In seismic full-waveform inversion (FWI), subsurface parameters are estimated by iteratively minimizing the difference between the modeled and the observed data. We have considered the problem of estimating the elastic constants of a fractured medium using multiparameter FWI and modeling naturally fractured reservoirs as equivalent anisotropic media. Multiparameter FWI, although promising, remains exposed to a range of challenges, one being the parameter crosstalk problem resulting from the overlap of Fréchet derivative wavefields. Parameter crosstalk is strongly influenced by the form of the scattering pattern for each parameter. We have derived 3D radiation patterns associated with scattering from a range of elastic constants in general anisotropic media. Then, we developed scattering patterns specific to a horizontal transverse isotropic (HTI) medium to draw conclusions about parameter crosstalk in FWI. Bare gradients exhibit crosstalk, as well as artifacts caused by doubly scattered energy in the data residuals. The role of the multiparameter Gauss-Newton (GN) Hessian in suppressing parameter crosstalk is revealed. We have found that the second-order term in the multiparameter Hessian, which is associated with multiparameter second-order scattering effects, can be constructed with the adjoint-state technique. We have examined the analytic scattering patterns for HTI media with a 2D numerical example. We have examined the roles played by the first- and second-order terms in multiparameter Hessian to suppress parameter crosstalk and second-order scattering artifacts numerically. We have also compared the multiparameter GN and full-Newton methods as methods for determining the elastic constants in HTI media with a two-block-layer model.

We have developed a new sparse-spike deconvolution (SSD) method based on Toeplitz-sparse matrix factorization (TSMF), a bilinear decomposition of a matrix into the product of a Toeplitz matrix and a sparse matrix, to address the problems of lateral continuity, effects of noise, and wavelet estimation error in SSD. Assuming the convolution model, a constant source wavelet, and the sparse reflectivity, a seismic profile can be considered as a matrix that is the product of a Toeplitz wavelet matrix and a sparse reflectivity matrix. Thus, we have developed an algorithm of TSMF to simultaneously deconvolve the seismic matrix into a wavelet matrix and a reflectivity matrix by alternatively solving two inversion subproblems related to the Toeplitz wavelet matrix and sparse reflectivity matrix, respectively. Because the seismic wavelet is usually compact and smooth, the fused Lasso was used to constrain the elements in the Toeplitz wavelet matrix. Moreover, due to the limitations of computer memory, large seismic data sets were divided into blocks, and the average of the source wavelets deconvolved from these blocks via TSMF-based SSD was used as the final estimation of the source wavelet for all blocks to deconvolve the reflectivity; thus, the lateral continuity of the seismic data can be maintained. The advantages of the proposed deconvolution method include using multiple traces to reduce the effect of random noise, tolerance to errors in the initial wavelet estimation, and the ability to preserve the complex structure of the seismic data without using any lateral constraints. Our tests on the synthetic seismic data from the Marmousi2 model and a section of field seismic data demonstrate that the proposed method can effectively derive the wavelet and reflectivity simultaneously from band-limited data with appropriate lateral coherence, even when the seismic data are contaminated by noise and the initial wavelet estimation is inaccurate.

Knowledge of changes in reservoir properties resulting from extracting hydrocarbons or injecting fluid is critical to future production planning. Full-waveform inversion (FWI) of time-lapse seismic data provides a quantitative approach to characterize the changes by taking the difference of the inverted baseline and monitor models. The baseline and monitor data sets can be inverted either independently or jointly. Time-lapse seismic data collected by ocean-bottom cables (OBCs) in the Valhall field in the North Sea are suitable for such time-lapse FWI practice because the acquisitions are of a long offset, and the surveys are well-repeated. We have applied independent and joint FWI schemes to two time-lapse Valhall OBC data sets, which were acquired 28 months apart. The joint FWI scheme is double-difference waveform inversion (DDWI), which inverts differenced data (the monitor survey subtracted by the baseline survey) for model changes. We have found that DDWI gave a cleaner and more easily interpreted image of the reservoir changes compared with that obtained with the independent FWI schemes. A synthetic example is used to demonstrate the advantage of DDWI in mitigating spurious estimates of property changes and to provide cross validations for the Valhall data results.

