Surface-wave seismograms are widely used by researchers to study Earth’s interior and earthquakes. To extract information reliably and robustly from a suite of surface waveforms, the signals require quality control screening to reduce artifacts from signal complexity and noise, a task typically completed by human analysts. This process has usually been done by experts labeling each waveform visually, which is time-consuming and tedious for large datasets. We explore automated approaches to improve the efficiency of waveform quality control processing by investigating logistic regression, support vector machines, k-nearest neighbors, random forests (RF), and artificial neural networks (ANN) algorithms. To speed up signal quality assessment, we trained these five machine learning methods using nearly 400,000 human-labeled waveforms. The ANN and RF models outperformed other algorithms and achieved a test accuracy of 92%. We evaluated these two best-performing models using seismic events from geographic regions not used for training. The results show the two trained models agree with labels from human analysts but required only 0.4% time. Although the quality assignments assessed general waveform signal-to-noise, the ANN or RF labels can help facilitate detailed waveform analysis. Our analyses demonstrate the capability of the automated processing using these two machine learning models to reduce outliers in surface-wave-related measurements without human quality control screening.

The Eastern United States has a complex geological history and hosts several seismic active regions. We investigate the subsurface structure beneath the broader eastern United States. To produce reliable images of the subsurface, we simultaneously invert smoothed P-wave receiver functions, Rayleigh-wave phase and group velocity measurements, and Bouguer gravity observations for the 3D shear wave speed. Using surface-wave observations (3-250 s) and spatially smoothed receiver functions, our velocity models are robust, reliable, and rich in detail. The shear-wave velocity models fit all three types of observations well. The resulting velocity model for the eastern U.S. shows thinner crust beneath New England, the east coast, and the Mississippi Embayment. A relatively thicker crust was found beneath the stable North America craton. A relatively slower upper mantle was imaged beneath New England, the east coast, and western Mississippi Embayment. A comparison of crust thickness derived from our model against four recent published models shows first-order consistency. A relatively small upper mantle low-speed region correlates with a published P-waves analysis that has associated the anomaly with a 75 Ma kimberlite volcanic site in Kentucky. We also explored the relationship between the subsurface structure and seismicity in the eastern U.S. We found earthquakes often locate near regions with seismic velocity variations, but not universally. Not all regions of significant subsurface wave speed changes are loci of seismicity. A weak correlation between upper mantle shear velocity and earthquake focal mechanism has been observed.

Monitoring nuclear reactors is an important safety and security task with growing requirements. We explore the possibility of using seismic and acoustic data for inferring the power level of an operating reactor. Continuous data recorded at a single seismo-acoustic station that is located about 50 m away from a research reactor was visualized and analyzed. The data show a clear correlation between seismo-acoustic features and reactor main operational states. We designed a workflow that includes two machine learning models to classify the reactor operational states (off, transition, and on) and estimate reactor power levels (10%, 30%, 50%, 70% and 90%). We applied and compared five machine learning algorithms for the reactor off-transition-on and four approaches for the power level classification. We also compared the performance of machine learning models trained with seismic-only, acoustic-only, and both types of data. Five-fold cross-validations were implemented to assure a thorough evaluation of the model performances. The results show the extreme boosting gradient algorithm worked best for the first model, while random forests performed best for the second model. Combining seismic and acoustic data leads to better performance than using a single type of data. Seismic data contributed more than acoustic data for both models. We reached an accuracy of 0.98 for reactor off and on. The accuracies for the transition state and power levels are less optimal. However, our results suggest seismic and acoustic data contain useful information about the transition state as well as power levels. Seismic and acoustic data could be integrated with other observations to improve monitoring performance.

