Time-lapse (4D) seismic: monitoring reservoir changes

The 4D acquisition setup

A 4D project requires TWO (or more) seismic surveys acquired at DIFFERENT calendar dates over the SAME area:

• Baseline: the pre-production or early-production survey. This is the reference. Every monitor survey is compared to it.
• Monitor: subsequent surveys at 1-5 year intervals. Each is compared to the baseline to reveal cumulative changes.

The central challenge: the two surveys have to be REPEATABLE. Small changes in the reservoir produce small 4D signals, often 5-15% of the baseline amplitude. If the acquisition geometry is even slightly different, that alone creates differences of the same magnitude, drowning the real signal. This is why 4D projects invest enormously in ACQUISITION REPEATABILITY: same boat, same streamer geometry, same source, same time of year, same water temperature, same season-dependent pressure effects. For permanent-seabed surveys (ocean-bottom nodes, OBN), the sensors stay in place for years, the gold standard for repeatability.

The quantitative repeatability metric is the NRMS (Normalized RMS) difference in a NON-RESERVOIR zone (where nothing should change). NRMS < 10% is excellent (well-acquired 4D); 10-25% is typical field data; > 35% usually means the 4D signal can’t be trusted.

Exercise, run the time-lapse

The widget below is a forward-modeling pipeline, not a colored diagram. A faulted anticline in depth is turned into elastic properties by real Gassmann fluid substitution with a stress-sensitive dry frame, traveltimes are integrated through each survey's own velocity model (so 4D time shifts EMERGE from the physics rather than being painted), reflectivity is convolved with a Ricker wavelet, and each survey carries its own noise plus per-trace static errors. The NRMS and δt numbers in the readout are then MEASURED from the two synthetics the way a 4D processor measures them.

• Start in Reservoir model (the truth) with the Waterflood dynamic and drag the years slider. The oil-water contact rises through the closure, leaving swept sand behind it, and stops dead at the sealed growth fault. Note the isolated compartment beyond the fault keeps its own DEEPER contact: two contacts across one fault is the classic evidence that the fault seals.
• Switch to Baseline survey. The reservoir is the bright soft event draped over the anticline; on the flanks, look for the FLAT SPOT where the contact cuts the sand, and find the second, deeper flat spot in the isolated compartment. The strong flat events below are the beacon markers used for time shifts.
• Switch to Monitor, then flip back and forth to the baseline. At realistic noise you can barely tell them apart: reservoir changes are a few-percent effect. That is why 4D works on differences, and why repeatability is everything.
• Open 4D difference. The swept zone is a hardening (red-over-blue) anomaly confined between the old and new contacts, and its edge stops at the fault. Now look BELOW the reservoir: the beacon events light up even though nothing changed down there, because reflectors MOVED in time. Every difference anomaly must pass three questions: is it above the noise floor, does its edge follow geology, and is it a time-shift artifact of a change above?
• Drag years to 0 with the noise up. The difference does not go blank; it shows the NRMS floor, and the readout confirms the reservoir "signal" now matches the overburden NRMS. This is what a false 4D anomaly looks like, and why NRMS in a zone that cannot have changed is the central QC number.
• Now push the noise slider toward its maximum with years at 5. Watch the flood front drown as NRMS climbs through the 10-25% typical band toward the > 35% untrustworthy band from the acquisition discussion above. The signal did not change, the floor rose.
• Switch the dynamic to Gas exsolution. Wood mixing makes a little gas collapse the fluid modulus, so the growing gas cap is the strongest softening anomaly in the widget, and in the 4D time shifts view the beacons record a POSITIVE shift: push-down, the wave slowed. Compare Depletion + compaction: almost no saturation change, a weaker hardening, and a NEGATIVE shift, pull-up.
• Now watch the trap. Go back to Waterflood in the shift view: it pulls up TOO, and at year 10 its shift is comparable to the gas case in size. The sign separates softening from hardening, but it does not separate a flood from compaction, because both stiffen the rock. What separates them is the RATIO of shift to amplitude: compaction produces the largest time shift of the four dynamics with the weakest amplitude anomaly, while the flood's anomaly is sharp, contact-bounded and amplitude-rich. That ratio is the observation quantitative pressure-saturation decomposition is built on.
• Finish with Injection pressure halo: softening around the injector from the pressure-up, hardening from the sweep, two signals of opposite sign in one image, and the MacBeth stress law makes the softening stronger than an equal depletion would stiffen. This is the pressure-saturation ambiguity that quantitative 4D decomposition exists to resolve.
• In the shift view at any setting, the dots are measured by cross-correlation and the dashed curve is the earth model's own prediction. Raise the noise and watch the measurement scatter around the truth: sub-millisecond time shifts demand excellent repeatability.

