Earthquake Prediction Research Articles

The Study on Anomalies of the Geomagnetic Topology Network Associated with the 2022 Ms6.8 Luding Earthquake

On 5 September 2022, the Ms 6.8 Luding earthquake occurred at 29.59°N and 102.08°E in China. To investigate the variations in geomagnetic signals before the earthquake, this study analyzes the geomagnetic data from nine stations around the epicenter. First, we apply the Multi-channel Singular Spectrum Analysis to reconstruct the periodic components of the geomagnetic data from multiple stations. Second, we employ K-means clustering to rule out the possibility of occasional anomalies caused by a single station. Subsequently, we construct a geomagnetic topology network considering the remaining stations. Network centrality is defined as a measure of overall network connectivity, where the higher the correlation between multiple stations, the greater the network centrality. Finally, we examine the network centrality 45 days before and 15 days after the Luding earthquake. The results show that several anomalies in network centrality are extracted about one week before the earthquake. We further validate the significance of the anomalies in terms of time as well as space and verify the utility of the centrality anomalies through the SEA technique. The anomalies are found to have a statistical correlation with the earthquake event. We consider that this study provides a new way and a novel observational perspective for earthquake precursor analysis of ground-based magnetic data.

Feasibility of Principal Component Analysis for Multi-Class Earthquake Prediction Machine Learning Model Utilizing Geomagnetic Field Data

Geomagnetic field data have been found to contain earthquake (EQ) precursory signals; however, analyzing this high-resolution, imbalanced data presents challenges when implementing machine learning (ML). This study explored feasibility of principal component analyses (PCA) for reducing the dimensionality of global geomagnetic field data to improve the accuracy of EQ predictive models. Multi-class ML models capable of predicting EQ intensity in terms of the Mercalli Intensity Scale were developed. Ensemble and Support Vector Machine (SVM) models, known for their robustness and capabilities in handling complex relationships, were trained, while a Synthetic Minority Oversampling Technique (SMOTE) was employed to address the imbalanced EQ data. Both models were trained on PCA-extracted features from the balanced dataset, resulting in reasonable model performance. The ensemble model outperformed the SVM model in various aspects, including accuracy (77.50% vs. 75.88%), specificity (96.79% vs. 96.55%), F1-score (77.05% vs. 76.16%), and Matthew Correlation Coefficient (73.88% vs. 73.11%). These findings suggest the potential of a PCA-based ML model for more reliable EQ prediction.

Multi‐Segment Earthquake Clustering as Inferred From 36Cl Exposure Dating, the Bet Kerem Fault System, Northern Israel

AbstractRecovering the seismic history of multiple segments within a fault system provides a spatiotemporal framework for the fault activity across the system. This kind of data is essential for improving our understanding of how faults interact during earthquake cycles and how they are distributed within a fault system. Bedrock fault scarps, reaching up to 10‐m height, are abundant across the Bet Kerem fault system, Galilee, northern Israel. Using the 36Cl exposure dating method, we recovered the last 30 ka scarp exhumation history of three fault segments from the Bet Kerem fault system. Results indicate that the three faults were active simultaneously in at least three distinguished activity periods, during which a minimum of 1.2 m of surface rupturing occurred in each period. The synchronized activity and total surface rupture at each activity period suggest that the three dated segments were ruptured simultaneously by the same earthquake. That is, a multi‐segment rupture earthquake and that each activity period included a cluster of at least two large multi‐segment earthquakes. The results also indicate a recurrence interval between clusters of 3.5–4.5 ka and the existence of a seismic super cycle with a recurrence interval of about 13 ka.

Machine Learning Predicts Earthquakes in the Continuum Model of a Rate‐And‐State Fault With Frictional Heterogeneities

AbstractMachine learning (ML) has been used to study the predictability of laboratory earthquakes. However, the question remains whether or not this approach can be applied in a tectonic setting where one may have to rely on sparse earthquake catalogs, and where important timescales vary by orders of magnitude. Here, we apply ML to a synthetic seismicity catalog, generated by continuum models of a rate‐and‐state fault with frictional heterogeneities, which contains foreshocks, mainshocks, and aftershocks that nucleate in a similar manner. We develop a network representation of the seismicity catalog to calculate input features and find that the trained ML model can predict the time‐to‐mainshock with great accuracy, from the scale of decades to minutes. Our results offer clues as to why ML can predict laboratory earthquakes and how the developed approach could be applied to more complex problems where multiple timescales are at play.

