28 Nov 2022
28 Nov 2022
Status: this preprint is open for discussion.

Complex fault system revealed from 3-D seismic reflection data with deep learning and fault network analysis

Thilo Wrona1,2, Indranil Pan3,4,5, Rebecca Bell6, Christopher A.-L. Jackson7, Robert Gawthorpe1, Haakon Fossen8, Edoseghe Osagiede1, and Sascha Brune2,9 Thilo Wrona et al.
  • 1Department of Earth Science, University of Bergen, Allégaten 41, N-5007 Bergen, Norway
  • 2GFZ German Research Centre for Geosciences, Telegrafenberg, 14473 Potsdam, Germany
  • 3Centre for Process Systems Engineering & Centre for Environmental Policy, Imperial College London, UK
  • 4The Alan Turing Institute, British Library, London, UK
  • 5School of Mathematics, Statistics & Physics, Newcastle University, UK
  • 6Basins Research Group (BRG), Department of Earth Science and Engineering, Imperial College, Prince Consort Road, London, SW7 2BP, UK
  • 7Department of Earth and Environmental Sciences, University of Manchester, Manchester, UK
  • 8Museum of Natural History, University of Bergen, Allégaten 41, N-5007 Bergen, Norway
  • 9Institute of Geosciences, University of Potsdam, Potsdam-Golm, Germany

Abstract. Understanding where normal faults are is critical to an accurate assessment of seismic hazard, the successful exploration for and production of natural (including low-carbon) resources, and for the safe subsurface storage of CO2. Our current knowledge of normal fault systems is largely derived from seismic reflection data imaging intra-continental rifts and continental margins. However, exploitation of these data is limited by interpretation biases, data coverage and resolution, restricting our understanding of fault systems. Applying supervised deep learning to one of the largest offshore 3-D seismic reflection data sets from the northern North Sea allows us to image the complexity of the rift-related fault system. The derived fault score volume allows us to extract almost 8000 individual normal faults of different geometries, which together form an intricate network characterised by a multitude of splays, junctions and intersections. Combining tools from deep learning, computer vision and network analysis allows us to map and analyse the fault system in great detail and a fraction of the time required by conventional interpretation methods. As such, this study shows how we can efficiently identify and analyse fault systems in increasingly large 3-D seismic data sets.

Thilo Wrona et al.

Status: open (until 26 Jan 2023)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on egusphere-2022-1190', Lukas Mosser, 22 Jan 2023 reply

Thilo Wrona et al.

Thilo Wrona et al.


Total article views: 235 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
174 54 7 235 2 3
  • HTML: 174
  • PDF: 54
  • XML: 7
  • Total: 235
  • BibTeX: 2
  • EndNote: 3
Views and downloads (calculated since 28 Nov 2022)
Cumulative views and downloads (calculated since 28 Nov 2022)

Viewed (geographical distribution)

Total article views: 222 (including HTML, PDF, and XML) Thereof 222 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
Latest update: 26 Jan 2023
Short summary
We need to understand where faults are to (1) assess their seismic hazard, (2) to explore for natural resources and (3) to store CO2 safely in the subsurface. Currently we still map faults manually using seismic data i.e. acoustic images of the subsurface. Mapping these images is however difficult and time-consuming. Here we show how to use deep learning and network analysis to accelerate and simplify fault mapping.