Search

In this paper, a scalable end-to-end tree boosting system called XGBoost has been applied for predicting the magnitude of an earthquake from the early part of earthquake waveform data. This model uses the features extracted from the early P wave phase of the records as an input. The model's effectiveness has been verified by using data on earthquakes occurring in the Eurasian plate of Japan Islands from 1996 to 2021. Feature engineering has given 29 new features identified from the early P wave phase of the record, which show a high correlation with the magnitude of an earthquake. The comparison of predicted and actual magnitude shows that a trained XGboost model, which uses a single input record for magnitude prediction, gives an average prediction error of 0.004 ± 0.57 for earthquakes in the test dataset. In contrast, the average prediction error of –1.1 ± 0.80 and –0.65 ± 0.69 has been obtained for the magnitude estimated from conventional τc and Pd methods using the same test dataset. It is further seen that the average predicted magnitude of a single earthquake of magnitude 4.5 and 6.1 (MJMA) obtained by using multiple nearfield records using XGBoost model is 4.58 ± 0.33 and 6.32 ± 0.29, which is close to the actual magnitude of the earthquake. The results presented in this paper clearly show that the structured data can be effectively used by complex machine learning or deep learning models to predict earthquake magnitude from single or multiple records. [ABSTRACT FROM AUTHOR]

Earthquake swarms are most commonly composed of small‐magnitude earthquakes. However, a recent study by Yoshida, Uchida, et al. (2023, https://doi.org/10.1029/2023GL106023) analyzed a swarm beneath the Noto Peninsula in Japan that, after more than two years of moderate‐magnitude seismicity, triggered the moment magnitude (Mw) 6.2 Suzu mainshock in May 2023. Based on high‐precision earthquake locations and a slip inversion of the mainshock, these authors found that the Mw 6.2 Suzu earthquake occurred on the updip extension of a fault that was active during the swarm, likely driven by increased fluid pressure. After publication of that paper, a much larger and more destructive Mw 7.5 event occurred nearby on 1 January 2024. These events underscore the potential for swarms to be precursors to large, damaging earthquakes. Forecasting the eventual evolution of swarms is currently very challenging but could be aided in the future by new observations and models. Plain Language Summary: Although earthquake swarms often only contain small earthquakes, occasionally they can trigger large, damaging earthquakes. Yoshida, Uchida, et al. (2023, https://doi.org/10.1029/2023GL106023) examined the triggering of a magnitude 6.2 earthquake in Japan, which caused extensive damage, following two years of smaller earthquakes. This sequence, like many swarms, may have been driven by elevated fluid pressure migrating upwards within the crust. After publication of that paper, a much larger and more destructive Mw 7.5 event occurred nearby. Forecasting the eventual evolution of ongoing swarms is currently very challenging but could be aided in the future by new observations and models. Key Points: Earthquake swarms can be precursors to damaging earthquakes, including the 1 January 2024, Mw 7.5 Noto Peninsula, Japan eventFluid pressure changes often drive swarm activity, and their degree of confinement may influence swarm evolutionDirect observations of fluids in fault zones combined with numerical modeling may aid swarm forecasting in the future [ABSTRACT FROM AUTHOR]

We studied the earthquake detection capacity of DONET (Dense Oceanfloor Network system for Earthquakes and Tsunamis) operating in the Nankai Trough, a target region monitored for future megathrust earthquakes. The focus of this paper was to evaluate the impact on this capacity from the malfunction of parts of the network. For this purpose, the completeness magnitude, above which all earthquakes are considered to be detected by a seismic network, was used. Then, a catalog that includes events observed by DONET was used. We found spatiotemporal variability of completeness magnitude, ranging from values below 1 in one of the areas where stations are densely deployed to values above 2 at the periphery and outside of the DONET area. We conducted a simulation computation for cases of malfunction of densely distributed stations. The results showed that completeness estimates in the area near the malfunctioning stations were about 1 magnitude larger. This implies that malfunction repair and/or replacement with new stations would be desirable because they pronouncedly affect earthquake monitoring. We then demonstrated an example of how to use the information of completeness magnitude as prior knowledge to compute the b value of the Gutenberg-Richter distribution. The result indicates the b value as a proxy that can help to image stress heterogeneity when there is a magnitude-6 class slow slip event on the Nankai Trough plate boundary. [ABSTRACT FROM AUTHOR]

