P-wave Arrival Times Research Articles

Modeling uncertainty in P-wave arrival-times retrieved from DAS data: case-studies from 15 fiber optic cables

Summary Distributed Acoustic Sensing (DAS) technology enables the detection of waves generated by seismic events, generally as uniaxial strain/strain rate time series observed for dense, subsequent, portions of a Fiber Optic Cable (FOC). Despite the advantages in measurement density, data quality is often affected by uniaxial signal polarization, site effects and cable coupling, beyond the physical energy decay with distance. To better understand the relative importance of these factors for data inversion, we attempt a first modeling of noise patterns affecting DAS arrival times for a set of seismic events. The focus is on assessing the impact of noise statistics, together with the geometry of the problem, on epicentral location uncertainties. For this goal, we consider 15 ”real-world” cases of DAS arrays with different geometry, each associated with a seismic event of known location. We compute synthetic P-wave arrival times and contaminate them with four statistical distributions of the noise. We also estimate P-wave arrival times on real waveforms using a standard seismological picker. Eventually, these five datasets are inverted using a Markov chain Monte Carlo (McMC) method, which offers the evaluation of the relative event location differences in terms of Posterior Probability Density (PPD). Results highlight how cable geometry influences the shape, extent, and directionality of the PPDs. However, synthetic tests demonstrate how noise assumptions on arrival times often have important effects on location uncertainties. Moreover, for half of the analyzed case studies, the observed and synthetic locations are more similar when considering noise sources that are independent of the geometrical characteristics of the arrays. Thus, the results indicate that axial polarization, site conditions, and cable coupling, beyond other intrinsic features (e.g., optical noise), are likely responsible for the complex distribution of DAS arrival times. Overall, the noise sensitivity of DAS suggests caution when applying geometry-only-based approaches for the a priori evaluation of novel monitoring systems.

3-D P-wave velocity structure of the upper mantle beneath eastern Indonesia from body wave tomography

Eastern Indonesia's tectonic setting is well known for its complexity and intense seismic activity. Controlled by several major and minor plates, including the Eurasian, Australian, and Pacific plates, this region is famous for its U-shaped subduction system beneath the Banda Arc. To better understand the architecture of the underlying structure in this region, we performed body-wave travel time tomography using ten years of catalog data provided by the Indonesian Agency for Meteorology, Climatology, and Geophysics. We utilize 9729 events in total, from which 46,446 P-wave arrival times were extracted. We used a double difference method to relocate the initial event catalog, which produced a pattern of seismicity consistent with a curved subduction system. Our tomographic model reveals a high velocity band between 90 and 240 km depth in the upper mantle, which is interpreted to be a concave dipping lithospheric slab that is parallel to the present-day Banda arc. Our results also show that lithosphere subducting from the north and south starts to collide at a depth of 300–350 km and becomes shallower further east. Apparent discontinuities in the high velocity band and a corresponding lack of seismicity supports the presence of a slab tear to the west of Seram. A dipping high velocity structure that is present from south to north beneath the island of Timor represents a subducting slab that dips more steeply beyond a depth of 150–200 km, which appears consistent with slab roll-back. Our tomographic model also shows evidence of back arc thrusting to the north of Sumbawa and Flores Islands in the form of a south-dipping higher velocity band at shallow depth. Furthermore, our tomographic models also reveal the possible presence of underthrust continental forearc in the form of a thin higher velocity anomaly that connects the backarc thrust and northward dipping lithosphere slab in the Timor area. Finally, a zone of low velocity above the higher velocity slab is clearly seen beneath Seram Island at a depth of ∼100 km and may represent a partial melting zone.

Application of machine learning methods for earthquake detection from high-density temporary observation seismic records on a volcanic island

We applied two machine learning models to detect earthquakes from records observed with seismometers temporarily installed on a volcanic island. The two models are based on different principles: one regards seismic waveforms as images, using a convolutional neural network (CNN) to determine the first arrival times of P-waves, S-waves. The other model regards seismic waveforms as series data. The model processes seismic waveforms as data in a specific order of noise, P-wave, and S-wave, similar to natural language.The purpose of this study is to present the results of using machine learning first arrival times identification models with two principles for noisy seismic waveforms, caused by sea waves and strong winds in volcanic islands, and to evaluate the effectiveness of machine learning models for noisy observation records.We created a Confusion Matrix using first arrival times determined by an expert and evaluated the detection performance of these two models using some metrics of the matrix. Additionally, we assessed accuracy of the model-identified first arrival times by generating a frequency distribution of the difference from the expert's detecting time.The study discovered that the model treating data as series had superior detection ability for noisy data compared to the one treating data as images and the accuracy of the first arrival time detection was also better for the series data model too.We compared the results obtained on this island with those obtained at the permanent station, which is considered to have less noise interference, described in Mousavi et al., 2020. It was found that the difference in detection ability between the two models is slight for data obtained at permanent stations with low noise interference, but that the difference in detection ability between the algorithms of the two models is significant in noisy environments.

