Stochastic Analysis: Probability Distributions, Sequential Gaussian Simulation, and Reservoir Uncertainty Quantification
Stochastic analysis is an analytical approach built around processes whose outcomes are governed by chance, in which each observation is treated as a single sample drawn from a probability distribution rather than as a fixed, deterministic value. The defining feature is that a stochastic model does not produce one answer; it produces a population of equally plausible answers, called realisations, each consistent with the input data but differing in the unmeasured details. This stands in direct contrast to deterministic analysis, which honours the data with a single smooth interpolation, such as kriging or hand-contouring, and reports one number. In the petroleum industry stochastic analysis is the backbone of modern geostatistical reservoir characterisation, where the goal is not just to estimate properties like porosity, permeability, and facies between sparse well control but to honestly represent how uncertain those estimates are. The most widely used engine is sequential Gaussian simulation, abbreviated SGS, which visits every uninformed cell of a reservoir grid in random order, computes the local conditional probability distribution from the surrounding data and previously simulated cells using a kriging system and a variogram model, then draws a random value from that distribution by Monte Carlo sampling. Because the draw is random, running SGS many times with different random seeds produces many different but equiprobable porosity or permeability volumes, each preserving the input statistics, the spatial correlation structure of the variogram, and the hard data at the wells, yet none artificially smoothed. The spread across that ensemble of realisations is the uncertainty. Reservoir engineers then carry a representative subset of realisations forward into flow simulation to produce a distribution of recovery outcomes, commonly summarised as P10, P50, and P90 estimates of original oil or gas in place and of recoverable reserves. This probabilistic framing feeds directly into volumetric reporting, development-plan economics, and reserve booking under standards such as the Canadian Oil and Gas Evaluation Handbook and NI 51-101. In the Western Canadian Sedimentary Basin, stochastic methods are applied across the heterogeneous McMurray oil sands, the laterally variable Cardium and Viking sands, and unconventional Montney and Duvernay resource plays, where a single deterministic map would hide the geological variability that ultimately controls well performance and recovery factor. By preserving that variability across an ensemble, stochastic analysis gives decision-makers a defensible range rather than a false-precision point estimate.
Key Takeaways
- Samples, not single values: Stochastic analysis treats each observation as one draw from a probability distribution, so its output is a population of equiprobable realisations rather than one deterministic answer. The variability across that population is the quantified uncertainty, which deterministic kriging or contouring deliberately smooths away.
- Sequential Gaussian simulation: SGS is the workhorse algorithm: it visits grid cells in random order, builds each cell's conditional distribution from nearby data and a variogram via kriging, then Monte Carlo samples a value. Repeating with new random seeds yields many porosity or permeability volumes that all honour the wells and the spatial statistics.
- Variogram carries the geology: The variogram model defines how strongly a property is correlated over distance and direction, encoding channel orientation, lateral continuity, and vertical layering. It is the single most important input because it controls how realistic the simulated heterogeneity looks between wells.
- P10, P50, P90 outcomes: Flowing a subset of realisations through reservoir simulation produces a distribution of recovery and in-place volumes, reported as percentile estimates. This probabilistic range underpins COGEH and NI 51-101 reserve booking far better than a single deterministic case can.
- Computationally heavy but essential: Each realisation solves a kriging system at every uninformed cell, so generating hundreds of high-resolution realisations is intensive. Practitioners manage cost with ranking, response-surface proxies, and Latin hypercube sampling to capture the uncertainty range with fewer full simulations.
Realisations Versus a Single Best Map
A deterministic porosity map drawn by kriging minimises estimation error and looks smooth, but that smoothness is a lie about the rock: it removes exactly the high and low extremes that control flow barriers and thief zones. Stochastic simulation deliberately reintroduces that texture. Each realisation is rougher than a kriged map and, taken alone, is less accurate at any one point, but the ensemble as a whole captures the realistic range of geological outcomes. Engineers do not ask which realisation is correct; they ask how widely the realisations disagree, because that disagreement is the honest measure of how little the sparse well data truly constrains the inter-well volume.
From Geomodel to Reserve Range
The workflow runs facies first, then petrophysical properties conditioned on facies, then fluid distribution. A typical study might build 100 to 500 equiprobable static realisations, rank them by gross rock volume or connected pore volume, and select a low, mid, and high case plus several intermediates for dynamic flow simulation. The resulting cumulative-production curves spread into a P10 to P90 fan that quantifies recovery uncertainty. This range, not a single number, is what a qualified reserves evaluator signs off on, and it directly shapes facility sizing, well-count decisions, and the economic threshold for sanctioning a WCSB development.
Fast Facts
Geostatistics was born not in oil but in South African gold mining, where engineer Danie Krige and mathematician Georges Matheron developed kriging in the 1950s and 1960s to estimate ore grades from sparse drill samples. The leap from deterministic kriging to stochastic simulation came in the 1980s when researchers realised that minimum-error estimates systematically understated variability, leading directly to the sequential simulation algorithms that the petroleum industry adopted wholesale in the 1990s for reservoir uncertainty work.
Related Terms
Stochastic analysis is the operating principle of geostatistics, the science of spatially correlated data, and it relies on Monte Carlo simulation to draw random samples from each conditional distribution. Its spatial behaviour is controlled by the variogram, which encodes correlation length and direction, and its end product is reserves estimates expressed as probabilistic P10, P50, and P90 ranges rather than single deterministic figures.
Real-World WCSB Scenario: Sizing a McMurray SAGD Pad
An oil sands operator evaluating a steam-assisted gravity drainage development in the McMurray Formation near Fort McMurray, Alberta, had 38 delineation wells across a township but needed to size central processing capacity for a 30-year project. The geomodelling team built 250 stochastic facies and porosity realisations honouring the inclined heterolithic stratification and mud-plug barriers, then ran flow simulation on a ranked subset at a study cost near CAD 480,000.
The realisations returned a P90 bitumen-in-place of 142 million barrels and a P10 of 218 million, a spread driven almost entirely by uncertain mudstone-barrier continuity between wells. The operator sized the plant to the P50 case and staged a second train against the upside, avoiding both an undersized facility and a stranded-capital overbuild that a single deterministic estimate could have caused.