3D Fault Imaging Enhanced by Frequency-Dependent CNN Models and Fault Network Analysis: A Barents Sea Example

This study explores the potential of deep learning techniques to enhance the definition of fault imaging in 3D seismic data. Acknowledging the limitations imposed by conventional ML methodologies applied to complex structural geological scenarios, it is proposed a new approach that integrates a frequency-dependent convolutional neural network (CNN) model with point-cloud-based fault network analysis. The aim of this integration is to reduce uncertainty around traditional full bandwidth predictions by improving the accuracy of structural interpretations. Utilising a real example from the Loppa High structure in the southwestern Barents Sea, it is demonstrated how a frequency-dependent CNN would outperform a full-bandwidth model in capturing large- scale faults featuring intricate branches, from which multiple variations in length, direction, and fault density can be identified. Moreover, the enhanced fault imaging and the findings of the fault network analysis are in alignment with the known structural complexity of the study area, which is characterised by multiple episodes of extension and structural inversion throughout geologic time. The study demonstrates the efficacy of this integrated approach, highlighting its potential to improve fault imaging as well as the structural consistency of subsurface predictions, with significant implications for exploration, reservoir characterisation, and the assessment of alternative energy resources.