Noise and Multiple Attenuation

Every noise problem was born in the field. Know how land and marine data are shot, and how the headers record it, before you judge what counts as signal.

You cannot attenuate what you cannot name; the taxonomy is the diagnosis that every filter choice downstream quietly assumes.

Signal-to-noise is the budget the whole flow spends against. Measure it on a raw gather once and the rest of this path becomes accounting.

Dead traces, polarity flips, and spikes are noise you can delete outright; editing first keeps every later algorithm from learning garbage.

Misaligned traces turn the stack from a noise attenuator into a smearing filter. Statics are noise control by alignment, solved before any subtraction is attempted.

Transform, mute, transform back: the recipe behind every denoise tool. Drive the f-k fan yourself and you will know the cost of every decibel you remove.

Spiking decon shapes the wavelet, predictive decon is your first shot at short-period multiples, and the spectrum is the proof that either one worked.

Surface-related or interbed, short-period or long: the class decides the algorithm, and misclassifying is how demultiple projects fail before they start.

The data predicts its own surface multiples, no earth model required. SRME is the default for a reason: the physics is nearly assumption-free and the cost is honest.

After NMO the primaries lie flat and the multiples still curve; the Radon domain turns that residual moveout into a distance you can mute across.

Every prediction is a few percent wrong in amplitude and a few milliseconds off in time. Adaptive subtraction is the step where the multiple actually leaves the data.

Interbed multiples never touch the surface, so SRME never sees them. Model-based prediction from the generating interface is the remaining honest tool, and it demands interpretation.

Swell, shipping, strum, and the biology you must work around: each marine family has a signature and a remedy, and half the remedy is decided at sea, not at the workstation.

Simultaneous shooting adds interference on purpose and trades it for speed. Deblending works only because the blend noise is incoherent in the right domain, a bet made at the design desk.

Ground roll is the single dominant noise on land and the geometry invited it. The field mitigations are the first pass of the very flow you now own.

Orthogonal, brick, zigzag, slanted: each land geometry answers a different noise question, and reading the layout tells you what the designer feared most.