Time series analysis and applications in data analysis, forecasting and prediction

Time series analysis is an essential field in data analysis, particularly within forecasting and prediction domains. Researching and building time series models play a crucial role in understanding and predicting the temporal dynamics of various phenomena. In mathematics, time series data is defined as data points indexed in chronological order and have a consistent time interval between consecutive observations. This can include data such as daily stock prices, annual national income, quarterly company revenue, and more. The advantage of time series data is that it can capture the state of a variable over time. In contrast, the world is constantly changing, and phenomena rarely remain static they typically exhibit variations over time. Therefore, time series data has highly practical applications and is used in various fields, including statistics, econometrics, financial mathematics, weather forecasting, earthquake prediction, electroencephalography, control engineering, astronomy, telecommunications, and signal processing. ARIMA, which stands for Auto Regressive Integrated Moving Average, is a widely used time series forecasting method in data science. It is a popular model for analyzing and predicting time-dependent data points. ARIMA combines autoregression, differencing, and moving averages to capture different aspects of time series data. In this paper, we study ARIMA, which is a significant model for analyzing and predicting time series data.

Megathrust locking encoded in subduction landscapes.

Locked areas of subduction megathrusts are increasingly found to coincide with landscape features sculpted over hundreds of thousand years, yet the mechanisms that underlie such correlations remain elusive. We show that interseismic locking gradients induce increments of irreversible strain across the overriding plate manifested predominantly as distributed seismicity. Summing these increments over hundreds of earthquake cycles produces a spatially variable field of uplift representing the unbalance of co-, post-, and interseismic strain. This long-term uplift explains first-order geomorphological features of subduction zones such as the position of the continental erosive shelf break, the distribution of marine terraces and peninsulas, and the profile of forearc rivers. Inelastic yielding of the forearc thus encodes short-term locking patterns in subduction landscapes, hinting that megathrust locking is stable over multiple earthquake cycles and highlighting the role geomorphology can play in constraining Earth's greatest source of seismic hazard.

The effects of precursory velocity changes on earthquake nucleation and stress evolution in dynamic earthquake cycle simulations

Seismic velocity changes in earthquake cycles have been observed over a wide range of timescales and may be a good indicator of the onset of future earthquakes. Understanding the effects of precursory velocity changes right before seismic and slow-slip events could potentially elucidate the onset and timing of fault failure. We use numerical models to simulate fully dynamic earthquake cycles in 2D strike-slip fault systems with antiplane geometry, surrounded by a narrow fault-parallel damage zone. By imposing S-wave velocity changes inside fault damage zones, we investigate the effects of these precursors on multiple stages of the seismic cycle, including nucleation, coseismic, postseismic, and interseismic stages. Our modeling results show a wide spectrum of fault slip behaviors including fast earthquakes, slow-slip events, and variable creep. One primary effect of the imposed velocity precursor is on the earthquake nucleation phase, and earlier onset of precursors causes earthquakes to nucleate sooner with a smaller nucleation size that is not predicted by theoretical equations. Furthermore, such precursors affect the nucleation of dynamic earthquakes and slow-slip events. Our results highlight the importance of short- and long-term monitoring of fault zone structures for better assessment of regional seismic hazard.

Unraveling the roles of fault asperities over earthquake cycles

Fault asperities can produce concentrated slips during large earthquakes and intensify damage. However, how asperities control fault behavior during other phases of the earthquake cycle remains poorly known. Here, we conduct friction experiment on a laboratory fault featuring two prominent geometric asperities to directly image their influences on long-term and short-term seismicity, and the nucleation and propagation of the mainshock. The laboratory observations and supporting numerical simulations reveal that one asperity located in the fault interior behaves firstly as a mechanical attractor to long-term seismicity, then a barrier to preseismic slow slip, and ultimately a source of large coseismic slip. In contrast, another asperity located near the fault margin primarily undergoes persistent aseismic slip throughout most phases of the earthquake cycle. These results provide new insights into how asperities partition strain across a broad spatiotemporal domain, establishing a physical link between long-term and short-term fault behaviors and the occurrence of large earthquakes.

