Attributes for reservoir characterization: putting it together

The reservoir characterization workflow (in five steps)

• Frame the question. What is the target? A specific stratigraphic interval, picked as a horizon and tied at a well? A trap-bounded prospect? A field already in production where you're looking for missed pay? The interpretation pipeline is shaped by what you're trying to find.
• Pick the structural framework. Map the key horizons (Sections 2.3, 2.4) and faults (Section 2.5). The structural picture is where everything else gets attached. Coherence and curvature (Sections 6.4, 6.3) help find the structures that horizon picking then commits to.
• Compute attributes at the target depth. RMS amplitude, envelope, spectral content, either as full volumes (Sections 6.1, 6.2) sliced for orientation, or as horizon-gated maps (Figure 6.6) read in a window hung from the target horizon.
• Look for direct evidence of hydrocarbons. The four classical DHIs (bright spot, dim spot, flat spot, polarity reversal) each have a multi-attribute signature. RGB blends (Section 6.5) make several DHIs visible in one picture.
• Tie to well control. Calibrate attribute responses against any wells that penetrate the target. Without wells, the interpretation is qualitative; with even one well, you can start quantifying. The seismic-to-well tie is where rock physics (Part 5) re-enters the workflow.

This is a loop, not a sequence. You revisit each step as the interpretation evolves, a wider gate reveals an unexpected anomaly, which prompts re-examining the structure, which re-positions the next attribute extraction.

Horizon-gated maps: a statistic in a window hung from a pick

So far we have displayed attributes as slices through 3D volumes: inlines, crosslines and time slices. A time slice is a map at one two-way time, so it follows a layer only where the layer is flat. The deliverable that goes into a reservoir model or a prospect ranking is a map of one interval: one value per trace, read in a window that follows the interpreted horizon wherever it dips or folds, ready to lay over a base map with the wells, the leases and the faults.

To make one, pick the horizon (Sections 2.3 and 2.4), hang a gate from the pick on every trace, from $c - L/2$ to $c + L/2$ about it (length $L$, centre offset $c$, so the gate can sit above, around or below the horizon), and reduce the $n$ samples $s_k$ inside it to one number:

The RMS and the mean magnitude measure the strength of the reflections in the gate; the maximum and the minimum read its largest peak and its deepest trough; the mean is useful only in a gate shorter than half a period, because over several periods a trace's peaks and troughs cancel. The samples can be the amplitude or any attribute volume (spectral power, coherence, dip, curvature): the gate does the same thing to all of them.

How long the gate should be is set by the wavelet, and the yardstick is the event's trough-to-trough period $T_d$ (Section 1.7, where half of it, $T_d/2$, is the tuning thickness):

• A gate of about $T_d/2$ or less reads one lobe of the reflection. It is the most specific measure of that reflector and the most fragile: a mis-pick of one sample slides it along the wavelet.
• A gate of one to two periods, centred on the pick, holds the whole wavelet of the reflector and little else, the usual choice for a single reflector.
• A gate of several periods maps an interval. It barely moves with the pick, but it averages in other reflectors, so a property of one bed is diluted by its neighbours. Use it for a package of beds, and say so.
• A gate hung off the pick, wholly above or below it, reads whatever cycles it cuts there, and a small change of pick moves it across them.

State the gate with the map ("RMS amplitude, 8 ms above to 8 ms below Top Reservoir"): the same horizon gives different maps with different gates, and a reader cannot judge a map whose gate is not written on it.

Reading Figure 6.6 on F3

Figure 6.6 tracks one reflector of the F3 subset itself, a weak peak near 1000 ms, from a seed at inline 400, crossline 600. Its careful tracker (search ±8 ms, with a waveform check) picks 99.4% of the 40,000 traces, between 980 and 1044 ms, and the picked event's trough-to-trough period is $T_d$ = 20.1 ms, so its tuning thickness is about 10 ms. The figure's headline divides the value at the probe by the map's median, its typical value, and repeats that with the pick moved 4 ms either way and both gate ends a sample out or in. Every number below is read from the figure.

