Explore the vast range of titles available.

Given the recent developments in machine-learning technology, its application has rapidly progressed in various fields of earthquake seismology, achieving great success. Here, we review the recent advances, focusing on catalog development, seismicity analysis, ground-motion prediction, and crustal deformation analysis. First, we explore studies on the development of earthquake catalogs, including their elemental processes such as event detection/classification, arrival time picking, similar waveform searching, focal mechanism analysis, and paleoseismic record analysis. We then introduce studies related to earthquake risk evaluation and seismicity analysis. Additionally, we review studies on ground-motion prediction, which are categorized into four groups depending on whether the output is ground-motion intensity or ground-motion time series and the input is features (individual measurable properties) or time series. We discuss the effect of imbalanced ground-motion data on machine-learning models and the approaches taken to address the problem. Finally, we summarize the analysis of geodetic data related to crustal deformation, focusing on clustering analysis and detection of geodetic signals caused by seismic/aseismic phenomena. Graphical Abstract

The 1 January 2024 Noto-Hanto (Noto Peninsula) earthquake (M JMA 7.6) generated strong ground motion, large crustal deformation and tsunamis that caused significant damage in the region. Around Noto Peninsula, both offshore submarine and partially inland active faults have been identified by previous projects: Ministry of Land, Infrastructure, Transport and Tourism (MLIT) and Japan Sea Earthquake and Tsunami Research Project (JSPJ). We inverted the tsunami waveforms recorded on 6 wave gauges and 12 tide gauges around Sea of Japan and the GNSS data recorded at 53 stations in Noto Peninsula to estimate the slip amount and seismic moment on each of active faults. The results show that the 2024 coseismic slips were 3.5 m, 3.2 m, and 3.2 m on subfaults NT4, NT5 and NT6 of the JSPJ model, located on the northern coast of Noto Peninsula and dipping toward southeast. A smaller slip, 1.0 m, estimated on NT8 on the southwestern end of the 2024 rupture, may be attributed to its previous rupture during the 2007 Noto earthquake. The total length of these four faults is ~ 100 km, and the seismic moment is 1.90 × 10 20 Nm (Mw = 7.5). Almost no slip was estimated on the northeastern subfaults NT2 and NT3, which dip northwestward, opposite to NT4–NT5–NT6, and western subfault NT8. Aftershocks including the M JMA 6.1 event occurred in the NT2–NT3 region, but they are smaller than the potential magnitude (Mw 7.1) those faults can release in a tsunamigenic earthquake. Similar features are also found for the MLIT model; the 2024 slip was only on F43 along the northern coast of Noto Peninsula, and northeastern F42 did not rupture, leaving potential for future event. Graphical Abstract

Slow earthquakes occur at deep and shallow plate boundaries along the Nankai Trough. Deep slow earthquakes are continuously distributed along the 30–40 km depth contours of the upper surface of the subducted Philippine Sea Plate. In contrast, shallow slow earthquakes occur in limited regions: Hyuga-nada, off Cape Muroto, and southeast off the Kii Peninsula. This review provides an overview of the up-to-date seismological, geodetic, geological, and experimental results in the shallow Nankai area for a unified understanding of the spot-like occurrence of shallow slow earthquakes. Shallow slow earthquakes tend to be distributed in transitional regions between the frictionally locked and stably sliding zones on the plate boundary. Based on geological and experimental studies, the lithology of incoming sediments and their friction coefficients can be variable along the Nankai Trough. Laboratory friction experiments revealed that sediments under shallow plate boundary conditions often exhibit positive (a − b) values, while negative (a − b) is possible via several processes. Subducted seamounts create complex fracture networks and stress shadows in their surrounding areas; however, not all subducted seamounts are related to shallow slow earthquake activities. This incomplete correlation suggests that alternative factors are required to explain the spot-like distribution of shallow slow earthquakes in the Nankai subduction zone. High pore fluid pressure conditions around shallow slow earthquake zones were interpreted based on seismological structural studies. In addition, ambient noise monitoring revealed temporal changes in seismic velocity structures associated with shallow slow earthquake migrations. This result suggests a close link between pore fluid migration and shallow slow earthquake episodes. Because transient changes in pore fluid pressure can lead to various slip behaviors, the episodic migration of pore fluid around the plate boundary could promote shallow slow earthquake activity along the Nankai Trough.

