Search

In areas of complex geology, strong impedance contrasts can generate highly elastic behavior leading to the breakdown of the acoustic assumption in FWI. In these cases, accurate Earth models cannot be obtained without considering the elastic wave equation. Many previous studies of elastic FWI typically focused on deep water areas where large salt structures generate elastic effects in the observed seismic data. In this article, we offer an alternative perspective on elastic FWI, highlighting its value over acoustic FWI in properly describing elastic effects observed in shallow-water environments, particularly around the chalk packages which are prevalent throughout the North Sea.

Erling Frantzen, Anna Lougon & Joe Zhou from our Earth Data team discuss how advanced geoscience technologies and data are successfully supporting energy companies in the search for hidden opportunities in the world’s hotspots.

Until now, noise attenuation and interpolation processes based on rank reduction needed spatially regular, or at least binned, data. Here, we show how the low-rank signal model in joint low-rank and sparse inversion (JLRSI), a recently proposed convex optimization framework for simultaneous random plus erratic noise attenuation and interpolation, can be extended to spatially irregular data by appropriately modifying the inverse problem formulation. Benefits of considering the true spatial locations of seismic traces for the quality of the signal reconstruction are illustrated on a three-dimensional regularization and interpolation example on real land data.

Brazilian pre-salt often exhibits geologically complex settings with many challenges for oil and gas exploration and production. One of these key challenges is the presence of igneous rocks. We analyze the ability of full-waveform inversion (FWI) to identify and detect this class of rock using an OBN data set from the Santos Basin. Comparisons between acoustic and elastic FWI are studied. Thanks to the improved physics, elastic FWI shows more accurate representations of the distribution and plumbing systems of igneous rocks in the pre-salt. Moreover, the resulting FWI Images show uplift over the classical reverse-time migration (RTM) images that have historically been used to image the field.

Full waveform inversion (FWI) is becoming an increasingly dominant force in velocity model building and seismic imaging, often providing us with unrivaled focusing and resolution of the subsurface image. The superior velocity models and seismic images achieved through FWI also enable serious and more meaningful discussions of uncertainties on FWI. While the Bayesian inference framework offers a general foundation for FWI uncertainty analysis, the high computational cost associated with classical algorithms, such as Monte Carlo methods, to sample the posterior distribution prohibits it from being applied to industrial-scale problems. The recent development of variational inference (VI) theory presents a promising alternative to traditional sampling algorithms, as it can generate reasonable estimations of the posterior distribution at a more affordable computational cost. In this abstract, we describe an FWI uncertainty analysis method based on a specific type of VI algorithm, the Stein variational gradient descent (SVGD). We demonstrate the efficacy and practicality of this method through 2D synthetic and 3D real data examples

Land seismic data in the Sultanate of Oman presents more and different challenges for full-waveform inversion (FWI) compared to marine data. This is mainly due to complex near-surface effects that create strong ground roll, which obscures already poor data quality, especially at lower frequencies. It also generates heavy multiple reverberations, particularly in the south of Oman, owing to very different interleaved lithologies that create shallow sharp velocity contrasts. This paper discusses the substantial improvements made to current methodologies in mitigating the effects of strong shallow heterogeneities. This results in more detailed high-resolution models that capture the complex velocity variations essential for accurate imaging.

Full-waveform inversion (FWI) has demonstrated tremendous potential to provide high-resolution subsurface models. However, obtaining an accurate FWI velocity model can still be challenging in steep continental slope shelf geologic environments. The uncertainties, often caused by shallow abrupt velocity contrasts, limited diving-wave penetration depth, and weaker illuminations instill less confidence in reservoir depth mapping. In this case study from offshore Senegal, we present a visco-acoustic dynamic resolution time-lag FWI (DR-TLFWI) workflow to construct a more accurate high-resolution model. The newly updated model further reduces residual undulations, which improves gather flatness and event continuity. The cumulative improvements increase the interpretation confidence and enable better reservoir delineation in this area.

Historically, Brazilian pre-salt fields have been imaged by narrow-azimuth towed-streamer (NATS) data. More recently, ocean bottom node (OBN) acquisitions have been employed to image and monitor already producing fields. In cases as such, 4D seismic monitoring is only possible via NATS/OBN hybrid pairs, where NATS and OBN data sets serve as base and monitor, respectively. This hybrid configuration implies low repeatability. Extracting subtle 4D changes in hard pre-salt carbonate reservoirs can be challenging. Considering this, we propose a joint 4D full-waveform inversion (FWI) formulation to overcome low repeatability between different seismic surveys. We demonstrate promising results using a field data set from Santos Basin, where realistic 4D anomalies can be identified.

Surface-related multiple elimination (SRME) is an effective and widely used tool for the attenuation of multiple energy in marine seismic data. However, removing multiples with SRME is still challenging in geologically complex areas, and we often observe residual multiples in such settings. Residual multiples can be caused by poor sampling of near offsets, low signal-to-noise ratio (S/N), and cross-talk among different orders of multiples. To remove strong residual multiples, we propose a targeted SRME flow where we demigrate the generators of the strongest residual multiples and predict an SRME model with demigrated data as input. We present applications of this method on two field data examples in the deepwater Gulf of Mexico, where the targeted residual multiples are effectively attenuated.