• A dim band. The RMS amplitude in a 16 ms gate centred on the pick shows a pale sinuous band on the right of the map, where the reflector dims: on the section its peak weakens and splits into two smaller ones. At inline 340, crossline 670 the gate reads 0.67 times the cube's RMS amplitude against a map median of 1.60: 0.42 of the median, in the lowest 0.2% of the map. Moving the pick or the gate ends keeps it between 0.37 and 0.47 of the median, still in the lowest 0.2%: the low belongs to the reflector, not to the extraction.
• The gate's length. A single sample on the pick puts the probe at the same extreme, 0.33 of the median, but a 4 ms shift lets it rise to 0.62 of the median and the 26th percentile, and a 4 ms mis-pick changes the whole map by 289% of the map's own standard deviation. For the 16 ms gate that change is 45%, for a 120 ms gate 16%; but the 120 ms gate holds six periods and several other reflectors, and the probe reads 0.92 of the median, at the 26th percentile, in the map's ordinary range: robust, and no longer about this reflector.
• The gate's position. Hung 12 to 28 ms below the pick, the same 16 ms gate puts the probe at 0.75 of the median and the 11th percentile, and anywhere from 0.68 to 1.1 times the median, from the 2nd to the 59th percentile, when the pick or the gate ends move.
• The statistic. The maximum in the centred 16 ms gate barely moves with a mis-pick (2% of the map's spread), because the picked peak stays inside the gate. The mean over an 80 ms gate is +0.11 times the RMS at the probe, where the RMS over the same gate is 1.12: the cycles cancel and the map takes both signs, so the figure ranks the probe instead of dividing by the median, and its rank swings from the 28th to the 95th percentile.
• A mis-pick. The loose tracker (search ±16 ms, no waveform check) leaves the careful one by more than half a period on 411 traces, most of them in a patch at the right edge of the survey where it takes the next peak down, a full period lower. There the map shows a patch with straight edges: 3.04 times the RMS at inline 325, crossline 696, 1.9 times the map's median and in its highest 0.06%, robustly so, and wrong. Only the time map and the section show it.
• No horizon. A flat 16 ms gate at the pick's mean time holds the horizon on only 52% of the map, because the horizon spans 64 ms; elsewhere it maps whatever reflectors cross that time, and the probe on the band rises to 0.84 of the median, at the 37th percentile.

From a gated map to a reservoir statement

A map is not yet a statement about the reservoir. Four checks turn it into one, and each leaves a number that belongs in the statement:

1. Quality-control the pick. Look at the time map for steps of one period and at sections across every anomaly, and report the share of traces tracked and how the gaps were treated.
2. Test the anomaly against the extraction. Recompute the map with the pick moved a sample either way and the gate ends moved a sample. An anomaly whose rank survives every one of these choices is a property of the data; one that comes and goes is a property of the gate. Figure 6.6 reports that spread for its probe.
3. Check conformance. Lay the map over the horizon's time contours and the fault map (see below). An anomaly that follows structure or a depositional outline is a candidate; one that follows nothing, or the edges of a tracking jump, is not.
4. Calibrate at wells. Only a well says what an anomaly is. Tie the well (Section 5.6), read the map at the well, and fit the relation between the gated value and the property it is meant to measure (net sand, porosity, fluid) over all the wells. Below the tuning thickness amplitude grows almost in proportion to thickness (Section 1.7), so a few wells can calibrate a thickness map; the scatter of the wells about the fit, together with the extraction spread of check 2, is the uncertainty to quote.

The F3 subset in this figure holds no well, so the most it supports is a statement of the second kind: along a pale sinuous band the reflector near 1000 ms reads less than half the map's median in a 16 ms gate, in the lowest 0.2% of the survey, and stays so when the pick or the gate ends move a sample. Whether the band is a sand-filled channel, a mud-filled one or something else again needs a well, and an honest map says so.

