High-quality datasets are critical for the development of advanced machine-learning algorithms in seismology.Here,we present an earthquake dataset based on the ChinArray Phase I records(X1).ChinArray Phase I was deplo...High-quality datasets are critical for the development of advanced machine-learning algorithms in seismology.Here,we present an earthquake dataset based on the ChinArray Phase I records(X1).ChinArray Phase I was deployed in the southern north-south seismic zone(20°N-32°N,95°E-110°E)in 2011-2013 using 355 portable broadband seismic stations.CREDIT-X1local,the first release of the ChinArray Reference Earthquake Dataset for Innovative Techniques(CREDIT),includes comprehensive information for the 105,455 local events that occurred in the southern north-south seismic zone during array observation,incorporating them into a single HDF5 file.Original 100-Hz sampled three-component waveforms are organized by event for stations within epicenter distances of 1,000 km,and records of≥200 s are included for each waveform.Two types of phase labels are provided.The first includes manually picked labels for 5,999 events with magnitudes≥2.0,providing 66,507 Pg,42,310 Sg,12,823 Pn,and 546 Sn phases.The second contains automatically labeled phases for 105,442 events with magnitudes of−1.6 to 7.6.These phases were picked using a recurrent neural network phase picker and screened using the corresponding travel time curves,resulting in 1,179,808 Pg,884,281 Sg,176,089 Pn,and 22,986 Sn phases.Additionally,first-motion polarities are included for 31,273 Pg phases.The event and station locations are provided,so that deep learning networks for both conventional phase picking and phase association can be trained and validated.The CREDIT-X1local dataset is the first million-scale dataset constructed from a dense seismic array,which is designed to support various multi-station deep-learning methods,high-precision focal mechanism inversion,and seismic tomography studies.Additionally,owing to the high seismicity in the southern north-south seismic zone in China,this dataset has great potential for future scientific discoveries.展开更多
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...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.展开更多
Template matching is a useful method to detect seismic events through waveform similarity between two signals.The traditional template matching method works well in detecting small tectonic earthquakes.However,the met...Template matching is a useful method to detect seismic events through waveform similarity between two signals.The traditional template matching method works well in detecting small tectonic earthquakes.However,the method has some difficulty when the signals have relatively low signal-to-noise ratios(SNRs)and simple shapes,e.g.a sinusoidal function.In this study,we modify the traditional template matching approach for this situation.We first construct a virtual three-component seismic station using vertical-component waveforms recorded by three stations.Next,we select a template event from the virtual station,and apply the traditional template matching.We then verify this method by detecting icequakes with simple waveforms on the Urumqi Glacier No.1 and compare the results with those from the short-term-averages over long-term-average(STA/LTA),the REST method,and traditional template matching method.It can be concluded that the modified template matching method using virtual stations has some advantages for seismic data with low SNRs.展开更多
The Sichuan-Yunnan area is located at the southeastern margin of the Tibetan Plateau,where tectonic movement is strong with deep and large faults distributed in a staggered manner,which results in strong seismic activ...The Sichuan-Yunnan area is located at the southeastern margin of the Tibetan Plateau,where tectonic movement is strong with deep and large faults distributed in a staggered manner,which results in strong seismic activities and severe earthquake hazards.Since the 21st century,several earthquakes of magnitude 7.0 or above occurred in this region,which have caused huge casualties and economic losses,especially the 2008 M_(s)8.0 Wenchuan earthquake.At present,earthquake monitoring and source parameter inversion,strong earthquake hazard analysis and disaster assessment are still the focus of seismological researches in the Sichuan-Yunnan region.Regional high-precision 3D community velocity models are fundamental for these studies.In this paper,by assembling seismic observations at permanent seismic stations and several temporary dense seismic arrays in this region,we obtained about 7.06 million body