conditional simulation, Oil & Gas Glossary

What Is Conditional Simulation?

Conditional simulation (also called stochastic simulation or geostatistical simulation) is a numerical modeling technique that generates multiple equally probable three-dimensional realizations of a reservoir property — such as porosity, permeability, or lithofacies — such that each realization exactly matches the measured values at well control points while reproducing the spatial variability structure encoded in a variogram. Unlike kriging, which produces a single smoothed estimate, conditional simulation generates an ensemble of models that spans the plausible range of geological outcomes, providing the P10, P50, and P90 inputs needed for probabilistic reserves estimation and development planning under uncertainty.

Key Takeaways

Each realization in a conditional simulation ensemble honors the conditioning data exactly at well locations — well-log-derived porosity or permeability at every sampled point is reproduced without error in every realization.
Sequential Gaussian Simulation (SGS) is the most widely used algorithm; it visits grid nodes in random order and draws property values from a local conditional probability distribution defined by kriging with the data and previously simulated nodes.
A typical reservoir model may use 50–200 realizations to characterize uncertainty; flow simulation on the full ensemble generates a production forecast distribution from which P10/P50/P90 recovery are extracted.
Conditional simulation preserves the variogram (spatial correlation length, range, sill, nugget) of the original data, meaning inter-well heterogeneity at the scale of the variogram range is realistically reproduced — a property kriging explicitly suppresses by smoothing.
Object-based simulation (for channels, lobes, turbidites) and multi-point statistics (MPS) are alternatives to SGS when the target geometry is non-Gaussian and cannot be captured by a two-point variogram alone.

Conditional Simulation vs. Kriging: Why Both Are Needed

Kriging is the best linear unbiased estimator of a property at an unsampled location given the surrounding data and a variogram model. It minimizes estimation variance, which means it produces the smoothest possible map that honors the data — a desirable property for a single "best-guess" model. However, this smoothing destroys the spatial variability structure of the actual reservoir. Kriged porosity maps show unrealistically gradual transitions between high- and low-porosity zones; kriged permeability maps understate the contrast between flow baffles and high-permeability streaks. When a single kriged model is used for flow simulation, the simulator sees a homogenized reservoir that overestimates sweep efficiency and underestimates early water breakthrough, leading to overly optimistic recovery forecasts.

Conditional simulation solves this by generating realizations that have the correct spatial variance at every scale captured by the variogram. Each realization looks geologically rough and heterogeneous — individual realizations may show high-permeability channels threading through lower-permeability matrix, shale baffles cutting across the reservoir, or porosity corridors aligned with depositional dip — consistent with what a real reservoir looks like at the scale of the model grid. The ensemble of realizations brackets the range of geological outcomes given the available well data, and running flow simulations on a representative subset of realizations (commonly 10–30 out of a 100-realization ensemble) produces a range of recovery predictions rather than a single deterministic number. The spread of that range directly quantifies reservoir geological uncertainty as seen in production outcomes.

In practice, both tools are used together. Kriging provides the E-type (expected value) map — the average of all realizations — which is the best single spatial estimate and is used for visualization and quick-look volumetrics. Conditional simulation provides the uncertainty envelope. The P90 realization (the pessimistic tail of the ensemble ranked by pore volume or recovery) feeds proved reserves estimates under SEC definitions; the P10 realization feeds upside resource assessments for investment decision analysis. This dual use of kriging for the mean and conditional simulation for uncertainty is now standard workflow in static reservoir modeling.

Fast Facts: Conditional Simulation

Primary algorithm: Sequential Gaussian Simulation (SGS)
Conditioning data: well-log measurements (porosity, permeability, Vshale, net-pay flags)
Typical ensemble size: 50–200 realizations per property
Spatial input required: variogram model (range, sill, nugget, anisotropy ratios)
Key difference from kriging: reproduces spatial variance; kriging suppresses variance through smoothing
Probabilistic outputs: P10 (optimistic), P50 (median), P90 (pessimistic) recovery scenarios
Alternative algorithms: Sequential Indicator Simulation (SIS) for categorical variables; object-based simulation for channels/lobes; MPS for complex patterns
Software platforms: Petrel (Schlumberger), IRAP RMS (Roxar/Emerson), Isatis (Geovariances), GSLIB (open-source)

Reservoir Modeling Tip:

The quality of a conditional simulation is only as good as its variogram model. Before running SGS, invest time in experimental variogram calculation from all available well data — both along-well and cross-well if interwell distances allow. Fit the variogram model carefully, paying attention to the nugget (small-scale variability below data resolution), the range (correlation length in each direction), and anisotropy (ratio of major to minor range). A poorly fitted variogram produces realizations that look statistically wrong even if they honor well data, leading to flow simulation results that cannot be trusted for reserves booking.