The four classical DHIs (Direct Hydrocarbon Indicators)

Hydrocarbons (especially gas) change the acoustic properties of rock dramatically: lower density, lower P-wave velocity, often stronger amplitude contrast. These changes leave detectable seismic signatures called Direct Hydrocarbon Indicators. None alone is conclusive; multi-attribute evidence is what builds confidence.

• Bright spot. A sand reservoir charged with gas has lower acoustic impedance than the surrounding shale, producing a stronger negative reflection at the top. Bright spots show as anomalously high amplitudes that conform to the reservoir geometry. Multi-attribute evidence: localized RMS anomaly (Section 6.1) PLUS amplitude conformance to a structural high or stratigraphic trap PLUS often a flat spot beneath. Dangerous false positive: thin-bed tuning (Section 1.7) can also produce bright amplitudes without any hydrocarbons.
• Dim spot. The opposite case: a gas-charged carbonate or stiff sand may have HIGHER impedance than the shale above, weakening the reflection. Dim spots are easy to miss because "no anomaly" is the visual clue. Multi-attribute evidence: anomalously low amplitude in a region where you'd expect normal reflectivity, plus the same conformance and structural-trap reasoning. Spectral attributes sometimes help, dim spots may show distinctive frequency shifts.
• Flat spot. A horizontal reflector cutting through dipping geology, marking a fluid contact (gas-water, gas-oil, or oil-water). The contact is horizontal because hydrostatic equilibrium dominates; the surrounding rock is dipping because of structure. When present a flat spot is the strongest single DHI, but not a unique one: a diagenetic front (such as the opal-A to opal-CT transition), a multiple, an igneous sill or a flat-lying depositional surface can also cut across dipping reflectors, so a flat spot should be flat in depth, conform to the closure and sit where the bright or dim amplitude stops. Look for them on inline sections through suspected reservoirs.
• Polarity reversal. A reflector that switches polarity laterally as you cross from a brine-filled to a hydrocarbon-filled section of the same reservoir. The brine response is one polarity; the hydrocarbon response (with different impedance contrast) is opposite. Subtle and easy to confuse with phase-rotation artefacts; cross-check against polarity convention (Section 2.2) and seismic-to-well ties.

An RGB blend (Section 6.5) of (RMS amplitude / spectral 15 Hz / spectral 30 Hz) is a powerful DHI screening tool. Bright zones with anomalous low-frequency content (red/yellow) flag candidate brights; structural conformance can be checked by comparing with a coherence/curvature blend.

Amplitude conformance: the most important visual check

An amplitude anomaly that doesn't conform to a structural or stratigraphic trap is suspicious. Conformance means the bright (or dim) zone's outline coincides with the closure of a trap, the boundary of the structural high, the edge of a fault block, the geomorphic outline of a channel. If the amplitude follows the trap boundary, it suggests the trap is filled with something acoustically distinctive (often hydrocarbons). If the amplitude crosses the trap boundary or sits in a region with no obvious trap, it's probably a depositional or processing artefact, not a hydrocarbon indicator.

Conformance is essentially a question of whether two maps look the same. Make the amplitude map (this section's widget). Make the structural map (a horizon-pick TWT contour map, or a coherence map outlining faults). Overlay them. If the amplitude follows the structure, you have a candidate prospect. If it doesn't, something else is causing the amplitude.

Tuning and reservoir thickness

Section 1.7's tuning curve told us that reflector amplitude depends on bed thickness in a non-monotonic way: as a bed thins from large thickness toward the tuning thickness, half the wavelet's trough-to-trough period $T_d$ (a quarter of the wavelength at the dominant frequency), its amplitude rises, to 1.45 times that of one interface at tuning, because the top and base reflections interfere constructively. Below tuning, amplitude rapidly decreases as thickness goes to zero.

For reservoir characterization this is both a help and a hazard:

• Help. Below tuning thickness, the amplitude depends on thickness, so amplitude maps are also (calibrated against well control) reservoir-thickness maps. This is one of the most-used quantitative interpretation results: net pay thickness derived from amplitude.
• Hazard. A bright spot caused by tuning of a sub-resolution shale-sand-shale package can look indistinguishable from a bright spot caused by hydrocarbons in a thicker reservoir. Spectral decomposition (Section 6.2) helps disambiguate, a tuned response shows specific frequency enhancement; a hydrocarbon response shifts the whole spectrum.

