Search

The desire for ultra-high density (UHD) seismic surveys is now becoming more achievable for future exploration and field development with the increasing availability of versatile nodal land systems. Acquisition geometry design using a higher density of sources as well as receivers considerably reduces the effects of spatial aliasing and also provides better subsurface illumination. By sampling the wavefield more densely, there are improved recordings of both signal and noise. This presents new opportunities for processing and imaging. We use a recent UHD nodal survey with nominal trace density approaching 200 times that of typical conventional cable-based surveys to discuss the challenges and rewards.

A high-resolution seismic image is of great importance to exploration and production in many ways, such as bypassing drilling hazards and identifying compartmentalized reservoirs. To achieve seismic resolution as high as possible, the conventional seismic imaging process takes more of a linear approach to deal with one or a few specific issues at a time, such as noise and multiple attenuation, source and receiver deghosting, velocity errors, illumination holes, and migration swings. Full-waveform inversion (FWI) Imaging models and uses the full-wavefield data, including primaries and multiples (ghost included) and reflection and transmission waves, to iteratively invert for the reflectivity together with velocity and thus is an elegant solution to resolve those issues in one (iterative) inversion. FWI Imaging has proven to be a superior method for providing seismic images of greatly improved illumination, S/N, focusing, and thus better resolution, over conventional imaging methods. We demonstrate with a towed-streamer data set and an OBN data set that FWI Imaging with a frequency close to the temporal resolution limit of seismic data (100 Hz or higher) can provide seismic images of unprecedented resolution from the recorded seismic data, which has been impossible to achieve with conventional imaging methods.

Images from Kirchhoff migration can suffer from uneven illumination and contamination by migration artefacts. One of the issues is that migration is not a true inverse operation – it is based on the adjoint of the forward modelling operator. In contrast, least squares migration approximates the inverse of the forward modelling and hence the impact of detrimental effects on the image can be reduced. Here, we describe benefits of a non-iterative Kirchhoff least squares method (migration deconvolution). We present a workflow and demonstrate that the method can be used to attenuate image artefacts, help balance image illumination, and increase clarity of AVO attributes.

4D time-lapse seismic imaging is typically performed using the same velocity model to migrate the baseline and monitor data. However, in complex cases where one producing reservoir sits above another, significant changes in the properties of the overlying reservoir can result in a 4D signal with associated 4D coda immediately below it, which would mask the 4D signal of reservoirs underneath. Resolving such an issue requires that the baseline and monitor data be migrated with separate models. We demonstrate how we have used a visco-acoustic full-waveform inversion to resolve 4D changes in velocity and absorption within an overburden gas-charged channel. The resulting 4D image shows minimisation of 4D coda below the overburden channel and the unveiling of 4D signals at deeper targets of interest.

Convolutional based deep neural networks can be used in addition to existing workflows, to improve turnaround or as a ‘guide’ for further processing. Whilst a lot of effort has been made to try to improve the DNN architecture for processing tasks or to understand their physical interpretation, the choice of the training-set is rarely discussed. For a good quality DNN result, the training-set must be representative of the variability (or statistical diversity) of the full dataset, and the question of the choice of this dataset for seismic data is discussed in this paper. We present two methods for the selection of the training set. The first one is based on proxy attributes and their clustering. Our clustering approach is not only using the clusters themselves but also the information on the distance to the centroid for the cluster definition. The other method is based on the data themselves. It starts from a predefined training set and then scans through the full dataset to identify additional training points that will be used to augment the initial training set.

Learning how to best mimic seismic processing algorithms or workflows with deep learning (DL) has become a very active field of research. However, seismic processing own particularities may necessitate adaptations of current DL methods. In this paper, we explain and illustrate how the different DL components can affect the outcome of a given seismic processing task. Among others, we show that the Unet neural network architecture ( Ronneberger et al., 2015 ) is naturally suited to learn how to “separate” the events into kinematics and their amplitudes, and how to use both information efficiently to perform the common image gathers preconditioning, skeletonization (or picks probability computation) and muting task. We also show how the convolution kernel shapes, the number of layers, the training cost function and the batch size can be adapted to specific data and seismic processing tasks.

Depth imaging of converted-wave (P-to-S) ocean-bottom seismic (OBS) data requires a depth model for both the P- and S-wave velocities. Building the S-wave velocity model is very challenging: conventional techniques include PP and PS image registration, or joint PP and PS tomography. These approaches are often impeded by the lack of a reliable PS image in the shallow part of the model due to the sparse-receiver acquisition of typical OBS surveys, and have limited resolution to deal with complex lateral velocity variations. We introduce a new full-waveform inversion technique to update the S-wave velocity using converted-wave reflection data recorded in the radial component of OBS surveys. Key aspects of the method include the use of acoustic Born-modeling, a robust objective function to handle kinematic and dynamic differences, and a layer-stripping strategy to simplify the non-linearity of the inversion problem. The proposed approach is validated on different synthetics, and demonstrated on a field data example, giving an improved S-wave velocity and better reflector continuity for PS imaging.

Carbonate velocity model building is challenging due to the complex geometry and sharp velocity contrasts associated with carbonates. Full-waveform inversion (FWI), together with long offsets, wide azimuth and good low frequency data, is known to be a powerful tool to address these challenges. Unfortunately, many vintage streamer datasets are handicapped by limited offsets and azimuth coverage, and a noisy low-frequency component. We used vintage streamer datasets acquired in the South China Sea to demonstrate that Time-lag FWI (TLFWI), together with other tools like dipconstrained tomography and well calibration, can overcome those shortcomings and produce a highresolution velocity field, leading to improved images. TLFWI uses a crosscorrelation cost function to mitigate amplitude mismatch and low signal-to-noise ratio problems. However, the carbonates being out of reach of diving waves can still be challenging to update with FWI, if the starting background velocity is far from the true model. In this case, an iterative FWI flow with well-constrained velocity updates inbetween offers a more reliable solution. The carbonate fracture system poses another challenge for estimating anisotropic parameters inside the carbonate layer. Here we use diffraction imaging to guide the fracture system identification, which helps to estimate an HTI system.

For long in seismic imaging, velocity model building and depth migration/inversion have produced information on the subsurface velocity model with no overlap in terms of resolved vertical wavelengths. The not covered wavelengths, among which the famous mid frequency gap, had then to be recovered in stratigraphic inversion by external information such as borehole data. The progresses in terms of acquisitions (long offset and low frequency) and imaging tools put us now in the situation of an overlap between all these processing/imaging/inversion approaches. FWI provides for example a velocity model building tool that covers potentially the full range of vertical frequencies in the area investigated by recorded diving waves. High resolution tomography from its side reaches vertical resolutions up to 6 Hz overlapping the resolution that can be obtained from depth migration and then stratigraphic inversion with low frequency data (down to 2.5 Hz). This new status has motivated investigations about improved ways of integrating these sources of information. We review here several of these attempts that allow taking advantage of the various approaches for the benefits of reliability and interpretability of the results. The estimation of uncertainties in ray based tomography is for example a precious add on for assessing the reliability of the final result of the imaging/inversion workflow.