Rock-physics primer for 4D signatures

Each reservoir dynamic has a characteristic elastic fingerprint. Here’s the rock physics of the main ones:

• Water replacing oil: brine is denser and stiffer than oil. Gassmann fluid substitution predicts Ip RISES ∼5-10%, Vp/Vs rises slightly. 4D amp-difference is POSITIVE (redder, stiffer). Time shift roughly neutral.
• Oil replacing gas (rare, during repressurization): Ip DROPS slightly (gas goes back into solution and compressibility of the mix falls back toward oil). Inverse of depletion.
• Solution gas exsolution (pressure drops below bubble point): dramatic drop in Ip (∼10-20% drop) because gas bubbles in the pore space enormously increase fluid compressibility. 4D amp-difference is strongly NEGATIVE. Signature is characteristic of PRESSURE DEPLETION.
• Reservoir compaction: effective stress rises as Pp drops; rock matrix stiffens. Vp and Vs both rise ∼5-10%, density rises ∼2%. Ip rises ∼7-12%. Reflectors below the reservoir pull UP (earlier arrival) because the wave spends less time in the stiffened rock.
• CO₂ injection: CO₂ (supercritical state) has low bulk modulus like natural gas. Ip drops similarly to gas exsolution. The plume edge is the key 4D anomaly. See Section 8.6 for the full story.

Each signature is a TELL. Skilled 4D interpreters can often distinguish water flood, depletion, and compaction from the amplitude + time-shift patterns alone, without needing additional logs. But quantitative predictions require FORWARD MODELING (taking your rock-physics template, applying the hypothesized dynamic, computing the expected 4D signature, and comparing to observed).

Quantitative 4D: pressure-saturation decomposition

Sometimes two dynamics happen simultaneously, waterflood AND compaction, or injection AND production. The 4D signal is a MIXTURE. Quantitative 4D (Q-4D) aims to DECOMPOSE the observed signal into pressure and saturation contributions:

Δ(Amplitude) = f(ΔP) + g(ΔS) + ε

Using pre-stack 4D data (angle-dependent 4D differences), you can separately solve for ΔP and ΔS per voxel. The math is essentially a 4D version of the pre-stack elastic inversion from Section 7.3, instead of inverting for (Ip, Is, ρ) at a fixed time, you invert for (ΔP, ΔS) between two time instants.

Outputs: pressure-change cube + saturation-change cube, volume by volume. Directly feeds reservoir-engineering models. The high-value deliverable of a mature 4D program.

Repeatability and common pitfalls

• Misaligned geometry. The baseline boat was 25 m off-line from the monitor boat. Result: 4D signal dominated by geometry-induced differences in fold, azimuth, offset coverage. Mitigation: cross-equalize with careful match-filter processing; use PRE-STACK methods that are less sensitive to geometry.
• Seasonal water column. Baseline acquired in January (cold water, fast sound speed), monitor in July (warm, slow). Travel times differ BEFORE anything reservoir-related. Correct via: time-of-year matching, or water-velocity time-lapse correction using measured profiles.
• Processing differences. The two surveys were processed on different software versions with slightly different demultiple parameters. Fixed: REPROCESS baseline and monitor TOGETHER through the same flow.
• Overburden changes. Regional fluid withdrawal in a field above your reservoir changes seismic velocities in the overburden. 4D signal in the reservoir can be shifted and scaled by overburden effects. Correct with: volumetric time-shift inversion, or explicit overburden geomechanical model.
• False 4D from NOISE. Low NRMS in the reservoir zone but HIGH NRMS in nearby shale. The noise floor could be producing false 4D "anomalies." Always compare reservoir 4D against shallow-overburden NRMS as a sanity check.
• Acquisition gaps. Missing line or infill zone with no monitor coverage. Clearly flag these areas; don’t over-interpret edge effects.

Applications in practice

• Norwegian Sea (Ekofisk, Valhall): world-class 4D programs in compaction-dominated chalk reservoirs. Time shifts up to 30 ms documented over decades. Drives repressurization + waterflood programs.
• Troll Field (North Sea): waterflood 4D revealed that gas was bypassing water in certain zones, changed development strategy dramatically.
• Sleipner (Norway): flagship CO₂ storage 4D project. Annual repeat surveys since 1996 clearly show the CO₂ plume growth within the Utsira aquifer. See Section 8.6 for the case study.
• Gulf of Mexico deep-water: 4D on permanent ocean-bottom-node surveys has become the standard for high-value subsea developments. Near-daily updates in some cases.
• Unconventional (shale): 4D in completion-induced fracture networks, microseismic 4D, and DAS (Distributed Acoustic Sensing) are rapidly maturing.