Time-lapse seismic data are widely used to monitor reservoir changes. Qualitative comparisons between baseline and monitor data sets or image volumes provide information about fluid and pressure effects within the reservoir during production. However, to perform real quantitative analysis of such reservoir changes, quantitative estimates of the elastic parameters are required as input parameters to rock-physics-based reservoir models. Full-waveform inversion has been proposed as a potential tool for retrieving subsurface properties, such as P- and S-wave velocities and density by fitting simulated waveforms to seismic data. An extension of this method to time-lapse applications seems straightforward, but, in fact, it requires more tailored processes such as double-difference waveform inversion (DDWI). We used realistic 2D synthetic pressure data examples to compare the performance of DDWI with that of two other inversion schemes: one using the same starting model for both inversions and the other starting the monitor inversion with the final baseline inversion model. The data simulation and inversion were based on acoustic theory. Although P-wave velocity changes were reliably recovered by each inversion method, DDWI was found to deliver the best results when perfectly repeated surveys were used. However, differencing the baseline and monitor data sets, as required by DDWI, could be found to be sensitive to the presence of survey nonrepeatability. To investigate the feasibility of using DDWI in practice, the dependence of DDWI on the quality of the baseline models and its robustness to survey nonrepeatability were studied with numerical tests. Various types of nonrepeatability were considered separately in the synthetic tests, including random noise, acquisition geometry mismatch, source wavelet discrepancy, and overburden velocity changes. A study of the correlation between the levels and types of nonrepeatability and the resulting contamination of the inversion results found that, for pressure data, DDWI was capable of inverting reliably for P-wave velocity changes under realistic survey nonrepeatability conditions.

Traditional amplitude variation with offset and azimuth (AVOAz) analysis for fracture characterization extracts fracture properties through analysis of reflection AVOAz to determine anisotropic parameters (e.g., Thomsen’s parameters) that are then related to fracture properties. The validity of this method relies on the basic assumption that a fractured unit can be viewed as an equivalent anisotropic medium. As a rule of thumb, this assumption is taken to be valid when the fracture spacing is less than $\lambda /10$. Under the effective medium assumption, diffractions from individual fractures destructively interfere and only specular reflections from boundaries of a fractured layer can be observed in seismic data. The effective medium theory has been widely used in fracture characterization, and its applicability has been validated through many field applications. However, through numerical simulations, we find that diffractions from fracture clusters can significantly distort the AVOAz signatures when a fracture system has irregular spacing even though the average fracture spacing is much smaller than a wavelength (e.g., $\ll \lambda /10$). Contamination by diffractions from irregularly spaced fractures on reflections can substantially bias the fracture properties estimated from AVOAz analysis and may possibly lead to incorrect estimates of fracture properties. Additionally, through Monte Carlo simulations, we find that fracture spacing uncertainty inverted from amplitude variation with offset (AVO) analysis can be up to 10%–20% when fractures are not uniformly distributed, which should be the realistic state of fractures present in the earth. Also, AVOAz and AVO analysis gives more reliable estimates of fracture properties when reflections at the top of the fractured layer are used compared with those from the bottom of the layer.

Automatic detection of low-magnitude earthquakes has become an increasingly important research topic in recent years due to a sharp increase in induced seismicity around the globe. The detection of low-magnitude seismic events is essential for microseismic monitoring of hydraulic fracturing, carbon capture and storage, and geothermal operations for hazard detection and mitigation. Moreover, the detection of micro-earthquakes is crucial to understanding the underlying mechanisms of larger earthquakes. Various algorithms, including deep learning methods, have been proposed over the years to detect such low-magnitude events. However, there is still a need for improving the robustness of these methods in discriminating between local sources of noise and weak seismic events. In this study, we propose a convolutional neural network (CNN) to detect seismic events from shallow borehole stations in Groningen, the Netherlands. We train a CNN model to detect low-magnitude earthquakes, harnessing the multi-level sensor configuration of the G-network in Groningen. Each G-network station consists of four geophones at depths of 50, 100, 150, and 200 m. Unlike prior deep learning approaches that use 3-component seismic records only at a single sensor level, we use records from the entire borehole as one training example. This allows us to train the CNN model using moveout patterns of the energy traveling across the borehole sensors to discriminate between events originating in the subsurface and local noise arriving from the surface. We compare the prediction accuracy of our trained CNN model to that of the STA/LTA and template matching algorithms on a two-month continuous record. We demonstrate that the CNN model shows significantly better performance than STA/LTA and template matching in detecting new events missing from the catalog and minimizing false detections. Moreover, we find that using the moveout feature allows us to effectively train our CNN model using only a fraction of the data that would be needed otherwise, saving plenty of manual labor in preparing training labels. The proposed approach can be easily applied to other microseismic monitoring networks with multi-level sensors.