Earthquake Dataset Research Articles

Improvements and Heterogeneities of the Global Centroid Moment Tensor Catalog

Abstract Earthquake catalogs are heterogeneous, especially those developed over long time spans. Changes in seismological monitoring, which provides the records on which these catalogs are based, are common. Typically, instruments and networks become more sensitive over time, allowing for the detection and characterization of smaller earthquakes. In pursuit of improvement, new methods for routine data analysis are occasionally introduced, modifying the procedures for catalog compilation. The resulting heterogeneities may not be evident to users, but they should be unveiled and considered in any application of the catalog, especially in statistical seismology, which analyzes large earthquake data sets. The Global Centroid Moment Tensor catalog is considered the most homogeneous database of global seismicity. However, a detailed analysis of its heterogeneities has been lacking. This work reviews changes in the catalog’s development from 1976 to 2023 and reveals how these have caused improvements and heterogeneities in the resulting data. Several periods are distinguished, separated by milestones in the methods employed for moment tensor inversion and catalog compilation, as well as by the advent of global broadband monitoring in 2004. These changes are shown to have caused variations in the catalog’s completeness and in the determinations of centroid depths, scalar seismic moments, and moment tensors. The magnitude of completeness is measured here in detail, both temporally and spatially. It has decreased over the years and shows spatial variations within each period, correlated to regional differences in network monitoring and compilation biases. Moment tensor determinations have been significantly different since 2004, resulting in a different frequency distribution of rake angles and a different dependence of the double-couple component as a function of rake. This work is expected to benefit all future uses of the catalog, enabling better characterization of seismicity properties and improved building and testing of models for earthquake occurrence.

Exploiting the synergy of SARIMA and XGBoost for spatiotemporal earthquake time series forecasting

AbstractEarthquakes are vibrations that occur on the surface of earth, generating fires, ground shaking, tsunamis, landslides and cracks. These incidents can cause severe damage and loss of life. Accurate earthquake forecasts are critical for anticipating and mitigating these hazards, which can avoid damage to buildings and infrastructure and save lives. To address the challenges given by earthquakes probabilistic nature, this paper presents a hybrid SARIMA–XGBoost approach to earthquake magnitude prediction. The suggested technique consists of a two‐step process: an exploration phase that uses exploratory data analysis, which includes descriptive statistics and data visualisation, and a prediction phase that focusses on forecasting future earthquakes. Using a large significant earthquake dataset spanning 1965–2023, the study intends to gain insights and lessons for more effective earthquake prediction methods. Further, in a comparison analysis, the results of SARIMA‐XGBoost model are compared to those of traditional ARIMA and SARIMA models. The results highlight the superior performance of the hybrid SARIMA–XGBoost model, showcasing a mean absolute error (MAE) of 0.038, a mean squared error (MSE) of 0.0040, and a root mean squared error (RMSE) of 0.068. These metrics collectively underscore the model's enhanced accuracy in forecasting earthquake magnitudes. The notably low values of MAE, MSE and RMSE indicate that our hybrid approach significantly improves prediction accuracy compared to alternative models. By integrating SARIMA's time series (TS) analysis with XGBoost's machine learning (ML) capabilities, the hybrid model reduces forecasting errors more effectively, demonstrating its clear advantage in precision.

A social context-aware graph-based multimodal attentive learning framework for disaster content classification during emergencies

In times of crisis, the prompt and precise classification of disaster-related information shared on social media platforms is of paramount importance for effective disaster response and public safety. During such critical events, people utilize social media as a medium for communication, sharing multimodal textual and visual content. However, due to the substantial influx of unfiltered and diverse data, humanitarian organizations face challenges in effectively leveraging this information. Numerous methods have been proposed for classifying disaster-related content, but these methods lack modeling users’ credibility, emotional context, and social interaction information, which is crucial for classification. In this context, we propose a method, CrisisSpot, that leverages a Graph-based Neural Network to comprehend intricate relationships between textual and visual modalities and Social Context Features to incorporate user-centric and content-centric information. We also propose Inverted Dual Embedded Attention, which captures both harmonious and contrary patterns present in the data to harness complex interactions and facilitate richer insights in multimodal data. We have developed a multimodal disaster dataset, TSEqD (Turkey-Syria Earthquake Dataset), which is a large annotated dataset for a single disaster event containing 10,352 data samples. Through extensive experimentation, CrisisSpot has demonstrated significant improvements, achieving an average gain of 9.45% and 5.01% in F1-score compared to the state-of-the-art methods on the publicly available CrisisMMD dataset and TSEqD dataset, respectively.

