In the realm of energy exploration, understanding the subsurface geology is crucial for locating potential oil and gas reserves. A team of researchers from King Abdullah University of Science and Technology (KAUST), including Ning Wang, Matteo Ravasi, and Tariq Alkhalifah, has developed a method to enhance seismic imaging, which is a key tool in this endeavor. Their work was recently published in the journal Geophysics.
Seismic imaging involves sending sound waves into the earth and analyzing the echoes that bounce back. This process helps create detailed pictures of the subsurface, but it can be hindered by multiple reflections, or “multiples,” which can obscure the true geological structure. The Upside-Down Rayleigh-Marchenko (UD-RM) method has emerged as a powerful technique to address this issue, particularly for ocean-bottom seismic data. However, its widespread use has been limited by the high computational cost required to estimate the focusing functions, which are essential for creating accurate images.
To overcome this challenge, the researchers proposed a self-supervised learning approach to accelerate the estimation of these focusing functions. They used a type of neural network called a U-Net, which was trained on a small subset of image points from within the target area. The network was tasked with predicting both the up- and down-going focusing functions from an initial estimate of the subsurface wavefields. Once trained, the network could generalize to other imaging locations, enabling direct prediction of the focusing functions.
The researchers validated their method on a synthetic dataset with both dense and sparse receiver sampling, using progressively fewer training points. In both cases, the resulting images closely matched those obtained from the conventional UD-RM method, but at a much lower computational cost. The method also significantly outperformed mirror migration, another seismic imaging technique, when the same input dataset was used. Finally, an application to the Volve field data confirmed the method’s robustness in practical scenarios.
This research underscores the potential of machine learning techniques to enhance seismic imaging, enabling more efficient and accurate exploration of subsurface resources. By reducing the computational cost of the UD-RM method, the proposed approach could facilitate its widespread application in the energy industry, ultimately aiding in the discovery and extraction of oil and gas reserves. The research was published in the journal Geophysics, a publication of the Society of Exploration Geophysicists.
This article is based on research available at arXiv.