Seismic Waveform Data Research Articles

A study on the effect of input data length on a deep-learning-based magnitude classifier

Abstract. The rapid characterisation of earthquake parameters such as its magnitude is at the heart of earthquake early warning (EEW). In traditional EEW methods, the robustness in the estimation of earthquake parameters has been observed to increase with the length of input data. Since time is a crucial factor in EEW applications, in this paper we propose a deep-learning-based magnitude classifier based on data from a single seismic station and further investigate the effect of using five different durations of seismic waveform data after first P-wave arrival: 1, 3, 10, 20 and 30 s. This is accomplished by testing the performance of the proposed model that combines convolution and bidirectional long short-term memory units to classify waveforms based on their magnitude into three classes: “noise”, “low-magnitude events” and “high-magnitude events”. Herein, any earthquake signal with magnitude equal to or above 5.0 is labelled as “high-magnitude”. We show that the variation in the results produced by changing the length of the data is no more than the inherent randomness in the trained models due to their initialisation. We further demonstrate that the model is able to successfully classify waveforms over wide ranges of both hypocentral distance and signal-to-noise ratio.

Seismic P‐wave first‐arrival picking model based on EQK‐IncResNet

SummarySeismic P‐wave first arrival picking is one of the critical problems in seismic source research. At present, the picking of seismic P‐wave first arrival mainly relies on the combination of traditional methods and manual picking. However, traditional algorithms have poor picking performance and low manual picking efficiency, which cannot meet the needs of research and application. Based on the deep learning network InceptionResNet‐V2, this paper proposes an improved network (EQK‐IncResNet) that can be used for seismic P‐wave first arrival picking. Compared with traditional methods, this model does not need to set the threshold manually and only needs to input the three‐component data of the seismic waveform data to intelligently identify the first‐arrival of the seismic P‐wave. This model has a lightweight structure and excellent feature extraction capabilities, which can effectively identify low signal‐to‐noise ratio data. The experimental results show that within the error thresholds of 0.1, 0.3, and 0.5 s, the hit rate of this method is 74.45%, 96.79%, and 98.68%, respectively. The average picking error is 0.031 s, which is better than the traditional methods such as AR‐AIC + STA/LTA and the mainstream deep learning methods such as GRU.

An Atypical Shallow Mw 5.3, 2021 Earthquake in the Western Corinth Rift (Greece)

AbstractModerate‐to‐large earthquakes in rifts may occur on leading boundary faults or inner antithetic faults. Here we show a rare case of the 2020–2021 seismic sequence in the Corinth rift, that culminated in the shallow rupture of the antithetic fault, neither preceded nor followed by the leading fault rupture. The hypocenter of the largest shock (Mw 5.3 of 17 February 2021) was located at ∼8 km depth. However, seismic waveform data, supported by satellite‐geodetic and tide gauge measurements, pointed to rupture at shallow depth (∼3 km), where no earthquakes were previously observed. We show that the earthquake most probably ruptured two orthogonal, conjugate fault segments: a weak nucleation phase occurred in the microseismically highly active sub‐horizontal detachment layer, followed – a few seconds later – by a larger, shallow moment release on a high‐angle, south‐dipping normal fault. The latter is the Mornos offshore fault, antithetic to the leading, north‐dipping Psathopyrgos fault. Our study presents the first instrumental/observational evidence of a very shallow Mw 5+ event in this rift – and one of the few reported worldwide. The depth limit of the main shallow slip patch coincides with the expected crossing of the Mornos fault with the Psathopyrgos fault, stressing the importance of fault segmentation and rooting inherited from the rift history. This unusual shallow slip in a depth range with little background seismicity and few aftershocks needs to be further investigated by dynamic modeling as a possible prototype of hazardous events in rift environments.

Could thermal pressurization have induced the frequency-dependent rupture during the 2019 Mw8.0 Peru intermediate-depth earthquake?

