Low-frequency seismic data are crucial for convergence of full-waveform inversion (FWI) to reliable subsurface properties. However, it is challenging acquire field with an appropriate signal-to-noise ratio in the low-frequency part spectrum. We have extrapolated from respective higher frequency components wavefield by using deep learning. Through wavenumber analysis, we find that extrapolation per shot gather has broader applicability than per-trace extrapolation. numerically simulate marine...

Building realistic and reliable models of the subsurface is primary goal seismic imaging. We have constructed an ensemble convolutional neural networks (CNNs) to build velocity directly from data. Most other approaches attempt map full data into 2D labels. exploit regularity acquisition train CNNs gathers neighboring common midpoints (CMPs) vertical 1D logs. This allows us integrate well-log inversion, simplify mapping by using labels, accommodate larger dips relative single CMP inputs....

Low-frequency signal content in seismic data as well a realistic initial model are key ingredients for robust and efficient full-waveform inversions. However, acquiring low-frequency is challenging practice active surveys. Data-driven solutions show promise to extrapolate given high-frequency counterpart. While being established synthetic acoustic examples, the application of bandwidth extrapolation field datasets remains non-trivial. Rather than aiming reach superior accuracy extrapolation,...

Full-waveform inversion (FWI) attempts to resolve an ill-posed nonlinear optimization problem retrieve the unknown subsurface model parameters from seismic data. In general, FWI fails obtain adequate representation of models with large high-velocity structures over a wide region, such as salt bodies and sediments beneath them, in absence low frequencies recorded signal, due nonlinearity nonuniqueness. We alleviate ill posedness associated data sets affected by using regularization. have...

Integrating advanced artificial intelligence (AI) into geoscience represents a pivotal moment, redefining how we approach exploration and interpretation of the earth's subsurface. Generative AI methods, such as large language models (LLMs), diffusion models, physics-informed learning, offer new ways to simulate, invert, interpret seismic data. LLMs are increasingly used in various tasks ranging from interpolation denoising direct inversion for subsurface properties. Promising attempts have...

Summary Full-waveform inversion (FWI) benefits in many ways from having low-frequency data. However, those are rarely available due to acquisition limitations. Here, we explore the feasibility of frequency-bandwidth extrapolation using an Artificial Neural Network (ANN) approach. The ANN is trained be a non-linear operator that maps high-frequency data for single source and multiple receivers Assuming point (delta function) both time space, train network on synthetic generated random...

Abstract Full‐waveform inversion (FWI) optimizes the subsurface properties of geophysical Earth models in such a way that modeled data, based on these properties, match observed data. The anisotropic whether monoclinic, orthorhombic, triclinic, or vertical transversally isotropic (VTI), subsurface, be it fractured reservoir core‐mantle boundary, are necessary to describe wave phenomena. There no principal limitations complexity anisotropy can inverted using FWI. However, question...

Low-frequency data are essential to constrain the low-wavenumber model components in seismic full-waveform inversion (FWI). However, due acquisition limitations and ambient noise it is often unavailable. Deep learning (DL) can learn map from high frequency updates of elastic FWI a update, producing an initial estimation as if was available low-frequency data. We train FusionNET-based convolutional neural network (CNN) on synthetic dataset produce set produced by with missing low frequencies....

Various parametrizations have been suggested to simplify inversions of first arrivals, or P waves, in orthorhombic anisotropic media, but the number and type retrievable parameters not decisively determined. We show that only six can be retrieved from dynamic linearized inversion waves. These are different needed describe kinematics Reflection-based radiation patterns P–P scattered waves remapped into spectral domain allow for our resolution analysis based on effective angle illumination...

Velocity model building is not straightforward in geologically complex environments. We train a convolutional neural network (CNN) to map full wavefields smooth subsurface parameter distributions address the problem. Specifically, cubes of neighboring CMP gathers are mapped into 1D vertical profiles simplify training phase and make it easier utilize well logs future applications. CNN using total one hundred thousand random models generated on-the-fly corresponding synthetic data. The...

Distributed acoustic sensing (DAS) has emerged as a transformational technology for seismic data acquisition. However, noise remains major impediment, necessitating advanced denoising techniques. This study pioneers the application of diffusion models, type generative model, DAS vertical profile (VSP) denoising. The network is trained on new generated synthetic dataset that accommodates variations in acquisition parameters. model applied to suppress and field DAS-VSP data. results...

ABSTRACT Full‐waveform inversion suffers from local minima, due to a lack of low frequencies in data. A reflector below the zone interest may, however, help recovering long‐wavelength components velocity perturbation, as demonstrated paper by Mora. With Born approximation for perturbation reference model consisting two homogeneous isotropic acoustic half‐spaces and assumption infinitely large apertures available data, analytic expressions can be found that describe spatial spectrum recorded...