Method for the P-wave arrival pickup of rock fracture acoustic emission signals under strong noise

This research aimed to investigate the accuracy of picking of P-wave arrival times in rock fracture acoustic emission signals. In order to simulate the mining scenario, Gaussian white noise and pulse noise were added to the data collected in the laboratory. Complete ensemble empirical mode decomposition with adaptive noise + Wavelet (CEEMDAN + Wavelet) was improved in this paper, where the Spearman rank correlation coefficient was adopted to effectively select intrinsic mode functions for denoising which retained the inherent characteristics of the rock fracture signal. The absolute amplitude and energy change rate of the envelope signal, calculated based on the Hilbert transform, were used as the input of the short term average/long term average (STA/LTA) normalization algorithm to pickup the P-wave arrival time. The reliability of this method was tested on 30 groups of recorded rock fracture laboratory data and 60 groups of added noise data. Taking the manual pickup results as the standard, the errors of CEEMDAN + Wavelet + STA/LTA + AIC (Akaike information criterion) method with the absolute amplitude of the signal as the input are all within 10 ms, and 86.67% of the results are within 5 ms. The method proposed in this paper effectively addressing the issue of false pickup caused by the sensitivity of AIC and traditional STA/LTA method for strong noise, and achieving relatively high accuracy and stability in processing low signal-to-noise ratio signals. This work contributes to monitor microscopic changes in rock bodies and is of great significance for the prediction and monitoring of geological disasters.

Clock-modeling-constrained Epoch Relative Positioning for GPS coseismic displacement estimation

High-rate GNSS technology is invaluable for capturing surface displacements. However, vertical displacement is strongly linked to zenith tropospheric delay (ZTD) and receiver clock bias, compromising accuracy and impeding the detection of weak seismic waves. Some GNSS stations, fortunately, feature high-stability oscillators, enabling receiver clock modeling instead of epoch-wise estimation. This paper introduces an epoch relative positioning approach (ERP)with receiver clock modeling (CMC-ERP) to enhance coseismic displacement accuracy. Applied to earthquakes in Japan and the United States, the method improves vertical displacement accuracy by 24% to 61%. Furthermore, P-wave arrival times are determined by processing displacement sequences extracted through both CMC-ERP and ERP methods using STA/LTA. Compared to theoretical arrival times calculated by TauP software, CMC-ERP accuracy is enhanced by 74%–99%, further confirming its effectiveness.

The MW5.5 earthquake on August 6, 2023, in Pingyuan, Shandong, China: A rupture on a buried fault

On August 6, 2023, a magnitude MW5.5 earthquake struck Pingyuan County, Dezhou City, Shandong Province, China. This event was significant as no large earthquakes had been recorded in the region for over a century, and no active fault had been previously identified. This study collects 1309 P-wave arrival times and 866 S-wave arrival times from 74 seismic stations less than 200 km to the epicenter to constrain the spatial distribution of the mainshock and its 125 early aftershocks by the double difference earthquake relocation method, and selects 864 P-waveforms from 288 stations located within 800 km of the epicenter to constrain the focal mechanism solution of the mainshock through centroid moment tensor inversion. The relocation and the inversion indicate, the Pingyuan MW5.5 earthquake was caused by a rupture on a buried fault, likely an extensive segment of the Gaotang fault. This buried fault exhibited a dip of approximately 75° to the northwest, with a strike of 222°, similar to the Gaotang fault. The rupture initiated at the depth of 18.6 km and propagated upward and northeastward. However, the ground surface was not broken. The total duration of the rupture was ∼6.0 s, releasing the scalar moment of 2.5895 × 1017 N·m, equivalent to MW5.54. The moment rate reached the maximum only 1.4 seconds after the rupture initiation, and the 90% scalar moment was released in the first 4.6 s. In the first 1.4 seconds of the rupture process, the rupture velocity was estimated to be 2.6 km/s, slower than the local S-wave velocity. As the rupture neared its end, the rupture velocity decreased significantly. This study provides valuable insights into the seismic characteristics of the Pingyuan MW5.5 earthquake, shedding light on the previously unidentified buried fault responsible for the seismic activity in the region. Understanding the behavior of such faults is crucial for assessing seismic hazards and enhancing earthquake preparedness in the future.

