Faults in 3D: how to see them, how to pick them

What a fault is

A fault is a fracture surface across which rocks on one side have moved relative to the rocks on the other side. The movement can be mostly vertical (normal or reverse faulting, in extensional or compressional regimes) or mostly horizontal (strike-slip). The amount of movement is the throw (vertical component) and heave (horizontal component). Together they describe the slip on the fault.

On seismic, a fault appears as an abrupt offset of reflectors. Layers on one side of the fault are at one time; on the other side they jump to a different time. The bigger the throw, the more obvious the offset, a 50 ms fault is visually prominent; a 5 ms fault might be at the limit of detection on a typical dataset.

What a fault looks like on seismic

Three visual signatures together identify a fault:

• Reflector offset. The horizon you're tracking steps abruptly up or down. Usually multiple horizons offset together, a fault cuts through a stack of layers, not just one.
• Acoustic shadow. The fault plane itself is often a diffracting surface that returns little coherent energy, so the seismic section shows a band of reduced amplitude or noisy character along the fault trace.
• Diffraction hyperbolas (unmigrated data) or migration artifacts (over-migrated data) cluster along fault planes, giving them a characteristic jagged appearance.

A good fault interpretation identifies the fault plane itself (where the rocks are broken) not just the offset, the plane is what you pick, because that's what the subsurface actually contains. The offsets on either side are the evidence, not the target.

Three fault types, three signatures

• Normal faults, the hanging wall (the block above the fault plane) moves down relative to the footwall. Most sedimentary basins have lots of these: extensional tectonics stretches the crust and produces down-dropped blocks. On seismic, normal faults dip at 55-70 degrees typically and show reflectors on the downthrown side clearly lower than those on the upthrown side.
• Reverse (thrust) faults, the hanging wall moves up relative to the footwall. These are compressional features, common in mountain belts and foreland basins. Low-angle thrusts (less than 30 degrees dip) can look superficially like bedding on cross-sections and are one of the hardest fault types to interpret correctly.
• Strike-slip faults, the two sides move horizontally past each other. On a cross-section the two sides might show similar stratigraphy at similar depths, which makes strike-slip difficult to spot on isolated inlines. Strike-slip faults are typically identified on time slices where the map-view offset of a stratigraphic feature is obvious.

Picking faults as polylines

Unlike a horizon (one time value per crossline), a fault trace on an inline is a line segment, typically just a few straight-line pieces that follow the fault plane from its shallow expression down to where it dies out (or leaves the volume). You pick a fault by clicking along its trace, laying down vertices to define a polyline. Each inline gets its own fault trace.

The picking workflow is simpler than horizon picking in one respect: you don't need to pick every crossline. Two or three vertices along a fault trace suffice for a straight fault; a listric (curved) fault might need five or six. The polyline between vertices is interpolated linearly, so vertex placement should sit where the fault direction changes.

On the other hand, fault picking is harder in one respect: you have to decide exactly where the fault plane is, and the seismic data often underdetermines it. The acoustic shadow along a fault is frequently several traces wide; the reflector offset gives you the approximate location but not the precise angle. Experienced interpreters consistently place fault traces in the middle of the shadow zone and reconcile the geometry across inlines.

