Benchmark on the accuracy and efficiency of several neural network based phase pickers using datasets from China Seismic Network

Seismic phase pickers based on deep neural networks have been extensively used recently, demonstrating their advantages on both performance and efficiency. However, these pickers are trained with and applied to different data. A comprehensive benchmark based on a single dataset is therefore lacking. Here, using the recently released DiTing dataset, we analyzed performances of seven phase pickers with different network structures, the efficiencies are also evaluated using both CPU and GPU devices. Evaluations based on F₁-scores reveal that the recurrent neural network (RNN) and EQTransformer exhibit the best performance, likely owing to their large receptive fields. Similar performances are observed among PhaseNet (UNet), UNet++, and the lightweight phase picking network (LPPN). However, the LPPN models are the most efficient. The RNN and EQTransformer have similar speeds, which are slower than those of the LPPN and PhaseNet. UNet++ requires the most computational effort among the pickers. As all of the pickers perform well after being trained with a large-scale dataset, users may choose the one suitable for their applications. For beginners, we provide a tutorial on training and validating the pickers using the DiTing dataset. We also provide two sets of models trained using datasets with both 50 Hz and 100 Hz sampling rates for direct application by end-users. All of our models are open-source and publicly accessible.     

基于优化参数的地震P、S波震相到时自动拾取及质量评估

准确拾取P、S波震相到时是深入开展地震波研究工作的基础,本文改进了自动拾取参数优化函数算法和质量评估方案,引入了拾取到时优化方案,使用基于参数优化的频带-带宽拾取算法、AICD拾取算法和峰度拾取算法对腾冲地区7个宽频带地震台站记录的地震资料开展了地震P、S波到时自动拾取,对拾取结果进行了优化和质量判定.结果表明：经参数优化、拾取优化后,采用3种方法自动拾取的P、S波到时与人工拾取到时的时差在0.1 s内的记录占比分别达到74.66%、70.98%.这些参数值均优于算法改进前的同类参数,证明了优化方法的可靠性.     

Enhancing seismic P phase arrival picking based on wavelet denoising and kurtosis picker

P phase arrival picking of weak signals is still challenging in seismology. A wavelet denoising is proposed to enhance seismic P phase arrival picking, and the kurtosis picker is applied on the wavelet-denoised signal to identify P phase arrival. It has been called the WD-K picker. The WD-K picker, which is different from those traditional wavelet-based pickers on the basis of a single wavelet component or certain main wavelet components, takes full advantage of the reconstruction of main detail wavelet components and the approximate wavelet component. The proposed WD-K picker considers more wavelet components and presents a better P phase arrival feature. The WD-K picker has been evaluated on 500 micro-seismic signals recorded in the Chinese Yongshaba mine. The comparison between the WD-K pickings and manual pickings shows the good picking accuracy of the WD-K picker. Furthermore, the WD-K picking performance has been compared with the main detail wavelet component combining-based kurtosis (WDC-K) picker, the single wavelet component-based kurtosis (SW-K) picker, and certain main wavelet component-based maximum kurtosis (MMW-K) picker. The comparison has demonstrated that the WD-K picker has better picking accuracy than the other three-wavelet and kurtosis-based pickers, thus showing the enhanced ability of wavelet denoising.     

东祁连山北缘断裂带基于深度学习的密集台阵地震事件快速检测与定位研究

为监测东祁连山北缘断裂带附近的地震活动性,布设包含240台短周期地震仪的面状密集台阵,进行约30 d的连续观测。首先使用基于深度学习的多台站地震事件检测算法（CNNDetector）进行地震事件检测,然后使用震相拾取网络（PhaseNet）对地震事件进行P波和S波到时拾取,其次使用震相关联算法（REAL）进行震相关联及初定位,最后使用双差定位（hypoDD）进行地震重定位,最终的精定位地震目录中共有517个地震。在密集台阵观测期间,中国地震台网正式地震目录中共有39个位于台阵内的地震事件,相比而言,密集台阵检测到大量小于0级的地震。因此通过布设密集台阵,可提高活动断裂微地震活动性的监测能力。与历史地震空间分布相比,密集台阵地震精定位分布具有较好的一致性,表现出更明显的线性分布特征。基于地震分布,发现研究区域存在与地表断层迹线走向不同的隐伏活跃断裂。     

基于深度学习的地震检测模型在区域台网的泛化性研究

将识别地震的深度学习算法PhaseNet应用于四川台网和首都圈台网,对该模型的泛化能力进行了测试和评估.首先利用2010年1月至2018年10月首都圈台网199个地震台站记录的29 328个事件(ML0～ML4)所对应的126761段事件波形,以及2019年4-9月四川及邻省部分台网227个地震台站记录的16595个事...     

