Back to search


Optimizing the Use of Marine Seismic Sources

Alternative title: Optimering av Marine Seismiske Kilder

Awarded: NOK 1.7 mill.

Project Manager:

Project Number:


Project Period:

2019 - 2021

Funding received from:


Volodya Hlebnikov's Ph.D. thesis, January 2022 Between 2019 and 2021 Volodya Hlebnikov did an industrial Ph.D. on seismic processing solutions based on Machine Learning, employed, and supervised by CGG with financial support from the Research Council of Norway with academic affiliation to the University of Oslo (UiO). The aim of seismic acquisition and processing is to obtain high quality three-dimensional images of the subsurface, mainly for oil and gas exploration, but also for other purposes such as engineering, monitoring of carbon storage or reducing geohazard. In Marine seismic acquisition a designated boat is towing several active sources, typically air-gun clusters, which emit acoustic wave-fields reflected from the water bottom and the rock layers underneath and detected by hydrophones in streamers that are several kilometers long. One of the major challenges in seismic processing is optimal reconstruction of the sparsely sampled wave field from the acquisition. The central theme of Volodya's Ph.D. thesis is this reconstruction (i.e., interpolation) of the wave-field both in-between the streamers, as well as between the sources using Machine Learning in the form of Convolutional Neural Networks. Access to state-of-the-art seismic processing tools and a wealth of data and experience in CGG have been a great advantage in Volodya's Ph.D. in the search for improved solutions to the interpolation problem. The main outcome of this research activity is a novel approach for interpolation of missing data in offset classes resulting in improved imaging, both in synthetic and real data examples. The approach is currently tested in CGG's standard workflow by experienced geophysicists, and a patent application "Modeling-based Machine Learning for seismic processing" was submitted in August 2021. Volodya's thesis, submitted 31st January 2022, consist in the following main elements (CNN stands for Convolutional Neural Network): 1: A detailed description of a wide range of marine noise types in towed streamer data. This includes some widely adopted techniques for attenuating noise, and some specific tricks applied in industrial de-noising. This work was published in Geophysics in 2021 and has the title "Noise types and their attenuation in towed marine seismic: A tutorial" 2: Development of CNN approaches for interpolation in-between streamers and shot points. These approaches are based on training of a NN based on a similarity between the spatial dimension of the seismic data, for instance along the streamers vs perpendicular to them, or between shot gathers and receiver gathers. The work demonstrates that a CNN can be trained in one dimension and used in another. A part of this work was published in an article in Geophysics in 2020 with title "Cross-streamer wavefield reconstruction through wavelet domain learning". This article was co-authored by another industrial Ph.D. student, Thomas Greiner from Lundin/UiO. 3: Development of CNN approaches for interpolation and regularization of irregularly and sparsely sampled offset classes Two main training approaches were tried: (1) using the conventional method as a "teacher" and (2) using de-migrated (i.e., modelled) data. Both approaches were tested with a range of neural networks, both 2D and 3D. Approach 2 provided improved interpolation and images compared to state-of-the-art methods. The general de-migration-based training approach to be used in several processing steps in addition to interpolation (for instance de-ghosting, de signature, de-bubbling and de-multiple) was submitted as a patent in August 2021. The following paper was submitted to Geophysical Prospecting in October 2021: "De-migration-based supervised learning for interpolation and regularization of 3D offset classes". The paper received positive feedback from the three reviewers who recommended publication after minor review.

Hovedresultatet av denne forskningsaktiviteten er en ny tilnærming for interpolering av manglende data i offsetklasser har ført til forbedret avbildning av undergrunnen, både i syntetiske og ekte dataeksempler. Metoden er testet i CGGs standardarbeidsflyt av erfarne geofysikere, og en patentsøknad «Modelleringsbasert maskinlæring for seismisk prosessering» ble sendt inn i august 2021. Denne metoden gir bedre resultater enn den klassiske interpolasjonsmetoden basert på Fouriertransformasjoner. Den er også mer tidseffektiv. Metoden er i ferd med å testes i CGG?s profesjonelle prosesseringsgrupper der den kan bane veien for å bruke neurale nettverk isteden for tradisjonelle dataprogrammer for å utføre interpolasjon av glisne seismiske data.

As a leader on the market, CGG offers state-of-the-art marine seismic acquisition technologies, developed to address specific imaging objectives. Seismic imaging is a geophysical technique that investigates the subsurface and gives vital information about its properties. Such features can give oil and gas companies a more astute indication if a prospect area contains hydrocarbons. The quality of the seismic data therefore is crucial. The marine seismic air-gun source plays a key role in seismic imaging. This project focuses on a novel use of marine seismic sources. Conventional marine seismic surveys today are conducted by a vessel towing two to three seismic sources combined in source arrays. The aim of this work, based on new research effort, is to move towards an acquisition setup where several distributed sources will be used instead. The actual configuration of this multi-source layout such as number of sources, positioning, size (volume), firing rate, steering and other criteria is of utmost importance. Practically speaking, these will be analyzed in detail by conducting modeling studies. In such a way, the R&D team will be able to understand the impact on the final seismic image. The multi-source concept aims to improve the image quality. This is possible as such kind of acquisition configuration brings the missing 'near offset' coverage. Additionally, by having more widely-spaced sources deployed, a larger subsurface area is covered as opposed to conventional survey. In this way, the acquisition efficiency is improved. Based on the work following this project, the knowledge and expertise about the multi-source acquisition will be improved. This knowledge and expertise is even more relevant in areas where dealing with subtle and complex prospects such as in the Barents Sea and North Sea. As a result, the anticipated outcome is to propose an 'optimal' source configuration for a specific geophysical imaging solution.

Funding scheme: