Abstract
The massive cost of 3D acquisition calls for methods to reduce the number of receivers by designing optimal receiver sampling masks. Recent studies on 2D seismic showed that maximizing the spectral gap of the subsampling mask leads to better wavefield reconstruction results. We enrich the current study by proposing a simulation-free method to generate optimal 3D acquisition by maximizing the spectral gap of the subsampling mask via a simulated annealing algorithm. Numerical experiments confirm improvement of the proposed method over receiver sampling locations obtained by jittered sampling.
Introduction
Acquiring 3D seismic data is essential for imaging the Earth's subsurface for energy exploration and carbon storage, yet the process is prohibitively expensive. To reduce costs, geophysicists often employ sparse sampling strategies, placing fewer receivers than a full grid requires. The missing data is subsequently reconstructed using matrix completion techniques. However, the success of this reconstruction is highly sensitive to the spatial distribution of the receivers.
Traditional methods for designing optimal survey layouts typically rely on computationally intensive wave simulations to evaluate performance. This paper introduces a mathematically rigorous, simulation-free framework that optimizes survey geometry by maximizing the "spectral gap" of the sampling mask. This approach allows for high-quality wavefield reconstruction without requiring prior knowledge of the underlying geology.
Methods
The researchers leverage a concept from graph theory: the spectral gap. For a binary sampling mask—a matrix where 1s represent receiver locations and 0s represent empty grid points—the Spectral Gap Ratio (SGR), defined as the ratio of the second-largest singular value to the largest ($\sigma_2/\sigma_1$), measures the connectivity of the sampling graph. A smaller SGR indicates a better-connected mask, which is theoretically more suitable for matrix completion.
Optimization Framework and the Single-Source Shortcut
The core of the methodology is a minimization problem for the SGR, subject to constraints such as the total number of receivers and the avoidance of empty rows or columns in the data matrix. A significant technical breakthrough presented is the "Single-Source Shortcut." The authors proved that if sources are fully sampled, the singular values of the massive global 3D sampling matrix are identical to those of a much smaller single-source receiver mask.
This discovery reduces the computational burden from massive matrices to a manageable $100 \times 100$ matrix, allowing the optimization to be performed using a simulated annealing algorithm on standard hardware in minutes.
Results
The proposed method was evaluated using the synthetic 3D Compass dataset, a standard benchmark featuring 1,681 sources and 10,000 potential receiver locations. To test the limits of the reconstruction, 90% of the receivers were removed.
The optimized mask achieved a significant reduction in the Spectral Gap Ratio, dropping from 0.507 (for standard jittered sampling) to 0.328. This mathematical improvement translated directly into superior reconstruction quality. When reconstructing a frequency slice at 16.8Hz, the optimized layout achieved a Signal-to-Noise Ratio (SNR) of 12.27 dB, compared to 10.88 dB for the jittered layout.
The approximately 1.4 dB gain in SNR represents a meaningful improvement in seismic processing, potentially enabling the identification of thin reservoir layers that would be missed using standard sampling techniques.
Discussion and Conclusion
The success of this method underscores that seismic reconstruction quality is driven by the connectivity of the sampling graph. By focusing on geometry rather than geology, the method remains robust across various geological settings.
While the current framework assumes fully sampled sources—common in marine Ocean Bottom Node (OBN) surveys—future work will focus on extending this optimization to joint source-receiver configurations. Additionally, the impact of different data matrix organizations (e.g., non-canonical folding) remains an area for further investigation. Ultimately, this simulation-free approach offers a computationally efficient path toward more cost-effective and accurate 3D seismic acquisition.
Summary updated