wave travel time data(including absolute and differential travel times)using a newly developed artificial intelligence body wave arrival time picking method and about 100,000 Rayleigh wave phase velocity dispersion data in the period range of 5-50 s from ambient noise cross-correlation technique.Based on this abundant dataset,we obtained the three-dimensional high resolution V_p and V_(s)model in the crust and uppermost mantle of southwest(SW)China by adopting the joint body and surface wave travel time tomography method considering the topography effect starting from the first version of community velocity model in SW China(SWChina CVM-1.0).Compared to SWChina CVM-1.0,this newly determined velocity model has higher resolution and better data fitness.It is accepted by the China Seismic Experimental Site as the second version of the community velocity model in SW China(SWChina CVM-2.0).The new model shows strong lateral heterogeneities in the shallow crust.Two disconnected low velocity zones are observed in the middle to lower crust,which is located in the Songpan-Ganzi block and the northern Chuandian block to the west of the Longmenshan-Lijiang-Xiaojinhe fault,and beneath the Xiaojiang fault zone,respectively.The inner zone of the Emeishan large igneous province(ELIP)exhibits a high velocity anomaly,which separates the two aforementioned low velocity anomalies.Low velocity anomaly is also shown beneath the Tengchong volcano.The velocity structures in the vicinity of the 2008 M_(s)8.0 Wenchuan earthquake,the 2013 M_(s)7.0Lushan earthquake and the 2017 M_(s)7.0 Jiuzhaigou earthquake mainly show high V_(p)and V_(s)anomalies and the mainshocks are basically located at the transition zone between the high and low velocity anomalies.Along with the segmentation characteristics of seismic activity,we suggest that areas with significant changes in velocity structures,especially in active fault zones,might have a greater potential to generate moderate to strong earthquakes.展开更多
基金funded by the National Key R&D Program of China (No. 2021YFC3000702)the Special Fund of the Institute of Geophysics, China Earthquake Administration (No. DQJB20B15)+2 种基金the National Natural Science Foundation of China youth Grant (No. 41804059)the Joint Funds of the National Natural Science Foundation of China (No. U223920029)the Science for Earthquake Resilience of China Earthquake Administration (No. XH211103)
文摘High-quality datasets are critical for the development of advanced machine-learning algorithms in seismology.Here,we present an earthquake dataset based on the ChinArray Phase I records(X1).ChinArray Phase I was deployed in the southern north-south seismic zone(20°N-32°N,95°E-110°E)in 2011-2013 using 355 portable broadband seismic stations.CREDIT-X1local,the first release of the ChinArray Reference Earthquake Dataset for Innovative Techniques(CREDIT),includes comprehensive information for the 105,455 local events that occurred in the southern north-south seismic zone during array observation,incorporating them into a single HDF5 file.Original 100-Hz sampled three-component waveforms are organized by event for stations within epicenter distances of 1,000 km,and records of≥200 s are included for each waveform.Two types of phase labels are provided.The first includes manually picked labels for 5,999 events with magnitudes≥2.0,providing 66,507 Pg,42,310 Sg,12,823 Pn,and 546 Sn phases.The second contains automatically labeled phases for 105,442 events with magnitudes of−1.6 to 7.6.These phases were picked using a recurrent neural network phase picker and screened using the corresponding travel time curves,resulting in 1,179,808 Pg,884,281 Sg,176,089 Pn,and 22,986 Sn phases.Additionally,first-motion polarities are included for 31,273 Pg phases.The event and station locations are provided,so that deep learning networks for both conventional phase picking and phase association can be trained and validated.The CREDIT-X1local dataset is the first million-scale dataset constructed from a dense seismic array,which is designed to support various multi-station deep-learning methods,high-precision focal mechanism inversion,and seismic tomography studies.Additionally,owing to the high seismicity in the southern north-south seismic zone in China,this dataset has great potential for future scientific discoveries.
基金jointly supported by the National Natural Science Foundation of China (No. 42074060)the Special Fund, Institute of Geophysics, China Earthquake Administration (CEA-IGP) (Nos. DQJB19B29, DQJB20B15, and DQJB22Z01)supported by XingHuo Project, CEA (No. XH211103)
文摘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.
