High performance velocity analysis for multi-AVO seismic data

Yankai Xu, Jiawei Li, Jiao Qi, Siyuan Cao, Hongwei Liu, W. Li, Hongduo ZhuResearch Square 2024 · DOI

TL;DR

Researchers have developed a new method called RBSL (RSVD-based Symbolic-Local semblance) to sharpen seismic velocity analysis. By combining a robust sign-correction technique with Singular Value Decomposition (SVD), the method successfully maps underground structures even when seismic signals flip polarity multiple times or are buried under heavy noise.

What Problem Does This Paper Address?

In seismic exploration, "velocity analysis" is the process of determining how fast seismic waves travel through different rock layers. This is crucial for creating clear images of the subsurface. The standard tool for this is semblance, which measures the coherence of signals across different recording sensors (offsets).

However, two major headaches plague geophysicists:

  1. AVO (Amplitude Variation with Offset): Sometimes, the signal's strength changes or its polarity flips (positive to negative) as the distance from the source increases. Conventional semblance assumes the signal stays consistent; when it flips, the math fails, and the "energy peaks" on the velocity map disappear.
  2. Noise: Real-world field data is messy. High noise levels can create "ghost" energy clusters, making it impossible to pick the correct velocity.

While previous methods like SBWS (Sign-Based Weighted Semblance) tried to fix polarity flips, they often fall apart when noise is introduced. This paper introduces a way to handle both "multi-AVO" (multiple polarity reversals) and high noise simultaneously.

Methods

The authors propose the RBSL method, which is a hybrid of three clever components:

  • Improved Symbolic Weighting ($S_k$): Instead of looking at a single trace to determine polarity, the algorithm looks at three adjacent traces. It uses a logic similar to a median filter to ensure that a random spike of noise doesn't trick the system into thinking a polarity reversal occurred.
  • Local Similarity: Rather than a global average, the method calculates trace-by-trace correlation coefficients. This provides much higher resolution than traditional "stacked" methods.
  • RSVD (Reduced Singular Value Decomposition) Weighting: This is the "noise killer." The algorithm performs SVD on the data within a small time window. Since the actual seismic signal is usually the most "spatially dependent" feature, it shows up in the first singular value. By calculating the ratio between the first and second singular values, the authors create a weighting function that boosts the signal and suppresses random noise.

Datasets

The authors validated their method using:

  • Synthetic Models: A five-layer reflector model and specific test cases featuring "No AVO," "Single AVO," and "Double AVO" (two polarity flips) to test the limits of the algorithm.
  • Field Data: Real seismic gathers consisting of 70 traces.
  • Modified Field Data: To prove robustness, they took real field data and artificially added "double AVO" reversals to see if the algorithm could still find the truth amidst real-world noise.

Key Insights and Ideas

  • The Power of Neighbors: The "aha moment" for polarity correction was moving from single-trace analysis to a three-trace window. This prevents the "jitter" seen in previous methods (like SBWS) where noise was mistaken for signal reversals.
  • SVD as a Mathematical Scalpel: By using the ratio of singular values ($R_r$), the researchers found a way to mathematically "ignore" noise. If the first singular value isn't significantly larger than the second, the system recognizes it as noise and down-weights that part of the spectrum.
  • Synergy: The symbolic correction fixes the physics problem (polarity flips), while the RSVD fixes the data quality problem (noise). Together, they produce a velocity spectrum with much tighter, more accurate energy clusters.

Key Results

  • Resolution: In synthetic tests with two closely spaced reflectors, RBSL was the only method capable of distinguishing the slight velocity differences between them.
  • Anti-Noise Performance: In noisy scenarios where conventional semblance and SBWS produced scattered, unreadable "clouds" of energy, RBSL maintained a continuous, sharp velocity curve.
  • Field Success: On real data with artificial AVO reversals, the RBSL method showed the "best velocity resolution" and successfully retained the velocity distribution where other methods showed significant energy deficits or lateral divergence.

Limitations and Future Work

  • Parameter Sensitivity: The RSVD weighting function relies on two constants ($a$ and $b$) that control the slope and threshold of noise suppression. In the paper, these were manually tuned for different datasets (e.g., $a=10, b=9$ for clean data vs. $a=8, b=6$ for noisy data). Future work could involve automating the selection of these parameters.
  • Computational Intensity: Performing SVD on every time window during a velocity scan is more computationally expensive than traditional semblance. While the paper focuses on "high performance" in terms of accuracy, the raw processing time for massive 3D datasets remains a practical consideration.

Reference: Xu, Y., et al. (2024). High performance velocity analysis for multi-AVO seismic data. Research Square / Acta Geophysica. DOI: 10.21203/rs.3.rs-4279087/v1

Summary updated

Abstract

<title>Abstract</title> Improving the velocity spectrum resolution is challenging because the field data contains some noise and polarity inversion in many cases. Although the AB semblance and local similarity method can deal with the seismic data with single polarity reversal, they do not perform as well for noisy and multi-AVO data. In this paper, we proposed a symbolic-weighted velocity analysis velocity spectrum analysis method to solve these problems. Firstly, to correct the polarity of multiple AVO seismic data, we adopt a technology using the symbolic information between adjacent traces. Then, considering the singular value distribution characteristics of seismic data in velocity scanning, the RSVD algorithm which is a modified mode of the singular value decomposition is used to improve the recognition sensitivity of hyperbolic events and suppress the influence of noise. Synthetic and field examples demonstrate that, compared with conventional semblance, local semblance and SBWS, the proposed method has higher precision and stronger anti-noise performance for the noisy multi-AVO seismic data.
BibTeX
@inproceedings{xu2024high,
  author = {Yankai Xu and Jiawei Li and Jiao Qi and Siyuan Cao and Hongwei Liu and W. Li and Hongduo Zhu},
  title = {{High performance velocity analysis for multi-AVO seismic data}},
  booktitle = {Research Square},
  year = {2024},
  doi = {10.21203/rs.3.rs-4279087/v1},
}

Loading related content...

Figure