QC: when your interpretation is wrong, and how you know

What can silently go wrong

Five failure modes account for most interpretation errors that make it into downstream workflows:

• Cycle skip — the horizon jumps one wavelet cycle (or more) across part of the survey. Most common failure; usually caused by a seed placed on the wrong peak, or by propagation across an area where two reflectors of similar brightness sit side-by-side.
• Sidelobe drift — the horizon gradually walks from the main peak onto a sidelobe. Subtle because the amplitude decrease is gradual; usually detected by a systematic well-mismatch rather than by eye.
• Wrong-side-of-fault tracking — auto-tracker crossed a fault undetected and is now picking the same reflector on the downthrown block, producing a horizon that's a hybrid of two different stratigraphic surfaces.
• Fault–horizon inconsistency — the horizon crosses through a mapped fault without showing the expected offset, or the fault is present on some inlines and disappears on others without structural reason.
• Depth-conversion errors — the interpretation is right in time but wrong in depth because the velocity model used for conversion is off. The interpreter may be correct and the downstream target may still be 200 m deeper than predicted.

Each of these has a characteristic signature that appears under the right QC check. An interpreter who runs all four QC categories below will catch nearly every instance of all five failure modes before any downstream workflow consumes the interpretation.

The four QC categories

(1) Near-well consistency. At every well location, your horizon picks should agree with the well's geologic markers within your expected depth uncertainty. Compute the synthetic seismogram for each well, tie it to the seismic, identify the expected horizon time at that well, compare to your pick. A systematic offset at one well is forgivable (local velocity anomaly, mis-tied synthetic). A systematic offset at multiple wells is the diagnostic signature of cycle skip or sidelobe drift — you have picked a reflector that is not the geologic surface you think it is.

(2) Map-view scrutiny. Display the horizon as a time map (or depth map after conversion). Look for:

• Smooth colour gradients = good. Real geology produces maps that look like smooth topography.
• Isolated colour spots in the middle of smooth areas = almost certainly pick errors. Inspect and fix.
• Sharp linear discontinuities = faults (if they match your fault interpretation) or pick drift (if they don't). Cross-reference with your structural interpretation.
• Patchwork zones with no coherent pattern = tracker walked onto noise. Delete and re-pick.

(3) Cross-section pass. Spend 10 minutes scrolling through every N-th inline and crossline (say every 10th on a typical survey), looking at the picks overlaid on the section. You are specifically looking for: picks that clearly sit on the wrong peak compared to neighbours; fault intersections where the horizon should terminate but keeps going; or areas where the reflector has obvious character change (perhaps a facies transition) that your picks ignored.

(4) Self-consistency. The interpretation as a whole should make geological sense:

• Faults should form closed networks (or at least terminate at boundaries that have some explanation).
• Horizons should terminate cleanly at fault boundaries, not cross through them.
• Structural relief on the horizon map should match the tectonic regime (extensional basins have half-graben geometries; thrust belts have hanging-wall anticlines).
• Stratigraphic relationships should be reasonable (a reservoir sand pinching out toward a salt margin is plausible; a reservoir sand of constant thickness across a basin boundary is not).

The self-consistency check is the hardest because it requires geological reasoning, not just data-pattern recognition. But it is also the check that catches failure modes the others miss — particularly the cases where your horizon picks and fault picks are each individually reasonable but together imply impossible geology.

When is it time to ship?

An interpretation is ready to hand off when:

• All wells tie within your stated uncertainty, with no systematic misfits.
• The horizon time maps and fault polygons are self-consistent when displayed together (no horizons crossing unmapped faults; no faults without associated horizon offsets).
• Residual pick anomalies have been either fixed or explicitly documented as areas of low confidence.
• The depth-conversion inputs (velocity model, well-tie residuals) are documented enough that someone else could reproduce or update them.

An interpretation is NOT ready to ship when any of the above fails. The temptation to ship early and "fix it later" is strong — resist it. Errors that escape into downstream work (depth conversion, volumetrics, drill plans) cost orders of magnitude more to fix than errors caught during the interpretation phase.

The discipline of stepping back

After every few hours of detailed picking, close your session notes, open the map view, and ask one question: does this make geological sense?

Not "are the picks in the right place" — you have been optimizing for that all day, and your sense of right is locally tuned. The question is whether the BIG PICTURE — this basin, this fault network, this sedimentary sequence — matches what the regional geology of the area should look like. If your basin suddenly has a fault network that looks unlike every other basin of the same tectonic type, the likely explanation is that your interpretation is wrong, not that this basin is unique.

Senior interpreters stop and do this check several times a day. Junior interpreters rarely do it at all. The habit is learnable; the rewards are enormous.

Practicing QC

The horizon and fault picker widgets from §2.3–2.5 serve as QC practice tools. Try this exercise:

• Go to §2.3 and auto-track a horizon across the trainer volume. Deliberately place ONE seed pick on the wrong cycle (say 30 ms too shallow).
• Press Time map. Observe how the bad seed shows up — as a zone of anomalously shallow time surrounded by correctly-picked areas.
• Find the bad seed without toggling Show reference. Look at the map, scan for the anomaly, zoom to the offending area by scrolling the inline slider until you find it.
• Delete the bad picks, re-seed correctly, regrade.

That five-minute drill is QC in miniature. The real thing on a real survey uses the same reasoning at much larger scale.