Discrimination Of Seismic Events Research Articles

Bayesian Network Inference for Low-Magnitude Nonnatural Seismic Event Discrimination

Abstract In response to the gaps in understanding the causal relationship between seismic waveform features and the types of seismic events, this research is focused on seismic events of low magnitude (ML≤3.0) in the North China region. Using the Bayesian network theory, we conduct an analysis to infer event types for natural earthquakes, artificial explosions, and mining collapses, and the outcomes achieved notable efficacy for the discrimination of seismic events. Through the analysis of seismic waveforms from 1818 events, we systematically extracted and quantified 55 features in temporal, spectral, and energy domains, which were then recoded as node variables for subsequent analysis. The new data set was subject to select nodes with strong associations to the node type. Subsequently, Bayesian network topologies were constructed using three different algorithms to reconstruct the custom network, calculating posterior probabilities and marginal probabilities. Simultaneously, an extensive evaluation with precision–recall curves of the network structure was carried out, encompassing accuracy, precision, recall, and F1-score. Ultimately, sensitivity analysis was performed on each node to reveal the extent of the influence of node variations on the inference of the node type. The findings showed that the sensitivity of discrimination of seismic events was notably high for several features, including high-frequency P/S spectral ratio values (11 to ∼20 Hz), central frequency, dominant frequency, average frequency, rise and decay average frequency, the real part of the complex cepstral coefficients, peak ground acceleration, and zero crossing. In the classification of natural earthquakes, artificial explosions, and mining collapses, it was observed that the probability of mining collapses was maximized when peak ground acceleration was less than 1526.08, and concurrently, the P/S spectral ratio (11 to ∼20 Hz) fell within the range of −0.25 to −0.02.

Cross-Regional Seismic Event Discrimination via Convolutional Neural Networks: Exploring Fine-Tuning and Ensemble Averaging

ABSTRACT The gradual densification of seismic networks has facilitated the acquisition of large amounts of data. However, alongside natural tectonic earthquakes, seismic networks also record anthropogenic events such as quarry blasts or other induced events. Identifying and distinguishing these events from natural earthquakes requires experienced interpreters to ensure that seismological studies of natural phenomena are not compromised by anthropogenic events. Advanced artificial intelligence methods have already been deployed to tackle this problem. One of the applications includes Convolutional Neural Networks (CNN) to discriminate different kinds of events, such as natural earthquakes and quarry blasts. In this study, we investigate the effects of ensemble averaging and fine-tuning on seismic event discrimination accuracy to estimate the potential of these methods. We compare discrimination accuracy of two different CNN model architectures across three datasets. This was done with the best models from an ensemble of each model architecture, as well as with ensemble averaging and fine-tuning methods. Soft voting was used for the CNN ensemble predictions. For the transfer learning approach, the models were pretrained with data from two of the datasets (nontarget regions) and fine-tuned with data from the third one (target region). The results show that ensemble averaging and fine-tuning of CNN models leads to better generalization of the model predictions. For the region with the lowest numbers of one event type, the combination of ensemble averaging and fine-tuning led to an increase in discrimination accuracy of up to 4% at station level and up to 10% at event level. We also tested the impact of the amount of training data on the fine-tuning method, showing, that to create a global model, the selection of comprehensive training data is needed.

Deep learning and transfer learning of earthquake and quarry-blast discrimination: applications to southern California and eastern Kentucky