Conditional simulation is also referred to as:

Stochastic simulation — emphasizes the random (stochastic) drawing of values from probability distributions during realization generation, as opposed to the deterministic interpolation of kriging.
Geostatistical simulation — highlights the geostatistical framework (variogram-based spatial statistics) underpinning the method, distinguishing it from object-based or process-based simulation approaches.
Monte Carlo reservoir simulation — used loosely in some corporate contexts to describe the broader process of running many flow simulations on stochastic property realizations, though strictly speaking Monte Carlo refers to the random sampling process rather than the spatial simulation method itself.
SGS realization — refers specifically to a single model generated by the Sequential Gaussian Simulation algorithm, the most common individual output of a conditional simulation workflow.

Frequently Asked Questions About Conditional Simulation

How many realizations are needed for a reliable P10/P50/P90 estimate?

The answer depends on the dispersion of the ensemble — how different realizations are from one another — and the accuracy needed for decision-making. For most reservoir properties, 50–100 realizations are sufficient to stabilize the P10, P50, and P90 of the pore volume distribution to within a few percent. However, if flow simulations are expensive (days per run for a large compositional model), fewer realizations may be practical. In that case, representative realization selection methods (such as proxy-based ranking or distance-based selection) are used to identify the subset of realizations that best spans the full ensemble for flow simulation, providing good uncertainty coverage with 10–30 runs rather than 100.

What is the difference between Sequential Gaussian Simulation and object-based simulation?

SGS works by filling a grid cell by cell, drawing porosity or permeability values from a local Gaussian distribution. It is flexible, fast, and works well when the geological architecture can be described by a variogram — for example, in sheet-like or mildly heterogeneous clastic reservoirs. Object-based simulation instead places geometrically defined geological bodies (channels, lobes, bars, shale drapes) into the model domain by sampling their dimensions, orientations, and positions from probability distributions derived from analogue data. Object-based methods are superior when the reservoir architecture is strongly non-Gaussian — braided channel systems, turbidite lobe complexes, or fractured carbonates — where the variogram cannot capture the multimodal geometry. The trade-off is that object-based models are harder to condition to dense well control because placing objects without violating well data becomes computationally challenging.

Does conditional simulation replace deterministic geological modeling?

No. Conditional simulation is a tool for capturing and communicating uncertainty, not a replacement for geological interpretation. The best practice is to build one or more deterministic base-case models informed by sequence stratigraphy, core analysis, seismic interpretation, and analogue data, then use conditional simulation to generate uncertainty realizations around that geological framework. The variogram, facies proportions, and geometric constraints fed into the simulation algorithm all carry the geologist's interpretation. A simulation run with no geological constraints produces statistically valid but geologically meaningless realizations. The strength of conditional simulation lies in its ability to honor the geological model structure while exploring the range of outcomes consistent with the sparse well data.

Why Conditional Simulation Matters in Oil and Gas

Modern reserves reporting under SEC, SPE-PRMS, and NI 51-101 frameworks explicitly requires probabilistic resource estimates expressed as P90/P50/P10 outcomes. Conditional simulation is the technical engine that converts sparse well data and seismic observations into defensible probability distributions for reserves. Beyond regulatory compliance, the ensemble outputs from conditional simulation drive capital allocation decisions: the P90 case determines the minimum economic threshold a project must exceed for sanction, while the P10 case defines the upside that justifies exploration risk. As fields mature and well spacing decreases, conditional simulation realizations update dynamically with new conditioning data, providing a continuously improving picture of reservoir heterogeneity and remaining upside that guides infill drilling and enhanced recovery decisions.

Conditional Simulation: Quantifying Reservoir Uncertainty in Geostatistics