Though there is now a substantial body of research on the relationship between nonprofits and the state, examples from non-Western countries have been lacking in the nonprofit literature in English. This paper attempts to fill this gap using nonprofit cases in three key districts in Fukushima after the Great East Japan Earthquake. By focusing on these cases, this study captures the complex nature of the relationship between the nonprofit sector and state in Japan during a period of increased activity of both the nonprofit sector and the state. The idea of the nonprofit sector, state, and market as relatively separate and with different interests, influences discussions on themes related to the nonprofit sector (e.g., business-like behavior). The study suggests how the Japanese case can contribute in uncovering assumptions that are embedded in the existing understanding of the nonprofit sector, based primarily on Western examples. [ABSTRACT FROM AUTHOR]

The Fukushima power station in Japan was affected by a major earthquake and tsunami in March 2011, inspecting the primary containment vessel remains difficult due to high radioactivity. Long-reach robot arms are useful in inspecting such hazardous environments, and a coupled tendon-driven mechanism enables realizing a long, light-weight, and thin manipulator. However, high elastic elongation of tendons due to gravity may lead to unstable joint control. In this paper, we introduce dynamics-based control as a feasible strategy for a long-reach tendon-driven robotic arm. Additionally, a planning method to identify the joint angle path ensuring stability is proposed. Considering stability analysis, the potential due to the tendon elasticity and gravity is evaluated and used as an index of joint stability. The rapidly exploring random tree is used as the planning algorithm. The effectiveness of the proposed method was demonstrated through the successful manipulation of a 5-kg payload by a 10-m long robotic arm. [ABSTRACT FROM AUTHOR]

The electron density (Ne), ion density (Ni), and electron temperature (Te) statistics recorded by the DEMETER satellite payload ISL (Instrument Sonde de Langmuir) were used to study the disturbance characteristics of the ionosphere before solid earthquakes of magnitude 6 or higher in Japan during the summer of 2005 – 2009 , to provide more information and methods for the coupling mechanism and short-range earthquake prediction. In this paper, the region of ± 10 ° of the epicenter is divided into 1 ° × 1 ° pixels, and the background field of each parameter is constructed using data without earthquakes and relatively calm space weather. We also define a measure of the perturbation intensity of ionospheric parameters relative to the background field during the occurrence of earthquakes. The analysis results of the four Japanese earthquakes from space and time show an excellent synchronization in the time and area of the anomalies in ionospheric parameters of the four cases. All four instances showed Ne abnormalities, and three showed Ni and Te anomalies, in which N e and Ni increased or decreased abnormally, while Te increased abnormally every time, and the anomalies mainly occurred about 9–12 days before the earthquake. This paper eliminates the influence of solar, geomagnetic, and satellite data defects on the experimental results as much as possible in data screening and method selection. The results partially agree with the conclusions reported in the existing literature, and the obtained anomalies are somewhat related to the ionospheric precursors of earthquakes. [ABSTRACT FROM AUTHOR]

This paper discusses the relationship between the vertical ground motion and ionospheric disturbances before the Kumamoto earthquake on 16 April 2016, in Kyushu, Japan, using the vertical ground motion measured by slant gauges widely distributed in Kyushu, and the NmF2 observed by ionosondes in Japan and another region. We provide evidence that vertical ground motion excites internal gravity waves (IGWs) that disturb changes in the ionospheric plasma density. From the spectral analysis results of the vertical ground motion data, the summation of various period (frequency) components analyzed from the original data of the slant gauge shows a possible correlation with the change of NmF2 before the earthquake. On the other hand, the influence of the geomagnetic disturbance on vertical ground motion seems to exist. However, we cannot confirm that vertical ground motion is influenced by the geomagnetic disturbance (Kp index) and that the earthquake is triggered by the geomagnetic disturbance. There are two conditions for the vertical ground motion to disturb variations in the ionospheric plasma density: (1) The effective vertical ground motion period should be shorter than 5 h. In addition, (2) vertical ground motion should continue to exist so that wave energy can be continuously injected into the atmosphere. A possible mechanism with which to modify the ionosphere is discussed. The results of this study can provide a basis for the future ionospheric precursors of earthquakes by using the vertical ground motion. [ABSTRACT FROM AUTHOR]