EQGraphNet: Advancing single-station earthquake magnitude estimation via deep graph networks with residual connections

Magnitude estimation is a critical task in seismology, and conventional methods usually require dense seismic station arrays to provide data with sufficient spatiotemporal distribution. In this context, we propose the Earthquake Graph Network (EQGraphNet) to enhance the performance of single-station magnitude estimation. The backbone of the proposed model consists of eleven convolutional neural network layers and ten RCGL modules, where a RCGL combines a Residual Connection and a Graph convolutional Layer capable of mitigating the over-smoothing problem and simultaneously extracting temporal features of seismic signals. Our work uses the STanford EArthquake Dataset for model training and performance testing. Compared with three existing deep learning models, EQGraphNet demonstrates improved accuracy for both local magnitude and duration magnitude scales. To evaluate the robustness, we add natural background noise to the model input and find that EQGraphNet achieves the best results, particularly for signals with lower signal-to-noise ratios. Additionally, by replacing various network components and comparing their estimation performances, we illustrate the contribution of each part of EQGraphNet, validating the rationality of our approach. We also demonstrate the generalization capability of our model across different earthquakes occurring environments, achieving mean errors of ±0.1 units. Furthermore, by demonstrating the effectiveness of deeper architectures, this work encourages further exploration of deeper GNN models for both multi-station and single-station magnitude estimation.

Using In-Building Observations of Small-to-Large Earthquakes to Predict the Seismic Response of Structures

ABSTRACT The main goal of this study is to evaluate the potential value of data from weak-to-moderate earthquakes for structural response analysis. Data recorded over 18 yr by the seismic network installed in the 12-stories Grenoble City Hall Building (France) is considered. The building response is analyzed in terms of intensity measures and engineering demand parameters, and then compared with strong earthquake data recorded in Japanese buildings. The uncertainties of structural response prediction are estimated and defined in terms of “within-building” and “between-building” components in the same way as the components of the ground-motion model. Data complementarity in the response model is observed between the weak-to-moderate (France) and the moderate-to-strong (Japan) earthquake datasets, disclosing nonlinear processes (associated with resonance period elongation) that are activated in buildings during low-to-strong motion. For example, fundamental frequency shifts are triggered at low values of both total structural drift amplitudes and equivalent strain rates (i.e., time derivative of structural drift). In addition, strain rate thresholds from 10−11 s−1 to 10−5 s−1 representing different structural conditions from undamaged to severely damaged buildings are observed to activate nonlinearities. This confirms the link between loading rates and structural conditions. Our results highlight the interest in instrumentation in buildings located in regions of weak-to-moderate seismicity, for (1) the development and calibration of realistic models for predicting the seismic response of structures, (2) for improving our understanding of the components of uncertainties in the risk assessment of existing buildings, and (3) to investigate physical processes activated in structures during seismic loading that influence their response.

Localization of Deformation on Faults Driven by Fluids During the L’Aquila 2009 Earthquake

AbstractCoseismic rupture and aftershock development on a fault plane are complex and heterogeneous processes. The Mw 6.1 L’Aquila 2009 normal faulting earthquake is a perfect case to explore how fault geometry and rheology influence the rupture process and aftershocks distribution. In this study, we use for the first time a dense set of earthquake data to obtain enhanced images of the causative normal fault structure to the kilometer scale. The hypocenter of the emergent onset of the mainshock took place within a low Vp/Vs volume, while the large coseismic slip occurred a few kilometers above, as the rupture propagated through a high Vp and high Vp/Vs fluid‐filled rock volume. The increase of Vp/Vs in the fault hanging wall during the sequence suggests a strong dehydration in the earthquake asperity, with an upward fluid pressure migration along the fault toward the host rock volume. We propose that the localization of deformation on the fault plane is favored by high fluid pressure, while the spreading of aftershocks on a wide volume around the fault is driven by the depletion of fluids from the slipped portion of the fault plane and migration to small segments within the fault host rocks.