SUMMARY The rupture process of earthquakes at intermediate-depth (∼70–300 km) have rarely been illuminated by a joint analysis of geodetic and seismic observations, hindering our understanding of their dynamic rupture mechanisms. Here we present detailed rupture process of the 2019 Mw8.0 Peru earthquake at the depth of 122 km depth, derived with a holistic approach reconciling InSAR and broad-band seismic waveform data. The joint inversion of InSAR observations and teleseismic body waves results in a finite rupture model that extends ∼200 km along strike, with unilateral rupture towards north that lasted for ∼60 s. There are four major slip patches in the finite fault model which are well corresponding to the position and timing of the sources in back-projection and multiple points source results. The largest slip patch, which occurred ∼40 s after the rupture initiation, had a longer and smoother rise time, and radiated much weaker high-frequency seismic waves compared to other smaller slip patches. This distinct frequency-dependent rupture could be explained by a strong dynamic weakening mechanism. We question whether thermal pressurization of pore free water rather than thermal run away could be such a mechanism. Our frequency content analysis could be generalized to study other earthquakes including those deeper than 300 km.

Seismic impedance inversion using a multi‐input neural network with a two‐step training strategy

ABSTRACTDeep learning has shown excellent performance in simulating complex nonlinear mappings from the seismic data to elastic parameters. However, seismic acoustic impedance estimated from a direct mapping from seismic waveform data to P‐wave impedance (single‐input network) is hampered by the limited frequency bands. In this paper, we propose to incorporate the low‐frequency impedance model to constrain the inversion (multi‐input network). We add a feature fusion layer to force the lateral smoothness. Besides, usually, a given seismic survey is likely to contain only a few well logs, which is insufficient for conventional deep‐learning ‐based methods to learn the complex mapping from seismic data to elastic parameters. The problem is compounded by the fact that a network trained with synthetic data (compensated for the lack of logs) cannot be directly used for field data. Therefore, we propose to use transfer learning to mitigate this issue. The multi‐input neural network is trained using synthetics and real data in two stages. We carry out experiments to demonstrate that the two‐step training multi‐input network approach has high accuracy in the time direction, excellent continuity in the lateral direction and favourable robustness. Synthetic and field data examples demonstrate that the proposed network can accurately predict impedance even with limited logging data, which provides a reference for oil and gas exploration in the actual production process.

The 2014 Zigui Earthquake Sequence near the Three Gorges Dam in China

AbstractSeismicity in the Three Gorges Reservoir (TGR) region increased noticeably as the water level of the reservoir rises, since the Three Gorges Dam (TGD) was built in 2003. Here, we determined moment tensors of six earthquakes in the 2014 Zigui earthquake sequence (ZGS) in the TGR region using local and regional broadband seismic waveform data. We also determined the focal depth of the mainshock using its teleseismic waveform data and then relocated the epicenter using its travel times at local stations. High-precision locations of 64 events in the sequence were obtained by combining the double-difference relative location result and the mainshock’s absolute location. The event locations and moment tensor solutions indicate that the ZGS occurred between depth 5 and 9 km on a previously unknown fault striking southwest and dipping ∼80° to the northwest. The event depth distribution and coulomb stress change estimation suggest that the ZGS were not induced directly by the reservoir water but triggered by the stress change from the annual reservoir water level variation. We estimated that the newly found fault has the potential for a magnitude 5.7 earthquake for which ground motion has a 16% probability to exceed the designed maximum intensity level at the TGD, though it would take more than 100 yr to accumulate the needed amount of slip.

The 2021 Mw7.4 Maduo earthquake: Coseismic slip model, triggering effect of historical earthquakes and implications for adjacent fault rupture potential

On 22 May 2021 (CST), an Mw7.4 earthquake struck Maduo County, Qinghai Province, China, which was the largest seismic event in China since the 2008 Mw7.9 Wenchuan earthquake. Several scientific questions associated with the event could be addressed: (1) what fault slip model can explain the Maduo earthquake? (2) what effects do historical earthquakes impose on the Maduo earthquake? and (3) what implications does the Maduo earthquake have for future rupture potential of adjacent tectonic faults? So we conduct a comprehensive study to answer the three questions by collecting satellite SAR images, GPS data, seismic waveform data, historical earthquakes, and aftershocks associated with the Maduo earthquake. The estimated fault slip model shows that the Maduo earthquake ruptures two faults in a manner of dominant sinistral strike-slip motion, with slip peaks of ~4.8 m occurring near the surface. The minor fault to the east dips to the south accommodating an obvious reverse slip, well consistent with reverse fault scarps, reverse faulting aftershocks, and significant upward surface displacements immediately south of this branch. Such a reverse slip is probably controlled by the motion of nearby major sinistral strike-slip faults (the Eastern Kunlun fault and the Maduo–Gande fault). Among 32 historical Mw≥ 6.0 earthquakes used in this study, we find that the 1937 Mw7.8 Huashixia earthquake may affect the Maduo earthquake most, delaying its occurrence by decreasing the Coulomb failure stress (CFS) at the hypocenter by > 1 bar and on the entire causative fault by an average of 0.68 bar. Besides, the Mw6.1 Yangbi earthquake, which occurred ~4.5 h ahead of the Maduo earthquake, appears to make little influence on the Maduo earthquake because it hardly perturbates the CFS at the hypocenter of the Maduo earthquake. Furthermore, the cumulative CFS change due to both the 32 historical earthquakes and the 2021 Maduo event indicates that the Tuosuo Lake–Maqu segment of the Eastern Kunlun fault may be at high risk of future rupture.

Real‐time deep‐learning inversion of seismic full waveform data for CO2 saturation and uncertainty in geological carbon storage monitoring

ABSTRACTDeep‐learning inversion has recently drawn attention in geological carbon storage research due to its potential of imaging and monitoring carbon storage in real time, significantly improving efficiency and safety of carbon storage operations. We present a deep‐learning full waveform inversion method that after the neural network has been trained can image CO2 saturation and its uncertainty in real time. Our deep‐learning inversion method is based on the U‐Net architecture with the neural network trained on pairs of synthetic seismic data and CO2 saturation models. Accordingly, our training establishes a mapping relationship between seismic data and CO2 saturation models and once fully trained directly estimates CO2 saturation as a function of subsurface location. We further quantify uncertainties of CO2 saturation estimates using the Monte Carlo dropout method and a bootstrap aggregating method. For this proof‐of‐concept study, the CO2 training models and data are derived from the Kimberlina 1.2 model, a hypothetical 3D geological carbon storage model that is constructed based on various geological and hydrological data from the Southern San Joaquin Basin, California. We perform deep‐learning inversion experiments using noise‐free and noisy training and test data sets and compare the results. Our modelling experiments show that (1) the deep‐learning inversion can estimate 2D distributions of CO2 fairly well even in the presence of Gaussian random noise and (2) both CO2 saturation imaging and uncertainty quantification can be done in real time. Our results suggest that the deep‐learning inversion method can serve as a robust real‐time monitoring tool for geological carbon storage and/or other time‐varying reservoir/aquifer properties that result from injection, extraction, and/or other subsurface transport phenomena.

Joint inversion of InSAR, local seismic and teleseismic data sets for the rupture process of the 2020 Mw 6.3 Yutian earthquake and its implications

On 25 June 2020, a Mw 6.3 earthquake struck Yutian county. The earthquake nucleated on a normal fault in the northwestern Tibetan Plateau. Through jointly inverting local seismic and teleseismic waveform data as well as Interferometric Synthetic Aperture Radar (InSAR) Line-of-Sight (LOS) displacement data, the spatiotemporal rupture properties of this earthquake were resolved to investigate and better understand regional crustal extension mechanisms. Inversion results show that this event ruptured along one main asperity concentrated within a depth range of 4.1–13.3 km with a maximum slip of ~0.9 m. The rupture front propagated initially around the hypocenter and then unilaterally toward the north. The whole process lasted approximately 8 s and released a total scalar seismic moment of 3.2 × 1018 Nm (Mw 6.3). The energy-based average stress drop was estimated in the range of 1.9–3.0 MPa. The radiation efficiency was estimated to be 10.0–15.8%, which is relatively low, demonstrating that most of the accumulated strain energy dissipated during rupturing. We suggest that the earthquake was possibly related to the nearby Cenozoic volcano in consideration of the dissipated rupture of the source and the west-dipping causative fault located on the western edge of the volcano. Both the westward motion of the western Kunlun block and the eastward motion of the Kunlun-Qaidam block were responsible for the EW-trending crustal extension in the Yutian district. We supposed that the 2020 normal faulting earthquake was primarily attributed to the different velocities of the Kunlun-Qaidam block moving eastward. Additionally, according to an analysis of coseismic Coulomb stress interactions, the previous 2008 Mw 7.1, 2012 Mw 6.2, and 2014 Mw 6.9 Yutian earthquakes contribute small Coulomb stress disturbance on the hypocenter of the 2020 Mw 6.3 Yutian earthquake which suggest that previous Yutian earthquakes have a limited effect on triggering the 2020 event.