Earthquake detection in the Jiangsu region,China using graphics-processing-unit-based Match & Locate and rapid earthquake association and location*

Earthquake detection and location are essential in earthquake studies, which generally consists of two main classes: waveform-based and pick-based methods. To evaluate the ability of two different methods, a graphics-processing-unit-based Match & Locate (GPU-M&L) method and a rapid earthquake association and location (REAL) method are applied to continuous seismic data recorded by 24 digital seismic stations from Jiangsu Seismic Network during 2013 for comparison. GPU-M&L is one of waveform-based methods by waveform cross-correlations while REAL is one of pick-based method to associate arrivals of different seismic phases and locate events through counting the number of P and S picks and travel time residuals. Twenty-six templates are selected from the Jiangsu Seismic Network local catalog by using the GPU-M&L. The number of newly detected and located events is about 2.8 times more than those listed in the local catalog. We both utilize a deep-neural-network-based arrival-time picking method called PhaseNet and a short-term/long-term average (STA/LTA) trigger algorithm for seismic phase detection and picking by applying the REAL. We then refine seismic locations using a least-squares location method (VELEST) and a high-precision relative location method (hypoDD). By applying STA/LTA and PhaseNet, 1006 and 1893 events are associated and located, respectively. The newly detected events are mainly clustered and show steeply dipping fault planes. By analyzing the performance of these methods based on long-term continuous seismic data, the detected catalogs by the GPU-M&L and REAL show that the magnitudes of completeness are 1.4 and 0.8, respectively, which are smaller than 2.6 given by the local catalog. Although REAL provides improvement compared with GPU-M&L, REAL is highly dependent on phase detection and picking which is strongly affected by signal-noise ratio (SNR). Stations at southeast of the study region with low SNR may lead to few detections in the same area.     

Generalization of PhaseNet in Shandong and its application to the Changqing M4.1 earthquake sequence

Waveforms of seismic events, extracted from January 2019 to December 2021 were used to construct a test dataset to investigate the generalizability of PhaseNet in the Shandong region. The results show that errors in the picking of seismic phases(P-and S-waves) had a broadly normal distribution, mainly concentrated in the ranges of-0.4–0.3 s and-0.4–0.8 s, respectively. These results were compared with those published in the original PhaseNet article and were found to be approximately 0.2–0.4 s l...     

基于深度学习到时拾取自动构建长宁地震前震目录

将深度学习到时拾取、震相关联技术与传统定位方法联系起来,构建一套连续波形自动化处理与地震目录自动构建流程,对于高效充分利用地震资料,提升微震检测能力具有十分重要的意义.我们应用最新发展的迁移学习震相识别技术、震相自动关联技术,对长宁Ms6.0地震震中附近21个台站震前半个月(6月1日-6月17日)的连续记录波形进行P、...     

Comparison of the earthquake detection abilities of PhaseNet and EQTransformer with the Yangbi and Maduo earthquakes

PhaseNet and EQTransformer are two state-of-the-art earthquake detection methods that have been increasingly applied worldwide. To evaluate the generalization ability of the two models and provide insights for the development of new models, this study took the sequences of the Yunnan Yangbi M6.4 earthquake and Qinghai Maduo M7.4 earthquake as examples to compare the earthquake detection effects of the two abovementioned models as well as their abilities to process dense seismic sequences. It has been demonstrated from the corresponding research that due to the differences in seismic waveforms found in different geographical regions, the picking performance is reduced when the two models are applied directly to the detection of the Yangbi and Maduo earthquakes. PhaseNet has a higher recall than EQTransformer, but the recall of both models is reduced by 13%–56% when compared with the results reported in the original papers. The analysis results indicate that neural networks with deeper layers and complex structures may not necessarily enhance earthquake detection performance. In designing earthquake detection models, attention should be paid to not only the balance of depth, width, and architecture but also to the quality and quantity of the training datasets. In addition, noise datasets should be incorporated during training. According to the continuous waveforms detected 21 days before the Yangbi and Maduo earthquakes, the Yangbi earthquake exhibited foreshock, while the Maduo earthquake showed no foreshock activity, indicating that the two earthquakes’ nucleation processes were different.     