The epidemic type aftershock sequence (ETAS) model is used as a baseline model both for earthquake clustering and earthquake prediction. In most forecast experiments, the ETAS parameters are estimated based on a short and local catalog, therefore the model parameter optimization carried out by means of a maximum likelihood estimation may be not as robust as expected. We use Bayesian forecast techniques to solve this problem, where non-informative flat prior distributions of the parameters is adopted to perform forecast experiments on 3 mainshocks occurred in Southern California. A Metropolis–Hastings algorithm is employed to sample the model parameters and earthquake events. We also show, through forecast experiments, how the Bayesian inference allows to obtain a probabilistic forecast, differently from one obtained via MLE. Graphical Abstract

We carried out seismic tomography study to reveal three-dimensional (3D) seismic velocity structures in the Noto peninsula, Japan, where swarm-like seismic activity started in December 2020. The obtained results reveal a highly heterogeneous structure in the crust. The most striking feature is the existence of a low-velocity anomaly in the lower crust beneath the Noto earthquake swarm. Although the data set used in this study cannot resolve the upper mantle structure, previous regional tomographic studies suggest that a low-velocity anomaly exists at depths of 50–150 km around the Noto peninsula that is probably interpreted as a fluid-rich region. We infer that fluids have been supplied from the uppermost mantle to the lower crust over a geological time scale and a large volume of fluids have accumulated below the seismogenic zone beneath the Noto peninsula. A further upward migration of fluids to the upper crust, which may have suddenly started in December 2020, probably triggers numerous earthquakes at depths of 10–15 km. Since major active faults exist at shallower extensions of the hypocenters of the Noto earthquake swarm, we consider that the earthquake swarm occurs along pre-existing and weak fault planes. Dense temporary seismic observations will highlight a smaller-scale (5–10 km) 3D seismic velocity model and finer hypocenter distribution, which provide additional information for better understanding of the generation mechanisms of the Noto earthquake swarm.

A combination of geological evidence (in the form of hydrothermal vein systems in exhumed fault systems) and geophysical information around active faults supports the localized invasion of near-lithostatically overpressured aqueous fluids into lower portions of the crustal seismogenic zone which commonly extends to depths between 10 and 20 km. This is especially the case for compressional–transpressional tectonic regimes which, beside leading to crustal thickening and dewatering through prograde metamorphism, are also better at containing overpressure and are ‘load-strengthening’ (mean stress rising with increasing shear stress), the most extreme examples being associated with areas undergoing active compressional inversion where existing faults are poorly oriented for reactivation. In these circumstances, ‘fault-valve’ action from ascending overpressured fluids is likely to be widespread with fault failure dual-driven by a combination of rising fluid pressure in the lower seismogenic zone lowering fault frictional strength, as well as rising shear stress. Localized fluid overpressuring nucleates ruptures at particular sites, but ruptures on large existing faults may extend well beyond the regions of intense overpressure. Postfailure, enhanced fracture along fault rupture zones promotes fluid discharge through the aftershock period, increasing fault frictional strength before hydrothermal sealing occurs and overpressures begin to reaccumulate. The association of rupture nucleation sites with local concentrations of fluid overpressure is consistent with selective invasion of overpressured fluid into the roots of major fault zones and with observed non-uniform spacing of major hydrothermal vein systems along exhumed brittle–ductile shear zones. A range of seismological observations in compressional–transpressional settings are compatible with this hypothesis. There is a tendency for large crustal earthquakes to be associated with extensive (L ~ 100–200 km) low-velocity zones in the lower seismogenic crust, with more local Vp/Vs anomalies (L ~ 10–30 km) associated with rupture nucleation sites. In some instances, these low-velocity zones also exhibit high electrical conductivity. Systematic, rigorous evaluation is needed to test how widespread these associations are in different tectonic settings, and to see whether they exhibit time-dependent behaviour before and after major earthquake ruptures.

Seismic phase picking is a fundamental task in seismology that is crucial for event detection and earthquake cataloging; however, manual analysis is impractical given the scale of modern seismic networks. We present SegPhase, a novel seismic arrival time picking model designed to efficiently process large-scale seismic data recorded by dense seismic networks in Japan. In contrast to conventional convolution-based models, SegPhase employs a hierarchical vision transformer structure that utilizes multi-head self-attention to dynamically focus on important waveform features, such as P- and S-wave onsets, noise, and coda waves. Compared to PhaseNet, the most widely used deep learning model, SegPhase improved arrival time match rates by ~ 11% and detected ~ 15% more events in continuous waveform tests, particularly enhancing the detection of small-magnitude events. Benchmark evaluations demonstrated that SegPhase achieved high classification performance in identifying P- and S-waves. We also examined the threshold of the output probability values when applying SegPhase to continuous waveforms for which the optimal threshold was unknown. By lowering the threshold to 0.1, we observed an increase in the number of detected events without noticeable changes in the hypocenter location error and observed–calculated discrepancies. This was achieved by more effectively utilizing high-probability picks, which further improved phase association. Based on these results, we recommend a threshold of 0.1 to enhance event detection while maintaining accurate arrival times. Our findings demonstrate that SegPhase enables robust arrival picking across diverse datasets and supports high-resolution seismic monitoring. Graphical Abstract