Simultaneous Seismic Deep Attribute Extraction and Attribute Fusion

Seismic attributes comprise an effective method for oil and gas reservoir characterization and prediction. Hundreds of seismic attributes have been introduced in the last 30 years. Among the seismic attributes targeting different reservoir features, the autoencoder (AE) receives a significant amount of attention, as it extracts deep attributes of seismic data, providing more details of seismic lateral features than other seismic waveform data and seismic attributes. However, data-driven deep attributes bring new challenges to interpretation as they lack the support of intrinsic physical mechanisms. Hence, a shared AE (S-AE) method is proposed in this article, which can extract seismic deep attributes and fuse traditional seismic attributes simultaneously. An S-AE is a revised version of an AE, which consists of an encoder and decoder. An S-AE takes the seismic waveform as the input of the encoder and obtains the deep attribute, and the decoder then transforms the deep attribute to reconstruct the seismic waveforms and attributes. In an S-AE, the network in front of the decoder is shared, while the networks after the decoder consist of independent layers. Such a network structure ensures the effect of reconstruction and associates seismic attributes with the extracted deep attribute, so as to achieve the purpose of attribute fusion and deep attribute extraction. The proposed S-AE method is compared with conventional seismic data fusion methods, such as RGB and principal component analysis, and the superiority of the S-AE is demonstrated in both synthetic and field applications.

Physics-Consistent Data-Driven Waveform Inversion With Adaptive Data Augmentation

Seismic full-waveform inversion (FWI) is a nonlinear computational imaging technique that can provide detailed estimates of subsurface geophysical properties. Solving the FWI problem can be challenging due to its ill-posedness and high computational cost. In this work, we develop a new hybrid computational approach to solve FWI that combines physics-based models with data-driven methodologies. In particular, we develop a data augmentation strategy that can not only improve the representativity of the training set but also incorporate important governing physics into the training process and therefore improve the inversion accuracy. To validate the performance, we apply our method to synthetic elastic seismic waveform data generated from a subsurface geologic model built on a carbon sequestration site at Kimberlina, California. We compare our physics-consistent data-driven inversion method to both purely physics-based and purely data-driven approaches and observe that our method yields higher accuracy and greater generalization ability.

Automatic tracking for seismic horizons using convolution feature analysis and optimization algorithm

Seismic horizon tracking is a fundamental aspect of seismic data interpretation. However, seismic horizons are typically obtained using manual tracking or a combination of manual tracking and traditional auto-tracking techniques, either of which is a time-consuming and error-prone process. To improve the efficiency and the accuracy of seismic horizon tracking, we developed a convolution feature analysis method on the basis of the traditional coherent technology combined with the Viterbi algorithm, and proposed a method for auto-tracking seismic horizons in complex exploration areas. Firstly, the local seismic waveform data of the target horizon passing through the drilling area have been extracted as the convolution kernel (i.e. the standard seismic trace). Then, the waveform data of each seismic trace in the whole region have been treated with the convolution to obtain the convolution feature in a sliding time window (i.e, the similarity seismic attribute profile). Finally, the convolution feature data have been taken as the inputs, and constraint optimization is performed for the automatic tracking of the seismic horizon. It has the ability to search for the maximum value by integrating the maximum similarity forward and searching for the shortest path method on synthetic and real seismic data. The obtained results show that the proposed method performs effectively for seismic horizons auto-tracking of low signal-to-noise ratio seismic data in complex exploration areas, and also traces the seismic horizons with good continuity and high accuracy. • Convolution feature analysis and optimization algorithm. • Auto-tracking seismic horizons in complex exploration areas. • The traced horizon exhibit good continuity and high accuracy.