Seismic evidence for melt-rich lithosphere-asthenosphere boundary beneath young slab at Cascadia

The Lithosphere-Asthenosphere Boundary (LAB) beneath oceanic plates is generally imaged as a sharp seismic velocity reduction, suggesting the presence of partial melts. However, the fate of a melt-rich LAB is unclear after these plates descend into the mantle at subduction zones. Recent geophysical studies suggest its persistence with down-going old and cold slabs, but whether or not it is commonly present remains unclear, especially for young and warm slabs such as in the Cascadia subduction zone. Here we provide evidence for its presence at Cascadia in the form of a large (9.8±\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\pm$$\\end{document}1.5%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}) decrease in shear-wave velocity over a very small (<3 km) depth interval. Similarly large and sharp seismic velocity reduction at the bottom of both old and young slabs, as well as along the base of oceanic plates before subduction, possibly represents widespread presence of melts. The melt-rich sub-slab LAB may strongly influence subduction dynamics and viscoelastic earthquake cycles.

Comprehensive quantitative determination of aquifer confinement based on tidal response of well water level and its application in North China

Aquifer confinement represents a pivotal property that significantly influences the vulnerability and contamination risk of groundwater resources. Several methods have been proposed for determining aquifer confinement by analyzing the response of well water level to Earth tides and atmospheric tides. In this study, we evaluated the performance of the existing single methods and put forward an optimized comprehensive approach. We compared the determination results of the three single methods with those of a comprehensive method using water-level data from 39 earthquake precursor monitoring wells in North China. The results demonstrate that the comprehensive method effectively determined aquifer confinement, significantly reducing the uncertainty associated with the three single methods. The application of the comprehensive method in North China reveals that aquifer confinement may undergo temporal variations during long-term continuous observation, especially in areas where the confining properties of aquifers may vary due to human activities and earthquakes. In such areas, the comprehensive method facilitates accurate assessment of groundwater vulnerability, as well as the potential dispersion of underground pollutants.

Application of Artificial Neural Network Methods to Anatolian Plate Earthquake Magnitude and Location Prediction

Same-region earthquakes usually have a pattern that is difficult to identify clearly. Therefore, time series analysis methods have been proposed for earthquake prediction. Our work attempts to predict three earthquake parameters in the Anatolian Peninsula using pure artificial neural network methods. An optimized BP-NN model and optimally hyper-parameterized LSTM Model have been designed to predict earthquake magnitude, latitude, and longitude. The two models are compared among each other and with previous works for their prediction performances, using four well-accepted metrics: Mean Squared Error, Mean Absolute Error, Median Absolute Error, and Standard Deviation. The time, depth, sun, and moon distances to Earth were identified as the most contributing factors in earthquake occurrence through analysis by five different feature extraction algorithms. The date harmed the prediction accuracy. The LSTM model outperformed the BP-NN Model in magnitude prediction with 0.062 MSE. Latitude predictions of both methods were satisfactory and close. However, BP-NN had lower error rates in latitude prediction. However, longitude prediction errors were large in both models. Therefore, our designs were not successful in predicting the exact location of the earthquakes. However, multi-variate, stacked LSTM models are promising in predicting Anatolian Peninsula earthquake magnitudes, but future work is necessary for location and timing predictions.

VLF Signal Noise Reduction during Intense Seismic Activity: First Study of Wave Excitations and Attenuations in the VLF Signal Amplitude

