4D time-lapse monitoring at a Sleipner CO2-injection target

Part 10 — Real-field capstones

Learning objectives

Pivot from static (3D) to dynamic (4D) seismic inversion
Recognise the time-lapse prior: only the plume changed between surveys
Apply PINN-augmented 4D inversion with regional + plume-shape priors
Visualise CO2 plume migration via Δv difference maps
Connect to production Sleipner monitoring (Chadwick et al 2010)

§10.5 inverted a STATIC sub-salt structure. §10.6 takes the next dimension: TIME. Sleipner (Norwegian North Sea, Equinor) is the canonical commercial-scale CO₂ injection target — running since 1996, ~1 Mt CO₂/year injected into the Utsira sand reservoir at 800-1100 m depth. Repeat 4-D seismic surveys since 1999 track plume migration in REAL TIME. The 4-D inversion problem: given baseline (pre-injection) and monitor (post-injection) surveys, infer the velocity DIFFERENCE Δv(x, z) caused by CO₂ saturation.

The Sleipner setup

Geology (this widget):

Caprock: Nordland Shale, V_p = 1.9 km/s, depth 0-500 m
Utsira sand: porous reservoir, V_p = 2.05 km/s, depth 500-1100 m. CO₂ injected here.
Basement: Oligocene shales, V_p = 2.8 km/s, depth >1100 m

The CO₂ plume is modelled as a Gaussian-shaped saturation cloud in the Utsira sand. Truth parameters: x_centre = 1.7 km, z_centre = 0.85 km, σ = 0.30 km, peak Δv = −0.25 km/s (a 12% velocity drop in the fully-saturated cell — upper end of the 5-15% range typical of free CO₂ above the dissolution limit). 8 surface sources × 16 surface receivers, σ_pick = 0.002 s on each survey (high-repeatability marine 4D, achievable with cable-based receivers; mimics Sleipner's actual 5-7% NRMS repeatability).

The time-lapse prior

The 4-D inversion problem WITHOUT a time-lapse prior would re-invert the entire velocity model (baseline + plume) from monitor data alone — multiplying parameter dimensionality by O(1000s of cells) and risking the optimizer confusing baseline structure with the plume.

The TIME-LAPSE prior says: between baseline and monitor surveys, only the CO₂ plume changed. Hold the baseline velocity model FIXED. Invert ONLY the plume parameters (x_centre, z_centre, σ, Δv_max) — 4 unknowns instead of thousands. This is the production Sleipner monitoring recipe (Chadwick et al 2010).

Mathematical statement:

\mathcal{L}_{\mathrm{4D}}(\Theta) = \frac{1}{2} \sum_{s, r} \bigl( (t_{\mathrm{mon, pred}} - t_{\mathrm{base, obs}})^{(s, r)}(\Theta) - (t_{\mathrm{mon, obs}} - t_{\mathrm{base, obs}})^{(s, r)} \bigr)^2 + \lambda \, \mathcal{L}_{\mathrm{phys}}(\Theta)

where $\Theta = (x_c, z_c, \sigma, \Delta v_{\max})$ are the plume parameters and $\mathcal{L}_{\mathrm{phys}}$ encodes regional priors:

Plume near injection well (x_c near x_inj = 1.5 km)
Plume in Utsira sand (z_c in [Z_caprock, Z_utsira_bot])
Plume amplitude PHYSICALLY NEGATIVE (CO₂ slows velocity, so Δv ≤ 0)
Plume size plausible (σ ∈ [0.1, 0.6] km)

Try it

Four panels (top-left → bottom-right):

Baseline velocity model: 3-layer cake (caprock / Utsira / basement), with the dashed white line marking the injection well at x = 1.5 km.
Monitor velocity (truth): same colormap, with a faint velocity DROP visible in the Utsira sand right of the injection well (the CO₂ plume).
Truth Δv map: monitor minus baseline. Red blob = velocity drop = CO₂ plume.
Recovered Δv map: monitor minus baseline AFTER PINN-aug 4D inversion. Should match truth Δv (modulo plume-shape limits of surface seismic).