An automatic arrival time picking algorithm of P-wave based on adaptive characteristic function

The accuracy of arrival time and picking speed are significant issues in terms of micro-seismic identification, location, travel-time tomography, as well as source mechanism. In order to detect small changes in signal amplitude and frequency with different signal-to-noise ratios (SNRs), the Allen's characteristic function (Allen's CF) was modified to an adaptive characteristic function (ACF) by relative time series and relative energy coefficients which was compared with the Allen's CF, Baer's CF, and Sedlak's CF. Then, an algorithm for S/L-AIC picker based on ACF for automatic P-wave arrival time picking was presented by the short-term-average to long-term-average ratios (STA/LTA) picker and the Akaike Information Criterion (AIC). In addition, the algorithm was applied to a dataset acquired by a local micro-seismic network at the Jing-Gong coal mine in Shanxi, China, which was compared with a two-step AIC picker, using the original manual selection as a reference to verify the reliability and robustness of the algorithm. Furthermore, the influence of five basic parameters on the accuracy of the S/L-AIC picker based on ACF was discussed by orthogonal tests. The results highlighted that: 81 % of the automatic picks processed by the S/L-AIC picker based on ACF were within +4 ms of deviation, while 66.3 % of the automatic picks processed by the two-step AIC picker were within the same error range; Compared with the other CFs, the ACF applied to the S/L-AIC algorithm performs better for low-SNR data and confirms more accurate and stable results. This study highlights that accurate automatic P-wave arrival time determination is feasible when using the S/L-AIC picker based on ACF.

An improved STA/LTA algorithm for P wave detection in the 2015 Nepal Earthquake

Automatic identification of seismic phases is one of the important tasks in earthquake rapid reporting and early warning. The STA/LTA method is currently the most widely used automatic pickup method. However, the accuracy and stability of its pickup results heavily depend on the selection of feature functions, time window lengths, and trigger thresholds, so it is to some extent impossible to achieve automatic pickup of P-wave arrival time. We present an improved algorithm for automatically picking up P-wave arrival times. In this scheme, the weight factor K is introduced to construct a new feature function, and a method for finding the maximum value on the STA/LTA curve corresponding to the P-wave arrival time is proposed. This method was applied to measure the arrival of the 2015 Nepal Earthquake to validate previous evaluations. The results show that our improved method has the advantage of eliminating the time wasted on adjusting the trigger threshold when picking up the P-wave, which is necessary for the traditional STA/LTA method, and overcoming the challenge of inaccurate prediction when weak seismic signals with a low signal-to-noise ratio occur.

MACHINE LEARNING FOR FAST AND RELIABLE SOURCE-LOCATION ESTIMATION IN EARTHQUAKE EARLY WARNING

We develop a random forest (RF) model for rapid earthquake location with an aim to assist earthquake early warning (EEW) systems in fast decision making. This system exploits Pwave arrival times at the first five stations recording an earthquake and computes their respective arrival time differences relative to a reference station (i.e., the first recording station). These differential P-wave arrival times and station locations are classified in the RF model to estimate the epicentral location. We train and test the proposed algorithm with an earthquake catalog from Japan. The RF model predicts the earthquake locations with a high accuracy, achieving a Mean Absolute Error (MAE) of 2.88 km. As importantly, the proposed RF model can learn from a limited amount of data (i.e., 10% of the dataset) and much fewer (i.e., three) recording stations and still achieve satisfactory results (MAE<5 km). The algorithm is accurate, generalizable, and rapidly responding, thereby offering a powerful new tool for fast and reliable source-location prediction in EEW.