Conditional generation of artificial earthquake waveforms based on adversarial networks

Earthquake waveforms are necessary for simulation, verification, and validation during the development of structural and earthquake engineering. However, due to the fact that the processes of earthquake preparation and generation are extremely complex and the human observations cover a relatively short period compared to large earthquake cycles, recorded earthquake waveforms are often limited and quite imbalanced. Consequently, the generation of artificial earthquake waveforms is meaningful and can produce useful earthquake data. In this study, a new method based on a generative approach is proposed by leveraging machine learning (ML) techniques. The generative models based on generative adversarial network (GAN) and conditional GAN (cGAN) are introduced, and the adversarial training between the generator and the discriminator is utilized to learn how to generate artificial earthquake waveforms. In addition, several criteria are exploited to terminate the adversarial training early in order to prevent modal collapse. Besides, different earthquake datasets and several earthquake features are used to verify the proposed method. The preliminary generation is carried out using the dataset recorded during the Chi-chi earthquake, and the generated samples show similar earthquake features to those extracted from the earthquake dataset. Subsequently, conditional generation is conducted using several sub-datasets recorded in different areas around Taiwan. The results demonstrate that the well-trained model can efficiently generate waveforms according to a pre-specified seismic intensity, with an overall accuracy of up to 90%. Although the model is reliable for Taiwan and in general to restricted subdomain of the probability distribution support, the proposed method shows its ability to produce new samples for a given class, even when the number of generated samples exceeds the number of training samples.

Selection of the structural severest design ground motions based on big data and random forest

When calculating the time history of the seismic response, the conventional approach only processes a limited number of ground motions because of the elevated computational costs associated with nonlinear time history analysis and the lack of useable earthquake data. To solve this problem, a selection method for the severest design ground motions based on big data and Random Forest is proposed. Firstly, to address the issue of insufficient data, the Stanford Earthquake Dataset including over one million ground motions, was obtained and processed into the data format that can be used in earthquake engineering. Then, a structural response prediction algorithm based on Random Forest is suggested to substitute the nonlinear time history analysis, and compared to Support Vector Machine and Extreme Gradient Boosting. Finally, under all ground motions in the Stanford Earthquake Dataset, the proposed model is used to predict the structural response, include a reinforcement concrete isolated continuous girder bridge and a reinforcement concrete six-layer frame. The severest design ground motions for both structures were determined based on ranked calculated structural response. The conclusions of this study are as follows: The processed dataset provides a solution for implementing intelligent algorithms in the field of seismic engineering; Random Forest optimized through Bayesian optimization performs well in predicting structural demand measures, demonstrating superior generalization performance with R2 of 0.918 and root mean square error of 0.214, compared to other models. For actual civil structures, the proposed method can determine the structural severest ground motions at two different fortification standards.

Seismogenic depth and seismic coupling estimation in the transition zone between Alps, Dinarides and Pannonian Basin for the new Slovenian seismic hazard model

Abstract. Seismogenic depth and seismic coupling are important inputs into seismic hazard estimates. Although the importance of seismic coupling is often overlooked, it significantly impacts seismic hazard results. We present an estimation of upper and lower seismogenic depth and expected hypocentral depth and seismic coupling in the transition zone between the Alps, Dinarides and Pannonian Basin, characterized by a complex deformation pattern, highly variable crustal thickness, and moderate seismic hazard, supporting the development of the 2021 seismic hazard model of Slovenia. The hazard model was based on three seismic source models: area source model, fault source model and smoothed seismicity (point) source model. We estimated the lower seismogenic depth using seismological and geological data and compared them. The seismological estimate was based on two regional earthquake catalogues prepared for this study. In the area source model, estimates of lower seismogenic depth from seismological data are deeper or equal to the ones derived from geological data, except in one case. In the fault source model, we analysed each fault individually and chose seismological lower depth estimates in 12 among 89 faults as more representative. The seismogenic thickness for each individual fault source was determined for seismic coupling determination. The seismic coupling was assessed by two approaches, i.e. we chose the most trusted value from the literature, and the value determined for each fault individually by using the approach based on the updated regional fault and earthquake data sets. The final estimate of seismic coupling ranges from 0.77 to 0.38. We compared the tectonic moment rate based on long-term slip rate using different values of seismic coupling with the seismic moment rate obtained from the earthquake catalogue. The analysis is done for the whole area, as well as for the individual area zones. The analysis of N–S components of estimated slip for the largest faults in the area of west Slovenia shows that the regional geological and geodetic shortening rates are comparable. The total activity rate of three global seismic source models is compared, which gives up to a 10 % difference. Our results contribute to a better understanding of the seismic activity in the region. The presented approach for seismic coupling estimation can be applied in cases where the total slip rate is given instead of its seismic part and can be used at regional or national level. The approach is also suitable for the cross-border harmonization of the European seismic hazard modelling data.

