Can you see the fault?

Why throw runs into the bed-thickness limit

A reflector cut by a fault gives two copies of the same event, one on each side, offset in time by the throw. That is the same geometry as the wedge in module 02 — two reflections a short time apart — so it fails at the same point: somewhere around a quarter of a wavelength.

Whether a fault is visible therefore depends on velocity, frequency and noise as much as on the size of the offset.

Try this — and what to watch for

1. Walk the throw down. The point: it puts a number on where your eye gives up. Start at 40 m and reduce it. Where does the offset stop reading as a clean step in the reflectors? Somewhere close to λ/4. Above that you can measure the throw off the section. Below it you can only say that something happened.
2. Keep going, and watch the trace panel. The point: this is what an attribute sees that you do not. Take the throw down to 5 m and compare the two traces. The step has gone from the section, but the traces still differ. Reflections interfering across the fault distort the waveform. Coherence and similarity attributes are built to find that residual, which is why an attribute can flag a fault that looks like nothing on the amplitude display.
3. Trade frequency against throw. The point: the same fault can be present or absent depending on the vintage of the data. Fix the throw at 12 m and sweep frequency from 15 Hz to 70 Hz. The geology never changes, and the fault goes from invisible to obvious. Whether you can see a fault depends on your bandwidth as much as on the rocks.
4. Make it deep. The point: it explains why deep fault maps are always simpler than shallow ones. Raise velocity to 5000 m/s and drop frequency to 15 Hz, roughly the change from 1 s to 4 s two-way time. λ goes past 300 m, and a 40 m throw that was obvious at shallow depth becomes marginal. Deep fault maps are less complete than shallow ones for reasons that have nothing to do with the interpreter.
5. Coarsen the acquisition. The point: resolution is limited sideways as well as vertically. Push trace spacing to 50 m with a shallow fault dip. The fault plane turns into a staircase. Trace spacing limits lateral detail the same way bandwidth limits vertical detail.
6. Push the dip past 90°. The point: normal and reverse faults look alike on seismic; the sense of slip is your interpretation, not the data's. Take it from 65° to 115° and watch the layer model. The plane now leans the other way, so the block that drops is the footwall rather than the hanging wall and the fault becomes reverse. A normal fault cuts section out at the fault; a reverse fault repeats it, and you can count the doubled beds in the model. On the seismic both look like an offset, which is the problem: sense of slip is a geological interpretation laid on top of what the data actually shows.
7. Thin the beds. The point: most reservoir intervals are tuned, and tracking a fault through tuned reflections is far harder. Set the thinnest bed to 8 m and the thickest to 20 m, with frequency at 25 Hz. Count the beds in the layer model, then count the reflectors in the seismic below it. They no longer match. The beds are below tuning, so the reflectors merge and interfere. Tracking a fault through tuned reflections is much harder than tracking one through clean, well-separated ones, and most reservoir intervals look like the former.
8. Run ten quiz cases on Mixed, then ten on Hard. The point: it tests the limit against your own judgement, without the answer visible. Keep score honestly. Roughly a third contain no fault at all, and throws are drawn as a fraction of the wavelength rather than in meters, so most of them sit near the limit instead of being comfortably above it. On Hard, only about one case in twenty is a clean resolved offset. Calling "no fault" on a throw below λ/16 is scored as correct, because past the detection limit that is the defensible answer. Scores are typically lower than expected, and correlated noise accounts for many of the false picks.