In volcanic regions, active earthquake swarms often occur in association with volcanic activity, and their rapid detection and analysis are crucial for volcano disaster prevention. Currently, these processes are ultimately left to human judgment and require significant time and money, making detailed real-time verification impossible. To overcome this issue, we attempted to apply machine learning, which has been successfully applied to various seismological fields to date. For seismic phase pick, several models have already been trained using a large amount of training data (mainly crustal earthquakes). Although there are some cases in which these models can be applied without any problems, regional dependence on pre-trained models has been reported. Since this study targets earthquakes in a volcanic region, applying existing pre-trained models may be difficult. Therefore, in this study, we compared three models; the publicly available trained model (model 0), a model which was trained with approximately 220,000 P- and S-wave onset reading data recorded at the Hakone volcano from 1999 to 2020 with initialized parameters (model 1) using the same architecture, and a model fine-tuned with the aforementioned Hakone data using the parameters of model 0 as initial values (model 2), and evaluated their phase identification performance for the Hakone data. As a result, the seismic phase detection rates of models 1 and 2 were much higher than those of model 0. However, small-amplitude signals are often missed when multiple seismic events occur within a detection time window. Therefore, we created training data with two earthquakes in the same time window, retrained the model using the data, and successfully detected events that previously would have been missed. In addition, it was found that more events were detected by setting the threshold to a low probability value for detection, increasing the number of seismic phase detections, and filtering by phase association and hypocenter location.

P-wave first-motion polarity is the most useful information in determining the focal mechanisms of earthquakes, particularly for smaller earthquakes. Algorithms have been developed to automatically determine P-wave first-motion polarity, but the performance level of the conventional algorithms remains lower than that of human experts. In this study, we develop a model of the convolutional neural networks (CNNs) to determine the P-wave first-motion polarity of observed seismic waveforms under the condition that P-wave arrival times determined by human experts are known in advance. In training and testing the CNN model, we use about 130 thousand 250 Hz and about 40 thousand 100 Hz waveform data observed in the San-in and the northern Kinki regions, western Japan, where three to four times larger number of waveform data were obtained in the former region than in the latter. First, we train the CNN models using 250 Hz and 100 Hz waveform data, respectively, from both regions. The accuracies of the CNN models are 97.9% for the 250 Hz data and 95.4% for the 100 Hz data. Next, to examine the regional dependence, we divide the waveform data sets according to the observation region, and then we train new CNN models with the data from one region and test them using the data from the other region. We find that the accuracy is generally high (\\[ \\] 95%) and the regional dependence is within about 2%. This suggests that there is almost no need to retrain the CNN model by regions. We also find that the accuracy is significantly lower when the number of training data is less than 10 thousand, and that the performance of the CNN models is a few percentage points higher when using 250 Hz data compared to 100 Hz data. Distribution maps, on which polarities determined by human experts and the CNN models are plotted, suggest that the performance of the CNN models is better than that of human experts.

We have been discussing the validity of using the horizontal-to-vertical spectral ratios (HVRs) as a substitute for S-wave amplifications after Nakamura first proposed the idea in 1989. So far a formula for HVRs had not been derived that fully utilized their physical characteristics until a recent proposal based on the diffuse field concept. There is another source of confusion that comes from the mixed use of HVRs from earthquake and microtremors, although their wave fields are hardly the same. In this study, we compared HVRs from observed microtremors (MHVR) and those from observed earthquake motions (EHVR) at one hundred K-NET and KiK-net stations. We found that MHVR and EHVR share similarities, especially until their first peak frequency, but have significant differences in the higher frequency range. This is because microtremors mainly consist of surface waves so that peaks associated with higher modes would not be prominent, while seismic motions mainly consist of upwardly propagating plain body waves so that higher mode resonances can be seen in high frequency. We defined here the spectral amplitude ratio between them as EMR and calculated their average. We categorize all the sites into five bins by their fundamental peak frequencies in MHVR. Once we obtained EMRs for five categories, we back-calculated EHVRs from MHVRs, which we call pseudo-EHVRs (pEHVR). We found that pEHVR is much closer to EHVR than MHVR. Then we use our inversion code to invert the one-dimensional S-wave velocity structures from EHVRs based on the diffuse field concept. We also applied the same code to pEHVRs and MHVRs for comparison. We found that pEHVRs yield velocity structures much closer to those by EHVRs than those by MHVRs. This is natural since what we have done up to here is circular except for the average operation in EMRs. Finally, we showed independent examples of data not used in the EMR calculation, where better ground structures were successfully identified from pEHVRs again. Thus we proposed here a simple empirical method to estimate S-wave velocity structures using single-station microtremor records, which is the most cost-effective method to characterize the site effects.