Hybrid Deep Learning Model for Earthquake Time Prediction

Earthquakes are one of the most dangerous natural disasters that have constantly threatened humanity in the last decade. Therefore, it is extremely important to take preventive measures against earthquakes. Time estimation in these dangerous events is becoming more specific, especially in order to minimize the damage caused by earthquakes. In this study, a hybrid deep learning model is proposed to predict the time of the next earthquake to potentially occur. The developed CNN+GRU model was compared with RF, ARIMA, CNN and GRU. These models were tested using an earthquake dataset. Experimental results show that the CNN+GRU model performs better than others according to MSE, RMSE, MAE and MAPE metrics. This study highlights the importance of predicting earthquakes, providing a way to help take more effective precautions against earthquakes and potentially minimize loss of life and material damage. This study should be considered an important step in the methods used to predict future earthquakes and supports efforts to reduce earthquake risks.

Empirical Earthquake Source Scaling Relations for Maximum Magnitudes Estimations in Central America

ABSTRACT Central America is a seismically active region where six tectonic plates (North America, Caribbean, Cocos, Nazca, Panama, and South America) interact in a subduction zone with transform faults and two triple points. This complex tectonic setting makes the maximum magnitude—Mmax—estimation a challenging task, with the crustal fault earthquakes being the most damaging in the seismic history of Central America. The empirical source scaling relations (ESSR) allow the Mmax of faults to be determined from rupture parameters. In this study, we use a dataset of well-characterized earthquakes in the region, comprising 64 events from 1972 to 2021 with magnitudes between Mw 4.1 and 7.7. The dataset incorporates records of rupture parameters (length, width, area, slip, and magnitude) and information on the faults and aftershocks associated. This database is an important product in itself, and through its use we determine which global relations fit best to our data via a residual analysis. Moreover, based on the best-quality records, we develop scaling relations for Central America (CA-ESSR) for rupture length, width, and area. These new relations were tested and compared with recent earthquakes, and logic trees are proposed to combine the CA-ESSR and the best-fit global relations. Therefore, we estimate the Mmax for 30 faults using the logic tree for rupture length, considering a total rupture of the fault and multifault scenarios. Our results suggest that in Central America rupture areas larger than other regions are required to generate the same magnitudes. We associate this with the shear modulus (μ), which seems to be lower (∼30% less) than the global mean values for crustal rocks. Furthermore, considering multifault ruptures, we found several fault systems with potential Mmax≥Mw 7.0. These findings contribute to a better understanding of regional seismotectonics and to the efficient characterization of fault rupture models for seismic hazards.

TXED: The Texas Earthquake Dataset for AI

Abstract Machine-learning (ML) seismology relies on large datasets with high-fidelity labels from humans to train generalized models. Among the seismological applications of ML, earthquake detection, and P- and S-wave arrival picking are the most widely studied, with capabilities that can exceed humans. Here, we present a regional artificial intelligence (AI) earthquake dataset (TXED) compiled for the state of Texas. The TXED dataset is composed of earthquake signals with manually picked P- and S-wave arrival times and manually picked noise waveforms corresponding to more than 20,000 earthquake events spanning from the beginning of the Texas seismological network (TexNet) (1 January 2017) to date. These data are a supplement to the existing worldwide open-access seismological AI datasets and represent the signal and noise characteristics of Texas. Direct applications of the TXED datasets include improving the performance of a global picking model in Texas by transfer learning using the new dataset. This dataset will also serve as a benchmark dataset for fundamental AI research like designing seismology-oriented deep-learning architectures. We plan to continue to expand the TXED dataset as more observations are made by TexNet analysts.

