The establishment of the High Sensitivity Seismograph Network (Hi-net) in Japan has led to the discovery of deep low-frequency tremors. Since such tremors are considered to be related to large earthquakes adjacent to tremors on the same subducting plate interface, it is important in seismology to investigate tremors before establishing modern seismograph networks that record seismic data digitally. We propose a deep learning method to detect evidence of tremors from seismogram images recorded on paper more than 50 years ago. In our previous study, we constructed a convolutional neural network (CNN) based on the Residual Network (ResNet) structure and verified its performance through learning with synthetic images generated based on past seismograms. In this study, we trained the CNN with seismogram images converted from real seismic data recorded by Hi-net. The CNN trained by fine-tuning achieved an accuracy of 98.64% for determining whether an input image contains tremors. The Gradient-weighted Class Activation Mapping (Grad-CAM) heatmaps to visualize model predictions indicate that the CNN successfully detects tremors without affections of a variety of noises, such as teleseisms. The trained CNN was applied to the past seismograms recorded at the Kumano observatory, Japan, operated by Earthquake Research Institute, The University of Tokyo. The CNN shows the potential to detect tremors from past seismogram images for broader applications, such as publishing a new tremor catalog.

Slow earthquakes are mainly distributed in regions surrounding seismogenic zones along the plate boundaries of subduction zones. In the Central American subduction zone, large regular interplate earthquakes with magnitudes of 7–8 occur repeatedly around the Nicoya Peninsula, in Costa Rica, and a tsunami earthquake occurred off Nicaragua, just north of Costa Rica, in 1992. To clarify the spatial distribution of various slip behaviors at the plate boundary, we detected and located very low frequency earthquakes (VLFEs) around the Nicoya Peninsula using a grid-search matched-filter technique with synthetic templates based on a regional three-dimensional model. VLFEs were active in September 2004 and August 2005, mainly near the trench axis, updip of the seismogenic zone. The distribution of VLFEs overlapped with large slip areas of slow slip events. Low frequency tremor signals were also found in high-frequency seismogram envelopes within the same time windows as detected VLFEs; thus, we also investigated the energy rates of tremors accompanied by VLFEs. The range of scaled energy, which is the ratio of the seismic energy rate of a tremor to the seismic moment rate of accompanying VLFE and related to the rupture process of seismic phenomena, was 10-9–10-8. The along-dip separation of shallow slow and large earthquakes and the range of the scaled energy off Costa Rica are similar to those in shallow slow earthquakes in Nankai, which shares a similar thermal structure along the shallow plate boundary.

Slow earthquakes are mainly distributed in the vicinity of seismogenic zones of megathrust earthquakes and relationships between both types of earthquakes are expected. We examined the activity of very low frequency earthquakes (VLFEs), classified as one type of slow earthquake, around Japan because they have the potential to clarify detailed spatiotemporal slip behaviors at the plate boundaries. The distribution of the shallow VLFE activity rate is heterogeneous along trench axes and exhibits an anticorrelation relationship with the spatial distribution of the interplate coupling ratio, whereas deep VLFEs are distributed only in weakly coupled areas and the spatial variation of the activity rate is small. Furthermore, VLFEs are mainly hosted by low seismic velocity anomalies. Thus, slow earthquakes can be triggered by decreased effective stress due to the high pore fluid pressure within regions with weak interplate coupling and their activity can be an indicator of interplate slip behavior.

Cross-correlation analysis was applied to long-term onshore broadband records from April 2004 to March 2021 to detect and relocate shallow very low frequency earthquakes (VLFEs) southeast off the Kii Peninsula, along the Nankai Trough, Japan. We then determined the moment rate functions of detected shallow VLFEs using the Monte Carlo-based simulated annealing method. According to this new comprehensive catalog, shallow VLFEs are widespread beneath the accretionary prism toe, but shallow VLFEs with large cumulative moments are localized around the western edge of the paleo-Zenisu ridge, which is subducted beneath southeast off the Kii Peninsula. Our results from long-term shallow VLFE catalog are well consistent with previous studies in this region, suggesting that heterogeneous structures and stress conditions due to the subducted paleo-Zenisu ridge promote the occurrence of shallow slow earthquakes. The relocated shallow VLFE epicenters illustrated three major episodes characterized by a similar activity area and five minor episodes characterized by different areas. The three major episodes exhibited slow frontal migration with different initiation locations, directions, and speeds, as well as several rapid reverse migrations. Episodes of minor activity were distributed in different locations within part of the area of major activity. Different patterns of shallow VLFE migration could reflect temporal changes in the pore-fluid distribution or stress conditions of the plate boundary.

