Capstone: Ultra-high-frequency near-surface imaging

The seismic processing techniques taught in Parts 0–9 scale from exploration (km-to-sub-km targets) down to engineering geotechnics (metre-scale targets) with relatively minor adjustments to the frequency band and sampling. This capstone walks a typical near-surface survey: locating voids, buried pipelines, and shallow faulting in the top 100 m using a 3 kHz source.

Project setup

Geotechnical survey before a proposed infrastructure project. 500 m 2D line, 2 m geophone spacing, 3 kHz accelerated weight drop source. Sample rate 0.1 ms, record length 200 ms. Targets: voids 2–5 m in diameter, pipeline diameter 0.3–1 m at 1–15 m depth, faults with 0.5–2 m throw. Low-SNR environment (surface traffic, construction, machinery noise within earshot).

The pipeline

Differences from exploration-scale seismic

• Bandwidth. 100–3000 Hz (vs 5–100 Hz for marine exploration). Thin-bed resolution of ≤10 cm depth is achievable.
• Data volume. Each shot ≤ 200 ms at 0.1 ms = 2000 samples. Small per-shot but many shots (~1000 for a dense line) and high throughput expected.
• Q is the enemy. At 3 kHz and Q=50, amplitudes attenuate by 24 dB in 100 ms. Q-compensation (§7.3) is aggressive.
• Near-surface directly IS the target. Refraction tomography isn't just a static correction; it IS the primary product. Every picked first break matters.

ML disproportionately valuable

Near-surface seismic is disproportionately benefited by ML because:

• Noise is heavy, variable, and coherent (traffic, machinery, wind). Classical f-x decon struggles; CNN denoisers excel.
• First-break pick counts are huge (every trace in every shot); ML automation saves days of interpreter time per project.
• Training data exists: many geotechnical companies have collected reference surveys for years. Pre-trained models are readily available.

PSDM at meter scale

Post-stack time migration is inadequate for metre-scale targets: the near-surface is strongly heterogeneous laterally, and time-domain velocities smooth over exactly the features we want to resolve. Pre-stack depth migration at 0.5–1 m grid resolution produces depth sections where 2 m voids appear as distinct low-amplitude anomalies against the background.

Where this goes next

§10.6 returns to 4D processing with the most stressful case: comparing 2006 legacy streamer data to 2024 OBN monitor data. Heterogeneous acquisitions force every trick in the Part 8 arsenal to produce a useful 4D difference.

References

• Yilmaz, Ö. (2001). Seismic Data Analysis (2 vols.). SEG.
• Claerbout, J. F. (1976). Fundamentals of Geophysical Data Processing. McGraw-Hill.
• Oppenheim, A. V., Schafer, R. W. (2009). Discrete-Time Signal Processing (3rd ed.). Prentice Hall.
• Sheriff, R. E., Geldart, L. P. (1995). Exploration Seismology (2nd ed.). Cambridge UP.