This study is a continuation of pilot research on the relationships between seismic activity and changes in very low frequency (VLF) signals starting a few minutes or a few dozen minutes before an earthquake. These changes are recorded in the time and frequency domains and their duration can be influenced not only by the strongest earthquake but also by others that occur in a short time interval. This suggests that there are differences in these changes in cases of individual earthquakes and during the period of intense seismic activity (PISA). In a recent study, they were validated in the time domain by comparing the amplitude noise reductions during the PISA and before earthquakes that occurred in the analysed periods without intense seismic activity (PWISA). Here, we analyse the changes in the VLF signal amplitude in the frequency domain during the PISA and their differences are compared to the previously investigated relevant changes during PWISA. We observe the signal emitted by the ICV transmitter in Italy and received in Serbia from 26 October to 2 November 2016 when 907 earthquakes occurred in Central Italy. The study is based on analyses of the Fourier amplitude AF obtained by applying the fast Fourier transform (FFT) to the values of the ICV signal amplitude sampled at 0.1 s. The obtained results confirm the existence of one of the potential earthquake precursors observed during PWISA: significantly smaller values of AF for small wave periods (they can be smaller than 10−3 dB) than under quiet conditions (the expected values are larger than 10−2 dB). Exceptions were the values of AF for wave periods between 1.4 s and 2 s from a few days before the observed PISA to almost the end of that period. They were similar or higher than the values expected under quiet conditions. The mentioned decrease lasted throughout the observed longer period from 10 October to 10 November, with occasional normalisation. It was many times longer than the decreases in AF around the considered earthquakes during PWISA, which lasted up to several hours. In addition, no significant wave excitations were recorded at discrete small values of the wave periods during the PISA, as was the case for earthquakes during PWISA. These differences indicate the potential possibility of predicting the PISA if the corresponding earthquake precursors are recorded. Due to their importance for potential warning systems, they should be analysed in more detail in future statistical studies.

Understanding earthquake precursors: from subcritical instabilities to catastrophic events

The collapse of man-made and natural structures is a complex phenomenon that has been studied for centuries. Existing models often focus on a ‘critical point’ where failure becomes imminent. This work presents a radically different perspective: large earthquakes may not arise from critical states, but instead develop dynamically from the subcritical regime as rare, extreme events. Our approach hinges on an extension of Onsager’s reciprocal theorem, allowing us to delve into this subcritical realm. We demonstrate that within such a regime, excitable systems, like those underlying earthquakes, are dynamically renormalised towards a nonlocal equilibrium. For these systems, the maximum entropy production of at least two interacting phases is used to replace the local equilibrium assumption for the subcritical state. Typically, dissipative processes at larger scales arrest these self-amplifying feedbacks. However, in rare instances, they can morph into intricate tensor networks of instabilities that ripple from microscopic scales to the entire system, culminating in an extreme event like a catastrophic earthquake. This novel framework offers a potentially deeper understanding of earthquake precursors and paves the way for exploring earthquake prediction based on the statistics of subcritical dynamics.

Association of land surface temperature anomalies from GOES/ABI, MSG/SEVIRI, and Himawari-8/AHI with land earthquakes between 2010 and 2021

Every year, earthquakes cause numerous casualties and billions of dollars in economic damages, and so far, it is impossible to predict them. Many researchers have studied the land surface temperature (LST) anomalies as promising potential earthquake precursors. In this work, more than 30,000 land earthquakes with a Magnitude (Mw) larger than 4 have been studied from 2010 to 2021. Global LST nighttime (0–6 am) data from Meteosat Second Generation (MSG), Himawari-8, and the Geostationary Operational Environmental Satellite (GOES) satellites have been used to evaluate the LST anomalies in the zones affected by earthquakes with spatial resolutions of 1, 2, and 4 km, respectively. The time series moving standard deviation (STD) and Interquartile (IQT) have been used to estimate the LST anomalies. The Confusion Matrix (CM), the receiver operating curve (ROC), and some other figures of merit are computed to assess the goodness of positive LST anomalies and detection thresholds as potential proxies of earthquake occurrence. A positive LST anomaly is typically observed a few days [1–7 d] before the earthquake’s occurrence, followed by a negative LST anomaly afterwards [1–3 d].

Extremely Low Frequency (ELF) Electromagnetic Signals as a Possible Precursory Warning of Incoming Seismic Activity

We analyzed a large number (77) of low-to-medium-magnitude earthquakes (M3.5–M6.5) that occurred within a period of three years (2020–2022) in the Southern half of Greece in relation to the ELF activity in that region and time period. In most cases, characteristic ELF signals appear up to 20 days before the earthquakes. This observation may add an important new element to the Lithospheric–Atmospheric–Ionospheric scenario, thus contributing to a better prediction of incoming earthquakes. We discuss the role of ELF observations in reliable seismic forecasting. We conclude that the magnitude of an earthquake larger than M4.0 and the distance of the epicenter shorter than 300 km from the recording site is needed for typical pre-seismic signals to be observed. Finally, we remark that a reliable prediction of earthquakes could result from an integrated project of multi-instrumental observations, where all the known variety of precursors would be included, and the whole data set would be analyzed by advanced machine learning methods.