The practical rule: never quote reservoir thickness from amplitude alone unless you have well control to calibrate the relationship for THIS reservoir in THIS data.

Beyond this section: AVO and inversion

Two bodies of technique carry reservoir characterization from maps of seismic response to maps of rock properties, and both are built on what Parts 5 and 7 teach:

• AVO (amplitude versus offset). How the reflection changes with the angle of incidence (Sections 5.4 and 5.5): fluid, lithology and porosity leave different intercepts and gradients, and a Class III response is the classic gas-sand signature. It needs pre-stack data with the offset-dependent amplitudes preserved through processing; this section works on the stack.
• Seismic inversion. Inverting the seismic for acoustic impedance, or with angle stacks for $V_P$, $V_S$ and density (Sections 5.6 and 7.3), turns a reflectivity volume into a property volume that no longer depends on the wavelet's conventions. Impedance maps and $\lambda\rho$ against $\mu\rho$ crossplots then feed the same gated mapping, calibrated at the same wells.

Both sit on top of the post-stack toolbox of this Part: the gate, the pick and the well tie matter for an impedance map exactly as they do for an RMS map.

Common reservoir-characterization pitfalls

• Cherry-picking attributes. The interpreter tries every attribute until they find one that "shows" the desired anomaly. With dozens of attributes, you can almost always find SOMETHING that matches your hypothesis. Discipline: pick the attributes you'll use BEFORE you look at the result, based on the geological hypothesis. Document the choice.
• Confirmation bias. An interpretation that found one DHI tends to find more, because the interpreter is now looking for support. Counter: actively search for evidence against the hypothesis. What would FALSIFY this interpretation? Look for that.
• Confusing tuning with hydrocarbons. The single most common mistake. A tuned thin bed produces amplitude effects that mimic gas. Always check thickness via spectral decomposition or, ideally, well control.
• Trusting attributes over wells. A bright spot that disagrees with a nearby well is almost always wrong (the well is much closer to ground truth than the seismic). The exception is when the well is far enough away that lateral facies changes can plausibly explain the discrepancy. Be honest about the well distance.
• Acquisition footprint masquerading as geology. Stripey amplitude or coherence patterns aligned with the acquisition shoot direction are almost certainly footprint, not geology. Always check this before reporting an anomaly.
• Ignoring uncertainty. Every attribute has a confidence; every map has zones where the data is good and zones where it isn't. Communicating uncertainty (e.g. confidence intervals on net pay, or "high-risk vs low-risk parts of the reservoir") is what distinguishes mature interpretation from naive overconfidence.

Closing Part 6

You now have the complete attribute toolkit:

• Amplitude attributes (Section 6.1) for "how loud is the reflectivity here"
• Frequency attributes (Section 6.2) for "what spectral colour is the wavelet"
• Geometric attributes (Section 6.3) for "how is the reflector tilted and bending"
• Coherence (Section 6.4) for "how similar to neighbours"
• Multi-attribute combination (Section 6.5) for integrating three at once into a colour image
• Horizon-gated maps (this section) for converting volumes into reservoir-delivery maps

Combined with the structural interpretation tools from Part 2 (horizons, faults, QC), this closes the foundational loop of post-stack quantitative interpretation. From here, the next stages of an interpreter's training go in two directions:

• Quantitative depth: rock physics (Part 5), AVO, pre-stack inversion, geomechanics. These extend the interpretive framework to predict ROCK properties from SEISMIC properties.
• Geological breadth: stratigraphic interpretation (Part 4), structural geology in depth (Part 3), basin analysis. These extend the interpretive framework to predict where the rock came from and how it ended up where it is.

Part 6 is one foundational block in the larger interpretive curriculum. Whatever you build next, you have what you need to read what a seismic volume is telling you.