In this paper, we explore the use of seismicity data on a single-station basis in site response characterization. We train a supervised deep-learning model, SeismAmp, to recognize and separate seismic site response with reference to seismological bedrock (VS  = 3.45 km s−1) in a broad frequency range (0.2–20 Hz) directly from single-station earthquake recordings (features) in Japan. Ground-truth data are homogeneously created using a classical multistation approach—generalized spectral inversion at a total number of 1725 sites. We demonstrate that site response can be reliably separated from single-station seismograms in an end-to-end approach. When SeismAmp is tested at new sites in both Japan (in-domain) and Europe (cross-domain), it achieves the lowest standard deviation among all tested single-station techniques. We also find that horizontal-to-vertical spectral ratio (HVSR) is not the optimal use of single-station recordings. The individual components of each record carry salient information on site response, especially at high frequencies. However, part of the information is lost in HVSR. SeismAmp could lead to improved site-specific earthquake hazard prediction in cases where recordings are available or can be collected at target sites. It is also a convenient tool to remove repeatable site effects from ground motions, which may benefit other applications, for example, improving the retrieval of seismic source parameters. Finally, SeismAmp is trained on data from Japan, future studies could explore transfer learning for practical applications in other regions. [ABSTRACT FROM AUTHOR]

Peak ground acceleration (PGA) is a key parameter used in earthquake early warning systems to measure the ground motion strength and initiate emergency protocols at major projects. The traditional P -wave peak displacement-dependent PGA prediction model (Pd-PGA model) tends to underestimate the PGA for large earthquakes because it cannot make full use of the fault continuity rupture information hidden in the time-varying process of ground motion. In this paper, a continuous PGA prediction long short-term memory (LSTM) neural network model is proposed. The model takes eight sequential features of stations that are proxies of the energy and other physical parameters as input and provides the recorded PGA at the station as the target output. A total of 5961 records from 119 earthquakes recorded by the Japanese Strong-Motion Earthquake Network (K-NET) in Japan are used to train the neural network and 3433 records from 73 earthquakes are used as the test set to verify the model's generalization ability. The results show that within the same data set, the residuals of the predicted PGA for the proposed model are smaller than those of the Pd-PGA model and that the problem of PGA underestimation is resolved. The prediction accuracy also improves with increasing sequence length, which indicates that the LSTM neural network learns the rules hidden in the time series. To further verify the model's generalization ability, the model performance is analyzed for an M 7.3 earthquake that was not included in the training or test data sets. The results show that the residuals of the predicted PGA for the event are consistent with those for the test data set, indicating that the model has good generalization ability. [ABSTRACT FROM AUTHOR]

The motivation of this study stems from a retaining wall made of waste tires. The wall did not suffer any earthquake or tsunami-induced damage during the 2011 Off the Pacific Coast of Tohoku Earthquake, Japan. This paper deals with a case study on this tire retaining wall. Results from a series of field and laboratory investigations are first described. A numerical simulation is also described in which whole tires were used as protective layers in an embankment. The results show that confining effect of each tire, friction between the tires, ductility, and damping inherent in tires contribute towards the excellent performance of the retaining wall during the earthquake loading. The case study also reveals that it is the isolation mechanism due to the vibration absorption capability of rubber particles in tires, which could protect tire retaining structure during the earthquake. [ABSTRACT FROM AUTHOR]

Shikoku island, southwestern Japan lies in the western Nankai Trough and showcases along-strike segmentation of slow earthquake behavior. Whether the spatial variation of tremor behavior reflects the regional differences in structure/source properties and how much such differences can be recognized by the seismic signals themselves are two questions addressed in this paper. Taking advantage of advanced methods in recognizing and classifying signals using machine learning approaches, we attempt to answer them by conducting signal classification experiments in Shikoku. Based on the tremor catalog from 1 June 2014 to 31 March 2015, the tremors recorded in four different areas were treated as different classes and segmented into 60-s-long signals. The number of tremors in four different areas (A to D, from west to east) reached 15,000, 31,000, 10,000, and 16,000, respectively. To efficiently distinguish between tremors from different areas, we applied a k-nearest neighbor (k-NN) classifier with Fisher's class separability criteria to select the optimal feature subset. The resulting classification performance reached more than 90% at all 12 stations. We further designed a triangle test to select the features that can better represent the differences in source properties between areas. We found that the most efficient features were associated with (1) the number of peaks in the temporal evolution of discrete Fourier transforms and (2) the energy distribution in the autocorrelation function (ACF). To match the difference in behavior revealed by the ACF, the size of the tremor zone, which mainly controls how long the seismic energy lasts in a tremor episode, was determined to be largest in Area B and smallest in Area C. The heterogeneity of the asperities in a tremor zone, which may control how spiky the tremor signals developed over time, was determined to be strong in Areas B and C. Together with previously documented variations in slow earthquake behavior in the same area, we finally propose a conceptual model that provides a better understanding of the regional differences in the tremor source properties in Shikoku, Japan. [ABSTRACT FROM AUTHOR]

