Capstone 6: CO₂-storage site characterization

Part 10 — Reservoir-characterization capstones

Learning objectives

  • Build a caprock-thickness uncertainty model from well measurements + variogram
  • Compute P(caprock failure) = P(min thickness in storage footprint < critical threshold)
  • Recognise that rare-event failure probabilities are TAIL-driven, not mean-driven
  • Translate realisation distributions into go / no-go regulatory decisions

The final reservoir-characterisation capstone closes the textbook with the most modern application: CO₂ geological storage. The decision-relevant question is not "what is the expected caprock thickness?" but "what is the PROBABILITY OF CONTAINMENT FAILURE?". This is a rare-event question driven by the LOWER TAIL of the thickness distribution — exactly the question geostatistical realisations are built to answer.

The integrity-rule framework

Storage-site integrity rests on the caprock seal. A typical regulatory criterion: caprock thickness must exceed a critical threshold t_crit (e.g., 30 m) EVERYWHERE within the CO₂ plume footprint. The failure event is:

Pfailure=P(minxfootprintthickness(x)<tcrit).P_{\text{failure}} = P(\min_{x \in \text{footprint}} \text{thickness}(x) < t_{\text{crit}}).

This is the probability that AT LEAST ONE cell is below threshold, NOT the probability that the mean is. Note the dramatic asymmetry: a site with mean = 50 m and SD = 12 m has very low P(mean < 30) but the P(min cell < 30) across a 32×32 grid is much higher — because the min has heavier left tail than the mean.

Why realisations matter HERE more than for OOIP

OOIP integrates over many cells — central-limit-style averaging reduces tail risk. Min thickness is the OPPOSITE — it's an extreme-value statistic that is MORE sensitive to tail behavior than to the mean. Generating realisations and counting failures gives the correct rare-event probability; closed-form solutions exist only for very specific assumptions (Gaussian min) but the spatial dependence makes those unreliable.

Workflow

  • Conditioning data: caprock thickness measurements at K wells (with measurement noise).
  • Variogram model: spherical, range determined from sequence-stratigraphic continuity and adjacent depositional analogs.
  • SGS realisations: 40+ realisations honoring well measurements and the variogram model.
  • Min-thickness per realisation: for each realisation, find the minimum thickness across the storage footprint.
  • P(failure) estimate: fraction of realisations with min thickness < t_crit. Repeat across alternative variogram parameters for sensitivity.

Co2 Storage Capstone DemoInteractive figure — enable JavaScript to interact.

Try it

  • Defaults (μ = 50, σ = 12, threshold = 30, range = 8). Mean is 50 m, threshold 30 m, safety margin 20 m — sounds safe. But ~5-15% of realisations show min thickness < 30 m somewhere. P(failure) is dramatically larger than P(mean < threshold) (which is near-zero).
  • Set σ = 4 (small thickness variability). P(failure) drops dramatically. Tight variability = small left tail = safe site.
  • Set σ = 25 (large variability). P(failure) explodes to 40%+. Heterogeneous caprock = thin spots somewhere = unacceptable risk.
  • Set range = 14 (long-correlation caprock). Thin spots become geographically larger (correlated thin zones). P(failure) is similar to short-range case but the FAILED REGIONS are larger and more identifiable.
  • Increase threshold to 40 (more conservative regulator). P(failure) shoots up. The cost of safety: more sites rejected.

A regulator demands P(failure) < 1% before approving CO₂ storage. The site characterisation shows P(failure) = 8%. The asset team proposes "we'll just monitor and shut in if leakage starts". What are the REGULATORY and TECHNICAL flaws of this proposal, and what does Bayesian decision-theory say should be the alternative?

What you now know

You can characterise CO₂-storage sites with statistical containment-failure probabilities. The same realisation-and-count machinery applies to: leak-detection-system design, monitoring-well placement, induced-seismicity risk, post-injection plume extent. CO₂ storage is one of the highest-impact applications of geostatistics — done right, it enables gigatonnes of permanent CO₂ sequestration; done wrong, it produces ineffective or unsafe projects.

This closes Part 10 — the six reservoir-characterisation capstones cover the major reservoir types and analytical decisions. Part 11 quizzes test the cumulative material; Part 12 provides the master-workflow reference card. You now have the complete OgbonLab Geostatistics curriculum.

References

  • IEAGHG (2019). "Caprock systems for CO₂ geological storage." IEAGHG report 2019-08.
  • Chadwick, R.A., Eiken, O., Lindeberg, E. (2009). "Best practice for the storage of CO₂ in saline aquifers." BGS / NORSAR / SACS.
  • Bachu, S. (2008). "CO₂ storage in geological media: role, means, status." Energy Procedia 1(1), 3743-3750.
  • Pyrcz, M.J., Deutsch, C.V. (2014). Geostatistical Reservoir Modeling, 2nd ed. (Closure: from petroleum to CO₂ applications.)
  • IPCC (2005). Special Report on Carbon Dioxide Capture and Storage. Cambridge University Press. (Foundational regulatory framework.)

This page is prerendered for SEO and accessibility. The interactive widgets above hydrate on JavaScript load.