Earthquake Early Warning Research Articles

DEEP LEARNING-BASED CLASSIFICATION OF SEISMIC EVENTS USING WAVEFORM DATA

Artificial Intelligence (AI)-enhanced seismology has evolved to its tremendous capability in analyzing real-time seismic data. This supports various tasks such as earthquake detection, phase detection, epicentral location prediction, time prediction of future earthquakes, phase picking, first motion polarity, early warning systems etc. This is crucial for the management of emergency hazard response spectra and post-disaster activities.

Electrokinetically propelled digital pendulum for seismic alert

An electrostatic induction pendulum (EIP) has been designed and developed for the first time for the innovative construction and maintenance of earthquake prewarning systems to address the existed issues of massive funding required, incomplete distribution even as well as the lack of coverage in remote or sparsely populated areas. This innovative EIP includes a pendulum and a static-electricity induced sensor which could detect subtle variations in electrical potential during motion. The sensor offers crucial insights into the vibration, damping, and resonance traits of the EIP, with the minimum detectable vibration intensity is 0.17 m/s2. Simultaneously, by introducing external vibrations varied with frequencies, the inherent frequency of the EIP can correspondingly be modified subtly leading energy distribution to be weakened, which lay a groundwork for earthquake monitoring and early warning. Furthermore, an earthquake monitoring and early warning system aided by EIP, data acquisition module, and machine learning is engineered for real-time analysis and decision-making. Once an earthquake occurs, the system automatically sends alerts to relevant departments or individuals to take protective measures in advance. This cost-effective, easily portable, and widely distributed earthquake monitoring solution complements existing systems and can be easily deployed even in remote areas.

Towards seismic risk reduction of critical facilities combining earthquake early warning and structural monitoring: a demonstration study

Towards seismic risk reduction of critical facilities combining earthquake early warning and structural monitoring: a demonstration study

Detection of Hidden Low-Frequency Earthquakes in Southern Vancouver Island with Deep Learning

Low-frequency earthquakes (LFEs) are small-magnitude earthquakes that are depleted in high-frequency content relative to traditional earthquakes of the same magnitude. These events occur in conjunction with slow slip events (SSEs) and can be used to infer the space and time evolution of SSEs. However, because LFEs have weak signals, and the methods used to identify them are computationally expensive, LFEs are not routinely cataloged in most places. Here, we develop a deep-learning model that learns from an existing LFE catalog to detect LFEs in 14 years of continuous waveform data in southern Vancouver Island. The result shows significant increases in detection rates at individual stations. We associate the detections and locate them using a grid search approach in a 3D regional velocity model, resulting in over 1 million LFEs during the performing period. Our resulting catalog is consistent with a widely used tremor catalog during periods of large-magnitude SSEs. However, there are time periods where it registers far more LFEs than the tremor catalog. We highlight a 16-day period in May 2010, when our model detects nearly 3,000 LFEs, whereas the tremor catalog contains only one tremor detection in the same region. This suggests the possibility of hidden small-magnitude SSEs that are undetected by current approaches. Our approach improves the temporal and spatial resolution of the LFE activities and provides new opportunities to understand deep subduction zone processes in this region.

Off-fault deformation feedback and strain localization precursor during laboratory earthquakes

Recent large-scale seismological observations have shown that off-fault strain localization and foreshock migration could serve as an early warning of an impending earthquake. However, this process is still largely unknown. In this study, state-of-the-art friction experiments were conducted in a oil-confined biaxial shear apparatus to investigate the link between stick-slip nucleation and off-fault deformation. Our findings indicate that there is a direct link between stick-slip nucleation and off-fault deformation, provided that the fault is conditionally unstable (a − b < 0). Inelastic off-fault deformation may trigger unstable slip by decreasing the stiffness of the surrounding rock volume, which favors earthquake nucleation. Additionally, the study presents laboratory observation of precursory strain localization around a fault during stick-slip cycles. These findings suggest that volumetric deformation processes could be a main factor in the nucleation of large ruptures and strain localization could be a reliable harbinger of large earthquakes.

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.

