Differentiating between natural earthquakes and quarry blasts is a labor-intensive task. Recent studies show that artificial neural network (ANN) can achieve good accuracy for these two types of seismic events based on a variety of engineered waveform features. Here, we use convolutional neural network (CNN) to achieve automatic classification with waveform inputs. The training and test datasets consist of 25100 blast recordings from 2010 to 2020 and 31542 earthquake recordings from 2019 to 2020, recorded by the Southern California Seismic Network (SCSN). We input the normalized waveforms to the CNN model and apply the grid search method to determine the optimal model configurations and hyperparameters. Our preliminary results reveal that the CNN can differentiate between these two event types. Next, we will focus on enhancing the accuracy of the CNN model and compare it with that of other methods. Updated results will be presented in the meeting.

Earthquakes generated from different mechanisms could have different rupture processes and thus potentially different moment release histories. While the generation of deep earthquakes (origin depth 100-700 km) is supposed to be different from shallow ones (0-100 km), it remains enigmatic whether intermediate-depth (100~300 km) and deep-focus earthquakes (400~700 km) have different mechanisms. Here, we train machine learning models to test the possibility of distinguishing shallow, intermediate-depth, and deep-focus earthquakes with their source time functions (STFs). Our results show that deep and shallow earthquakes are distinguishable with STFs at an accuracy of 87%±1.1%, and intermediate-depth and deep-focus earthquakes are distinguishable at an accuracy of 68%±2.7%. We find that two STF features, i.e., scaled duration and scaled variance, are critical in both classification tasks, which is consistent with previous findings that rupture duration of same-size earthquakes is generally shorter at deeper depth. Our results indicate that at least part of intermediate-depth and deep-focus earthquakes have different mechanisms; the indistinguishable portion of events could either have the same mechanisms, or have different mechanisms but no recognizable STF differences.

Central Taiwan is a seismic active region that also experiences landslides and typhoons. 5 devastating earthquakes, the 2009 M = 6.2 Nantou, 2010 M = 6.4 Jiashian, 2013 M = 6.2 Nantou in March, 2013 M = 6.5 Nantou in June and 2016 M = 6.6 Meinong occurred in central mountainous area after (3 - 78 months) the 2009 Typhoon Morokat hit the same area. The identified close proximity in space and time between the 2009 Typhoon Morokat and earthquakes suggest the occurrence of cascading hazards. Triggering relations between typhoor or landslide induced surface erosion and earthquakes remain poorly understood. We propose to test the hypothesis that typhoon or landslide induced surface erosion resulted in surface unloading which may promote the rupture of earthquakes. We also probe the contribution of other sources of erosions, including interseismic erosion and large earthquake induced erosion to earthquake triggering. In this study, we present a comprehensive analysis of stress perturbations induced by four sources of erosional unloading (background, a large earthquake, and two typhoons). Our results indicate that surface erosion promoted the initiation and rupture propagation of the 2016 Meinong event, rupture propagation of the 1999 Chi-Chi mainshock, and future failure on the Chukou fault, Chelungpu fault, and flat decollement of Central Taiwan at ~ 4 – 15 km in depth. The cumulative stress perturbations due to surface erosion suggest a dominant role of background erosion with an order of 0.1 bar and secondary effect from earthquakes or typhoons with an order of 0.01 bar. Our study demonstrates a prominent earthquake triggering mechanism related to short-term background, large earthquakes, or wet typhoons induced surface erosion.

Delineation of physical factors that contribute to earthquake triggering is a challenging issue in seismology. We analyze hydrological modulation of seismicity in Taiwan using groundwater level data and GNSS time series. In western Taiwan, the seismicity rate reaches peak levels in February-April and drops to its lowest values in July-September, exhibiting a direct correlation with annual water unloading. The elastic hydrological load cycle may be the primary driving mechanism for the observed synchronized modulation of earthquakes, as also evidenced by deep earthquakes in eastern Taiwan. However, shallow earthquakes in eastern Taiwan (<18 km) are anti-correlated with water unloading, which is not well explained by either hydrological loading, fluid transport or pore pressure changes, and suggests other time-dependent processes. The moderate correlation between stacked monthly trends of large historic earthquakes and present-day seismicity implies a modestly higher seismic hazard during the time of low annual hydrological loading. Our analysis of interannual time series of earthquakes and water loading in 1994-2018 also shows more frequent small earthquakes in the dry years.