We constructed a catalog of volcanic deep low-frequency (DLF) earthquakes across 52 regions in Japan to investigate their seismicity based on three analyses: relocation, classification, and detection. Relocation and classification analyses are based on waveform correlation, and detection analysis is conducted using the matched filter technique. We found that DLF earthquakes in many regions are spatially clustered at approximately 5 km intervals between the lower limit of crustal earthquakes and Moho discontinuity. Based on temporal seismicity patterns, DLF earthquake groups in each region can be classified into episodic and non-episodic types. Episodic groups consist of seismic swarms and quiescence. In some episodic groups, DLF earthquakes have constant recurrence interevent times or increasing interevent times as a function of time. Swarms of DLF earthquakes sometimes are associated with volcanic activity at the surface, such as eruptions or crustal deformations in some regions. The spatiotemporal characteristics of DLF earthquake groups may be linked to the movement of magmatic fluids. The discrete vertical separation of DLF earthquake groups may reflect small-scale heterogeneities, such as injected magma. Periodic activity patterns may be caused by volcanic mechanisms, such as second boiling. The variety of DLF earthquake patterns may suggest that multiple mechanisms may trigger DLF earthquakes, such as complex underground structures and volcanic processes, rather than a single mechanism.

Low-frequency earthquakes are peculiar energy-release events mostly occurring at the transition between the seismogenic and the freely creeping zones of a subducting slab. The source characterization of these events is of fundamental importance to understand physical processes that govern the slow out of equilibrium evolution of the subduction interface that may lead to the generation of large, destructive earthquakes. Nevertheless, their source mechanisms still remain unclear. Here, we estimate the source parameters of ~23,000 low-frequency earthquakes continuously detected from 2013 to 2015 in Shikoku, Japan. We show that a cubic moment-duration scaling characterizes these events, suggesting a self-similar process as for regular earthquakes. However, their high-frequency fall-off suggests an omega-cube decay in contrast to the omega-squared model of earthquakes. Source characteristics do not change when low-frequency earthquake bursts occur during the analyzed three years. On the other hand, we observe a coherent along-strike variation of the product of stress drop and the cube of rupture velocity, possibly related to a weaker behavior of tremor patches in central Shikoku. Secondary microseismic noise and network-dependent completeness magnitude lead to missing event detections that do not allow discriminating between Gutenberg-Richter event size distribution and any deviation from it. Our findings suggest that the same observational limits might affect worldwide detection of low-frequency earthquakes.

Slow earthquakes are generally distributed in regions surrounding seismogenic zones along the plate boundaries of subduction zones. In the Costa Rica subduction zone, large regular interplate earthquakes with a magnitude of 7–8 occur repeatedly, and a tsunami earthquake occurred in the northern part in 1992. To clarify the spatial distribution of various slip behaviors at the plate boundary in the Costa Rica subduction zone, we detected and located very low frequency earthquakes (VLFEs) using a grid-search matched-filter technique with synthetic templates based on a regional three-dimensional model. VLFEs were activated in September 2004 and August 2005, and most of the VLFEs were located near the trench axis at a depth range of 5–10 km, the updip of the seismogenic zone. The spatial distribution of VLFEs complements the slip areas of large earthquakes and the tsunami earthquake. Low frequency tremor signals were also found in high-frequency seismogram envelopes within the same time windows of detected VLFEs; thus, we also investigated the energy rates of tremors accompanied by VLFEs. The range of scaled energy, which is the ratio of the seismic energy rate of a tremor to the seismic moment rate of accompanying VLFE, was 10-9–10-8. This value is similar to that in shallow slow earthquakes in the Nankai subduction zone. The similarity of characteristics and distribution of shallow slow earthquakes in the Costa Rica and Nankai subduction zones may be due to common tectonic features, such as age, temperature, or the presence of accretionary prisms.