Capstone 2: Carbonate reservoir with facies modeling

Carbonate reservoirs differ from clastic shelves: porosity is BIMODAL, high in grainstones and packstones, low in mud-dominated facies. A two-step workflow (facies first, porosity within facies) captures this geology far better than a single Gaussian SGS on porosity alone.

Why facies-first matters

If you ignore facies and run pure SGS on porosity in a carbonate, the model has the right average φ but the WRONG variance structure: realizations look smoothly varying instead of sharply transitioning. OOIP estimates miss the connectivity and compartmentalisation that drive flow performance. The two-step approach captures the geometry first, then the within-facies property variation.

Workflow stages

• Facies-indicator variogram: fit the variogram on the BINARY grainstone indicator. Range typically shorter than for porosity (facies switch faster than continuous attributes).
• SISIM / indicator simulation: produce 20+ FACIES realizations honoring the target grainstone fraction (e.g., 40%).
• Per-facies porosity SGS: in each realization, fill grainstone cells with φ ~ N(0.22, 0.05²); fill mudstone cells with φ ~ N(0.06, 0.02²). Separate range parameters per facies are common.
• OOIP rollup: per realization, compute Σ cells φ(c) × thickness × area × (1 - Sw) / Bo. Histogram across realizations gives P10/P50/P90.

Uncertainty in carbonate OOIP comes from TWO sources combined:

• FACIES UNCERTAINTY, how much grainstone vs mudstone, and where
• WITHIN-FACIES POROSITY UNCERTAINTY, variation around the facies-conditional mean

Try it

• Defaults (40% grainstone, range = 8, μ_g = 0.22, μ_m = 0.06). Note the bimodal porosity in realization 1, clearly distinct grainstone (high) and mudstone (low) regions. OOIP P90/P10 ratio reflects the combined facies + within-facies uncertainty.
• Increase grainstone fraction to 0.6. OOIP shifts upward (more high-φ rock). Notice the facies map becomes dominated by grainstone.
• Set mean φ_grainstone = 0.30 (very high) and μ_mud = 0.02 (very low). The bimodal contrast becomes extreme; OOIP becomes very sensitive to the grainstone fraction.
• Reduce facies range to 3 (small-scale heterogeneity). The facies maps become much patchier; realizations look very different from each other. OOIP P90/P10 ratio rises.
• Compare against a hypothetical pure-φ SGS with mean = 0.4·0.22 + 0.6·0.06 = 0.124 and sd around 0.08. The mean is the same, but the realizations would look smooth, losing the geological meaning entirely.

An asset team has 25 wells, 10 of which are in grainstone-dominated facies. The first 20% of the reservoir thickness is logged as grain-dominated, then transitions to mud-dominated. How should you encode this VERTICAL TREND in the facies modeling workflow, and what is the consequence of ignoring it?

What you now know

You can run a two-step carbonate workflow: facies first, then per-facies porosity. The same approach scales to multi-facies (3+ rock types) and to 3D grids. The widget's indicator-then-Gaussian approach is the basis of SISIM in real geostatistical software, the GSLIB SISIM program implements this exactly, with proper variogram-driven sequential indicator simulation.

Mark capstone complete →

References

• Pyrcz, M.J., Deutsch, C.V. (2014). Geostatistical Reservoir Modeling, 2nd ed. Oxford. (Carbonate facies modeling chapter.)
• Lucia, F.J. (2007). Carbonate Reservoir Characterization: An Integrated Approach, 2nd ed. Springer.
• Deutsch, C.V., Journel, A.G. (1998). GSLIB. SISIM module reference.
• Strebelle, S. (2002). "Conditional simulation of complex geological structures using multiple-point statistics." Math. Geol. 34(1).
• Pyrcz, M.J., Boisvert, J.B., Deutsch, C.V. (2008). "Multipoint and object-based modeling of meandering river reservoirs." Petroleum Geoscience 14, 119-135.