From Labquakes to Megathrusts: Scaling Deep Learning Based Pickers Over 15 Orders of Magnitude

AbstractThe application of machine learning techniques in seismology has greatly advanced seismological analysis, especially for earthquake detection and seismic phase picking. However, machine learning approaches still face challenges in generalizing to data sets that differ from their original training setting. Previous studies focused on retraining or transfer‐learning models for these scenarios, but require high‐quality labeled data sets. This paper demonstrates a new approach for augmenting already trained models without the need for additional training data. We propose four strategies—rescaling, model aggregation, shifting, and filtering—to enhance the performance of pre‐trained models on out‐of‐distribution data sets. We further devise various methodologies to ensemble the individual predictions from these strategies to obtain a final unified prediction result featuring prediction robustness and detection sensitivity. We develop an open‐source Python module quakephase that implements these methods and can flexibly process input continuous seismic data of any sampling rate. With quakephase and pre‐trained ML models from SeisBench, we perform systematic benchmark tests on data recorded by different types of instruments, ranging from acoustic emission sensors to distributed acoustic sensing, and collected at different scales, spanning from laboratory acoustic emission events to major tectonic earthquakes. Our tests highlight that rescaling is essential for dealing with small‐magnitude seismic events recorded at high sampling rates as well as larger magnitude events having long coda and remote events with long wave trains. Our results demonstrate that the proposed methods are effective in augmenting pre‐trained models for out‐of‐distribution data sets, especially in scenarios with limited labeled data for transfer learning.

Microseismicity at the Time of a Large Creep Event on the Calaveras Fault is Unresponsive to Stress Changes

The potential relationship between surface creep and deeper geological processes is unclear, even on one of the world’s best-studied faults. From June to August 2021, a large creep event with surface slip of more than 16 mm was recorded on the Calaveras fault in California, part of the San Andreas fault system. This event initially appeared to be accompanied by along-fault migration of seismicity, suggesting it penetrated to depth. Other studies have suggested that surface creep events are likely a shallow feature, unrelated to deep seismicity. To provide more detail on the relationship between earthquakes, surface creep, and potential aseismic slip at seismogenic depth, we tripled the number of earthquakes in the Northern California Earthquake Catalog in the region of the creep event for all of 2021. This was accomplished by implementing earthquake detection techniques based on both template matching (EQCorrscan) and AI-based automatic earthquake phase picking (PhaseNet). After manual inspection, the detected earthquakes were first located using Hypoinverse and subsequently relocated via GrowClust. Our enhanced catalog indicates that the spatiotemporal pattern of earthquakes here is not strongly influenced by the creep event and is better explained by structural heterogeneity than transient stress changes, indicating a decoupling of seismicity rate and surficial creep on this major fault.

Near Real‐Time Earthquake Monitoring in Texas Using the Highly Precise Deep Learning Phase Picker

AbstractArtificial intelligence (AI) seismology has witnessed enormous success in a variety of fields, especially in earthquake detection and P and S‐wave arrival picking. It has become widely accepted that DL techniques greatly help routine seismic monitoring by enabling more accurate picking than traditional pickers like STA/LTA. However, a completely automatic AI‐driven earthquake monitoring framework has not been reported due to the concerns of potential false positives using DL pickers. Here, we propose a novel AI‐facilitated near real‐time monitoring framework using a recently developed deep learning (DL) picker (EQCCT) that has been deployed in the Texas seismological network (TexNet). For the West Texas area, TexNet's seismic monitoring relies on the EQCCT picker to report earthquake events. For earthquakes with a magnitude above two, the picks are further validated by analysts to output the final TexNet catalog. Due to the fast‐increasing seismicity caused by continuing oil&amp;gas production in West Texas, this AI‐facilitated framework significantly relieves the workload of TexNet analysts. We show the mean absolute error (MAE) of automatic magnitude estimation for the magnitude‐above‐two earthquakes is smaller than 0.15 in West Texas and MAEs of hypocenter locations within 2.6 km in both distance and depth estimates. This research provides more evidence that DL pickers can play a fundamental role in daily earthquake monitoring.

Smart Green Concrete: Innovation of Concrete Materials from Fly ash Waste and Lapindo Mud Integrated Smart E. crassipes Coir Geotextile to Realize Indonesia's Sustainable Infrastructure

Pariaman City, situated near the Semangko fault and the Indo-Australian plate subduction zone, is highly prone to earthquakes and tsunamis. To address this, the Earthquake Buddy Application (EDY App) was developed as an innovative solution for disaster mitigation. EDY App utilizes Geographic Information Systems (GIS) to provide real-time earthquake vulnerability maps, early warnings, and evacuation route suggestions to Temporary Evacuation Sites (TES). It also offers education on earthquake preparedness and facilitates donations for relief efforts. The application’s main contribution lies in its integration of comprehensive disaster management tools into one platform. By providing critical information during emergencies, the app empowers users to make informed decisions, which can reduce casualties in high-risk areas. The app also encourages community preparedness by offering accessible education on earthquake mitigation. EDY App is designed to create a disaster-resilient society in Pariaman, ensuring that residents are better equipped to respond to natural disasters through real-time alerts and evacuation support, ultimately enhancing local preparedness and reducing the potential impact of future earthquakes.