In this paper, we explore the Western Pacific subduction impact on the geodynamics of the Asian continent. The data on migration of slow strain and earthquakes from the Nankai, Japan and Kuril-Kamchatka segments of the Western Pacific subduction zone deep into mainland Asia are analyzed. The calculations performed on five profiles, crossing the Kuril Islands, the Japanese Archipelago and Sakhalin Island toward the Asian continent, have revealed the transverse migration of earthquakes from the Japan–Kuril-Kamchatka subduction zone. The velocities of hypocenter migration of M ≥ 4.5 earthquakes from the Kuril-Kamchatka Trench via northern and central Sakhalin vary from 6 to 17 km/year, on average, at different depths. The profiles crossing the islands of Hokkaido and Sakhalin show the M ≥ 4.earthquake migration from the Kuril and Japan trenches at velocities of 8–27 km/year. [ABSTRACT FROM AUTHOR]

A modified k-nearest-neighbors method is introduced that provides an efficient nonlinear estimate of the intensity of a point process (field) based on locations of events. The method is applied to perform a detailed statistical analysis of the spatial structure of the seismic intensity field (in the context of this paper, the term "intensity" refers to the intensity of the seismic field, and should not be mistaken for the intensity of ground motion caused by an earthquake). The proposed method requires neither a preliminarily delineated area nor a normalization procedure for the estimates. In contrast to many interpolation methods, estimates based on the proposed method are statistically justified. The "uncertainty relation" between the spatial smoothing effect and random errors is established for the method in an explicit form. A procedure for choosing the number of nearest neighbors k* that controls the effective radius of smoothing is suggested. The proposed method is applied to analyze the seismic intensity field in two seismogenic regions surrounding the Kuril Islands and Japan during the period from 1904 to 2011. Spots of increased seismic activity in these regions are localized by the new method, and some quantitative statistical characteristics of these spots are determined and discussed. [ABSTRACT FROM AUTHOR]

Rupture size is a fundamental earthquake source parameter that is challenging to infer independently from far‐field seismological observations. Here, we develop a novel observational constraint on source size based on the decay rate of wavefield coherence across a seismic array. For a given earthquake, waveform coherence decays with increasing interstation distance or, more precisely, with increasing projection difference defined as the difference between the takeoff vectors associated to the two stations projected along the rupture direction. We find that coherence generally falls off with projection difference faster for earthquakes of larger magnitudes. The magnitude dependence of the coherence decay rate can be explained by a finite source effect: larger source sizes cause larger differences of phase delays between waves arriving from different parts of the rupture at different stations, hence a stronger spatial variability of the wavefield, resulting in a breakdown of waveform coherence. Assuming a 1‐D Haskell's source model, the rupture size can be estimated from the coherence decay rate. We apply this method to USArray data of earthquakes in three subduction zones, the Sea of Okhotsk, South America and Japan. The source sizes inferred from the coherence decay patterns are consistent with scaling relations intermediate between width‐saturated L‐models and quasi equi‐dimensional rupture models. Our observation captures a unique pattern of array waveform coherence and demonstrates the potential of utilizing waveform coherence to study earthquake source parameters. Plain Language Summary: In this paper, we observe that for a given earthquake, the similarity of the waveform (coherence) recorded by stations in a far‐field array decays with increasing interstation distance. The waveform coherence generally falls off with interstation distance faster for earthquakes of larger magnitudes and therefore, larger rupture sizes. We develop a novel method using array waveform coherence to estimate the earthquake rupture size, and apply this method to USArray data of earthquakes in three geological regions: the Sea of Okhotsk, South America and Japan. The source sizes inferred from the coherence decay patterns are consistent with previous studies using different methods. According to our observation, there is a potential to infer earthquake rupture independently from far‐field observations by waveform coherence. Key Points: Waveform coherence decays as a function of interstation distance faster for large earthquakes (Mw > 7)The waveform coherence pattern is governed by earthquake source sizeRupture sizes inferred from waveform coherence satisfy empirical scaling relations [ABSTRACT FROM AUTHOR]