Finite source properties of large strike-slip earthquakes

SUMMARY Earthquake ruptures are complex physical processes that may vary with the structure and tectonics of the region in which they occur. Characterizing the factors controlling this variability would provide fundamental constraints on the physics of earthquakes and faults. We investigate this by determining finite source properties from second moments of the stress glut for a global data set of large strike-slip earthquakes. Our approach uses a Bayesian inverse formulation with teleseismic body and surface waves, which yields a low-dimensional probabilistic description of rupture properties including the spatial deviation, directivity and temporal deviation of the source. This technique is useful for comparing events because it makes only minor geometric constraints, avoids bias due to rupture velocity parametrization and yields a full ensemble of possible solutions given the uncertainties of the data. We apply this framework to all great strike-slip earthquakes of the past three decades, and we use the resultant second moments to compare source quantities like directivity ratio, rectilinearity, average moment density and vertical deviation. We find that most strike-slip earthquakes have a large component of unilateral directivity, and many of these earthquakes show a mixture of unilateral and bilateral behaviour. We notice that oceanic intraplate earthquakes usually rupture a much larger width of the seismogenic zone than other strike-slip earthquakes, suggesting these earthquakes may often breach the expected thermal boundary for oceanic ruptures. We also use these second moments to resolve nodal plane ambiguity for the large oceanic intraplate earthquakes and find that the rupture orientation is usually unaligned with encompassing fossil fracture zones.

Statistical inference for mixed jump processes by Markov switching model with application to identify seismicity levels

Describing the non-stationarity of mixed stochastic jump processes presents a formidable challenge. This study employs a mixed compound Poisson process by Markov switching model (CPMSM) to address the issue of overdispersed data of random jumps. The Poisson rate is governed by a continuous-time Markov chain in this model. By assuming independence of the compound Poisson process and jump sizes, we present a generalized algorithm to estimate the parameters of CPMSM and try to predict the future behavior of this stochastic system. To investigate the potential application of this model and the accuracy of our algorithm, we provide a numerical simulation example in the context of the auto insurance payout scenario. The results suggest that CPMSM is an effective method of describing jump behavior controlled by random counts with mode switching via a Markov chain. Furthermore, we demonstrate a specific application of this model in identifying seismicity levels of Greek earthquake data and identify four hidden states. Finally, we compare its fitting effectiveness with the ETAS model using two public earthquake datasets in Japan and Iran.

Burr Puzzle-Inspired Seismic Metamaterials

Two novel seismic metamaterials models, trophy and apple core, inspired by an ancient Chinese puzzle are proposed in this study. Bandgaps below 15[Formula: see text]Hz are achieved by exclusively using concrete as the base material in the designs. Vibration modes are investigated to clarify the bandgap formation. Through comparison, we found that under the same volume, the apple core model can provide a lower frequency bandgap than the trophy model. Results from transient simulations and band structures are in good agreement, validating the accuracy of the simulation results. To broaden the bandwidth, the radii of the apple core models are varied and arranged in a gradient pattern. Transient analyses using a Gaussian wave packet and Taiwan’s Chi-Chi earthquake dataset are conducted to verify the effectiveness of the gradient apple core seismic metamaterial.

A novel seismic random noise suppression method based on wavelet threshold and Lipschitz

The seismic random noise suppression is very important in the processing and interpretation of seismic data. Wavelet transform is an effective tool to remove random noise. Aiming at the problems of signal propagation point omission and misselection, insignificant denoising effect and poor stability of the traditional wavelet maximum mode (WMM) method, this paper proposes an improved wavelet mode maximum value (IMWMM) denoising method. The Lipschitz of wavelet transform with signal and noise has different features used to distinguish signal and noise. Comparing the propagation points of adjacent scales and removing the points with reduced mode values achieve the separation of seismic signal and noise. The synthetic data, Stanford Earthquake Dataset (STEAD) data and real-field seismic signal experimental results prove the validity of the IMWMM method for both suppressing random noise and preserving seismic signal. The real-field seismic data came from Wenchuan experimental results present that the proposed IMWMM method has effectiveness of random noise suppression with SNR above 28.4527 dB, that has improvement for quality of seismic data.

