Land acquisition: sources, receivers, geometry

Part 0 built the math. Part 1 walks onto the field. Every processing step later depends on what the acquisition crew did, so we spend five sections making sure we understand the data before we try to improve it.

1. Two sources you will meet

• Dynamite. A buried explosive. The source wavelet is approximately an impulse in time, short and broadband. The recorded trace looks almost directly like the convolutional model, $w(t) \\ast h(t) + n(t)$: the wavelet convolved with the earth’s reflectivity, plus noise. Used in rugged terrain, jungle, swamp, anywhere a vibrator truck cannot drive or would shake things it should not.
• Vibroseis. A truck with a baseplate that presses on the ground and sweeps through a range of frequencies over several seconds. The source is a long sinusoidal chirp, not an impulse. The recorded trace looks like a smeared-out mess, because it is the earth’s reflectivity convolved with a multi-second sweep.

Vibroseis is the dominant land source today. Safer, cheaper per shot, far better environmental footprint. But it comes with one catch: the recording is not usable until you correlate it with the pilot sweep.

2. The correlation trick

Cross-correlating the recorded trace $r(t)$ with the pilot sweep $p(t)$ gives

where $h(t)$ is the earth’s impulse response and $\\star$ denotes cross-correlation. The result is the earth’s impulse response convolved with the autocorrelation of the sweep, $k(\\tau) = (p \\star p)(\\tau)$, the Klauder wavelet. For a broadband sweep, two octaves or more, $k$ is a short, zero-phase wavelet, and the correlated trace looks like an impulsive-source recording after all.

The sweep has two levers, and neither moves the other. A longer sweep puts more energy into the ground: the peak $k(0)$ is that energy, so it grows in proportion to the length $T$, while correlated white noise grows only as $\\sqrt{T}$, and the signal-to-noise ratio gains about 3 dB for each doubling. A broader sweep shortens the Klauder wavelet and improves temporal resolution. In Figure 1.1 you correlate a buried record with its pilot, then pull each lever in turn.

At the opening setting, a 10 to 40 Hz sweep 8 s long, correlation gathers each buried sweep into a Klauder wavelet 20 ms across its central lobe and lifts the strongest reflector from −10 dB in the raw record to +22 dB. That wavelet is too broad for two reflectors 12 ms apart, and they merge into one peak. Raising the top of the sweep to 80 Hz narrows the wavelet to 11 ms and splits the pair, although the two peaks sit 14 ms apart, farther than the reflectors, because the two wavelets interfere. Lengthening the sweep from 2 to 16 s raises the S/N by about 9 dB and leaves the central lobe exactly as it was.

Hertz are not the whole story. The same 40 Hz of band swept from 5 to 45 Hz spans three octaves and gives a compact wavelet; swept from 60 to 100 Hz it spans less than one, and the wavelet rings, with side lobes only 1 dB below its peak that look like reflectors. The amplitude spectrum stays nearly flat inside the sweep band and drops off outside it: frequencies you did not sweep through cannot be recovered. And because $k$ is an autocorrelation, correlated vibroseis data are zero phase, while a dynamite wavelet is close to minimum phase, one reason the two are deconvolved differently later.

3. Receiver: the geophone

A geophone is a coil wound around a magnet on a spring. When the ground moves, the coil moves relative to the magnet, inducing a voltage proportional to velocity. Its natural frequency is typically 4 to 10 Hz. Above it the response is flat; below it the velocity response falls off at about 12 dB per octave, so for frequencies under 5 Hz you need a special low-frequency phone or an accelerometer. An accelerometer (usually a MEMS sensor) measures ground acceleration, and its response stays flat down to very low frequencies; the price is an instrument noise floor that is usually higher than a geophone’s.

Real surveys group several geophones (6 to 24) at each recording station. The group acts as a spatial array: reflections arrive nearly vertically, so they reach every geophone in the group at the same time and add coherently, while ground roll travels horizontally with a short wavelength, so it reaches the geophones out of phase and largely cancels when their outputs are summed, provided the array is about as long as the ground-roll wavelength. This is the first noise attenuation in any land data flow, done in hardware, before a single byte is recorded.

4. Survey geometry terms

• Shot point (SP). Where a source fires. Usually on a regular grid (shot line + shot interval).
• Receiver station (RS). Where a group of geophones records. Also on a regular grid.
• Group interval. Spacing between receiver stations. Sets the spatial Nyquist, from Section 0.5.
• Source-receiver offset. Horizontal distance between SP and RS. Decides how much of each reflection hyperbola is recorded, and so how much moveout the data carry for velocity analysis.
• Spread. All active receivers recording one shot. Modern 3D has 1,000-10,000+ channels per shot.
• Template. The relative shot ↔ receiver pattern that “rolls” across the survey as you shoot.

5. What “3D” actually means

A 2D line lays shots and receivers along one profile. You get a 2D section under that line, nothing to the side. Cheap, sparse, lots of assumptions.

A 3D survey lays receivers and shots on a grid, producing a full 3D data cube. You sample cross-line variation, you can image 3D structures, and you can migrate properly. Every capstone in Parts 5-10 of the Interpretation course was 3D.

6. What matters downstream

For a processor opening a new dataset, the acquisition facts that change your workflow are:

• Source type. Vibroseis → correlate first, then everything else. Dynamite → skip correlation.
• Record length. Vibroseis records are sweep-time longer; processing cuts to “listening time” after correlation.
• Geometry regularity. Regular grids make migration straightforward; irregular surveys need interpolation or regularization (covered in Part 4 and Part 9).
• Spread pattern. Sets offset and azimuth distributions, which decide AVO, 4D, and anisotropy feasibility.

Where this goes next

Section 1.2 looks offshore. Airguns, streamers, OBN, ghost notches, a different physics and a different set of processing challenges.

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.