Prediction Equations for Peak-Ground Accelerations and Velocities in Northeast Japan Using the S-net Data

S-net is a seafloor observation network for earthquakes and tsunamis around the Japan Trench, comprising 150 observatories with seismometers and pressure gauges. The region has been known to experience massive earthquakes, and several magnitude 6 and 7 class earthquakes have occurred after the network was established in 2016. This study constructed ground motion prediction equations (GMPEs) for horizontal peak ground accelerations (PGAs) and peak ground velocities (PGVs) using the S-net data and revealed that the GMPEs can be used to predict the PGAs and PGVs at the land stations where measured S-wave velocities are available. We used a relatively short time window of the S-net records from the viewpoint of earthquake early warning but included S waves. Data from earthquakes of magnitudes between Mw 5.5 and Mw 7.4 were used. The construction of the GMPEs was achieved in two steps. First, regression analysis was conducted for each event data, and mean site residual was obtained over the available records at each S-net site. Second, the data were adjusted by the mean site residuals, and stratified regression analysis, which decouples the source and path factors, was performed. Finally, we applied the GMPEs to predict PGAs and PGVs at the KiK-net sites on land. We determined that the residuals at the KiK-net sites were systematically biased with Vs30 (average S-wave velocity in the upper 30 m). We obtained correction factors for the bias and demonstrated that the PGAs and PGVs at the KiK-net sites could be predicted reasonably well.

Seismic Local Buckling Limits for Hollow Structural Section and Built-Up Box Columns

Recently completed research on seismic local buckling limits for steel hollow structural section (HSS) and built-up box columns is featured. These National Center for Research on Earthquake Engineering (NCREE) studies are led by Dr. Chung-Che Chou and Dr. Tung-Yu Wu in the Department of Civil Engineering at National Taiwan University (NTU). Dr. Chou also serves as the NCREE Director. Dr. Chou’s research has focused on seismic testing, analysis, and design for steel and post-tensioned self-centering structures. Some recent work includes studies on hybrid simulation of a full-scale steel moment frame, a post-tensioned self-centering brace, novel prediction models for early earthquake warnings, and earthquake reconnaissance work in eastern Taiwan. Dr. Chou’s numerous accolades include the Awards for Excellent Research and Technology Transfers for his leadership on a research team developing a sandwiched buckling-restrained brace and a self-centering brace. Dr. Wu’s research interests include collapse behavior of cold-formed HSS columns under seismic loading, subwavelength seismic metamaterial structures, crack growth in railway crossings under high wheel-rail impacts, and seismic resilience of steel buildings. In addition to multiple scholarships and fellowships, Dr. Wu’s honors include the Raymond C. Reese Research Prize, awarded by the American Society of Civil Engineers (ASCE) for papers representing notable achievements in research with impact on design. The National Science and Technology Council is supporting this research on seismic local buckling limits for HSS and built-up box columns. Selected highlights from both projects are presented, along with a preview of future research tasks.

Design and fabrication of an accelerometer structure using glass fiber reinforced epoxy resin for seismic P-wave early warning

Earthquake early warning represents a vital and effective strategy for mitigating earthquake-related losses. Among the existing sensors for earthquake early warning, MEMS accelerometers involve complex manufacturing process, leading to high research and development costs and long cycles. Traditional mechanical sensors are bulky, costly, hence not suitable for large-scale, high-density monitoring network. In this study, a low-cost accelerometer using glass fiber reinforced epoxy resin is proposed for longitudinal seismic waves (P-waves) early warning. Both theorical and finite element analysis are carried out to define and optimize the initial structure of the accelerometer. A pre-deformation structure is implemented into the spring suspension system to offset the influence of the sensor’s gravity, so that a self-balanced state in the direction of gravity is achieved. The simulation results show that the accelerator can achieve an eigenfrequency of 12.67 Hz, a Q-value of 49.2, and mechanical thermal noise of 1.72×10-9g/Hz. Based on the simulation results, the accelerometer’s structure is processed using a computer numerical control (CNC) milling machine, and tested by collecting the voltage signals when the accelerometer’s structure undergoes an impact in the direction of the gravity. The measured eigenfrequency is around 10 Hz, and the signal attenuation rate closely aligns with the simulation results. The accelerometer structure has dimensions within 6 cm and a relatively simplified manufacturing process, making it a cost-effective solution for the development of large-scale, high-density earthquake early warning systems.