Since the appearance and evolution of earthquake ionospheric precursors are expected to show a nonlinear and complex behaviour, the use of nonlinear predictor models seems more appropriate. This paper proposes a new approach based on deep learning as a powerful tool for extracting the nonlinear patterns from a time series of ionospheric precursors. A Long Short-Term Memory (LSTM) network as a type of Recurrent Neural Network (RNN) was used to investigate 52 six-month time series, deduced from the three Swarm satellite (Alpha (A), Bravo (B) and Charlie (C)) measurements, including electron density (Ne), electron temperature (Te), magnetic scalar and vector (X, Y, Z) components, Slant and Vertical Total Electron Content (STEC and VTEC), for day and night periods around the time and location of a seismic event. This new approach was tested on a strong Mw = 7.1 earthquake in Japan on 13 February 2021, at 14:07:50 UTC by comparing the results with two implemented methods, i.e., Median and LSTM methods. Furthermore, clear anomalies are seen by a voting classification method 1, 6, 8, 13, 31 and 32 days before the earthquake. A comparison with atmospheric data investigation is further provided, supporting the lithosphere–atmosphere–ionosphere coupling (LAIC) mechanism as a suitable theory to explain the alteration of upper geolayers in the earthquake preparation phase. In other words, using multi-method and multi-precursor analysis applied to 52 time series and also to the orbit-by-orbit investigation, the observed anomalies on the previous day and up to 32 days before the event in normal solar and quiet geomagnetic conditions could be considered as a striking hint of the forthcoming Japan earthquake. [ABSTRACT FROM AUTHOR]

A new risk-based tsunami early warning method is developed using Random Forest (RF) with an extensive tsunami monitoring network (S-net) deployed off the Northeastern region of Japan. To consider a wide range of possible tsunami waves that may occur in the future, the RF model is developed using simulated 4000 tsunami wave time series at the S-net sensors. The response variable is the total aggregate loss of buildings caused by the tsunami, and the explanatory variables include earthquake information (magnitude, epicenter latitude, and epicenter longitude) and tsunami wave amplitudes at the S-net sensors. Unlike the conventional tsunami early warning method for predicting the tsunami wave amplitude along the shoreline, the response variable adopted in this study is a tsunami risk metric that reflects the tsunami impact to people and assets in coastal areas. The RF model is suitable for predicting highly nonlinear and scattered tsunami loss due to its characteristics of nonparametric regression and ensemble learning. The result indicates that compared with conventional linear regression-based algorithms, RF predicts the tsunami loss significantly better, reducing the mean square error by 90%. Furthermore, RF does not rely heavily on earthquake information, making it useful for announcing early warning for tsunamis generated by other sources. • A new risk-based tsunami early warning algorithm is developed. • A random forest method is particularly suitable for forecasting highly nonlinear tsunami loss. • The random forest method outperforms a conventional method of multiple linear regression method. • A case study application considers an extensive ocean bottom sensor network in northeastern Japan (S-net). • The dataset and the MATLAB code are included in the paper. [ABSTRACT FROM AUTHOR]

• Curie point depth analysis by magnetic data reveals thermal structure of the Japanese islands. • The map of Curie depth was upgraded from the previous hand-writing map to a digital map. • Drillhole data provides local temperature data, while the map of Curie depth provides regional thermal structure. • It may be possible to estimate heat sources by combining Curie depths, drillhole data, and seismic data. This paper shows results of Curie point depth (CPD) analysis of the Japanese Islands using magnetic data. The analysis method is Bouligand et al. (2009) based on fractal theory. This analysis result was combined with the cutoff depth of ordinary earthquakes, low frequency earthquakes and drillhole data. CPD by magnetic data (Dcmag) performed in the case of 50 km window size ranges 6–9 km deep in volcanic areas and is deeper than 9 km in non-volcanic areas. The results are concordant with the previous study by Okubo et al. (1989). Cutoff depths calculated by JMA unified catalog are 10 km or shallower in the volcanic area, deeper than 20 km in the forearc side and concordant with Dcmags. Dcmags across the Kakkonda GP gradually become shallower from 6.8 km west Akita-Komagatake toward to 4.6 km near Iwate volcano. The depth-to-temperature curves from drillhole database around the Kakkonda GP show that thermal gradients between Iwate volcano and the Kakkonda GP range 0.157 - 0.222 K/m. Assuming that these thermal gradients continue from the bottom of hole to the deeper part, the depths reaching 580 °C (Dchole) are determined. Dchole is concordant with Dcmag in the west side of the Kakkonda GP and much shallower than Dcmag in the east side. This suggests that Dcmags may miss to delineate a local high temperature area smaller than 50 km square in the east because of the average effect of CPD analysis. Dchole is shallower than 3 km west Iwate volcano and deepens at about 5 km west the Kakkonda GP. Combining the low frequency earthquakes data from the JMA unified earthquake catalog and the drillhole data of the Kakkonda geothermal field, we find the Curie point depths are consistent with those data. [ABSTRACT FROM AUTHOR]