Below the panels, a summary box reports truth plume parameters, recovered plume parameters, plume migration distance from injection well, and parameter L² error.

Expected behaviour: with the time-lapse prior, the 4-parameter inversion recovers the plume location to within 100-200 m and the peak Δv to within 0.05 km/s on most seeds. The recovered plume migrating ~0.2 km up-dip from the injection well mirrors the real Sleipner plume's lateral migration along the Utsira-Nordland Shale boundary (Bickle 2009).

Why time-lapse seismic works for CO₂

The seismic Δv signature of CO₂ is characterised by Gassmann fluid substitution. The dry-frame V_p depends on the rock matrix (porosity, mineralogy); fluid saturation modulates V_p via the Biot-Gassmann coupling. Adding CO₂ (compressible) to brine-saturated sand DROPS V_p by 5-15% depending on saturation. The drop is OBSERVABLE on time-lapse seismic if pre-injection and post-injection surveys are repeatable enough — and Sleipner runs DEEP-TOWED streamer arrays + ocean-bottom-cable receivers specifically tuned for time-lapse repeatability.

4D AMPLITUDE inversion. Travel-time alone has limited shape resolution; production codes invert the full pre-stack reflection AMPLITUDE difference between surveys, recovering Δv with much better lateral resolution.
Multiple monitor surveys. Sleipner has been monitored at 1999, 2001, 2002, 2004, 2006, 2008, 2010, 2013... — each produces a snapshot of the plume's growth. Joint inversion across all snapshots constrains plume-MIGRATION dynamics.
Repeatability metrics. Production 4D quotes NRMS (normalised RMS) repeatability between surveys; <10% NRMS is needed for reliable Δv recovery. Sleipner achieves 5-7% NRMS with cable-based receivers.
Mass-balance verification. Combine 4-D seismic Δv with rock-physics models to estimate CO₂ MASS in the plume. Compare to known injection volume from the plant. Sleipner mass-balance closes within ±10% over 25 years of operation.

Other CO₂ storage projects monitored 4D

Snøhvit (Barents Sea, started 2008): CO₂ injection into the Tubåen formation; monitored similarly to Sleipner.
In Salah (Algeria, 2004-2011): different setup — CO₂ injected into a producing gas field via lateral wells. Halted due to caprock-integrity concerns flagged by 4D + InSAR monitoring.
Quest (Alberta, Canada, started 2015): Shell's commercial CCS for the Scotford upgrader.
Northern Lights (Norway, started 2024): EU-wide CO₂ transport + storage hub; 4D monitoring planned.

What §10.7 will do

§10.7 closes the forward-operator loop and runs full eikonal-based BASIN tomography — body-wave first-arrival inversion of 2-D velocity over a sedimentary basin from many seismic events to a regional receiver array. Where this widget uses straight-ray as a small-perturbation linearisation around the baseline, §10.7 generalises to the eikonal forward operator (replaceable by PINN-eikonal at production scale).

References

Chadwick, A., Williams, G., Delepine, N., et al (2010). Quantitative analysis of time-lapse seismic monitoring data at the Sleipner CO₂ storage operation. The Leading Edge 29(2), 170–177. The standard Sleipner 4D paper.
Arts, R., Eiken, O., Chadwick, R.A., Zweigel, P., van der Meer, B., Kirby, G.A. (2004). Seismic monitoring at the Sleipner underground CO₂ storage site. Energy 29(9-10), 1383–1392. First decade of Sleipner monitoring.
Bickle, M.J. (2009). Geological carbon storage. Nat. Geosci. 2, 815–818. The geology of Sleipner-class storage.
Furre, A.K., Eiken, O., Alnes, H., Vevatne, J.N., Kiær, A.F. (2017). 20 years of monitoring CO₂-injection at Sleipner. Energy Procedia 114, 3916–3926. The two-decade retrospective.
Lumley, D. (2010). 4D seismic monitoring of CO₂ sequestration. The Leading Edge 29(2), 150–155. 4D monitoring methodology overview.