Search

Generation of inaccurate multiple models that are subsequently adapted to the data can easily cause signal damage where dip discrimination between primaries and multiples is low. This is often diagnosed by appreciable distortion of the multiple model during adaption. Using data from onshore Oman, pre-stack multiple modelling and attenuation performed with variations of the predictive deconvolution and correlation-convolution methods produced accurate multiple models that undergo little distortion during adaption. The methods appear safe for pre-stack use and for both structural and AVO analysis.

We put forward an optimized process in which dynamic warping is repeated with offset dependent constraint to ensure the traveltime difference reliable for the inversion. With the better reliably detected time-shifts, we can better manage the cycle-skipping problem when starting at a relatively higher frequency. The synthetic data from the Marmousi model as well as one challenging OBC dataset from Indonesia are used to demonstrate this method.

Velocity model building in the shallow overburden is difficult in the central North Sea. Shallow waters is tough for reflection tomography, shallow gas pockets cause velocity errors and poor imaging and anisotropy is a guess at best with little shallow well control. We present a 3-parameter FWI workflow addressing velocity, absorption factor (Q), and epsilon. This improves the imaging and positioning of complex injectite structures around 2.5 km depth. We also seek an increased level of detail in the final velocity model for use as a background to AVO inversion.

Surface related demultiple continues to be a major challenge in shallow water environments. While multiples generated by the water bottom reflection can often be adequately attenuated with targeted demultiple approaches, multiple reflections from other shallow events can be more challenging to remove. Deconvolution can help to mitigate this problem, but applications are generally limited to 2D or work on one sailline at a time. We describe a 3D multi-sailline deconvolution-like approach that uses surface datum deghosted data to derive a multiple prediction operator in the image domain using one-way wavefield extrapolation operators. Synthetic and real data results show how the method may predict multiples from shallow multiple generators and may outperform targeted demultiple approaches.

Illustration of how machine learning application is used for Seismic global QC and detailled clusterization of data for tailored processing. Data reduction with deep neural network shows great efficiency to limit memory issue.

Reprocessing of available seismic dataset using high-end processing technology was seen to be the most economical solution for search of deeper, pre-carbonate targets in a marginal field. The vintage data suffered from amplitude loss due to absorption by gas bodies and carbonate layers causing poor reflector continuity especially at target zone. Removal of noise and multiples beneath the carbonate layer was also essential to image the deeper fault systems properly. Even with narrow azimuth and limited offset data, which has illumination issues, current advanced imaging technologies were able to unmask the reservoir.

We illustrate a new case study based on our land FWI workflow. We also demonstrate the impact of optimal transport (OT) FWI on the resulting velocity model, which aims to mitigate cycle skipping. We demonstrate how it corrects several localized failures with conventional FWI and constructs the geologically consistent velocity model.

Placing marine sources on the top of the seismic streamer spread improves resolution and reservoir inversion in shallow and intermediate depth targets. This is due to the abundance of near offset data and high level of illumination compared to conventional marine seismic. A drawback of this top source solution is the lack of long offsets, which are important for deeper imaging and AVO. In this paper, we present a combined solution with both sources on top and in the front of the spread. We deploy the front sources from the vessel that also is towing the streamers, while a separate source vessel is towing the top sources. This gives an operationally efficient top-to-bottom solution to the seismic imaging and inversion problem solving both high and low targets. We will show examples of real data with up to six top sources and describe the solutions to the deblending challenges.

Shallow stratigraphy in Southern Oman is characterized by the presence of a shallow anhydrite layer causing a strong velocity inversion which makes seismic imaging particularly tedious. This known shallow sharp velocity inversion cannot be easily captured with reflected waves based techniques or even acoustic full waveform inversion. We propose to recover it applying multi-wave inversion, an approach combining information from first breaks of P waves and from dispersion curves picked on ground-roll. In addition, an unsupervised machine learning technology is used to improve the quality of surface wave dispersion curve picks, crucial for the reliability of the multi-wave inversion results. With this innovative approach, the joint inversion of first breaks and surface waves leads to an enhanced and high resolution P-wave velocity model of the near surface along with an improvement in the depth imaging.