Coseismic landslides represent a major cascading hazard associated with high-magnitude earthquakes in mountainous regions. Although recent evidence suggests that landslide hazard can persist and even increase in the years immediately after an earthquake, our understanding of how patterns of landsliding and specifically how runout from rainfall-triggered reworking via debris flows, evolves through time and space remains limited. To address this, we established a systematic multi-temporal landslide inventory covering the full rupture area of the 2015 M_w 7.8 Gorkha, Nepal earthquake, extending across pre-, co- and post-seismic periods (2014-2020). We map landslides bi-annually from satellite imagery to record changes to existing landslides and new failures. We use the mapped landslides as source locations for a spatially distributed empirical runout model, Flow-R, that routes debris downslope simulating plausible ‘worst case’ scenarios. Our results show that even 4.5 years after the earthquake, the number and area of mapped landslides remains higher than on the day of the earthquake. Whilst coseismic landslides persist through time, we also document spatial shifts in the density of landsliding that develop from new but dispersed post-seismic landslides. Comparing modelled runout predictions to our sequential mapping of landslides, we find that 14% (>80 km²) of the total modelled coseismic runout hazard area is realised within 4.5 years of the earthquake, and so represents a portion of post-earthquake landslide hazard that can be foreseen. More broadly, we demonstrate that the persistence of coseismic landslides combined with the occurrence of new post-seismic landslides means that the potential runout hazard also remains well above coseismic levels by the end of the time series. These findings are particularly important for informing a comprehensive and precautionary approach to building into earthquake preparedness planning the likely nature of post-earthquake landslide hazard and risk.

Detailed image of subsurface structure is needed to understand the dynamics of tectonic plates and the processes that occur in the Earth's interior such as convection currents. Seismic tomographic techniques can be used to produce images that describe the structure of the 3D seismic wave velocity in the Earth's mantle globally. The phase data such as P, pwP, pP, S, and sP can be used to image the structure of crust, lithosphere, and Earth's mantle. One of the complex subduction zones resulting from plate tectonic dynamics is the Banda Arc in the eastern part of Indonesia. The Banda Arc is the result of the collision between the volcanic arc and the Australian continent. The Banda Arc region is marked by active seismicity in the eastern part of Indonesia.  This study used the catalog data of the International Seismological Center-Engdahl, van der Hilst, Buland (ISC-EHB) from 1964 to 2015. During this period, 5,243 earthquake events were obtained from 314 global seismic stations with a total of 19.533 phases ( P (16.941 phases), pP (30 phases), pwP(26 phases), S (2.516 phases), and SP (30 phases)). The preliminary result of our travel time tomographic imaging provides a good description of subsurface structure in the crust and mantle down to a depth of 650 km beneath the Banda Arc. More detailed subsurface structure information is being used as input for mapping disaster-prone areas and will be presented during the meeting.

Earthquakes in oceanic subduction zones are closely related to dehydration of subducting slabs. Here we present an integrated study on retrograded eclogites in Yezhai village of the Dabie ultra-high pressure (UHP) metamorphic belt, China. The eclogites have experienced amphibolite facies retrograde metamorphism and partial melting, and developed recumbent folds, tectonic boudins and fault breccias. Quartz veins were accumulated at the fold hinge zone, in which migmatitic eclogites appear as breccias. Zircons from host migmatitic eclogites, eclogite breccias and quartz veins are euhedral and have a core-rim structure. Their rare earth element (REE) distribution is characterized by light REE depletion, heavy REE enrichment, positive Ce anomaly and negative Eu anomaly. The U-Pb ages of 755–722 Ma for magmatic core of zircons indicate their origin of Neoproterozoic protoliths, whereas the ages of 242–221 Ma of metamorphic zircon rims from quartz veins yield timing of fluid flux and earthquakes. The Ti-in-zircon thermometer suggest equilibrium temperature from ~763 ° C and 814 ° C. In addition, EBSD analyses reveal that quartz in both eclogite migmatite and quartz veins show prism <c> slip and prism <a> slip, representing deformation temperature at 650–750℃ and 400–450℃. After high-temperature deformation, rapid healing of the fractures by crystallization of quartz veins will allow overlapping of mid-temperature ductile deformation in both migmatitic eclogites and quartz veins. Breakdown of phengite during exhumation of UHP metamorphic rocks could release water and facilitate partial melting and folding of eclogites. We propose that Si-rich fluids precipitated during melt crystallization and accumulated at the fold hinge, leading to a rapid increase in pore pressure. Therefore, the pulsed fluid flux of UHP rocks during exhumation can trigger shallow to intermediate-focus earthquakes in subduction channels. Such melting-associated hydraulic embrittlement process could also occur during subduction and dehydration of slabs.