机器学习在地震检测与震相识别的应用综述

在地震学研究中地震检测与震相识别是最基础的环节,其拾取速度和精度直接影响其在地震精确定位以及地震层析成像中的应用效率和精度。近年来,机器学习在地震学领域中引起广泛关注。机器学习可以改进传统地震检测和震相识别方法,使它们能达到更加准确,识别率更高的效果。把机器学习方法按照监督学习和无监督学习分类介绍,并对机器学习方法流程进行总结,并对目前在地震检测与震相识别方面应用较为广泛的机器学习方法(卷积神经网络、指纹和相似性阈值、广义相位检测、PhaseNet、模糊聚类)进行综述。结果表明:机器学习在地震事件检测和震相识别将会是主要的手段。数据驱动的机器学习在地震学中的应用和物理模型的联合运用将是未来的发展趋势。     

基于深度卷积神经网络的地震震相拾取方法研究

地震震相拾取是地震数据自动处理的首要环节,包括了信号检测、到时估计和震相识别等过程,震相拾取的准确性直接影响到后续事件关联处理的性能,影响观测报告的质量.为了提高震相拾取的准确性,进而提高观测报告质量,本文采用深度卷积神经网络方法来解决震相拾取问题,构建了多任务卷积神经网络模型,设计了分类和回归的联合损失函数,定义了基于加权的分类损失函数,以三分量地震台站的波形数据作为输入,同时实现对震相的检测识别和到时的精确估计.利用美国南加州地震台网的200万条震相和噪声数据对模型进行训练、验证和测试,对于测试集中直达波P、S震相识别的查全率达到98%以上,到时估计的标准偏差分别为0.067s,0.082s.利用迁移学习和数据增强,将模型用于对我国东北地区台网的6个台站13000条数据的训练、验证和测试中,对该数据集P、S震相查全率分别达到91.21%、85.65%.基于迁移训练后的模型,设计了用于连续数据的震相拾取方法,利用连续的地震数据对该算法进行了实际应用测试,并与国家数据中心和中国地震局的观测报告进行比对,该方法的震相检测识别率平均可达84.5%,验证了该方法在实际应用中的有效性.本文所提出的方法展示了深度神经网络在地震震相拾取中的优异性能,为地震震相和事件的检测识别提供了新的思路.     

General seismic wave and phase detection software driven by deep learning

We developed an automatic seismic wave and phase detection software based on PhaseNet, an efficient and highly generalized deep learning neural network for P- and S-wave phase picking. The software organically combines multiple modules including application terminal interface, docker container, data visualization, SSH protocol data transmission and other auxiliary modules. Characterized by a series of technologically powerful functions, the software is highly convenient for all users. To obtain the P- and S-wave picks, one only needs to prepare three-component seismic data as input and customize some parameters in the interface. In particular, the software can automatically identify complex waveforms (i.e. continuous or truncated waves) and support multiple types of input data such as SAC, MSEED, NumPy array, etc. A test on the dataset of the Wenchuan aftershocks shows the generalization ability and detection accuracy of the software. The software is expected to increase the efficiency and subjectivity in the manual processing of large amounts of seismic data, thereby providing convenience to regional network monitoring staffs and researchers in the study of Earth's interior.     

Improving automatic location procedure by waveform similarity analysis: An application in the South Western Alps (Italy)

The accuracy of automatic procedures for locating earthquakes is influenced by several factors such as errors in picking seismic phases, network geometry, modeling errors and velocity model uncertainties. The main purpose of this work is to improve the performances of the automatic procedure employed for the “quasi-real-time” location of seismic events in North Western Italy by developing a procedure based on a waveform similarity analysis and by using only one seismic station.To detect “earthquake families” a cross-correlation technique was applied to a data set of seismic waveforms recorded in the period 1985-2002, in a small test area (1600 km²) located in the South Western Alps (Italy). Normalized cross-correlation matrices were calculated using about 2700 seismic events, selected on the basis of the signal to noise ratio, manually picked and located by using the Hypoellipse code. The waveform similarity analysis, based on the bridging technique, allowed grouping about 65% of the selected events into 80 earthquake families (multiplets) located inside the area considered. For each earthquake family a master event is selected, manually re-picked and re-located by using Hypoellipse code. Having chosen a reference station (STV) on the basis of the completeness of the available data set, an automatic procedure has been developed with the aim of cross-correlating new seismic recordings (automatically picked) to the waveforms of the events belonging to the detected families. If the new event is proved to belong to a family (on the basis of the cross-correlation values), its hypocenter co-ordinates are defined by the location of the master event of the associated family. The performance of the proposed procedure is tested and demonstrated using a data set of 104 selected earthquakes recorded in the period January 2003-June 2004 and located in the test area. The automatic procedure is able to locate, associating events with the multiplets detected by the waveform similarity analysis, about 50% of the test events, almost independently of the accuracy of the automatic phase picker and without the biasing of the network geometry and of the velocity model uncertainties.     

