking, similar waveform searching, focal mechanism analysis, and paleoseismic record analysis. We then introduce studies related to earthquake risk evaluation and seismicity analysis. Additionally, we review studies on ground-motion prediction, which are categorized into four groups depending on whether the output is ground-motion intensity or ground-motion time series and the input is features (individual measurable properties) or time series. We discuss the effect of imbalanced ground-motion data on machine-learning models and the approaches taken to address the problem. Finally, we summarize the analysis of geodetic data related to crustal deformation, focusing on clustering analysis and detection of geodetic signals caused by seismic/aseismic phenomena. Graphical Abstract

show abstract

Section: Event Detection/classification and Arrival Time Pickingmentioning

confidence: 99%

Section: Event Detection/classification and Arrival Time Pickingmentioning

confidence: 99%

Recent advances in earthquake seismology using machine learning

Kubo,

Naoi,

Kano

2024

Earth Planets Space

View full text Add to dashboard Cite

show abstract

“…1(b)]. For in situ earthquake waveforms, we detected and manually checked the P and S arrivals using the RED-PAN model [30]. In total, we identified 348 rockfall events recorded by more than two stations, comprising 750 sets of three-component waveforms; 1399 sets of three-component rockfall waveforms labeled with only one station; 193 car-induced events with 495 sets of threecomponent waveforms; 280 sets of engineering signals with 455 three-component waveforms; and 1834 earthquake events with 5324 three-component waveforms.…”

Section: Datamentioning

confidence: 99%

“…2(a)]. The two encoders separately map the Z score standardized input data to a high-dimensional feature space with a series of recurrent-residual convolution (RRC) blocks [30], [31], [32]. Let R enc ( j) wave and R enc( j) spec be the jth level RRC layer of waveform encoder and spectrogram encoder, R enc( j=5) wave ∈ R (10 * 30) and R enc( j=5) spec ∈ R (1 * 8 * 30) are the last layer of both encoders in this study, which are also the inputs for the feature fusion block.…”

Section: B Single-station Detection Modelmentioning

confidence: 99%

RockNet: Rockfall and Earthquake Detection and Association via Multitask Learning and Transfer Learning

Liao

Lee

Wang

et al. 2023

IEEE Trans. Geosci. Remote Sensing

Self Cite

View full text Add to dashboard Cite

Seismological data plays a crucial role in timely slope failure hazard assessments. However, identifying rockfall waveforms from seismic data poses challenges due to their high variability across different events and stations. To address this, we propose RockNet, a deep-learning-based multitask model capable of detecting both rockfall and earthquake events at both the single-station and local seismic network levels. RockNet consists of two submodels: the single-station model, which computes waveform masks for earthquake and rockfall signals and performs earthquake P and S phase picking simultaneously on single-station seismograms, and the association model, which determines the occurrences of local seismic events by aggregating hidden feature maps from the trained single-station model across all stations. Since the rockfall data is relatively scarce and may not be sufficient to train a deep-learning model effectively, we augment the dataset with abundant nonrockfall data and add additional tasks to promote shared interpretability and robustness. RockNet is trained and tested on a local dataset collected from the Luhu tribe in Miaoli, Taiwan, achieving macro F1-scores of 0.983 and 0.990 for the single-station model and the association model, respectively. Furthermore, we evaluate RockNet on an independent dataset collected from the Super-Sauze unstable slope region in France, and it demonstrates good generalization performance in discriminating earthquake,

show abstract

“…In recent years, deep learning-based earthquake detection methods represented by U-net have been rapidly developed (Zhu and Beroza, 2019;Liu et al, 2020;Ross et al, 2020;Jiang Y. et al, 2021). At present, the better deep learningbased seismic phase picking algorithms are developed based on the U-net structure, such as PhaseNet (Zhu and Beroza, 2019), Unet_cea (Zhao et al, 2019), APP (Liu et al, 2020), RED-PAN (Liao et al, 2022). U-net (Ronneberger et al, 2015) was first applied in the field of medical image segmentation.…”

Section: Introductionmentioning

confidence: 99%

CSESnet: A deep learning P-wave detection model based on UNet++ designed for China Seismic Experimental Site

Fan²,

Jiang³

et al. 2023

Front. Earth Sci.

View full text Add to dashboard Cite

Accurate detection of P-wave arrivals has important applications in real-time seismic data processing, such as earthquake monitoring and earthquake early warning. The Sichuan and Yunnan regions, where the China Seismic Experimental Site (CSES) is located, has frequent strong earthquakes and large amount small earthquakes, resulting in serious earthquake disasters. In this paper, we modify the UNet++ network structure and use 490,000 event waveform data and 78,000 noisy data from the CSES as the data set, and analyze the effects of the training set quality, labeled data and loss function on the model performance to obtain a new P-wave detection model-CSESnet. The recall, precision and F1 score of this model are 94.6%, 85.4% and 89.7%, respectively. The tests in Beijing Capital Circle (BCC) indicates the performance of the CSESnet decrease little and has good generalization. The test in Luxian M6.0 earthquake shows that CSESnet can also predict the P-wave arrival times of large earthquakes and process strong motion data very well. CSESnet provides a new detection model to improve the earthquake detection capability in CSES.

show abstract

RED-PAN: Real-Time Earthquake Detection and Phase-Picking With Multitask Attention Network

Cited by 7 publications

References 48 publications

Recent advances in earthquake seismology using machine learning

Recent advances in earthquake seismology using machine learning

RockNet: Rockfall and Earthquake Detection and Association via Multitask Learning and Transfer Learning

CSESnet: A deep learning P-wave detection model based on UNet++ designed for China Seismic Experimental Site

Contact Info

Product

Resources

About