An Overview of Traditional and Next-Generation Earthquake Early Warning Systems

The Earthquake Early Warning System generates rapid and effective warnings after detecting an earthquake, before the arrival of destructive waves to the areas that may be affected. In this way, it is aimed to minimize the loss of life and property. In this study, traditional early warning systems that are frequently used in the world are discussed. The details of the next-generation early warning system, which has recently produced successful outputs, are discussed, and its advantages over traditional early warning systems are mentioned. The recent developments of this system are also analyzed. Next-generation early warning systems consist of a new array-based algorithm composed of two modules for real-time epicenter detection. The detection of an earthquake occurring in the Marmara Sea (Türkiye) with a next-generation early warning system using array-based location methodology is given as an example. Next-generation early warning systems have advantages over traditional ones, such as lower cost and time gains of up to minutes.

Why Do Great Continental Transform Earthquakes Nucleate on Branch Faults?

Abstract The five Mw≥7.8 continental transform earthquakes since 2000 all nucleated on branch faults. This includes the 2001 Mw 7.8 Kokoxili, 2002 Mw 7.9 Denali, 2008 Mw 7.9 Wenchuan, 2016 Mw 7.8 Kaikōura, and 2023 Mw 7.8 Pazarcık events. A branch or splay is typically an immature fault that connects to the transform at an oblique angle and can have a different rake and dip than the transform. The branch faults ruptured for at least 25 km before they joined the transforms, which then ruptured an additional 250–450 km, in all but one case (Pazarcık) unilaterally. Branch fault nucleation is also likely for the 1939 M 7.8 Erzincan earthquake, possible for the 1906 Mw∼7.8 and 1857 Mw∼7.9 San Andreas earthquakes, but not for the 1990 Mw 7.7 Luzon, 2013 Mw 7.7 Balochistan, and 2023 Mw 7.7 Elbistan events. Here, we argue that because fault continuity and cataclastite within the fault damage zone develop through cumulative fault slip, mature transforms are pathways for dynamic rupture. Once a rupture enters the transform from the branch fault, flash shear heating causes pore fluid pressurization and sudden weakening in the cataclastite, resulting in very low dynamic friction. But the static friction on transforms is high, and so they are usually far from failure, which could be why they tend to be aseismic between, or at least for centuries after, great events. This could explain why the largest continental transform earthquakes either begin on a branch fault or nucleate along the transform at locations where the damage zone is absent or the fault continuity is disrupted by bends or echelons, as in the 1999 Mw 7.6 İzmit earthquake. Recognition of branch fault nucleation could be used to strengthen earthquake early warning in regions such as California, New Zealand, and Türkiye with transform faults.

Detection of Small Earthquakes by Waveform Envelope Using Machine Learning

ABSTRACT Analyzing seismic data has helped reveal most of our knowledge about the Earth’s interior. However, the number of moderate-to-large earthquakes is limited; and most small earthquakes, although essential for monitoring the dynamic process between major earthquakes, are difficult to detect from earthquake recordings. In this study, we present a detection method that catches the envelope pattern of seismic data in machine learning. The waveform envelope presents reliable features and a low-frequency pattern for any event rather than being dominated by details of noise and signals. We first apply a fully connected neural network to extract data envelopes from recordings and then a convolutional neural network to detect events using the envelopes as input. Our method is tested against the previously published method on seismic data in Japan. The new approach identifies the largest number of small events. In the testing dataset, the precision and recall of the approach for events are 98.73% and 96.54%, respectively, and those for noise are 97.41% and 99.05%, respectively. We demonstrate that the approach performs well in different signal-to-noise ratios (SNRs) and filter frequency band tests. For events with low SNRs (0–6 dB), the detection accuracy of the approach is approximately 10% higher than a published deep learning method. Besides most of the events in the earthquake catalog, the process also detects more events from continuous data that are not documented in the catalog.