An ObsPy Library for Event Detection and Seismic Attribute Calculation: Preparing Waveforms for Automated Analysis

We have implemented an extension for the observational seismology <em>obspy</em> software package to provide a streamlined tool tailored to the processing of seismic signals from non-earthquake sources, in particular those from deforming systems such as glaciers and landslides. This <em>seismic attributes</em> library provides functionality to: (1) download and/or pre-process seismic waveform data; (2) detect and catalogue seismic events using multi-component signals from one or more seismometers; and (3) calculate characteristics (‘attributes’/‘features’) of the identified events. The workflow is controlled by three main functions that have been tested for the breadth of data types expected from permanent and campaign-deployed seismic instrumentation. A selected STA/LTA-type (short-term average/long-term average), or other, event detection algorithm can be applied to the waveforms and user-defined functions implemented to calculate any required characteristics of the detected events. The code is written in Python 2/3 and is available on GitHub together with detailed documentation and worked examples.

Estimation of Seismic Moment Tensors Using Variational Inference Machine Learning

AbstractWe present an approach for rapidly estimating full moment tensors of earthquakes and their parameter uncertainties based on short time windows of recorded seismic waveform data by considering deep learning of Bayesian Neural Networks (BNNs). The individual neural networks are trained on synthetic seismic waveform data and corresponding known earthquake moment‐tensor parameters. A monitoring volume has been predefined to form a three‐dimensional grid of locations and to train a BNN for each grid point. Variational inference on several of these networks allows us to consider several sources of error and how they affect the estimated full moment‐tensor parameters and their uncertainties. In particular, we demonstrate how estimated parameter distributions are affected by uncertainties in the earthquake centroid location in space and time as well as in the assumed Earth structure model. We apply our approach as a proof of concept on seismic waveform recordings of aftershocks of the Ridgecrest 2019 earthquake with moment magnitudes ranging from Mw 2.7 to Mw 5.5. Overall, good agreement has been achieved between inferred parameter ensembles and independently estimated parameters using classical methods. Our developed approach is fast and robust, and therefore, suitable for down‐stream analyses that need rapid estimates of the source mechanism for a large number of earthquakes.

Site characterizations of BMKG seismic network, Indonesia, based on HVSR analysis

In the last few years BMKG (Agency for Meteorology Climatology and Geophysics) has increased the number of seismic stations significantly. Until mid-2020, when this study was conducted, the number of BMKG stations has reached 339 units. Development of the network is aimed to improve the data quality, speed of earthquake data processing and information dissemination to the public, accuracy of the hypocenter, as well as the magnitude. The objective of this study is to identify the site characteristics of the network. We analysed the noise recorded at BMKG stations throughout Indonesia using the HVSR (Horizontal to Vertical Spectral Ratio) method. The results of HVSR analysis were used to classify the site conditions of each station. We got the number of stations with the classification of hard rock, rock, hard soil, medium soil, and soft soil, 47, 57, 52, 30, and 84, respectively. This site condition represents the stations characteristics and affects the quality of seismic waveform data.

Focal Mechanism and Tectonic Stress Field of Small Earthquakes in Southwest Margin of Dongting Basin

Based on the broadband seismic waveform data recorded by Hunan digital seismic network, 41 focal mechanism solutions of small earthquakes with M L ≥1.8 in the southwest margin of Dongting Basin from 2009 to 2017 were obtained by integrating the P-wave first-motion and maximum amplitude ratio methods. The parameters of focal mechanism solutions were analyzed using mathematical statistics and systematic cluster analysis. The results showed that the seismic properties in this area are mainly thrust type ( 30/41 ), and the principal compressive stress axis in the focal mechanism solution parameters is NNW and NE, while the tensile stress axis is SSW. Compared to previous studies, the inversion results of the focal mechanism of small earthquakes in the southwest margin of Dongting Basin are in good agreement with the focal mechanism solutions of regional historical strong earthquakes, indicating that the focal mechanism solutions of small earthquakes can represent the regional tectonic stress field in this area.