SUMMARY Discrimination between tectonic earthquakes (EQs) and quarry blasts is important for accurate EQ cataloguing and seismic hazard analysis. However, reliable classification of these two types of seismic events is challenging with no prior knowledge of source parameters. Here, we applied deep learning to perform this classification task in southern California and eastern Kentucky. Since the two regions differ significantly in available labelled data, class imbalance and waveform characteristics, we adopted different strategies for them. We directly trained a convolutional neural network (CNN) for southern California due to its data abundancy. To alleviate the class imbalance, the blast data were augmented by repeated sampling. The model for California yields F1-scores of >83.5 per cent when estimated by individual stations and >98.1 per cent by network average (i.e. averaging the CNN’s outputs on all available stations for each event). As eastern Kentucky has a much smaller data size, we apply transfer learning to the pre-trained California model to fit the Kentucky data. The transfer-learned model yields F1-scores of >86.9 per cent when estimated by individual stations and >96.7 per cent by network average. The transfer-learned model outperforms the model re-trained from scratch for the Kentucky data. Gradient-weighted class activation mapping shows the S onset and the S long-period coda are important to identify EQs and blasts, respectively. By visual inspections of the seismograms, the source locations, the origin time and the P-wave polarities, we verified that most of the events falsely predicted by our models are actually mislabelled by seismic analysts. Our models thus show great potential in helping seismic analysts find those mislabelled events which remain hidden in the historical catalogue. Our results demonstrate that deep learning can achieve high accuracy in seismic event discrimination and that transfer learning is effective and efficient to generalize deep learning models across different regions.

Discrimination of Seismic Events in the North Korean Test Site and Surrounding Area Using MDAC Spectral Ratio

ABSTRACT An event of ML 3.7 occurred near the Punggye-ri nuclear test site on 11 February 2022. It was stronger than the known historical earthquakes around the North Korean test site (NKTS). How to accurately identify whether this event and future events near NKTS are explosions or natural earthquakes is worth paying attention to. For moderate and small earthquakes, the regional P/S spectral ratio has the advantage of identifying explosions and earthquakes. However, the P/S spectral ratio depends on magnitude and ray path. Magnitude and distance amplitude correction (MDAC) was proposed and demonstrated to be an effective method to correct magnitude and path effect. Different from the traditional MDAC method, we use explosions to build the MDAC model to reduce the influence of the source radiation pattern. After building the MDAC model, we analyze the Pg/Lg and Pn/Lg spectral ratios of 6 nuclear tests and 60 earthquakes at the NKTS from 2013 to 2021. The results show that the spectral ratio residuals of the 6 nuclear tests are mainly distributed around 0 after correction, whereas the spectral ratio residuals of earthquakes are obviously <0 above 3 Hz and are clearly distinguished from explosions as expected. The Pg/Lg and Pn/Lg spectral ratios of the ML 3.7 earthquake near the NKTS on 11 February 2022 exhibit similar characteristics to historical natural earthquakes. We further discuss the influence of magnitude on P/S spectral ratios. In the NKTS, the differences in the P/S ratio between earthquakes and explosions are similar before and after magnitude correction at 4–8 Hz. Nevertheless, the magnitude correction increases by ∼27% of the difference between explosions and natural earthquakes at 6–9 Hz to better separate explosions from natural earthquakes.

A two-step algorithm for acoustic emission event discrimination based on recurrent neural networks

We present an algorithm for seismic event discrimination and event approximate location based on multi-station seismograms. A deep learning approach was applied using a two-step algorithm: (i) signal onsets were identified in individual tracks based on the use of long-short-term memory neural network layers; (ii) if a sufficient number of onsets were reliably identified, a preliminary location was determined. We adopted a “reverse location approach” where the time sense of a seismogram is reverted and the origin time is predicted using a neural network approach based on previously determined onsets. Successful location or origin time prediction also served as a feedback for confirming previous onset identification.The method was tested using a data set of Acoustic Emission generated from the uniaxial loading of a Westerly Granite specimen. Accuracy of the method was better than 97%. Discriminated events were automatically located and their seismic moment tensor was determined. Both types of results were in good agreement with the baseline data set.With respect to the particular nature of processed data, we provide a demo code which shows examples presented in the article. In addition, a detailed description of the algorithm, including the control parameter values, is provided in the text. Based on this information the method can be applied on any data.

Imbalanced Seismic Event Discrimination Using Supervised Machine Learning.

