We analyzed 49,592 teleseismic receiver functions recorded by 278 CEArray stations to image the mantle transition zone (MTZ) beneath the South China Block to understand origins of deep velocity anomalies and their potential links to subduction and intraplate volcanism. We employed a fast-marching method and a high-resolution 3-D velocity model (FWEA18) derived from full waveform inversion in computing P-to-S conversion times to better image the 410-km and 660-km discontinuities. Our results indicate that the common-conversion-point stacking of receiver functions using 3-D conversion times yielded better migration images of the two discontinuities. The images revealed a slightly depressed 410-km with a few small uplifted patches, and showed that the 660-km beneath the western Yangtze Craton is depressed by 10-25 km, which is likely caused by the stagnant Paleo-Pacific slab. The 660-km beneath the southern Cathaysia Block has a 5-15 km high plateau with a topographic low at its central part. The lateral dimension of the topographic low is ~150 km and located beneath the central Pearl River Mount Basin near Hong Kong. We speculate the topographic low occurs within the Hainan plume with a temperature excess of ~300-400 K and is caused by the garnet phase transition. The displaced deep plume enters the MTZ and spreads nearly horizontally at the base. The plume evolves into two channels with a minor one toward the northeast and a major one toward the southwest, which keep moving upward to the 410-km. The southwest channel is likely the source that feeds the Hainan volcanoes.

Reverse Time Migration (RTM) Surface Offset Gathers (SOGs) are demonstrated to deliver more superior residual dip information than ray-based approaches. It appears more powerful in complex geological settings, such as salt areas. Still, the computational cost of constructing RTM SOGs is a big challenge in applying it to 3D field data. To tackle this challenge, we propose a novel method using dips of local events as a guide for RTM gather interpolation. The residual-dip information of the SOGs is created by connecting local events from depth-domain to time-domain via ray tracing. The proposed method is validated by a synthetic experiment and a field example. It mitigates the computational cost by an order of magnitude while producing comparable results as fully computed RTM SOGs.

Monitoring of in-situ, stress-induced, seismic velocity change provides an increasingly important contribution to the study of the earthquake nucleation process. Continuous Active-Source Seismic Monitoring (CASSM) with borehole sources and sensors has proven to be a very effective tool to monitor seismic velocity and to identify its temporal variations at depth. Since June 2017, we have been operating a crosswell CASSM field experiment at the San Andreas Fault Observatory at Depth (SAFOD) where a previous CASSM experiment identified the two seismic velocity reductions approximately 10 and 2 hours before micro-earthquakes. The ultimate goal of our experiment is to continuously monitor tectonic stress for the San Andreas Fault near seismogenic depth. Our active-source experiment makes use of two boreholes drilled at the SAFOD project site. A piezoelectric source and a three-component accelerometer have been installed in the SAFOD pilot and main holes, respectively, at about 1 km depth. A seismic pulse is generated by the piezoelectric source four times per second, and waveforms are recorded with a 48 kHz sample rate, with recordings summed for 1 to 10 minutes to capture seismic velocity changes at a high-temporal resolution. Since deployment in June 2017, and as of July, 2019, local seismicity has not been above our current threshold of detection. However, we have identified a velocity reduction at the SAFOD site (0.5 microsecond change in crosswell travel time, measured in a coda window) possibly induced by dynamic stress changes from the distant 6 July 2019 M 7.1 Ridgecrest earthquake, California. We will characterize and report the co-seismic change and post-seismic recovery process for this remotely triggered velocity change. We will also report on the overall status of this unique CASSM experiment.

We develop an automatic workflow for denoising the fundamental mode surface wave from ambient noise cross correlations (ANCs) calculated for a dense linear array. The surface wave signal traveling between each station pair is first enhanced through three-station interferometry. Then, phase travel times at different periods are determined in the frequency domain. The proposed array-based method is applied to a 1.6-km-long dense linear nodal array crossing surface traces of the San Jacinto fault near Anza, California. Surface wave signals in ANCs of the nodal array are significantly enhanced after denoising, particularly at high frequencies (> 2 Hz). Phase travel times are extracted reliably in the period ranges of 0.3-1.3 s and 0.3-1.6 s for Rayleigh and Love waves, respectively. The corresponding period-dependent phase velocity profiles derived from the eikonal equation reveal high-resolution details of fault zone internal structures beneath the array. A broad (500-1000 m) low-velocity zone that narrows with increasing period is observed, illuminating a flower-shaped structure of the San Jacinto fault damage zone.

Machine learning algorithm is applied to shear wave velocity (Vs) inversion in surface wave tomography, where a set of 1-D Vs profiles and the corresponding synthetic dispersion curves are used in network training. Previous studies showed that performances of a trained network depend on the input training dataset with limited diversity and therefore lack generalizability. Here, we present an improved semi-supervised algorithm-based network that takes both model-generated and observed surface wave dispersion data in the training process. The algorithm is termed Wasserstein cycle-consistent generative adversarial networks (Wcycle-GAN). Different from conventional supervised approaches, the GAN architecture extracts feature from the observed surface wave dispersion data that can compensate the limited diversity of the training dataset generated synthetically. The cycle-consistency enforces the reconstruction ability of input data from predicted model using a separate data generating network, while Wasserstein metric provides improved training stability and enhanced spatial smoothness of the output Vs model. We demonstrate improvements by applying the Wcycle-GAN method to 4076 pairs of fundamental mode Rayleigh wave phase and group velocity dispersion curves obtained in Southern California. The final 3-D Vs model from the best trained network shows large-scale features that are consistent with the surface geology. Our Vs model has smaller data misfits, yields better spatial smoothing, and provides sharper images of structures near faults in the top 15 km, suggesting the proposed Wcycle-GAN algorithm has stronger training stability and generalization abilities compared to conventional machine learning methods.

Current machine learning based shear wave velocity (Vs) inversion using surface wave dispersion measurements utilizes synthetic dispersion curves calculated from existing 3-D velocity models as training datasets. It is shown in the previous studies that the performances of the resulting networks are dependent on the diversity of the training data. We present an improved semi-supervised deep learning algorithm-based method that incorporates both observed and synthetic surface wave dispersion curves in the network training process. The algorithm is termed Wasserstein cycle-consistent generative adversarial networks (Wcycle-GAN), which combines the architecture of cycle-consistent GAN with Wasserstein loss metrics in optimization. Different from conventional supervised deep learning approaches, the GAN architecture also extracts structural information from the observed surface wave dispersion data in the training process that may improve generalization of the resulting network. The cycle-consistent loss addresses soft constraints on the trained neural networks to be reversible and thus reduces the variance of the trained networks. The Wasserstein metric provides weaker topology for convergence and improves spatial continuity of the predicted shear velocity (Vs) models. We demonstrate these improvements by applying the Wcycle-GAN to 4066 fundamental mode Rayleigh wave phase and group dispersion curves obtained in Southern California (SC). In general, the 3-D Vs model predicted by the best training Wcycle-GAN is consistent with previous surface wave tomography studies of SC in the overlapping area, but with smaller data misfit, yields better spatial smoothing, and provides improved images of structures near faults and in the top 5 km. Our results indicate that the proposed Wcycle-GAN algorithm has strong training stability and generalization abilities.