Accurate and efficient seismic data interpolation in the principal frequency wavenumber domain

Benfeng Wang, Wenkai LuJournal of Geophysics and Engineering 2017 · DOI

Abstract

Seismic data irregularity caused by economic limitations, acquisition environmental constraints or bad trace elimination, can decrease the performance of the below multi-channel algorithms, such as surface-related multiple elimination (SRME), though some can overcome the irregularity defects. Therefore, accurate interpolation to provide the necessary complete data is a pre-requisite, but its wide applications are constrained because of its large computational burden for huge data volume, especially in 3D explorations. For accurate and efficient interpolation, the curvelet transform- (CT) based projection onto convex sets (POCS) method in the principal frequency wavenumber (PFK) domain is introduced. The complex-valued PF components can characterize their original signal with a high accuracy, but are at least half the size, which can help provide a reasonable efficiency improvement. The irregularity of the observed data is transformed into incoherent noise in the PFK domain, and curvelet coefficients may be sparser when CT is performed on the PFK domain data, enhancing the interpolation accuracy. The performance of the POCS-based algorithms using complex-valued CT in the time space (TX), principal frequency space, and PFK domains are compared. Numerical examples on synthetic and field data demonstrate the validity and effectiveness of the proposed method. With less computational burden, the proposed method can achieve a better interpolation result, and it can be easily extended into higher dimensions.

Introduction

In seismic exploration, "perfect" data is rarely achieved. Physical obstacles, budget constraints, and equipment failure lead to missing traces, resulting in irregularly sampled datasets. This irregularity is a major obstacle for downstream processing tasks like imaging or noise elimination, which require dense, regularly sampled grids. While many interpolation methods exist, they often face a trade-off between accuracy and computational cost. Wave-equation methods are accurate but computationally massive, while prediction filters struggle with complex, curved seismic events. Sparse transform methods, such as those using Curvelets, are effective but often too slow for the massive data volumes involved in 3D exploration.

This paper introduces a highly efficient method for filling in missing seismic data by performing interpolation in the Principal Frequency Wavenumber (PFK) domain. By combining the Curvelet Transform with a "Projection onto Convex Sets" (POCS) algorithm, the proposed method achieves significant speedups while simultaneously improving reconstruction accuracy.

The PFK Domain and POCS Framework

The core innovation lies in the transition from the standard time-space (TX) domain to the PFK domain. By selecting only the "principal" frequency (PF) components that contain the signal, the data volume is immediately reduced by at least 50%. This dimensionality reduction is the primary driver of the method's efficiency.

Comparison of seismic data in (a) time-space (TX), (b) frequency-space (FX), and (c) frequency-wavenumber (FK) domains.

In the PFK domain, the spatial irregularity of the original data is transformed into incoherent noise. Because the Curvelet Transform (CT) is specifically designed to represent curved edges (like seismic reflections) very sparsely, it is exceptionally effective at separating the coherent signal from this incoherent noise in the PFK domain. This makes the interpolation more accurate than in the standard TX domain.

Algorithm Implementation

The proposed method utilizes an iterative POCS framework. In this process, the model iteratively "guesses" the missing data, transforms it to check for sparsity using Curvelets, and then "projects" it back to ensure the known data points remain unchanged.

Flowchart of the POCS-based interpolation algorithm in the PFK domain.

As shown in the flowchart, the algorithm involves taking the FFT along the time axis, selecting the PF band, and then iteratively applying the Curvelet thresholding operator ($T_{\lambda}$) in the PFK domain. This iterative refinement continues until the model converges or reaches a set number of iterations ($N$).

Key Results and Performance

The PFK method was evaluated against traditional Time-Space (TX) and Principal Frequency-Space (PFX) methods using both synthetic and field datasets:

  • Efficiency: On synthetic data, the PFK method demonstrated a roughly 6x increase in efficiency, taking only 17.92 seconds compared to 111.03 seconds for the TX domain method.
  • Accuracy: The PFK method achieved the highest Signal-to-Noise Ratio (SNR). For synthetic data, it reached 27.88 dB, significantly higher than the TX method's 24.45 dB.
  • Visual Fidelity: Residual plots confirmed that the PFK method left behind the least amount of unrecovered signal, particularly in complex curved areas, and effectively suppressed the "ringing" artifacts often seen in PFX-domain interpolation.

Conclusion

By operating in the PFK domain, the proposed POCS-based interpolation method offers a "just right" solution for seismic data reconstruction. It provides the accuracy required for complex geological structures while maintaining the efficiency necessary for large-scale 3D datasets. Future work will focus on extending this framework to handle simultaneous interpolation and denoising in real-world, noisy environments.

Summary updated

Abstract

Seismic data irregularity caused by economic limitations, acquisition environmental constraints or bad trace elimination, can decrease the performance of the below multi-channel algorithms, such as surface-related multiple elimination (SRME), though some can overcome the irregularity defects. Therefore, accurate interpolation to provide the necessary complete data is a pre-requisite, but its wide applications are constrained because of its large computational burden for huge data volume, especially in 3D explorations. For accurate and efficient interpolation, the curvelet transform- (CT) based projection onto convex sets (POCS) method in the principal frequency wavenumber (PFK) domain is introduced. The complex-valued PF components can characterize their original signal with a high accuracy, but are at least half the size, which can help provide a reasonable efficiency improvement. The irregularity of the observed data is transformed into incoherent noise in the PFK domain, and curvelet coefficients may be sparser when CT is performed on the PFK domain data, enhancing the interpolation accuracy. The performance of the POCS-based algorithms using complex-valued CT in the time space (TX), principal frequency space, and PFK domains are compared. Numerical examples on synthetic and field data demonstrate the validity and effectiveness of the proposed method. With less computational burden, the proposed method can achieve a better interpolation result, and it can be easily extended into higher dimensions.
BibTeX
@article{wang2017accurate,
  author = {Benfeng Wang and Wenkai Lu},
  title = {{Accurate and efficient seismic data interpolation in the principal frequency wavenumber domain}},
  journal = {Journal of Geophysics and Engineering},
  year = {2017},
  doi = {10.1088/1742-2140/aa82dc},
}
Figure