The discrimination between earthquakes and artificial explosions is a significant issue in seismic analysis to efficiently prevent and respond to seismic events. However, the discrimination of seismic events is challenging due to the low incidence rate. Moreover, the similarity between earthquakes and artificial explosions with a local magnitude derives a nonlinear data distribution. To improve the discrimination accuracy, this paper proposes machine-learning-based seismic discrimination methods—support vector machine, naive Bayes, and logistic regression. Furthermore, to overcome the nonlinear separation problem, the kernel functions and regularized logistic regression are applied to design seismic classifiers. To efficiently design the classifier, P- and S-wave amplitude ratios on the time domain and spectral ratios on the frequency domain, which is converted by fast Fourier transform and short-time Fourier transform are selected as feature vectors. Furthermore, an adaptive synthetic sampling algorithm is adopted to enhance the classifier performance against the seismic data imbalance issue caused by the non-equivalent number of occurrences. The comparisons among classifiers are evaluated by the binary classification performance analysis methods.

Discrimination of Seismic Events (2006–2020) in North Korea UsingP/LgAmplitude Ratios from Regional Stations and a Bivariate Discriminant Function

AbstractTwo events of magnitude (mb) 3.6–3.8 occurred in southern North Korea (NK) on 27 June 2019 and 11 May 2020. Although these events were located ∼330–400 km from the known nuclear test site, the fact that they occurred within the territory of NK, a country with a recent history of underground nuclear tests, made them events of interest for the monitoring community. We used P/Lg ratios from regional stations to categorize seismic events that occurred in NK from 2006 to May 2020, including these two recent events, the six declared NK nuclear tests, and the cavity collapse and triggered earthquakes that followed the 3 September 2017 nuclear explosion. We were able to separate the cavity collapse from the population of nuclear explosions. However, based on P/Lg ratios, the distinction between the earthquakes and the cavity collapse is ambiguous. The performed discriminant analyses suggest that combining Pg/Lg and Pn/Lg ratios results in improved discriminant power compared with any of the ratio types alone. We used the two ratio types jointly in a quadratic discriminant function and successfully classified the six declared nuclear tests and the triggered earthquakes that followed the September 2017 explosion. Our analyses also confirm that the recent southern events of June 2019 and May 2020 are both tectonic earthquakes that occurred naturally.

Seismic Event Discrimination Using Deep CCA

Detection and discrimination of seismic events have important implications. Precise detection of earthquakes may help prevent collateral damage and even save lives. On the other hand, the ability to identify explosions reliably not only helps prevent false alarms but also is crucial for monitoring nuclear experiments. In this work, we present a method for automatic discrimination of seismic events. A neural network is trained to fuse information from multiple seismic channels into a correlated space. Then, we augment the minority class to avoid class imbalance. Finally, a tree-based classifier is used to estimate the nature of the suspected event. We apply the proposed approach to 1609 events collected in Israel and Jordan. Our framework demonstrates improved precision and recall scores.

Combined Rayleigh- and Love-Wave Magnitudes for Seismic Event Discrimination and Screening Analysis

Abstract Love waves have the potential to aid in discrimination for anomalous explosion events. We develop a calibrated mathematical formulation for an explosion discriminant that combines Rayleigh‐ and Love‐wave magnitude values and employs an error model that correctly partitions variances among events and stations separately. The discriminant is calibrated using a global data set of 124 earthquakes and 26 nuclear explosions and applied to the May 2009 Democratic Republic of North Korea (DPRK) announced nuclear test, as well as the calibration data set. All 26 explosions were correctly identified; only 6 earthquakes were incorrectly identified as explosions. Compared to an analogous treatment using only Rayleigh data, the combined discriminant improves the DPRK event p ‐value only nominally but reduces the number of false positives in the calibration data set by 70%, with no additional false negatives. While not dramatically improving the discrimination power for anomalous events, such as the 2009 DPRK test, the combined discriminant proposed here offers improved screening capabilities for typical events. Online Material: Earthquake and explosion calibration data set.