Automatic Identification of Mantle Seismic Phases Using a Convolutional Neural Network

AbstractTypical seismic waveform data sets comprise hundreds of thousands to millions of records. Compilation is performed by time‐consuming handpicking of phase arrival times, or signal processing algorithms such as cross‐correlation. The latter generally underperform compared to handpicking. However, differences in picking methods creates variations in models and interpretation of Earth's structure. Here, we exploit the pattern recognition capabilities of Convolutional Neural Networks (CNN). Using a large handpicked data set, we train a CNN model to identify the seismic shear phase SS. This accelerates, automates, and makes consistent data compilation, a task usually completed by visual inspection and influenced by scientists' choices. The CNN model is employed to identify precursors to SS generated by mantle discontinuities. It identifies precursors in stacked and individual seismograms, producing new measurements of the mantle transition zone with quality comparable to handpicked data. This rapid acquisition of high‐quality observations has implications for automation of future seismic tomography studies.

Tempo-spatial characteristics of repeating seismic events in the middle of Tianshan orogenic belt

Based on the seismic waveform data recorded in the stations of the Xinjiang seismic network from 2009 to 2017, the repeating earthquakes in the middle of Tianshan orogenic belt and its periphery in Xinjiang were determined and relocated by using the waveform correlation technique and the master event approach. The results show that 11 618 of the 30 181 events are repeating events, which consist of 2395 groups of doublets and clusters, accounting for 38.5% of the total events. According to the statistical results of the distance between doublets before and after repeating events relocation, it is estimated that the system location error in the research area is about 5—10 km. In addition, combined with the latest source classification results in this area, the results show that repeating earthquakes of different source types have different spatial and temporal distribution characteristics. Repeating quarry blasts appear mostly as clusters, 93.6% of them occur during the daytime, and they also exhibit a seasonal pattern with more events in summers and fewer ones in winters. Tectonic earthquakes occurred in various thrust faults in the Tianshan orogenic belt, and occurred randomly at any time of the day, and the monthly tectonic event frequency is relatively stable during the studied period. Repeating induced earthquake locations indicate that most of them are located near large gas/oil fields and water reservoirs, but some also geographically overlay tectonic earthquakes in some regions. The occurring time characteristics of induced earthquakes are similar to those of tectonic earthquakes, that is random distribution within 24 hours.

An ObsPy library for event detection and seismic attribute calculation: preparing waveforms for automated analysis

We have implemented an extension for the observational seismology obspy software package to provide a streamlined tool tailored to the processing of seismic signals from non-earthquake sources, in particular those from deforming systems such as glaciers and landslides. This seismic attributes library provides functionality to: (1) download and/or pre-process seismic waveform data; (2) detect and catalogue seismic events using multi-component signals from one or more seismometers; and (3) calculate characteristics ('attributes'/'features') of the identified events. The workflow is controlled by three main functions that have been tested for the breadth of data types expected from permanent and campaign-deployed seismic instrumentation. A selected STA/LTA-type (short-term average/long-term average), or other, event detection algorithm can be applied to the waveforms and user-defined functions implemented to calculate any required characteristics of the detected events. The code is written in Python 2/3 and is available on GitHub together with detailed documentation and worked examples.

Pycheron: A Python-Based Seismic Waveform Data Quality Control Software Package

Abstract Supplementing an existing high-quality seismic monitoring network with openly available station data could improve coverage and decrease magnitudes of completeness; however, this can present challenges when varying levels of data quality exist. Without discerning the quality of openly available data, using it poses significant data management, analysis, and interpretation issues. Incorporating additional stations without properly identifying and mitigating data quality problems can degrade overall monitoring capability. If openly available stations are to be used routinely, a robust, automated data quality assessment for a wide range of quality control (QC) issues is essential. To meet this need, we developed Pycheron, a Python-based library for QC of seismic waveform data. Pycheron was initially based on the Incorporated Research Institutions for Seismology’s Modular Utility for STAtistical kNowledge Gathering but has been expanded to include more functionality. Pycheron can be implemented at the beginning of a data processing pipeline or can process stand-alone data sets. Its objectives are to (1) identify specific QC issues; (2) automatically assess data quality and instrumentation health; (3) serve as a basic service that all data processing builds on by alerting downstream processing algorithms to any quality degradation; and (4) improve our ability to process orders of magnitudes more data through performance optimizations. This article provides an overview of Pycheron, its features, basic workflow, and an example application using a synthetic QC data set.