基金financially supported by the National Key R&D Program of China(No.2018YFC1504200)the LU JIAXI International Team Program from the KC Wong Education Foundation and CAS(No.GJTD-2018-12)National Natural Science Foundation of China(Nos.41661164035 and 41704066).
文摘Template matching is a useful method to detect seismic events through waveform similarity between two signals.The traditional template matching method works well in detecting small tectonic earthquakes.However,the method has some difficulty when the signals have relatively low signal-to-noise ratios(SNRs)and simple shapes,e.g.a sinusoidal function.In this study,we modify the traditional template matching approach for this situation.We first construct a virtual three-component seismic station using vertical-component waveforms recorded by three stations.Next,we select a template event from the virtual station,and apply the traditional template matching.We then verify this method by detecting icequakes with simple waveforms on the Urumqi Glacier No.1 and compare the results with those from the short-term-averages over long-term-average(STA/LTA),the REST method,and traditional template matching method.It can be concluded that the modified template matching method using virtual stations has some advantages for seismic data with low SNRs.
基金supported by the National Natural Science Foundation of China(Grant Nos.42004034,U1839205,42125401)the Special Fund of the Institute of Geophysics,China Earthquake Administration(Grant No.DQJB22Z01)the National Key R&D Program of China(Grant No.2021YFC3000602)。
文摘The Sichuan-Yunnan area is located at the southeastern margin of the Tibetan Plateau,where tectonic movement is strong with deep and large faults distributed in a staggered manner,which results in strong seismic activities and severe earthquake hazards.Since the 21st century,several earthquakes of magnitude 7.0 or above occurred in this region,which have caused huge casualties and economic losses,especially the 2008 M_(s)8.0 Wenchuan earthquake.At present,earthquake monitoring and source parameter inversion,strong earthquake hazard analysis and disaster assessment are still the focus of seismological researches in the Sichuan-Yunnan region.Regional high-precision 3D community velocity models are fundamental for these studies.In this paper,by assembling seismic observations at permanent seismic stations and several temporary dense seismic arrays in this region,we obtained about 7.06 million body wave travel time data(including absolute and differential travel times)using a newly developed artificial intelligence body wave arrival time picking method and about 100,000 Rayleigh wave phase velocity dispersion data in the period range of 5-50 s from ambient noise cross-correlation technique.Based on this abundant dataset,we obtained the three-dimensional high resolution V_p and V_(s)model in the crust and uppermost mantle of southwest(SW)China by adopting the joint body and surface wave travel time tomography method considering the topography effect starting from the first version of community velocity model in SW China(SWChina CVM-1.0).Compared to SWChina CVM-1.0,this newly determined velocity model has higher resolution and better data fitness.It is accepted by the China Seismic Experimental Site as the second version of the community velocity model in SW China(SWChina CVM-2.0).The new model shows strong lateral heterogeneities in the shallow crust.Two disconnected low velocity zones are observed in the middle to lower crust,which is located in the Songpan-Ganzi block and the northern Chuandian block to the west of the Longmenshan-Lijiang-Xiaojinhe fault,and beneath the Xiaojiang fault zone,respectively.The inner zone of the Emeishan large igneous province(ELIP)exhibits a high velocity anomaly,which separates the two aforementioned low velocity anomalies.Low velocity anomaly is also shown beneath the Tengchong volcano.The velocity structures in the vicinity of the 2008 M_(s)8.0 Wenchuan earthquake,the 2013 M_(s)7.0Lushan earthquake and the 2017 M_(s)7.0 Jiuzhaigou earthquake mainly show high V_(p)and V_(s)anomalies and the mainshocks are basically located at the transition zone between the high and low velocity anomalies.Along with the segmentation characteristics of seismic activity,we suggest that areas with significant changes in velocity structures,especially in active fault zones,might have a greater potential to generate moderate to strong earthquakes.