An unsupervised learning algorithm: application to the discrimination of seismic events and quarry blasts in the vicinity of Istanbul

Abstract. The results of the application of an unsupervised learning (neural network) approach comprising a Self Organizing Map (SOM), to distinguish micro-earthquakes from quarry blasts in the vicinity of Istanbul, Turkey, are presented and discussed. The SOM is constructed as a neural classifier and complementary reliability estimator to distinguish seismic events, and was employed for varying map sizes. Input parameters consisting of frequency and time domain data (complexity, spectral ratio, S/P wave amplitude peak ratio and origin time of events) extracted from the vertical components of digital seismograms were estimated as discriminants for 179 (1.8 < Md < 3.0) local events. The results show that complexity and amplitude peak ratio parameters of the observed velocity seismogram may suffice for a reliable discrimination, while origin time and spectral ratio were found to be fuzzy and misleading classifiers for this problem. The SOM discussed here achieved a discrimination reliability that could be employed routinely in observatory practice; however, about 6% of all events were classified as ambiguous cases. This approach was developed independently for this particular classification, but it could be applied to different earthquake regions.

On the potential of recording earthquakes for global seismic tomography by low‐cost autonomous instruments in the oceans

We describe the development and testing of an autonomous device designed to revolutionize Earth structure determination via global seismic tomography by detecting earthquakes at teleseismic distances in the oceans. One prototype MERMAID, short for Mobile Earthquake Recording in Marine Areas by Independent Divers, was constructed and tested at sea. The instrument combines two readily available, relatively low‐cost but state‐of‐the‐art components: a Sounding Oceanographic Lagrangian Observer, or SOLO float, and an off‐the‐shelf hydrophone, with custom‐built data logging hardware. We report on the development of efficient wavelet‐based algorithms for the detection and discrimination of seismic events and analyze three time series of acoustic pressure collected at a depth of 700 m in pilot experiments conducted offshore San Diego, CA. In these tests, over 120 hours of data were gathered, and five earthquakes, of which one was teleseismic, were recorded and identified. Quantitative estimates based on these results suggest that instruments of the MERMAID type may collect up to a hundred tomographically useful teleseismic events per year. The final design will also incorporate a Global Positioning System receiver, onboard signal processing software optimized for low‐power chips, and high‐throughput satellite communication equipment for telemetered data transfer. With these improvements, we hope to realize our vision of a global array of autonomous floating sensors for whole‐earth seismic tomography.

A Quantitative Study of the Site Effects Observed at the GERESS Array

Site effects were studied for the individual sites of the 25 elements of the German Experimental Seismic System (GERESS) array based on the traditional spectral ratio (TSR) technique and horizontal-to-vertical ratios (HVRs) for earthquakes and microtremors, commonly referred to as Nakamura9s technique. An understanding of these site effects is required for various applications, in particular for the discrimination of seismic events. The spectral ratio technique identifies three stations (A0, A3, B3) as showing prominent peaks in the frequency band from 1 to 17 Hz up to a factor of 30, with A0 and B3 having the largest site effect at a frequency of 7 Hz. These stations are situated in the immediate vicinity of the array center. A number of other stations, such as A1, B1, B4, B5, C1, C3–C5, D2, D5, and D7–D9, display a broad frequency range preferably above 10 Hz, where the spectral ratios indicate up to 10 times higher spectral amplitudes compared to the reference site, C2. The remaining stations, A2, B2, C6, C7, D1, D3, D4, and D6, show no noticeable site effects. Further study considers the appropriateness for the selection of C2 as reference site for the TSR technique and the comparison of TSR and HVR techniques for the three-component elements of the array. Although typically site response refers to the site amplification of horizontal ground motions, the assessed site response of vertical ground motions seems to yield comparable results. The spectral variance, a parameter based on the smoothness of stacked array spectra from the best 17 array elements of the GERESS array, has been used successfully to distinguish between local earthquakes and mining explosions. It is shown that the spectral variance estimate is considerably improved using all 25 array elements and applying a site response correction.

Study of Regional Surface Waves and Frequency-dependent M s : m b Discrimination in the European Arctic

Accurate discrimination of seismic events with a regional network requires detailed knowledge of the propagation characteristics of seismic waves in the region. At present, such propagation characteristics are reasonably well known for P and S waves in the European Arctic, however much work remains to be done regarding surface wave propagation and magnitude estimation.Regional long-period or broadband seismic data in digital form has been available in the European Arctic for only a few years. In order to assess regional surface wave propagation, and in particular to evaluate the M s :m b discriminant at regional distances, it is therefore necessary to take advantage of the historic analog recordings. The station APA in Apatity forms a unique source of such data, with high-quality long-period seismic recordings of regional earthquakes and nuclear explosions dating back about 30 years.