Reliability of the automatic procedures for locating earthquakes in southwestern Alps and northern Apennines (Italy)

Reliable automatic procedure for locating earthquake in quasi-real time is strongly needed for seismic warning system, earthquake preparedness, and producing shaking maps. The reliability of an automatic location algorithm is influenced by several factors such as errors in picking seismic phases, network geometry, and velocity model uncertainties. The main purpose of this work is to investigate the performances of different automatic procedures to choose the most suitable one to be applied for the quasi-real-time earthquake locations in northwestern Italy. The reliability of two automatic-picking algorithms (one based on the Characteristic Function (CF) analysis, CF picker, and the other one based on the Akaike’s information criterion (AIC), AIC picker) and two location methods (“Hypoellipse” and “NonLinLoc” codes) is analysed by comparing the automatically determined hypocentral coordinates with reference ones. Reference locations are computed by the “Hypoellipse” code considering manually revised data and tested using quarry blasts. The comparison is made on a dataset composed by 575 seismic events for the period 2000–2007 as recorded by the Regional Seismic network of Northwestern Italy. For P phases, similar results, in terms of both amount of detected picks and magnitude of travel time differences with respect to manual picks, are obtained applying the AIC and the CF picker; on the contrary, for S phases, the AIC picker seems to provide a significant greater number of readings than the CF picker. Furthermore, the “NonLinLoc” software (applied to a 3D velocity model) is proved to be more reliable than the “Hypoellipse” code (applied to layered 1D velocity models), leading to more reliable automatic locations also when outliers (wrong picks) are present.     

国内外不同抗震设计规范中场地分类方法的内在关系研究

目前国内外不同行业（如建筑、铁路、公路、构筑物、电力、水利等）在抗震设计的场地分类方面,主要考虑各自的应用特色和具体需求,尚无一套所有相关行业统一认可的标准和依据。本文归纳和对比了国内外多种行业规范在场地土和场地类别划分方面的依据和方法,并讨论和分析了各规范之间的主要异同和内在关系。在此基础上,指出我国在抗震设计方面,应制定一种兼顾多行业需求的通用型场地分类规范,以更好地适应各类工程建设的需求。     

Optimal filtering high-resolution seismic reflection data using a weighted-mode empirical mode decomposition operator

Gas-hydrate related processes in deep-water marine settings exist on spatial scales that challenge conventional seismic reflection profiling to successfully image them. The conventional approach to acoustic identification of buried hydrates is to use advanced, cost-prohibitive survey techniques and highly customized software to define subsurface structure wherein compositional changes may be modeled and/or interpreted. This study adopts a different approach derived from recent theoretical advancements in signal processing. The method consists in optimal filtering high resolution, single-channel seismic reflection data using the Empirical Mode Decomposition (EMD). The time series is decomposed in sub-components and the noisy portions are suppressed adopting the technique that we referred as Weighted Mode(s) EMD. The optimal filtering greatly improves the resolution and fidelity of the seismic data set.High Resolution single channel seismic profiles, acquired over a carbonate\hydrates site in the northern Gulf of Mexico, manipulated in such way, show a complex, shallow subsurface, and suggest potential evidence for buried gas hydrates.     

基于学习型超完备字典的地震数据去噪(英文)

基于变换基函数的方法,是地震去噪处理中最常用的技术之一,它利用地震数据在某种基函数变换域内的稀疏性和可分离性来达到剔除噪声的目的。但传统的做法是事先选定一组固定的变换基并在对应域内进行处理,其效果往往并不十分令人满意。为了探索新的改进方法,我们引入学习型超完备冗余字典,即根据地震模型数据进行学习和训练,以寻求最优的稀疏表示字典,而不是只选用固定的变换基。本文在字典学习中融入全变差最小化策略以压制伪吉布斯现象。我们选用离散傅里叶变换作为初始变换,并以随机噪声为例,对单一的全局变换、未经学习的超完备冗余字典和学习型超完备冗余字典做了比较。结果表明,利用经过训练的超完备冗余字典,在对地震数据进行稀疏表示的同时,也达到了有效去除噪声的目的,可视性和信噪比都得到了明显提高。我们也比较了均匀和不均匀字典子块的效果,结果表明,不均匀的字典子块更利于地震数据去噪。     

