Geophysical Prospecting最新文献

We present deep learning (DL) networks for three-dimensional (3D) joint inversion of active seismic full waveform and passive seismic traveltime data to image reservoirs and their properties and quantify imaging uncertainties. Active seismic full-waveform data can provide high-resolution monitoring images but are collected only intermittently because of their high acquisition cost. In contrast, passive seismic data can be gathered at relatively low cost between regular active surveys, although their imaging quality can be compromised by factors such as low signal-to-noise ratios and limited ray coverage of the target. Although these datasets are routinely acquired together at CO₂ storage sites, their combined inversion within a 3D DL framework has not been previously demonstrated. To our knowledge, this is the first study to address this gap, combining the strength of both data types. For efficient data storage and DL training with large 3D seismic datasets, we use a 3D data matrix in which a random number of passive seismic traveltime data are stored as parabolic envelopes using one-hot encoding and a 3D full-waveform data matrix in which multiple shot gathers are summed. Two network architectures are evaluated: a single-encoder U-Net for single-data type inversion and a dual-encoder U-Net for joint inversion of active and passive seismic data. We also evaluate the single-encoder U-Net for joint inversion by concatenating full-waveform data and traveltime data. We propose a systematic approach for selecting an optimal dropout rate that balances regularization during training and Monte Carlo dropout-based uncertainty quantification during prediction by examining the correlation coefficient between standard deviation and prediction error, along with the training misfit, across a range of dropout rates. 3D DL inversion experiments include five different network configurations, with evaluations under ideal, noisy and dropout-enabled conditions. Both model and data uncertainties are assessed, as well as their combined effects. Across all conditions, the networks consistently predict accurate CO₂ saturation models with low prediction errors, such as a structural similarity index of 0.993 and CO₂ difference of 1.1%. Uncertainty estimates show strong spatial correlation with prediction errors, confirming the effectiveness of the proposed dropout selection approach. The results demonstrate that our DL approach, utilizing compact data representations and appropriate uncertainty quantification, yields accurate subsurface images under various inversion conditions and provides valuable insights into the reliability of predictions.

As demand for large-scale seismic data interpretation tasks increases, machine learning-based horizon autotracking methods have gained attention in the geological and geophysical fields. Although such methods have demonstrated time- and cost-efficiency in large-scale data interpretation, studies on the expansion of interpreted horizons into the reservoir characterization process are relatively limited. Hence, a reservoir characterization process that can incorporate the machine learning-interpreted horizons and their structural uncertainties into the reservoir uncertainty assessment is necessary for an efficient reservoir modelling process. The proposed workflow consists of various modelling processes, including horizon construction where machine learning-interpreted horizons are used instead of manually interpreted horizons, facies modelling and petrophysical modelling. The modelling algorithms are based on stochastic methods: sequential indicator simulation for facies models and Gaussian random function simulation for petrophysical properties. Each modelling process incorporates variables such as variogram parameters, facies ratios and modified porosity values. The results show promising performance in incorporating machine learning-interpreted horizons into the uncertainty quantification process and analysing their impact by capturing the influence of structural uncertainties of horizons in the final reservoir pore volume.

The Marchenko method is a powerful tool for reconstructing full-wavefield Green's functions using surface-recorded seismic data. These Green's functions can then be utilized to produce subsurface images that are not affected by artefacts caused by internal multiples. Despite its advantages, the method is computationally demanding, primarily due to the iterative nature of estimating the focusing functions, which links the Green's functions to the surface reflection response. To address this limitation, an optimization-based solver is proposed to estimate focusing functions in an efficient way. This is achieved by training a network to approximate the forward modelling problem on a small subset of pre-computed focusing function pairs, mapping final up-going focusing functions obtained via the conventional iterative scheme to their initial estimates. Once trained, the network is fixed and used as the forward operator within the Marchenko framework. For a given target location, an input is initialized and iteratively updated through backpropagation to minimize the mismatch between the output of the fixed network and the known initial up-going focusing function. The resulting estimate is then used to compute the corresponding down-going focusing function and the full Green's functions based on the Marchenko equations. This strategy significantly reduces the computational cost compared to the traditional Marchenko method based on the conventional iterative scheme. Tests on a synthetic model, using only 0.8% of the total imaging points for training, show that the proposed approach accelerates the imaging process while maintaining relatively good imaging results, which is better than single scattering imaging. Application to the Volve field data further demonstrates the method's robustness and practicality, highlighting its potential for efficient, large-scale seismic imaging.