Evolution of shallow volcanic seismicity in the hydrothermal system of La Soufrière de Guadeloupe following the April 2018 Mlv 4.1 earthquake

La Soufrière de Guadeloupe volcano is characterized by seismo-volcanic activity dominantly linked to an active hydrothermal system. This microseismicity is shallow, mainly triggered in swarms and characterized by repeating earthquakes. Four recurrent families of Volcano-Tectonic (VT) repeaters have been identified and the main repeater accounts for 80% of detections. Stacking repeating seismic waveforms produces MASTER events with a high Signal-to-Noise Ratio (SNR) and thanks to the deployment of a temporary seismic nodal array, their absolute locations can be better constrained. Because we observe systematic positive residuals of P-wave arrival times at almost all stations, we investigate potential bias in the velocity model in the shallow part of the dome. By increasing the P-wave velocity and decreasing the VP/VS ratio from 1.8 to 1.69, we reduce these residuals and improve the local velocity model. We use this new velocity model to locate each VT event relatively to is own MASTER hypocenter. This procedure offers a new image of the hydrothermal seismic activity that we locate under the acid lake of the summit Tarissan crater, along a sub-vertical conduit. We also define a linear relationship between the logarithm of the peak amplitude of seismic events and their duration to obtain a pseudo local magnitude. The approach enables to accurately and automatically estimate the magnitude during the event detection. We show that after the occurrence in April 2018 of the largest earthquake (Mlv 4.1) since the last phreatic eruption in 1976–1977, the swarm frequency and amplitude increased significantly, and a new family of VT repeater emerges, accounting for up to 14% of detections. We show a quiescence in the main repeaters activity in the month following the April 2018 earthquake, suggesting a significant stress release within the volcanic system, which is likely related to dynamic stress changes rather than static stress variations caused by the earthquake. Moreover, we detect the emergence of a new family of repeaters 3 months after the event, located ∼100 m above the main repeater family, that could be related to a fluid pressure increase induced by the regional hydrological cycle, superimposed to the internal forcing driven by the hydrothermal system. Finally, using a statistical approach, we highlight seasonal periodicities in the number of events and the released seismic moment at La Soufrière de Guadeloupe, with a dominant peak of seismic activity in October–November and a second, less significant, peak of in April.

Three-dimensional velocity structure of a coal mine revealed by induced microseismic traveltime data

Early detection of rockburst risk areas is a prerequisite for ensuring safe production in coal mines. A reliable and efficient detection method is essential for identifying stress distribution and disturbances in coal and rock masses during mining. In this study, we employed seismic traveltime tomography to study the three-dimensional velocity structure of a continuous mining and excavation area of a coal mine by inverting a number of P-wave arrival times from the induced micro-earthquakes (M < 2.0). The response characteristics and stress evolution of the inversion areas at different mining stages were investigated by examining the obtained time-lapse velocity structure. The microseismic (MS) dataset was partitioned into two clusters in a temporal sequence, with each cluster containing 367 relocated events. A series of checkerboard and restoring resolution tests, as well as field investigations, confirmed the resolution, robustness, and reliability of the tomographic results. In addition, absolute velocity change patterns were proposed to assess and predict rockburst risk areas caused by mining activities. The results showed that high-velocity zones could be attributed to roof subsidence, stress transfer after coal and rock mass rupture, and areas with special geological conditions in isolated working face situations. The coupled effects of stress evolution and geological conditions may lead to more pronounced stress concentrations during the mining process. Furthermore, induced events occur not only in areas between high and low stress but also near the isoline edges of pattern changes. Our established velocity patterns have significant potential for predicting hazardous areas with a timeliness of up to 10 days.

Ancient slabs beneath Arctic and surroundings: Izanagi, Farallon, and in-betweens

A detailed 3-D tomographic model of the whole mantle beneath the northern hemisphere (north of ~ 30°N latitude) is obtained by inverting a large amount of P-wave arrival time data (P, pP, and PP) to investigate transition of subducted slabs beneath Eurasia–Arctic–North America. We apply an updated global tomographic method that can investigate the whole mantle 3-D structure beneath a target area with high resolution comparable to that of regional tomography. The final tomographic model is obtained by performing independent calculations for 12 different target areas and stitching together the results. Our model clearly shows the subducted Izanagi and Farallon slabs penetrating into the lower mantle beneath Eurasia and North America, respectively. In the region from Canada to Greenland, a stagnant slab lying below the 660-km discontinuity is revealed. Because this slab has a texture that seems to be due to subducted oceanic ridges, the slab might be composed of the Farallon and Kula slabs that had subducted during ~60–50 Ma. During that period, a complex rift system represented by division between Canada and Greenland was developed. The oceanic ridge subduction and hot upwelling in the big mantle wedge above the stagnant slab caused a tensional stress field, which might have induced these complex tectonic events.