3D Seismic survey design using mixed-radix based algorithm inversion

The determination of three-dimensional geometry and acquisition parameters, the seismic acquisition survey design, is constantly subject of studies in obtaining data with the highest seismic quality, operational efficiency and cost minimization. In this paper, we propose a methodology for inverting geometry parameters of three-dimensional orthogonal land seismic surveys based on a direct search method using a mixed-radix based algorithm. In this algorithm, the search space is discretized on a mixed-radix base, which depends on the extreme values and the search resolution of each parameter. We will show how to reparametrize the orthogonal acquisition geometry elements in order to obtain the independents and integers parameters that are necessary to construct the mixed-radix base. For the optimization purpose, we define an objective function to contemplate target parameters associated with the elements of the acquisition geometry directly related to the geophysical and operational constraints. Taking in account that the mathematical functions and the objective function we define for the problem have no significant computational cost, all model space parameters are fast and efficiently tested. We applied the algorithm, using as input data, provided by a one-line roll orthogonal reference geometry, assuming a pair of geological objectives as shallow and deep targets. All selected models that meet both the proposed objectives and the constraints are organized by decreasing order of fitness so that with the mixed-radix inversion algorithm we found not only the best model, but also a set of suitable models. Likewise, with the best set of geometries, it is possible to establish a direct comparison between them, analysing their adherence to the technical and operational requirements according to the availability and degree of detail of each one. We show the top 10 best results as a table, allowing a direct comparison between all aspects of these geometries, and we summarize the results showing graphically the fitness of all selected geometries and the inverted geometry elements for the 1000 best geometries. These graphical displays provide a direct way to understand how each model behaves as the fitness decreases. The algorithm is very flexible and its application can be extended to any environment and type of acquisition geometry, and in any phase study of an area be it regional, exploratory or development.     

Updating default depths in the ISC bulletin

The International Seismological Centre (ISC) publishes the definitive global bulletin of earthquake locations. In the ISC bulletin, we aim to obtain a free depth, but often this is not possible. Subsequently, the first option is to obtain a depth derived from depth phases. If depth phases are not available, we then use the reported depth from a reputable local agency. Finally, as a last resort, we set a default depth.In the past, common depths of 10, 33, or multiples of 50 km have been assigned. Assigning a more meaningful default depth, specific to a seismic region will increase the consistency of earthquake locations within the ISC bulletin and allow the ISC to publish better positions and magnitude estimates. It will also improve the association of reported secondary arrivals to corresponding seismic events.We aim to produce a global set of default depths, based on a typical depth for each area, from well-constrained events in the ISC bulletin or where depth could be constrained using a consistent set of depth phase arrivals provided by a number of different reporters.In certain areas, we must resort to using other assumptions. For these cases, we use a global crustal model (Crust2.0) to set default depths to half the thickness of the crust.     

中国东部及其邻区上地幔顶部Pn波速度结构及各向异性

中国东部地处欧亚板块东南部,紧邻西太平洋板块俯冲带,有着复杂的地质构造和深部结构,是国内外学者研究的重点区域.本文采用Pn波层析成像方法反演得到了中国东部及其邻区上地幔顶部Pn波速度结构及各向异性.研究中应用的Pn到时数据主要来自多种地震观测报告,并特别补充了东北流动台阵、华北流动台阵以及区域地震台网记录的地震事件,拾取了大量高精度的Pn到时数据,最终挑选出2049个台站记录的24072次地震的240814条Pn波到时数据.结果显示,中国东部地区上地幔顶部平均速度为8.06 km·s^-1,速度变化范围为7.81~8.32 km·s^-1.东北地区东部表现为显著的低速异常,华北克拉通中、东部Pn呈现低速,而西部地区则表现为高速异常,华南地块主体表现为高速.反演结果还揭示江汉盆地、下辽河盆地、二连盆地及海拉尔盆地都显示出高速,而四川盆地和松辽盆地内部则呈现出不均匀的结构特征.四川盆地的高速异常显示出明显的分块现象,这可能是该盆地在沉积前具有不同的基底物质;松辽盆地的北部呈现为高速,而南部却表现为低速异常,这一特征与松辽盆地南、北部分别为高、低热流相对应,暗示盆地南部的岩石圈已经历了改造.研究进一步揭示Pn波低速区和高、低速过渡带的各向异性也较为强烈,而大部分强震都发生在这些区域之上的地壳内,说明这些部位容易发生变形而应力集中或产生应力差.