Earthquake prediction technique: a comparative study

Earthquakes are one of the most dangerous natural disasters facing humans because of their occurrence without warning and their impact on their lives and property. In addition, predicting seismic movement is one of the main research topics in seismic disaster prevention. In geological studies, scientists can predict and know the locations of earthquakes in the long term. Therefore, about 80% of the major global earthquakes lie along the Pacific Ring belt, known as the Ring of Fire. Machine learning methods have also been used for short-term earthquake prediction, and studies have applied the random forest method to determine the factors that precede earthquakes. The machine learning method was based on various decision trees, each of which predicted the time to the nearest oscillation. The third group of scientists used the hybrid prediction method, which combines machine learning and geological studies. This research deals with a review of most of the geological studies and machine learning techniques applied to earthquake data sets, which showed a total lack of prediction of potential earthquakes through one approach, so studies designed by geologists were combined with machine learning.

Fractal Slope-Based Seismic Wave Detection Method

ABSTRACT Automatic P-wave arrival detection is the first task in an earthquake early warning systems. This study proposes a novel detection method for this based on a fractal slope (FS). We improved the calculation method of the fractal dimension to increase the calculation speed and proposed a continuous algorithm. Furthermore, we applied FS in conjunction with the short-term average over the long-term average (STA/LTA), named STA/LTA + FS. We designed orthogonal experiments with different parameters and selected a total of 40,020 sets of seismic waves from the Japanese dataset to test the best parameters. A total of 45,302 sets of seismic waves from the STanford EArthquake dataset and the Chinese dataset were selected to test the generality of the proposed method. The results show that the mean error in detection time of the proposed method is +0.042 s for different datasets. In addition, STA/LTA + FS is robust over a wide range of signal-to-noise ratio, epicentral distance, and magnitude, with the percentage of timing errors below 0.5 s higher than 95%.

Earthquake data augmentation using wavelet transform for training deep learning based surrogate models of nonlinear structures

Machine learning methods have gained traction in the civil engineering community for analysis of civil infrastructures. A major field of application is in development of surrogate models for non-linear dynamic analysis for civil infrastructures. These data-driven models generally consist of a large number of parameters which need to be estimated from available data. However, due to limited availability of data in the field of earthquake engineering, often, it is difficult to develop stable models that can predict structural response by taking into account both material and earthquake uncertainty. To this end, the authors propose a discrete wavelet transform (DWT) enabled surrogate modeling framework for prediction of nonlinear structural response while accounting for material and earthquake uncertainty. DWT is used for earthquake data augmentation and feature extraction, which, along with constitutive material parameters, are used to train a deep neural network (DNN). It is shown that DWT can efficiently augment existing small earthquake datasets while the DNN can capture the non-linear structural response. The framework is validated by applying it to successfully predict structural response of 3DOF nonlinear spring–mass–damper system and non-linear three-story steel moment-resisting frame for unseen earthquakes with a median error less than 12%.

Extending the Frequency Band of Surface‐Wave Dispersion Curves by Combining Ambient Noise and Earthquake Data and Self‐Adaptive Normalization

AbstractDispersion curves of surface waves are widely used in subsurface structural inversions. However, the non‐uniqueness problem in inversion is inevitable with limited data. Extending the dispersion curves in frequency is helpful for providing better constraints. The dispersion spectra extracted from ambient noise and earthquake data do not overlap and can be combined to provide complementary information. Besides, because multiple modes can be present at the same frequencies in the dispersion spectrum with highly non‐uniform energies, dispersion curves with lower energies are hidden by those with higher energies, thus the extracted dispersion spectrum is often not fully exploited by simple normalization. We propose to use a self‐adaptive normalization technique to uncover the hidden information and extend the dispersion curves in frequency. Both ambient noise and earthquake data recorded by the USArray Transportable Array are processed by simple and self‐adaptive normalizations, and a Markov chain Monte Carlo Bayesian inversion is conducted to estimate the velocity structure. The inversion results reveal the importance of extending dispersion curves and a significant improvement by the combined ambient noise and earthquake datasets and the self‐adaptive normalization technique.