The stress tensor is an important property for upper crustal studies such as those that involve pore fluids and earthquake hazards. At tectonic plate scale, plate boundary forces and mantle convection are the primary drivers of the stress field. In many local settings (10s to 100s of km and <10 km depth) in tectonic plate interiors, we can simplify by assuming a constant background stress field that is perturbed by local heterogeneity in density and elasticity. Local stress orientation and sometimes magnitude can be estimated from earthquake and borehole-based observations when available. Modeling of the local stress field often involves interpolating sparse observations. We present a new method to estimate the 3D stress field in the upper crust and demonstrate it for Oklahoma. We created a 3D material model by inverting multiple types of geophysical observations simultaneously. Integrating surface-wave dispersion, local travel times and gravity observations produces a model of P-wave velocity, S-wave velocity and density. The stress field can then be modeled using finite element simulations. The simulations are performed using our simplified view of the local stress field as the sum of a constant background stress field that is perturbed by local density and elasticity heterogeneity and gravitational body forces. An orientation of N82˚E, for the maximum compressive tectonic force, best agrees with previously observed stress orientations and faulting types in Oklahoma. The gravitational contribution of the horizontal stress field has a magnitude comparable to the tectonic contribution for the upper 5 km of the subsurface.

The important task of tracking seismic activity requires both sensitive detection and accurate earthquake location. Approximate earthquake locations can be estimated promptly and automatically; however, accurate locations depend on precise seismic phase picking, which is a laborious and time-consuming task. We adapted a deep neural network (DNN) phase picker trained on local seismic data to meso-scale hydraulic fracturing experiments. We designed a novel workflow, transfer-learning aided double-difference tomography, to overcome the three orders of magnitude difference in both spatial and temporal scales between our data and data used to train the original DNN. Only 3,500 seismograms (0.45% of the original DNN data) were needed to re-train the original DNN model successfully. The phase picks obtained with transfer-learned model are at least as accurate as the analyst’s, and lead to improved event locations. Moreover, the effort required for picking once the DNN is trained is a small fraction of the analyst’s.

Enhanced Geothermal Systems could provide a substantial contribution to the global energy demand if their implementation could overcome inherent challenges. Examples are insufficient created permeability, early thermal breakthrough, and unacceptable induced seismicity. Here we report on the seismic response of a meso-scale hydraulic fracturing experiment performed at 1.5 km depth at the Sanford Underground Research Facility. We have measured the seismic activity by utilizing a novel 100 kHz, continuous seismic monitoring system deployed in six 60 m-length monitoring boreholes surrounding the experimental domain in 3-D. The achieved location uncertainty was on the order of 1 m, and limited by the signal-to-noise ratio of detected events. These uncertainties were corroborated by detections of fracture intersections at the monitoring boreholes. Three intervals of the dedicated injection borehole were hydraulically stimulated by water injection at pressures up to 33 MPa and flow rates up to 5 L/min. We located 1933 seismic events during several injection periods. The recorded seismicity delineates a complex fracture network comprised of multi-strand hydraulic fractures and shear-reactivated, pre-existing planes of weakness that grew unilaterally from the point of initiation. We find that heterogeneity of stress dictates the outcome of hydraulic stimulations, even when relying on theoretically well-behaved hydraulic fractures. Once hydraulic fractures intersected boreholes, the boreholes acted as a pressure relief and fracture propagation ceased. In order to create an efficient sub-surface heat exchanger, production boreholes should not be drilled before the end of hydraulic stimulations.

Seismic sensors and seismic imaging have been widely used to monitor the geophysical properties of the subsurface. As subsurface engineering techniques advance, more precise monitoring systems are required. Seismic event catalogs and seismic velocity structures are two of the major outputs of seismic monitoring systems. Although seismic event catalogs and velocity structure are often studied separately, published reports suggest constraining them simultaneously can lead to better results. We conducted a double-difference seismic tomography analysis to constrain both the seismic event locations and the 3D seismic velocity structure. Passive seismic data collected from a geothermal research project in Lead, South Dakota were used to image a 3D volume on the scale of tens of meters. Specifically, around 18,500 P-wave and 8,900 S-wave arrival times from 1,874 seismic events were used. Checkerboard tests showed that the observed data can image the seismically active region well. We compared tomography results with fixed seismic event locations against those with updated event locations. Tomography results with updated event locations showed better fits to the observations and improved the seismic event catalog, showing sharper patterns compared to the original one. These patterns helped us monitor the seismically active fractures since the seismic events were mostly due to hydraulic stimulations. Two parallel fractures revealed by the updated seismic event catalog spatially correlated with independent borehole temperature observations. The average seismic velocity values of the well-constrained volume agreed to the first order with core sample measurements and active-source seismic surveys.