Dip steering: why the window has to follow the bed

Below is a single, perfectly continuous reflector with no fault, no noise and no change of any kind along it. Two analysis windows sit on the same sample. The gray one is horizontal. The crimson one is tilted to follow the bed. Both read the same traces; they differ only in where on each trace they read.

At zero the two windows are the same window. Move the dip either way and they separate; a negative dip simply tilts the bed the other way, and the horizontal window is equally wrong about it.

Two intervals of time decide the outcome. The first is how much the reflector's arrival time changes from one end of the window to the other — the dip multiplied by the width of the aperture. The second is the period of the wavelet, the time from one peak to the next.

While the first is a small fraction of the second, the traces still broadly overlap and the window sees one event. As it grows toward half a period, the far end of the window is reading a trough where the near end reads a peak, so the window is comparing a peak against a trough. That is the whole mechanism, and it explains why the failure point depends on the wavelet rather than on anything geological.

Below are five snapshots of the same five traces, read straight across at five different bed dips. The reflector is perfectly continuous in every one of them — nothing is broken and nothing has changed but the dip. Watch the traces slide out of line, and watch the score fall as they do. A dip-steered window would show you the first picture every single time, whatever the dip.

Under each snapshot is the number that explains it: how much time shift has built up across the window, as a fraction of one period of the wavelet. That single quantity decides everything here.

Now a folded section cut by a single fault. Coherence is computed at every sample twice, with identical windows and identical arithmetic, differing only in whether the window is steered. Dark is low coherence.

1 The seismic a gently folded sequence cut by a normal fault whose plane leans across the section

2 Coherence, dip-steered the window follows the bed

3 Coherence, horizontal window the same window, never steered

The same comparison on a time slice

A section shows the fault as a line, which is a hard place to judge how much damage an unsteered window does. A map view is much clearer. These two panels are horizontal slices through a small 3D volume: a sequence of reflectors carried over a monoclinal flexure — a one-sided step in the structure, flat on both sides and steepening through the middle — cut by one planar fault. The fault is identical in both panels. Everything else on the right is the structure appearing in the attribute.

These two panels use a fixed patch of 3 by 3 traces, not the aperture set by traces used in the header; that slider drives the section panels above. Window height is shared. The values quoted in the exercises for these maps therefore refer to a three-trace window.

Both panels degrade, but not in the same way. The unsteered one was already dark on the flanks before any noise was added.

Modules 01 and 02 spent a good deal of effort on how a dip estimate goes wrong. The cost of that error is measured here. The slider adds a fixed error to every dip estimate before the window is steered by it, so the window still tilts, just not by the right amount.

1 What the error does to the window a bed dipping 2 ms per trace, with both windows on it

The teal window is steered by the correct dip and lies along the bed. The crimson one is steered by the dip plus your error, so it still tilts but cuts across. At a large enough error it is no better aligned than a horizontal window would have been, which is the point of the last exercise.

2 Coherence, steered by the wrong dip same gray scale as step 3

An unsteered attribute does not fail visibly. It produces a volume that looks entirely plausible, full of lineaments and fabric, and the lineaments are contours of the structure rather than anything broken. There is no error message and no obvious artifact, only a plausible image of the wrong thing.

What it changes in practice

• Fault picking. On the unsteered volume in step 3, the darkest thing on a row is the fault only about half the time. Anyone auto-tracking it extracts fold flanks as faults over much of its length, and those picks look like perfectly good faults until someone checks them against the seismic.
• Stratigraphic work. Bleed-through in map view follows the structural grain, which is exactly what makes it convincing as stratigraphy. The time slice in step 3 shows a broad band along the flexure that reads like a depositional feature and is nothing of the kind.
• Every attribute above coherence. Curvature, shape index, amplitude curvature and aberrancy are all computed in the steered frame. The dip volume is not just an input to them; it is the coordinate system they are expressed in.
• Structure-oriented filtering. Steered, it smooths along bedding and sharpens faults. Wrongly steered, it smooths across bedding, removing the discontinuities it was meant to preserve, and the output looks cleaner rather than worse.

And why the dip volume deserves real effort

Steering is not a switch that is either on or off. Step 4 shows a half-millisecond per trace error — smaller than the cost of coarsening the dip search step in module 01 — already eating into the fault contrast, and a 2 ms per trace error landing in the same place as not steering at all. Since the volume is still labeled dip-steered, that failure is harder to spot than the honest unsteered version.

That is the practical argument for the order of this course. Modules 01 and 02 spend a long time on getting dip right because the quality of the dip field is carried, undiminished, into everything computed from it.