The scattering angle between the source and receiver wave-fields can be utilized in full-waveform inversion (FWI) reverse-time migration (RTM) for regularization quality control or to remove low frequency artifacts. access information is costly as relation local image features angles has non-stationary nature. For purpose of a more efficient extraction, we develop techniques that utilize simplicity based filters constant-velocity background models. We split velocity model into several...

When present in the subsurface, salt bodies impact complexity of wave-equation-based seismic imaging techniques, such as least-squares reverse time migration and full-waveform inversion (FWI). Typically, Born approximation used every iteration least-squares-based inversions is incapable handling sharp, high-contrast boundaries bodies. We have developed a variance-based method for reconstruction velocity models to resolve issues caused by Our main idea lies retrieving useful information from...

Summary Full-waveform inversion (FWI) is a technique which solves the ill-posed seismic problem of fitting our model data to measured ones from field. FWI capable providing high-resolution estimates model, and handling wave propagation arbitrary complexity (visco-elastic, anisotropic); yet, it often fails retrieve high-contrast geological structures, such as salt. One reasons for failure that updates at earlier iterations are too smooth capture sharp edges salt boundary. We compare several...

The initial quantification of data quality is an important step in seismic acquisition design, including the choice sensing strategy. signal-to-noise ratio (SNR) often drives distributed acoustic (DAS) parameters vertical profiling (VSP). We compare this established approach for assessment with metrics comparing DAS products to available well logs. First, we create kinematic and dynamic derived from original data, such as interval velocity amplitude P-wave arrivals. Next, quantify using log...

ABSTRACT Full‐waveform inversion, a popular technique that promises high‐resolution models, has helped in improving the salt definition inverted velocity models. The success of inversion relies heavily on having prior knowledge salt, and using advanced acquisition technology with long offsets low frequencies. Salt bodies are often constructed by recursively picking top bottom from seismic images corresponding to tomography combined flooding techniques. process is time consuming highly prone...

Time-lapse seismic data acquisition is an essential tool to monitor changes in a reservoir due fluid injection, such as CO 2 injection. By acquiring multiple surveys the exact same location, authors can identify by analyzing difference data. However, analysis be skewed near-surface seasonal velocity variations, inaccuracy, and repeatability parameters, other inevitable noise. The common practice (cross equalization) address this problem uses part of which are not expected design matching...

Summary Building subsurface models requires the incorporation of geological knowledge that often comes in form text. Such is typically first converted into mathematical formulation and then used as part an inverse problem a penalty or constraint. Stable diffusion can be for conditional image generation thus digitizing information spatially varying priors later inversion. Here we show how such generated conditioned using well logs. In particular, create new small dataset where extract patches...

Missing low-frequency content in seismic data is a common challenge for inversion. Long wavelengths are necessary to reveal large structures the subsurface and build an acceptable starting point later iterations of full-waveform inversion (FWI). High-frequency land particularly challenging due elastic nature Earth contrasting with acoustic air at typically rugged free surface, which makes use low frequencies even more vital We propose supervised deep learning framework bandwidth...

Full-waveform inversion (FWI) includes migration and tomography modes. The tomographic component of the gradient from reflection data is usually much weaker than component. To use mode to fix background velocity errors, it necessary extract gradient. Otherwise, will be dominated by mode. We have developed a method based on nonstationary smoothing raw By analyzing characteristics scattering angle filtering, wavenumber at given frequency seen smaller that Therefore, low-wavenumber-pass...

Distributed acoustic sensing (DAS) technologies are now becoming widespread, particularly in vertical seismic profiling (VSP). Being a spatially densely sampled recording of the wavefield, DAS data provide an extended measurement compared with point geophone VSP. We have developed basic theory that enables intuitive geophysical understanding amplitudes using concepts kinetic and potential energy their fluxes. start by relating measurements to energy, respectively. use this relationship...

Training datasets consisting of numerous pairs subsurface models and target variables are essential for building machine learning solutions geophysical applications. We apply an iterative style transfer approach from image processing to produce realistically textured based on synthetic prior models. The key idea is that content texture representations within a convolutional neural network are, some extent, separable. Thus, one can be transferred match the another image. demonstrate examples...

Summary Deep learning can be used to help reconstruct low frequencies in seismic data, and directly infer velocity models simple cases. In order succeed with deep learning, a good training set of is critical. We present new way design random that are statistically similar given guiding model. Our approach based on shuffling the coefficients wavelet packet decomposition (WPD) model, allowing us replicate realistic textures from synthetic generate realistically BP 2004 Marmousi II for neural...