During seismic data processing, strong single-frequency interference noise often affects data quality. Traditional methods for single-frequency interference identification are typically conducted in the frequency domain, primarily by searching for abnormal peaks in the frequency spectrum of each trace. However, when the interference amplitude in the frequency domain is weak relative to the entire seismic frequency, the identification process becomes significantly more challenging. To address this issue, this article proposes a time-domain approach for identifying single-frequency interference. First, frequency analysis is performed on seismic data containing single-frequency interference to obtain the initial frequency $�f_0$, and sine and cosine signals with an amplitude of 1 at this frequency are then generated. Next, the seismic data are normalized to balance amplitude differences across different datasets in the time domain. After normalization, deep-time seismic data are cross-correlated with the generated sinusoidal and cosine signals, and correlation coefficient R is computed to determine whether suppression is necessary. On the basis of theoretical simulations and field data analysis, suppression is considered necessary when R > 0.001. Finally, for identified single-frequency interference, a hierarchical approximation method based on a cross-correlation objective function is employed to search for interference frequencies with finer step sizes near the initial frequency and calculate the corresponding amplitudes. The interference signal is subsequently subtracted in the time domain to achieve interference suppression. Through synthetic experiments and the application analysis of various field seismic data (including single-shot and stacked profiles), the proposed method demonstrates high efficiency and accuracy in identifying and suppressing single-frequency interference.

Staggered-grid finite-difference (FD) schemes are widely used in numerical simulation of seismic wave propagation. The traditional explicit staggered-grid scheme adopts the second-order temporal and explicit high-order spatial FD, so it easily suffers from significant temporal dispersion and the spatial operator–length saturation effect. The recently developed implicit time–space-domain high-order staggered-grid scheme for 2D acoustic wave simulation overcomes those two weaknesses effectively, yielding high-order accuracies in both time and space and thus better suppressing the numerical dispersion. However, the involved FD coefficients are generally determined by Taylor-series expansion (TE) or the least-squares (LS) method and still cannot effectively control spatial dispersion at the large wavenumber range. Adopting the same FD stencil, we alternatively determine the FD coefficients using a combination of TE and the Remez optimization algorithm. The temporal accuracy–related coefficients are determined by the TE of the time–space-domain dispersion relation, whereas the implicit spatial FD coefficients are calculated by using the Remez exchange optimization algorithm to effectively extend the effective wavenumber range while achieving a high-order temporal accuracy and consequently enhance the overall modelling accuracy. The newly optimized scheme is then extended into a 3D case. Dispersion and numerical analyses validate that the proposed new schemes better suppress the spatial dispersion while maintaining the high-order temporal accuracy and outperform the existing TE- and LS-based FD schemes. To further improve the modelling efficiency, the variable–operator–length strategy is combined. Numerical examples of the 2D and 3D complicated models validate the effectiveness of the combined scheme in reducing the operator length and consequently improving the modelling efficiency.

Vertically transverse isotropic (VTI) acoustic wave equations are widely used to simulate wave propagation in VTI media. A commonly used acoustic VTI wave equation can be derived by setting the vertical shear-wave velocity to zero in the elastic VTI wave equation. However, the resulting acoustic VTI wave equation has a non-symmetric propagation operator, which leads to the operator of the adjoint equation in full-waveform inversion (FWI) being different from that of the forward equation. Consequently, two separate sets of code are required for simulating the forward and adjoint wavefields. To simplify code implementation, we propose a symmetric-form acoustic VTI equation. This new formulation allows both the forward and adjoint equations in FWI to share the same operator, enabling a unified code for both the forward and adjoint wavefield simulation and streamlined implementation. In addition, although both the symmetric and non-symmetric formulations yield the same gradient, the adjoint wavefield from the non-symmetric equation shows weaker amplitudes in deeper regions compared to that from the symmetric equation. As a result, FWI using the non-symmetric formulation may suffer from insufficient compensation when employing a spatial preconditioner based on an approximated diagonal pseudo-Hessian, leading to slower convergence. Numerical examples using the Marmousi2 and BP anisotropic models, as well as an ocean-bottom cable (OBC) field data set, demonstrate that the proposed symmetric-form acoustic VTI FWI achieves better inversion results and faster convergence than its non-symmetric counterpart.