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.

Machine learning technique in the north zagros earthquake prediction

Studying the changes in seismicity, and the potential of the occurrences of large earthquakes in a seismic zone is not only extremely important from the aspect of seismological research, but it is additionally significant in the decisions of crisis management. Since, nowadays Machine learning techniques have proven the high ability for analyzing information, and discovering the relations among the parameters, in this research were tested some of these techniques for the earthquake prediction. For analysis, the north Zagros seismic catalogue was selected. A region that is an active seismic zone, and large cities are located there. Moreover, nine seismic parameters were used to study the possibility of large earthquake prediction for 1 month using three different Machine Learning (ML) techniques (Artificial Neural Network (ANN), Random Forest, and Support Vector Machine (SVM)). The accuracy of prediction models was evaluated using four different statistical measures (recall, accuracy, precision, and F1-score). The results showed that the (ANN) method is more accurate than other methods. Based on three investigated methodologies, greater accuracy results have been produced to forecast the earthquakes with bigger scale earthquakes about the completeness of the seismic catalogue in large magnitude. These achievements promise the possibility of successful prediction in a short period, which is hopeful for better crisis management performance.

A Bibliometric Review of Earthquake and Machine Learning Research

This article presents a bibliometric review of earthquake research and its integration with machine learning techniques. Over the past two decades, there has been a growing interest in using machine learning to enhance earthquake prediction and research. The review collected 1172 scholarly articles from the Web of Science database, focusing on the keywords "earthquake" and "machine learning." Machine learning has shown promise in improving earthquake forecasting models and aiding decision-making in disaster management, infrastructure design, and emergency response. However, it is noted that the application of machine learning in earthquake engineering is still in its early stages and requires further exploration. Key findings of this review include the increasing importance of certain keywords in earthquake and machine learning research, such as "prediction," "neural network," "classification," "logistic regression," and "performance." These keywords highlight the central areas of research focus within this field. The review also identifies research trends and gaps, including the need for more exploration of large-scale, high-dimensional, nonlinear, non-stationary, and heterogeneous spatiotemporal data in earthquake engineering. It emphasizes the necessity for novel machine learning algorithms tailored specifically for earthquake prediction and analysis. Furthermore, it highlights the need for addressing uncertainty in earthquake research and improving forecasting models. The review underscores the growth in interest and collaboration in earthquake research and machine learning, evident in the increasing number of scholarly contributions over the years. In summary, this bibliometric review highlights the importance of accurate forecasting and the potential of machine learning techniques in advancing this field.

Identification of Possible Short-Term Ionospheric Earthquake Precursors for Seismic Events with Intermediate Hypocentrals Depths by Measuring the Standard Parameters of the Mid-Latitude Es Layer

Identification of Possible Short-Term Ionospheric Earthquake Precursors for Seismic Events with Intermediate Hypocentrals Depths by Measuring the Standard Parameters of the Mid-Latitude Es Layer

Research on sensitized Fiber Bragg Grating temperature sensor based on bimetal three-substrates.

Temperature is one of the most important physical quantities in the field of earthquake precursor observation. Aiming at the problem of low sensitivity in the Fiber Bragg Grating (FBG) temperature sensor, the sensitized FBG temperature sensor based on bimetal three-substrates is proposed. Through theoretical analysis of the bimetallic model, the structural parameters of the sensor are optimized, and the sensor is simulated and analyzed with ANSYS. Then, the sensor is developed according to the simulation results, and the temperature test system is built to test the performance of the sensor. The results show that the sensitivity of the temperature sensor is 49.3 pm/°C, which is about 4.9 times that of the bare FBG sensor, and the linearity is over 0.999. The research results provide a reference for developing the same type of sensors and further improving the sensitivity of FBG temperature sensors.