Picking horizons: methodology and practice

What a "horizon" is and is not

A seismic horizon is a locus of picks, one pick per trace (or per picked trace), chosen so they all follow the same reflector. The horizon is, in principle, a surface. In practice it is a cloud of picks that you later interpolate into a surface.

Important: a horizon is not the reflector itself. The reflector is a physical boundary in the earth; the horizon is your interpretation of where that reflector sits at every location on your survey. The horizon can be wrong in many ways (picked the wrong peak, followed a different reflector across a fault, picked noise where there is no reflector). Treating the horizon as ground truth rather than as your best interpretation is a recipe for overconfidence.

The four-stage picking workflow

• Choose your reflector. Before picking, decide which reflector you want to trace. Usually this is determined by a well tie: the well log says "Top Reservoir is at 2340 m", the synthetic seismogram computed from the log says "this corresponds to a strong peak at 1870 ms TWT", so that's the event you follow. Without a well tie, you can still pick horizons, but the name attached to the horizon ("top carbonate") is a geologic hypothesis, not a fact.
• Seed a few key inlines. Place picks on 3-5 widely-spaced inlines manually, ideally inlines you understand well (near wells, across obvious structures). This is your "skeleton". Every subsequent pick gets cross-checked against the skeleton for consistency.
• Propagate. Fill in the gaps between seed inlines. This is where auto-tracking helps (Section 2.4 covers it). For now, do it manually, walk through inlines at regular spacing and make sure your picks stay on the same reflector as the skeleton.
• QC and fix. Display the horizon as a time map (shading by the pick time across the survey). Geologic surfaces should look smooth, with coherent structural patterns. If your horizon map has isolated bright or dark spots that don't match the reflectors on nearby slices, those are pick errors. Go back, inspect, and fix. Section 2.6 covers QC in detail.

The discipline is: never skip from stage 1 to stage 3 without seeding. A horizon picked by scrolling from one end of the survey to the other, clicking on "the brightest reflector at each location," almost certainly drifts off the intended surface somewhere along the way because the "brightest" reflector changes as you cross faults or go through a stratigraphic pinchout.

Snap-to-peak and its limits

Every interpretation tool provides snap-to-peak: you click near a reflector and the software refines your click to the nearest local extremum (peak or trough) within a small window. In Figure 2.3 it is the nearest one of the same sign as the trace under your click, so a click on a peak stays on that peak. This does two things: it removes click-position noise (a hand click lands several milliseconds either side of the crest; a snapped pick, interpolated between samples, lands on it), and it enforces that you are always picking an extremum, which is what a reflector event actually is.

When snap-to-peak fails: (1) in low signal-to-noise zones where there IS no clear extremum near your click; (2) when two reflectors are closer than your snap window, causing the snap to jump between them unpredictably; (3) when the wavelet is not zero-phase and the "extremum" does not sit at the actual reflector position. If you see a run of picks with large time scatter, or the picks bouncing between two cycles, snap-to-peak is confusing you rather than helping. Turn it off in those zones and pick manually.

Common picking mistakes, and how to catch them

• Cycle skip. You pick the reflector on one inline, then on the next inline you accidentally pick the peak one cycle shallower or deeper. The horizon map shows a step. Fix: display the map, scroll through the steps, re-pick the offending inlines.
• Crossing a fault. Your reflector is offset by a fault; you didn't notice, and the picks walk from one side of the fault onto a reflector on the other side. Fix: display coherence or dip attributes (Part 6) which highlight fault planes and let you stop your picks at fault boundaries.
• Following sidelobes. The wavelet has sidelobes on either side of its main peak. Your pick drifts onto a sidelobe because of local amplitude variation, then keeps following the sidelobe. Fix: a well tie will reveal a consistent time bias; re-pick with reference to the well.
• Noise masquerading as reflector. In low-SNR zones you can pick "reflectors" that are actually processing artifacts. Fix: check the horizon continuity against known wells; if a pick in a low-SNR zone disagrees with the nearest well, it is almost certainly noise.

Exercise

• Press Clear horizon and find the brightest continuous reflector near 700 ms on inline 1046. That is the target.
• Place six to eight picks across the inline with snap on, spread out rather than clustered. Plate (b) scores each one as you place it: pick time minus the true time of the reflector it matches, with one period of the wavelet marked either side.
• Step to other inlines (the Inline control, the arrow buttons above the section, or the arrow keys with the section focused) and pick the same reflector on three or four more across the volume. Pick both sides of the fault near crossline 2059 when you reach it.
• Turn on Reference reflectors and set Time shown in (a) to Reflectors. Thin dotted lines show the true times of all five reflectors the survey holds, the target, the pinchout top, the facies boundary and the two doublet reflectors a period apart, each named beside the section unless one of your horizons already follows it. Every one is a real event you can pick, and New horizon builds them side by side.
• Read the table, and press Grade for the written grade. The bench works out which of the five reflectors your picks sit on and scores you against that one, so a second horizon is not marked against the target. It separates bias, your picks sitting consistently to one side, from scatter either side of the reflector, and names picks on the wrong event: half a cycle onto the opposite polarity is a different mistake from a full cycle skip, with a different fix.
• Watch plate (c). A horizon is a map, and the surface you build is what anyone else will use. Dots are your picks, everything between them is interpolation whose weight tapers to zero at 18 bins, so a pick enters the map smoothly rather than as a step, and hatching marks ground you have not picked near. Switch it to Misfit and one mis-picked trace shows as a bullseye, where on a section it is one dot slightly out of line. Plate (d) shows every horizon you have picked as a surface at its true depth relative to the others, with the vertical exaggeration stated, because a couple of hundred milliseconds of relief across a survey 4 km wide is a flat sheet at true scale.

Aim for Good or Excellent: an RMS within 8 ms, or within one 4 ms sample. A clean pick of a different reflector is graded against that reflector and scores well, so a poor grade means cycle skips or picks that wander between events. If yours is poor, turn on Reference reflectors, delete the bad picks (shift-click them, or set A click on (a) to Deletes and tap them), pick again and regrade.