Due to the complexity of earthquakes, predicting their magnitude, timing and location is a challenging task because earthquakes do not show a specific pattern, which can lead to inaccurate predictions. But, using Artificial Intelligence-based models, they have been able to provide promising results. However, few mature studies are dealing with large earthquake prediction, especially as a regression problem. For these reasons, this paper investigates an attention-based LSTM network for predicting the time, magnitude, and location of an impending large earthquake. LSTMs are used to learn temporal relationships, and the attention mechanism extracts important patterns and information from input features. The Japan earthquake dataset from 1900 to October 2021 was used because it represents a highly seismically active region known for its large earthquakes. The results are examined using the metrics of MSE, RMSE, MAE, R-squared, and accuracy. The performance results of our proposed model are significantly better compared to other empirical scenarios and a selected baseline method, where we found that the MSE of our model is better by approximately 60%. • An attention-based LSTM network for earthquake prediction. • Clustering seismic regions for location prediction. • Time, magnitude, and location of earthquakes of magnitude >6 are predicted. • Two different seismic regions are studied: Japan and the northern Red Sea. • Discussion and comparison of results. [ABSTRACT FROM AUTHOR]

Topography results from the interplay between rock uplift and erosion, and studying river responses to active tectonics may enable the rates and spatial distribution of rock uplift to be inferred. Normalized channel steepness index (k sn) is widely used to infer rock uplift rates, and rock erodibility plays a significant role in determining the variation of channel steepness in response to external forces. Various measures may be applied to evaluate rock erodibility; however, since they are essentially qualitative, it is difficult to quantify relationships between rock mass strength and channel forms. In this study, we inferred long-term rock uplift rates along the Futagawa fault, an active normal-dextral fault, and the Idenokuchi fault, an active normal fault, in southwestern Japan, based on morphological characteristics of rivers across substrates with contrasting erodibility. The widely used detachment-limited model predicts that, when the rock uplift rate is spatially uniform, the ratio of erodibility of different rock types can be written by an inverse ratio of their k sn values. We employed this theoretical prediction to quantitatively calibrate the effects of rock mass strength on channel steepness. Within the study area, channel steepness was larger in the eastern part where volcanic rocks were exposed than in the western part where less resistant sedimentary rocks predominated. We calculated ratios of substrate erodibility using drainage basins close to the Futagawa fault that experienced similar rock uplift. Calibrated k sn exhibited a clear difference between the eastern and the western part of the fault, suggesting that contrasting erodibility alone cannot explain the observed channel steepness. We also confirmed that the observed channel steepness could not be explained without differential rock uplift, even when considering the combined effects of channel width, precipitation, and erosion process. The distribution of long-term rock uplift rates inferred from calibrated k sn and vertical slip rates of the fault compiled from published papers agree with the distribution of vertical displacement of the earthquake that occurred on the fault in 2016, suggesting that earthquakes similar to the 2016 event have occurred repeatedly in the study area. • We calibrated channel steepness based on the erodibility of substrate materials. • Spatial distribution of long-term uplift is inferred from bedrock river morphology. • Long-term uplift is similar to the vertical displacement of the 2016 earthquake. [ABSTRACT FROM AUTHOR]