Developing and evaluating a risk-informed decision support system for earthquake early warning at a railway bridge

Earthquake early warning (EEW) systems provide timely information on the arrival of strong seismic waves at a site. Such information can help mitigate the consequences of earthquakes on the operation of infrastructure assets. EEW-informed mitigation actions should stem from risk-based decision-making protocols, and EEW benefits should be evaluated for the range of possible rupture scenarios that affect an asset of interest. This paper addresses these challenges specifically for railway bridges by: (1) developing a risk-informed EEW decision support system (DSS) for these assets; and (2) quantifying the effectiveness of the proposed EEW-DSS in mitigating seismic risks for railway bridges across relevant rupture scenarios. The proposed EEW-DSS combines information on site-specific seismic hazard, time-dependent EEW algorithm outputs, probabilistic seismic demand modeling, damage/derailment fragilities, and seismic loss models. These modeling components compose a multi-criteria decision-making framework. Value of information theory is then proposed to estimate the loss-mitigation benefits of the EEW-DSS, accounting for varied stakeholder risk priorities as well as dynamic lead-time/accuracy tradeoffs related to EEW performance. A railway bridge with multiple piers is adopted for the case study, which demonstrates the importance of risk-based, uncertainty-informed decision-making to the overall effectiveness of EEW in potentially reducing seismic losses.

Earthquake Magnitude Estimation through Mixed-Effects Ground-Motion Modeling of Early P-Wave Arrivals

Abstract A regional earthquake early warning system (EEWS) warrants potential predictive models to accurately extract earthquake parameters like magnitude and intensity from the first few seconds of a P-wave arrival. In this study, a maiden predictive model depicting the relationship between peak displacement amplitude (Pd) and magnitude (ML) is proposed for the western Himalayan region through a mixed-effects regression and compared with those from similar tectonic regimes. This model for EEWS is derived from the vertical-component waveforms with a high signal-to-noise ratio, using three different time-window lengths (Td) of 1, 2, and 3 s, just after the P onset. Waveforms from 83 earthquakes in the magnitude (ML) range of 3 and 5.5 registered at 27 strong motion seismic stations are used for this purpose. The hypocentral distance range varies between 5 and 264 km. A comparative analysis between the models obtained through linear and linear mixed-effects (lme) regression reveals that the latter is robust. It is observed that the intra-event uncertainties are significantly reduced after site corrections and contribute more toward the total variabilities, compared to the inter-event uncertainties. Based on the results from this study, it is emphasized that the local site effects should be incorporated while developing the predictive models for EEWS. Importantly, the displacement magnitude (Mpd) derived from Pd values, accurately matches with ML, even for the data not used to derive the model, lending credence to the final model. A scaling relation between the peak ground velocities (PGV) and Pd values is also established to evaluate the seismic hazard levels. Advocating that the adapted models should be calibrated for a targeted region, the dissimilarities among different models and the implications from epistemic uncertainties are also discussed in the present study.

Rapid and Resilient LoRa Leap: A Novel Multi-Hop Architecture for Decentralised Earthquake Early Warning Systems.

We introduce a novel LoRa-based multi-hop communication architecture as an alternative to the public internet for earthquake early warning (EEW). We examine its effectiveness in generating a meaningful warning window for the New Zealand-based decentralised EEW sensor network implemented by the CRISiSLab operating with the adapted Propagation of Local Undamped Motion (PLUM)-based earthquake detection and node-level data processing. LoRa, popular for low-power, long-range applications, has the disadvantage of long transmission time for time-critical tasks like EEW. Our network overcomes this limitation by broadcasting EEWs via multiple short hops with a low spreading factor (SF). The network includes end nodes that generate warnings and relay nodes that broadcast them. Benchmarking with simulations against CRISiSLab's EEW system performance with internet connectivity shows that an SF of 8 can disseminate warnings across all the sensors in a 30 km urban area within 2.4 s. This approach is also resilient, with the availability of multiple routes for a message to travel. Our LoRa-based system achieves a 1-6 s warning window, slightly behind the 1.5-6.75 s of the internet-based performance of CRISiSLab's system. Nevertheless, our novel network is effective for timely mental preparation, simple protective actions, and automation. Experiments with Lilygo LoRa32 prototype devices are presented as a practical demonstration.

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.