Seismic evidence for collisional tectonics of the North China and Yangtze blocks in the Tongbai-Dabie orogenic belt

The Tongbai-Dabie orogenic belt was formed in the Middle to Late Triassic through continental collision between the North China Block and the Yangtze Block. To reveal the collisional tectonics of the two blocks, we apply teleseismic tomography to study the 3-D P-wave velocity structure down to 400 km depth beneath the Tongbai-Dabie orogenic belt using 37,251 first P-wave arrival times of 1423 teleseismic events recorded at 99 seismic stations in the study region. Our tomography reveals three obvious high-velocity anomalies. One is dipping southward beneath the Tongbai-Dabie orogenic belt, reflecting the southward subduction of the North China Block. The second is visible down to 100 km depth beneath the eastern Tongbai-Dabie orogenic belt, which reflects the Yangtze Block. The third appears at depths of 300–400 km, which is interpreted as the retained broke-off paleo-Tethyan oceanic slab. On the basis of our tomographic results and previous findings, we propose a two-stage subduction model showing that the Yangtze Block subducted northward in the Triassic, and then the North China Block subducted southward in the Jurassic beneath the Tongbai-Dabie orogenic belt.

A Preliminary Tomography Inversion Study on the Palu Koro Fault, Central Sulawesi Using BMKG Seismic Network

The Palu Koro Fault, which frequently causes solid and destructive earthquakes, is the most exciting aspect of Central Sulawesi’s tectonics. We used local earthquake data from the area in this preliminary study. Our preliminary investigation entails analyzing local seismic data and estimating the velocity structure and hypocenter location using a tomographic approach that employs P-wave arrival times from the Agency for Meteorology, Climatology, and Geophysics earthquake observation network from 2010 to 2019. We use a 1-D initial velocity structural model in this tomographic inversion method, and the results are evaluated using the checkerboard test, derivative weight sum, and ray hit count. To the west and east of the Palu Koro Fault area, preliminary tomographic inversion results show a relatively high seismic velocity. These areas are interpreted as West Sulawesi Mandala and East Sulawesi Mandala Poso Fault and Wekuli Fault. The low-velocity anomaly is found near the Palu Koro Fault. Because of the high-speed thrusting, the relocated hypocenters also clustered around the Palu Koro Fault area. In the following activity, we will determine the speed structure of the S wave, which is expected to provide better geological information in the interpretation.

Anisotropic tomography of the East Japan subduction zone: influence of inversion algorithms

SUMMARY An important element of seismic tomography is the inversion process. In this work, we use P-wave arrival times of local earthquakes recorded at onshore and offshore seismic stations in East Japan to investigate the influence of two well-known inversion algorithms (LSQR and L-BFGS-B) on anisotropic tomography. Our synthetic tests show that a large damping parameter in the LSQR algorithm can lead to a stable and fast convergence, but it can result in many small value disturbances. The L-BFGS-B algorithm, which has second-order convergence, could converge fast to the optimal solution without damping regularization, but an inappropriate bound on the unknown parameters makes them hard to be recovered fully and causes strong trade-off between isotropic velocity and azimuthal anisotropy. If appropriate control parameters are adopted, the two inversion algorithms lead to almost the same results, though the L-BFGS-B provides a more efficient convergence and leads to a slightly better fit to the data than LSQR does. The two algorithms are applied to investigate the 3-D P-wave velocity (Vp) structure and azimuthal anisotropy of the East Japan subduction zone. Our results show that high-Vp anomalies and trench-normal fast-velocity directions (FVDs) exist in the forearc crust beneath the Pacific Ocean off South Hokkaido, which may reflect a cold and hydrated forearc crust with aligned microcracks or fractures. Significant low-Vp anomalies and trench-parallel FVDs exist at 40–80 km depths beneath Hokkaido, reflecting a water-rich mantle wedge with aligned B-type olivine. In the subducting Pacific slab, strong anisotropy with trench-parallel FVDs is revealed, reflecting localized horizontal bending of the slab.

