Marine acquisition: airguns, streamers, OBC/OBN

Marine surveys look deceptively simple: tow a source and a long cable of receivers behind a boat. Physically, the setup is cleaner than land; computationally, it imposes a signature filter that shapes the bandwidth of every trace and needs a dedicated processing step to remove. That filter is the ghost.

1. The standard towed-streamer setup

• Source. An array of airguns towed at 5 to 12 m depth. Each gun releases compressed air into the water, producing a bubble that collapses and re-oscillates. Arrays are tuned (sub-arrays at different depths, different volumes) so the bubble harmonics destructively interfere and the effective source is a short broadband pulse.
• Streamers. Long, neutrally buoyant cables (5 to 10 km) holding pressure-sensitive hydrophones at 6 to 15 m depth. A modern survey tows 8 to 20 streamers in parallel, 50 to 100 m apart.
• Recording. Each hydrophone digitizes 2 to 8 s of data after each shot. Thousands of channels × thousands of shots → terabytes per survey.

2. The ghost

The sea surface is an almost perfect mirror for seismic waves: acoustic impedance of air is essentially zero, so the reflection coefficient at the water-air interface is close to $R = -1$. Every pulse emitted by the source travels downward and upward; the upward one bounces off the surface and comes back down, inverted in polarity, delayed by the extra round trip through the water above the source: $\\Delta t\_s = 2z\_s/c$ at vertical incidence, or $2z\_s\\cos\\theta/c$ for a wave leaving at angle $\\theta$ from vertical, which is 8 ms for a 6 m tow. That delayed inverted copy is the source ghost.

The same thing happens at the receiver side: every upcoming wave arrives at a hydrophone, and a delayed, inverted copy, having bounced off the surface, arrives shortly after. That is the receiver ghost.

3. Ghost operator in the frequency domain

In the frequency domain, the source ghost multiplies the pulse by

where $z\_s$ is the source depth and $c$ is the speed of sound in water ($c \\approx 1500$ m/s). The magnitude is

which, at vertical incidence, is zero at

The notch at $n = 0$ is there whatever the tow depth. For a wave travelling at angle $\\theta$ from vertical every $z$ becomes $z\\cos\\theta$, so the notches move up to $f\_n = n\\,c/(2z\\cos\\theta)$. And the sea is never a perfect mirror: a rough surface has $|R|$ a little below 1, and its notches are only partly deep.

The receiver ghost does the same at the streamer depth, creating its own notch family. The total ghost response is the product of the source and receiver ghosts. In the figure you set the two tow depths, tilt the arrival and roughen the sea, then read the notches and the low-frequency loss off the recorded spectrum.

The figure opens on a common tow, the source at 6 m and the streamer at 8 m. The first notch is the streamer’s, at 93.75 Hz, and the two ghosts together take 21.5 dB at 5 Hz. Pick the shallow tow, 5 and 6 m, and the first notch rises to 125 Hz, above most of the signal band, but the low end pays: 25.6 dB lost at 5 Hz. Pick the over-deep tow, 20 and 22 m, and the loss at 5 Hz falls to 2.8 dB while the first notch drops to 34.09 Hz, punching a hole in the middle of the band. Tilt the arrival to 30° and the streamer’s notch climbs to 108.3 Hz, because the extra path is only $2z\\cos\\theta$. Every survey designer weighs this trade.

4. Why low frequencies matter

Low frequencies are the first thing you lose to the ghost. They are also the hardest to recover by deconvolution, because the ghost response falls to zero as $f \\to 0$ (it is about $4\\pi f z/c$ at low frequency): dividing by a number that small amplifies noise far more than signal. And low frequencies are exactly what FWI and broadband impedance inversion need most (Sections 6.2 and 7.4). This is why modern survey design agonizes over the first notch.

5. Ways around the trade: over/under, multi-sensor and de-ghosting

• Over/under cables. Two streamers at different depths; their ghost notches do not coincide, so a least-squares combination fills the notches of either, everywhere except where both ghosts are small at once (and always at 0 Hz).
• Multi-sensor streamers. Record pressure with hydrophones and vertical particle motion with motion sensors at the same station. The surface ghost has opposite polarity in the two records, so their notches interleave (pressure at $n\\,c/(2z)$, vertical motion at $(n+\\tfrac12)\\,c/(2z)$). Scale and sum them and the receiver ghost cancels while the primaries add: broadband recording without tuning the tow depth. The source ghost stays.
• Wave-equation de-ghosting. Forward-model the ghost and invert it out in the processing flow.

6. OBC and OBN: put the receivers on the seafloor

If you lay receivers on the seafloor instead of towing them, you move the receiver-side ghost out of the signal band: in 500 m of water it arrives about 667 ms after the primary, a separate down-going wave that P-Z summation removes (or mirror imaging uses) rather than a notch. The source is still towed, so its ghost remains. Also, you can record three-component or four-component data (pressure + three velocity components), enabling shear-wave processing and PS imaging.

• Ocean-Bottom Cable (OBC). Receivers on a cable laid from a boat. Limited length, repeated deployments.
• Ocean-Bottom Node (OBN). Autonomous pod receivers dropped on the seafloor by ROV, retrieved days or weeks later. Deeper water, better density, but expensive.

OBC/OBN surveys cost far more than towed-streamer per square kilometer, but they are the standard for subsalt imaging and high-value reservoir monitoring (time-lapse). Part 10 has a capstone on Gulf of Mexico OBN imaging through salt.

7. The processing impact

Every marine processing flow has these acquisition-specific early steps:

• Source signature deconvolution (remove the bubble tail).
• De-ghosting (source and/or receiver side, depending on acquisition).
• De-multiple (surface-related multiples, covered in Part 4).
• De-noise (swell noise, tidal noise, covered in Section 1.4).

Where this goes next

Section 1.3 is the bookkeeping section: SEG-Y format, trace headers, and how shots and receivers turn into CMP bins. Not glamorous, but every processor spends at least a week per year debugging header problems, and this is where you learn what to check.

References

• Sheriff, R. E., Geldart, L. P. (1995). Exploration Seismology (2nd ed.). Cambridge UP.
• Yilmaz, Ö. (2001). Seismic Data Analysis (2 vols.). SEG.
• Claerbout, J. F. (1976). Fundamentals of Geophysical Data Processing. McGraw-Hill.