Urban subsurface monitoring requires high temporal‐spatial resolution, low maintenance cost, and minimal intrusion to nearby life. Distributed acoustic sensing (DAS), in contrast to conventional station‐based sensing technology, has the potential to provide a passive seismic solution to urban monitoring requirements. Based on data recorded by the Stanford Fiber Optic Seismic Observatory, we demonstrate that near‐surface velocity changes induced by the excavation of a basement construction can be monitored using existing fiber optic infrastructure in a noisy urban environment. To achieve satisfactory results, careful signal processing comprising of noise removal and source signature normalization are applied to raw DAS recordings. Repeated blast signals from quarry sites provide free, unidirectional, and near‐impulsive sources for periodic urban seismic monitoring, which are essential for increasing the temporal resolution of passive seismic methods. Our study suggests that DAS will likely play an important role in urban subsurface monitoring.

The lack of low-frequency signals in seismic data makes the full-waveform inversion (FWI) procedure easily fall into local minima leading to unreliable results. To reconstruct the missing low-frequency signals more accurately and effectively, we have developed a data-driven low-frequency recovery method based on deep learning from high-frequency signals. In our method, we develop the idea of using a basic data patch of seismic data to build a local data-driven mapping in low-frequency recovery. Energy balancing and data patches are used to prepare high- and low-frequency data for training a convolutional neural network (CNN) to establish the relationship between the high- and low-frequency data pairs. The trained CNN then can be used to predict low-frequency data from high-frequency data. Our CNN was trained on the Marmousi model and tested on the overthrust model, as well as field data. The synthetic experimental results reveal that the predicted low-frequency data match the true low-frequency data very well in the time and frequency domains, and the field results show the successfully extended low-frequency spectra. Furthermore, two FWI tests using the predicted data demonstrate that our approach can reliably recover the low-frequency data.

We have evaluated a field test in the city of Singapore to assess the feasibility of the passive seismic survey for bedrock depth determination and to further investigate the optimal acquisition parameters. The ambient noise field, dominated by urban traffic noise, is recorded passively for the application of seismic interferometry. Spectral analysis indicates that the traffic-induced noise by local roads is concentrated between 3 and 25 Hz. We use multiple signal classification beamforming for wavefield direction of propagation analysis. We apply seismic interferometry to retrieve the surface-wave part of the Green’s functions, based on which surface-wave dispersion relations are extracted and further inverted for 1D S-wave velocity profiles. Subsequently, we compare the inversion results from seismic interferometry with borehole logs at multiple sites in Singapore and establish that the bedrock depths are well-determined using passive seismic methods within a maximum error of 3 m. By investigating the convergence of the crosscorrelograms, we ascertain that the best compromise of cost, efficiency, and accuracy for a passive site investigation in Singapore can be achieved in 15 min in the morning of a working day using an array as short as 30 m with six vertical geophones, although these parameters should be reinvestigated at other sites and other times. The success of this case study demonstrates that accurate near-surface site investigation can be achieved with faster acquisition, fewer receivers, and a smaller acquisition footprint compared with conventional methods, all of which improve the efficiency particularly in a highly developed urban environment.

The development of digital rock physics relies on the availability of high‐quality 3D digital rock images, which can be directly obtained with X‐ray micro‐Computed Tomography (μCT). However, X‐ray μCT is hampered by its high expenses, small sample size (several millimeters in diameter) and low resolution (in micron scale). Although Scanning Electron Microscope (SEM) provides higher resolution on larger rock samples, it only images the 2D rock surface structure. Thus, 3D digital rock reconstruction from 2D cross‐section images becomes promising in saving imaging cost for μCT scan and improving image quality by enabling the incorporation of SEM images in 3D digital rock reconstruction. Here, we propose a machine learning method to reconstruct 3D digital rocks from 2D cross‐section images taken at large constant intervals along the axial direction of the rock sample. The key idea is to train a Progressive Growing Generative Adversarial Network (PG‐GAN) to generate high‐quality gray‐scale cross‐section images, and then reconstruct the 3D digital rock by linearly interpolating the inverted latent vectors corresponding to the sparsely scanned images. We apply our method to reconstructing a large‐size high‐resolution 3D image of an Estaillades carbonate rock sample. We demonstrate that both the reconstructed image and the extracted pore network are visually indistinguishable from the ground truth. Overall, our method achieves nine times speedup of the imaging process, and greater than 4,500 times compression of the image data for the Estaillades carbonate rock sample. The PG‐GAN can enlarge the digital rock repository and enable efficient imaging editing in its linear latent space.

Time‐lapse or 4D seismic survey is a crucial monitoring tool for CO₂ geological sequestration. Conventional time‐lapse interpretation provides detailed characterization of CO₂ distribution in the storage unit. However, manual interpretation is labor‐intensive and often inconsistent throughout the long monitoring history, due to the inevitable changes in seismic acquisition and processing technology and interpreter's subjectivity. We propose a neural network (NN)‐based interpretation method that translates baseline and monitoring seismic images to the probability of CO₂ presence. We use a simplified 3D U‐Net, whose training, validation and testing are all based on the Sleipner CO₂ storage project. The limited labels for training are derived from the interpreted CO₂ plume outlines within the internal sandstone layers for 2010. Then we apply the trained NN on different time‐lapse seismic data sets from 1999 to 2010. The results suggest that our NN‐based CO₂ interpretation has the following advantages: (a) high interpretation efficiency by automatic end‐to‐end mapping; (b) robustness against the processing‐induced mismatch between the baseline and time‐lapse inputs, relaxing the baseline reprocessing demands when compared to newly acquired or reprocessed time‐lapse data sets; and (c) inherent interpretation consistency throughout multiple vintage data sets. Testing results with crafted time‐lapse images unveil that the NN takes both amplitude difference and structural similarity into account for CO₂ interpretation. We also compare 2D and 3D U‐Nets under the scenario of sparse 2D labels for training. The results suggest that the 3D U‐Net provides more continuous interpretation at the cost of larger computational resources for training and application.