Capstone: 4D repeatability challenge

Most 4D projects use a baseline and monitor acquired with matched equipment. This capstone is the hard case: 18 years between a 2006 legacy streamer baseline and a 2024 OBN monitor, with different sources, geometries, bandwidths and processing environments. Every technique of Part 8 has to work, and even then the achievable NRMS is much higher than for a dedicated 4D survey.

Project setup

A legacy field under about 1100 m of water. Baseline: 2006 narrow-azimuth streamer, 8 km cables, peak frequency about 25 Hz, no usable signal below 5 to 6 Hz. Monitor: 2024 OBN, 400 m node spacing, four-component, peak frequency about 25 Hz with usable signal down to about 1.5 Hz. Eighteen years of production have caused compaction, fluid-contact movement and pressure changes that the 4D processing should reveal. Primary deliverables: a compaction (time-shift) map and a fluid-movement map.

NRMS between a baseline trace $b$ and a monitor trace $m$ in a window is $\\mathrm{NRMS}=200\\,\\mathrm{RMS}(m-b)/(\\mathrm{RMS}(m)+\\mathrm{RMS}(b))$ percent: 0 % is a perfect repeat, about 141 % is two uncorrelated traces of equal energy, and 200 % is a polarity flip.

The pipeline

Differenced as delivered, the two volumes in the figure disagree by 154 % NRMS in the quiet overburden, past the 141 % of two unrelated traces: they differ in water column, wavelet, azimuth and migration before they differ in the reservoir. The joint flow brings the NRMS to 28 %, inside the 25 to 40 % band for this kind of project, and the ladder in (d) shows which stages paid for it. At that floor the crest's compaction time shift, measured at 3.3 ms, stands far above its scatter off the reservoir in (e), and the gross sweep stands out in (b); lower $\\Delta Z/Z$ to 2 % and the swept zone sinks into the residual.

Why this is the hardest 4D case

• Geometry mismatch. Streamers move; nodes stay put. Streamer and node surveys share no traces, so there is no trace-to-trace pairing: both are regularised onto a common grid (5D interpolation), and the node data are decimated to the streamer's offsets and azimuths before anything is matched. What regridding cannot restore is the part of the wavefield that depended on exactly where each source and receiver stood; that positioning floor grows with $\\Delta S + \\Delta R$.
• Wavelet mismatch. The 2006 air-gun array and the 2024 array differ in volume, depth and measured signature, and the streamer carries a receiver ghost that the sea-floor nodes do not. Designature, deghosting and Wiener cross-equalisation, designed in a window where production changed nothing, above the geomechanical halo, are mandatory.
• Bandwidth mismatch. The legacy data have no signal below 5 to 6 Hz; the monitor has signal down to 1.5 Hz. Spectra cannot be matched below the legacy cutoff, and the report must say that these frequencies are absent from the baseline.
• Water column. Tide and water velocity differ between the surveys. Through 1100 m of water a 4 m/s change in water velocity is about 4 ms, a whole-trace static computed from each survey's measured water-velocity profiles and tide.
• Processing environment mismatch. The legacy data were processed in 2006 with 2006 tools and conventions (possibly with AGC somewhere), the monitor with 2024 tools. Both must be reprocessed jointly from the raw gathers and migrated with one operator and the baseline velocity model; comparing the archived outputs is no shortcut.

Realistic NRMS targets

• Dedicated 4D (matched acquisition and joint processing): under 15 %.
• Legacy versus modern, joint-reprocessed: 25 to 40 %. This project; areas above about 40 % count as unmatched.
• Legacy versus modern, best-effort matching only: 50 to 70 %. Too noisy for quantitative interpretation.

What this project can deliver

• Compaction maps. Eighteen years of production stretch the overburden above compacting reservoirs, producing time shifts of a few milliseconds (typically 2 to 10 ms). Time shifts are measured over whole windows by warping, so they read through the noise floor; their slope with time, the time strain $\\partial \\Delta t/\\partial t \\approx \\Delta z/z - \\Delta v/v$, maps where the overburden stretched.
• Gross fluid-contact movement. Lateral sweep of oil-water and gas-oil contacts over hundreds of metres, and vertical contact rise of several tens of metres, are detectable.
• Pressure depletion (qualitative). Velocity increases in depleted reservoirs as effective stress rises, and decreases in the stretched overburden above them; visible on 4D, but with about 20 % quantitative uncertainty.

What this project cannot deliver

• Detailed fluid fronts with a precision better than 100 m.
• Quantitative saturation changes: rock-physics inversion needs a repeatability this dataset does not have.
• Reliable thin-bed 4D: tuning effects dominate the residual.

Part 10 closes here

Six capstones, six project archetypes: land vibroseis, OBN deep-water salt, WAZ FWI, CCS monitoring, geotechnical near-surface and legacy-to-modern 4D. Each uses techniques from Parts 0 to 9 in different combinations, with different stages on the critical path. The skill of a seismic processor is not knowing a single flow but recognising which combination of techniques is right for each project.

What Part 11 provides

Part 11 is the quiz bank: about ten questions are drawn per previous part (sampled from a larger pool), plus a 50-question final exam drawing from all eleven parts. Use it to check your understanding, section by section.

References

• Yilmaz, Ö. (2001). Seismic Data Analysis (2 vols.). SEG.
• Sheriff, R. E., Geldart, L. P. (1995). Exploration Seismology (2nd ed.). Cambridge UP.
• Claerbout, J. F. (1976). Fundamentals of Geophysical Data Processing. McGraw-Hill.
• Kragh, E., Christie, P. (2002). Seismic repeatability, normalized rms, and predictability. The Leading Edge, 21, 640.
• Hatchell, P., Bourne, S. (2005). Rocks under strain: strain-induced time-lapse time shifts are observed for depleting reservoirs. The Leading Edge, 24, 1222.