Exercise

• Start on inline 1038, where the largest fault, F3, carries 42 ms of throw and stands at least 13 crosslines clear of its neighbours. Drag the five handles of the stick in (a) into the gap between the offset ends and watch the signed offset fall towards zero. Switch the display to Coherence if you want the plane to show itself: where the reflectors are broken, a trace stops resembling its neighbour, so the plane shows as a dark seam.
• Pick new sticks by clicking down a plane and pressing Finish stick. Each belongs to the fault highlighted in the chips above the section, so start a New fault before picking a different one. Naming faults yourself is the point: a fault is a surface you decide to join together, and the figure tells you when you joined two that are not one. Carry forward copies a stick to the next inline as a dashed guess, because where a fault strikes across the inlines its trace moves from one section to the next; drag it onto the plane and it counts as a pick. The figure's third exercise carries one fault five inlines without adjusting it: the copies end up 3.3 crosslines left of the plane on average while agreeing with each other perfectly. As you step through the inlines the trace of one fault should change smoothly, and the table's largest unexplained jump is how far a trace moves beyond what its fault does. Every other tool and shortcut is in the help line under the section.
• Watch plate (b), throw against inline for every fault in the survey. Throw rises from one tip to a maximum and dies at the other: G rises from about 7 ms at inline 1009 to 34 ms around inline 1046 and falls back to 7 ms at 1083, with no trace beyond. Ticks mark the inlines you have picked, and ticks across only the middle of an ellipse mean the surface is not finished. A profile that falls towards zero and recovers is two faults overlapping at a relay. The fourth exercise joins F5 and F3 under one name: the throw measured from its picks falls from 18 ms at inline 1007 to 6 ms at 1015, where F5 is dying out, and is back at 37 ms by 1019 on F3, which overlaps it 4 crosslines away. Every stick is accurate, and the figure still says the fault is two.
• Press Grade for the full report in words. The headline is the signed offset: an error with the same sign at every vertex is the signature of picking along the offset reflectors instead of through them, the mistake this section warns about above all others.
• Toggle Show the true fault planes to reveal every fault that crosses the inline, heavier lines for larger throw. Each line stops where its fault stops offsetting a reflector, so a short line is a small fault dying out, not an unfinished answer. Use it to check a pick, not to make one.
• Plates (c) to (g) measure your faults from your own picks, not from an answer key: where each stick crosses the reference horizon, the horizon is fitted on either side and the offset read off, as with a ruler. Plate (c) places each fault on the throw-length plot against the survey's 31 faults and the scaling band $T = 0.03\,L$; (d) compares its throw along strike with a complete ellipse; (e) and (g) fit the cumulative frequency, how many faults carry at least a given value, with a power law, an exponential, a lognormal and a straight line, all scored the same way. Read the margin before believing any of them: on the survey's 31 throws the exponential leads the power law by 0.018 in $R^2$ (0.976 against 0.959), and on its lengths by 0.005, far too little to tell them apart, so the figure declines to name a distribution. A ruler read across a trace also takes in any other fault or bend inside its window; the figure takes out the steps it can separate from each fault's own, and the notes in (g) say when a neighbour sits too close for that and what a plain ruler would have read.
• Not every bend is a break. Where a fault reaches its tip its displacement dies to zero, so the reflectors sag across several traces instead of stepping across one. That drape is real structure, but it is not a fault plane. If a feature never resolves into a clean offset on any inline, you are almost certainly on the tip of a fault whose plane is somewhere else in the volume.

Aim for a signed offset within about 1.6 crosslines of zero and an RMS distance of 2.5 crosslines or less over all your vertices. A larger, one-sided offset usually means your vertices sit on the offset reflector ends rather than on the plane between them.

Pitfalls to avoid

• Picking the offset, not the fault. Beginners sometimes draw a polyline along the shifted reflectors rather than the fault plane. The reflectors are the evidence that a fault exists; the fault itself is the plane cutting through them. Always pick between the offset reflectors, not on them.
• Fault looking wrong on different inlines. If your fault pick changes dramatically from one inline to the next, something is wrong. Real faults are 3D surfaces whose traces should vary smoothly across inlines. Abrupt changes mean you picked something else on one of the inlines, or the fault you identified is actually part of a fault network with separate segments.
• Acquisition footprint mimicking faults. Linear features that appear periodic and parallel to the acquisition direction are often acquisition footprint, not real faults. Check whether the "fault" corresponds to any stratigraphic offset, if reflectors pass through it without disruption, it's not a fault.
• Sideswipe masquerading as a fault. A reflector from outside the line (on unmigrated 2D) can look like a fault. Migrated 3D data mostly eliminates this, but very steep dips and edge-of-survey areas still suffer.
• Missing the small ones. Below about a quarter wavelength, $\lambda/4$, a fault no longer shows a clean offset, but it still exists and can still bend, dim or phase-shift the reflector. For reservoir-scale work such faults can matter for flow, and attributes such as coherence and curvature (Part 6) reveal some of them.

From 2D traces to 3D fault surfaces

A fault picked on just one inline is a line. Pick the same fault on many inlines and you have a set of lines that together outline the 3D fault surface. The quality of a 3D fault interpretation depends on picking enough inlines to constrain the surface. For a simple planar fault, 4-6 inlines evenly spaced across the fault's extent are enough; for complex fault networks (intersecting faults, changing strike, branching), you need denser coverage.

In production workflows the picks from individual inlines are triangulated into a fault plane, which is then used to constrain horizon picking (horizons stop at fault boundaries), reservoir volumetrics (fault blocks are the compartments), and flow modelling (faults can be sealing or conducting). Plate (h) of Figure 2.5 does that joining for you, and the distinction it draws is the one that matters: your sticks are solid lines, the sheet lofted between them is translucent. One is what you measured and the other is what was assumed between your measurements. A surface built from three sticks across sixty inlines looks convincing from a distance and is mostly interpolation, which is easier to see once it is a shape you can orbit than it ever is on a section.