In this paper, we propose a location-dependent earthquake prediction based on recurrent neural network algorithms. The location-dependent prediction consists of clustering the seismic dataset based on its geographical parameters (longitude and latitude) using the K-Means algorithm. In addition, we divide each cluster into two specific subsets. The first set consists of events with an average magnitude between 2 and 5, and the second set contains seismic events with magnitudes >5. Clustering allows our models to focus on each region independently. Associated with splitting methods, it helps them to learn the specific trends of each cluster with accurate performance. Moreover, in our work, the large earthquakes with few events are trained independently and remain unaffected by the other dominant events. We implement our models using the powerful recurrent neural network algorithms: Long Short Memory (LSTM), Gated Recurrent Network (GRU), and their hybrid model (LSTM-GRU). They are tested using data related to three different regions: Morocco, Japan, and Turkey. We evaluate the performance according to three metrics: mean absolute error, mean squared error, and root-mean-square error. Compared to other works in the literature, our models generally perform well, especially in predicting large earthquakes. - A location-dependent prediction using a geographic clustering by K-means algorithm. - Earthquake prediction using recurrent neural network algorithms: LSTM and GRU. - Three different seismic regions are studied: Japan, Turkey, and Morocco. - Discussion and comparison of results. [ABSTRACT FROM AUTHOR]

Anomalies in Schumann resonance (SR) (or global electromagnetic ELF (extremely low frequency) resonance due to global lightning activity in the Earth-ionosphere cavity) attract lot of attention in possible association with earthquakes (EQs). The first anomalies were observed in Japan for EQs in Taiwan when the distance between observatory and the EQ epicenter was a few Mm (1 Mm = 1000 km). Recently, Hayakawa et al. (2019, 2020a, b) have suggested a new SR anomaly related to the nearby (a few hundred km) EQs unlike the previous distant EQs. This paper presents the SR anomalies observed in the vicinity of Nagoya-city for two relatively close (∼1000 km) successive EQs of the magnitude around 7 that occurred offshore the Tohoku prefecture in Japan. The anomaly is characterized by the enhancement in amplitudes of SR modes, and this abnormal behavior was detected before and after each of the two EQs. Observational data were interpreted using the seismogenic perturbations of the lower ionospheric conductivity. Model computations imply the full wave solution of ELF electromagnetic problem in form of Riccati equation and the 2D telegraph equations. We show that observed anomalous alterations in the SR spectra might be attributed to two types of seismogenic modifications: the compression or the expansion of the vertical profiles of mesosphere conductivity over the EQ epicenter. • New type of SR anomaly was found for two nearby successive EQs of M~7 in Japan a few hundred kilometers away from observer. • The anomaly is observed as enhancements in intensity of all SR modes emerging prior and after each EQ. • Experimental data are interpreted by scattering at seismogenic ionospheric perturbation located above the EQ epicenter. • Model computations imply the full wave solution (Riccati equation) and use the 2D telegraph equations. • Seismogenic ionosphere modifications are used of compression or expansion type to match with the observational results. [ABSTRACT FROM AUTHOR]

The purpose of this paper is to discuss the lithosphere–atmosphere–ionosphere coupling (LAIC) effects with the use of multiparameter precursor observations for two successive Japanese earthquakes (EQs) (with a magnitude of around 7) in February and March 2021, respectively, considering a seemingly significant difference in seismological and geological hypocenter conditions for those EQs. The second March EQ is very similar to the famous 2011 Tohoku EQ in the sense that those EQs took place at the seabed of the subducting plate, while the first February EQ happened within the subducting plate, not at the seabed. Multiparameter observation is a powerful tool for the study of the LAIC process, and we studied the following observables over a 3-month period (January to March): (i) ULF data (lithospheric radiation and ULF depression phenomenon); (ii) ULF/ELF atmospheric electromagnetic radiation; (iii) atmospheric gravity wave (AGW) activity in the stratosphere, extracted from satellite temperature data; (iv) subionospheric VLF/LF propagation data; and (v) GPS TECs (total electron contents). In contrast to our initial expectation of different responses of anomalies to the two EQs, we found no such conspicuous differences of electromagnetic anomalies between the two EQs, but showed quite similar anomaly responses for the two EQs. It is definite that atmospheric ULF/ELF radiation and ULF depression as lower ionospheric perturbation are most likely signatures of precursors to both EQs, and most importantly, all electromagnetic anomalies are concentrated in the period of about 1 week–9 days before the EQ to the EQ day. There seems to exist a chain of LAIC process (cause-and-effect relationship) for the first EQ, while all of the observed anomalies seem to occur nearly synchronously in time for the send EQ. Even though we tried to discuss possible LAIC channels, we cannot come to any definite conclusion about which coupling channel is plausible for each EQ. [ABSTRACT FROM AUTHOR]