Deep learning has been widely used to invert geophysical properties due to the availability of training data and an increased computing power. In particular, Bayesian deep learning is commonly applied to estimate the uncertainty of rock properties, which is essential for risk management and decision-making. However, the selection of appropriate prior parameters, such as the standard deviation in the Gaussian prior distribution placed on neural parameters including neural weights and biases, is crucial for training Bayesian neural networks (BNN), as it significantly impacts the prediction performance of the trained models. In this research, we introduce a hierarchical structure to the BNN, and the mean and standard deviation in the Gaussian prior placed on neural parameters are randomly drawn from hyper-priors, thus excluding the preliminary tuning runs with trial values. Compared to traditional BNN, a consistent prediction accuracy is achieved with an estimate of aleatoric and epistemic uncertainties when using the hierarchical Bayesian networks with different hyper-priors, thereby making them more robust against the choice of prior parameters. In addition, we apply the statistical sampling technique to reduce the overall size of the training data, which can proportionally decrease the training time of the deep learning models when large amounts of training data are available.

Seismic data interpolation using convolutional neural networks (CNNs) suffers from accuracy limitations due to the inter-band interference across different frequency bands, which negatively affects subsequent inversion and interpretation. To address this limitation, we propose a multi-band strategy that first decomposes the seismic data into multiple sub-bands through frequency filtering. Independent CNN models are then used to process each specific frequency band to isolate spectral interference. We focus on regularly missing shots interpolation, assuming that dense receiver arrays are available during seismic acquisition with sparse shots. As for the training data preparation, the spatial reciprocity of Green's function is considered, which guarantees the similarity between common shot gathers (CSGs) and common receiver gathers (CRGs). The available dense CSGs are used to train networks using the multi-band-assisted training strategy. The resulting optimized independent models are then employed to reconstruct missing shots in sparse CRGs for each frequency band separately. Interpolated multi-band data are finally fused by summation to obtain the full-band result. Numerical experiments on synthetic and field data demonstrate that the proposed multi-band-assisted training strategy provides superior interpolation accuracy compared to traditional full-band training, particularly in mitigating cross-band interference.

This study describes an integrated landslide monitoring program in the landslide-prone Tizzano Val Parma region (Italy) using traditional and innovative geoelectrical techniques, namely, electrical resistivity tomography (ERT), induced polarization (IP) and self-potential (SP) methods. Both the conventional fixed-base and the unconventional sparse gradient array configurations were adopted. The use of analytic signal amplitude (ASA) technique enabled for a better recognition of primary SP anomaly sources, for the sparse gradient arrays, providing useful insights in delineating areas of interest. The region faces recurrent landslides due to geological and geomorphological factors, leading to high hydrogeological instabilities and environmental risks. Borehole stratigraphy reveals a complex lithology of sandstones, clayey marl, and coarse materials. The ERT–IP survey provides insights into various landslide types, identifying distinct domains including complex quiescent, active and undetermined landslides. Active fault evolution is observed, indicating potential risk zones. Sparse gradient SP monitoring captures short-term electrokinetic anomalies and stable long-term variations between 2022 and 2023. Both fixed-base and sparse gradient SP monitoring highlight displacement anomalies towards the valley, suggesting potential landslide movements. Interpretation of SP maps allowed to identify preferential water flow directions which denote probable accentuating risks during intense rainfall events. This study emphasizes the significance of integrated geoelectrical monitoring for early landslide detection. Non-conventional and conventional SP arrays provide insights into anomaly repeatability and stability. Comparison of ERT–IP results with borehole information enables the extrapolation of geological characteristics, yielding a holistic understanding of subsurface structures and potential risk zones. Integration of these methodologies contributes to effective landslide management, underscoring the dynamic nature of landslide-prone regions and the necessity for ongoing risk assessment and monitoring.