Seismic source and structure estimation in the western Mediterranean using a sparse broadband network

We present a study of regional earthquakes in the western Mediterranean geared toward the development of methodologies and path calibrations for source characterization using regional broadband stations. The results of this study are useful for the monitoring and discrimination of seismic events under a comprehensive test ban treaty, as well as the routine analysis of seismicity and seismic hazard using a sparse array of stations. The area consists of several contrasting geological provinces with distinct seismic properties, which complicates the modeling of seismic wave propagation. We started by analyzing surface wave group velocities throughout the region and developed a preliminary model for each of the major geological provinces. We found variations of crustal thickness ranging from 45 km under the Atlas and Betic mountains and 37 km under the Saharan shield, to 20 km for the oceanic crust of the western Mediterranean Sea, which is consistent with earlier works. Throughout most of the region, the upper mantle velocities are low which is typical for tectonically active regions. The most complex areas in terms of wave propagation are the Betic Cordillera in southern Spain and its north African counterparts, the Rif and Tell Atlas mountains, as well as the Alboran Sea, between Spain and Morocco. The complexity of the wave propagation in these regions is probably due to the sharp velocity contrasts between the oceanic and continental regions as well as the the existence of deep sedimentary basins that have a very strong influence on the surface wave dispersion. We used this preliminary regionalized velocity model to correct the surface wave source spectra for propagation effects which we then inverted for source mechanism. We found that this method, which is in use in many parts of the world, works very well, provided that data from several stations are available. In order to study the events in the region using very few broadband stations or even a single station, we developed a hybrid inversion method which combines Pnl waveforms synthesized with the traditional body wave methods, with surface waves that are computed using normal modes. This procedure facilitates the inclusion of laterally varying structure in the Green's functions for the surface waves and allows us to determine source mechanisms for many of the larger earthquakes (M > 4) throughout the region with just one station. We compared our results with those available from other methods and found that they agree quite well. The epicentral depths that we have obtained from regional waveforms are consistent with observed teleseismic depth phases, as far as they are available. We also show that the particular upper mantle structure under the region causes the various Pn and Sn phases to be impulsive, which makes them a useful tool for depth determination as well. Thus we conclude that with proper calibration of the seismic structure in the region and high‐quality broadband data, it is now possible to characterize and study events in this region, both with respect to mechanism and depth, with a limited distribution of regional broadband stations.

Seismic event discrimination

abstract Discrimination of small events begins with detection, association, and location. Recent advances in detection include use of updated measures of the noise variance to control the false alarm rate; and updating the noise amplitude spectrum, N , for application of the optimum detection filter S / N ** 2 where S is the signal spectrum. Postdetection processes of interest include determination of signal azimuth and emergence angle from linear three-component processing. Association programs have recently been improved by implementation of techniques taking advantage of information provided by arrays. Location advances include modification of the a priori travel-time residual covariance matrix to account for the fact that travel-time residuals are correlated at stations near to each other, simultaneous determination of location and phase velocity when using regional phases, and direct use of phases other than P , e.g., Pg , Lg , S , pP , and PKPBC . According to most source theories, small earthquakes and explosions have the same, flat, displacement spectrum for P waves at teleseismic frequencies. Aside from location, discrimination therefore rests on focal mechanism differences and generation of shear waves by earthquakes. The M s : m b discriminant rests on the shear wave generation; but there is overlap of explosions and earthquakes at low magnitudes for dip-slip earthquakes. Low frequency (0.3 Hz) P waves are weak from explosions due to the surface pP reflections; P / pP amplitude ratios for earthquakes vary over the focal sphere in a way impossible for a pair of explosions. Regional discrimination shows promise in the ratio of maximum amplitudes before and after Sn , and in the observed fact that shear phases from explosions detonated near high impedance contrasts are of lower frequency than those from earthquakes in the same region, possibly due to the generation of the explosion shear phases by P to S scattering. Recent experimental analyses show that there is almost no decoupling at high frequencies, in agreement with earlier theories. These frequencies may be observable at regional distances.