APPRAISAL OF THE ABAJI-ABUJA (NIGERIA) ML 2.25 EARTH-TREMOR OF 10th JANUARY 2020

This research aims at determining the epicentre, magnitude and energy dissipated during the earth-tremor which took place at Abaji-Abuja (Nigeria) on 10th January 2020 at 13:46:20 UTC. The event was recorded by Volksmeter II VMII-2RU broadband Seismographs located at the Earthquake and Space Weather laboratory, Obafemi Awolowo University, Ife, Osun state, Nigeria. The computed P-wave and S-wave arrival times and lag times were 35.8seconds, 36.1seconds and 33.3seconds recorded by three Seismographs Ife (1), Ife (2) and Ife (3) respectively. The WinQuake software was used for the analysis. Epicentre distances computed were 330.3 km, 333.5 km and 303.2 km from the three seismograms, using trilateration method. The results gave the epicentre location at 8049’N and 6047’E (Gurdi town, Yaba District, Abaji LGA, Abuja, Nigeria),focal depth of 5 km, and average epicentral distance of 322.3km from Ile-Ife. The tremor has a Local Magnitude of 2.25 MLand Coda Magnitude of 3.7Md, and the dissipated energy of 149.6 × 10-12 KJ. The implication of these results is that a local magnitude of 2.25 ML of 149.6 × 10-12KJ of energy serves as an indicator to the future occurrence of another earthquake within the said region. Necessary precautionary measures should be taken when carrying out geological and construction works within the Abaji-Abuja environs. The massive rock blasting and quarrying must have reactivated the faults within the area.

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

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.

The CGAS Deep Learning Algorithm for P-Wave Arrival Time Picking of Mining Microseismic Events

The CGAS Deep Learning Algorithm for P-Wave Arrival Time Picking of Mining Microseismic Events

Traveltime-based microseismic event location using artificial neural network

Location of earthquakes is a primary task in seismology and microseismic monitoring, essential for almost any further analysis. Earthquake hypocenters can be determined by the inversion of arrival times of seismic waves observed at seismic stations, which is a non-linear inverse problem. Growing amounts of seismic data and real-time processing requirements imply the use of robust machine learning applications for characterization of seismicity. Convolutional neural networks have been proposed for hypocenter determination assuming training on previously processed seismic event catalogs. We propose an alternative machine learning approach, which does not require any pre-existing observations, except a velocity model. This is particularly important for microseismic monitoring when labeled seismic events are not available due to lack of seismicity before monitoring commenced (e.g., induced seismicity). The proposed algorithm is based on a feed-forward neural network trained on synthetic arrival times. Once trained, the neural network can be deployed for fast location of seismic events using observed P-wave (or S-wave) arrival times. We benchmark the neural network method against the conventional location technique and show that the new approach provides the same or better location accuracy. We study the sensitivity of the proposed method to the training dataset, noise in the arrival times of the detected events, and the size of the monitoring network. Finally, we apply the method to real microseismic monitoring data and show that it is able to deal with missing arrival times in efficient way with the help of fine tuning and early stopping. This is achieved by re-training the neural network for each individual set of picked arrivals. To reduce the training time we used previously determined weights and fine tune them. This allows us to obtain hypocenter locations in near real-time.

GTUNE: An Assembled Global Seismic Dataset of Underground Nuclear Test Blasts

AbstractFrom catalogs of available declassified underground nuclear explosions, we compiled a comprehensive seismic waveform and event catalog termed GTUNE (Georgia Tech Underground Nuclear Explosions). Nuclear blast seismic records are sourced from previously prepared published datasets and openly available waveforms from online sources. All seismic traces were assembled into a user-friendly format compatible with most python-based machine learning (ML) packages. The GTUNE dataset includes the raw seismogram time series, event coordinates and origin time, sampling rate, station metadata, channel, epicentral distance, and P-wave arrival time from the origin dataset when available and otherwise identified using a tuned automated picker. This is the first openly available comprehensive global underground nuclear blast seismic dataset and consists of 28,123 vertical-component waveforms from 774 nuclear test blasts between 1961 and 2017 recorded between 0 and 90 epicentral degrees. For stations where data are not directly included due to data-sharing restrictions, the mechanisms to acquire and process these data are included. In this article, we describe various steps involved in data collection and quality control to ensure accurate labels, and present summary properties of the catalog and data set. The catalog was initially developed for applications with ML methods but can be used for a wide range of studies such as source physics, earth structure, and event detection methodological development.