Uncertainty

Key Takeaways

• The distinction between reducible and irreducible uncertainty is the foundation of sensible uncertainty management — reducible uncertainty is uncertainty that additional data (seismic acquisition, appraisal wells, laboratory testing, production history) can meaningfully narrow; irreducible uncertainty is inherent to the system and cannot be reduced regardless of additional data (future oil prices, long-term reservoir performance beyond the current analog, the global demand trajectory for hydrocarbons); the economically rational response to reducible uncertainty is to invest in data acquisition up to the point where the cost of additional data equals the value of the uncertainty reduction it provides (the value of information, or VOI, calculation); the rational response to irreducible uncertainty is to design projects and portfolios that are robust to the expected range of outcomes — investments that work at $50 oil and are enhanced at $80 oil, rather than investments that only work at prices at the high end of the range.
• Monte Carlo simulation is the standard computational method for propagating uncertainty through petroleum engineering calculations — rather than using single point estimates for each uncertain parameter (porosity = 18%, net pay = 45m, recovery factor = 35%), the Monte Carlo method samples randomly from the probability distributions assigned to each uncertain parameter in each simulation run, calculates the outcome (reserve volume, NPV, production profile) for that combination, and repeats hundreds of thousands of times to build a statistical distribution of possible outcomes; the resulting distribution shows not just the most likely outcome (P50) but the full range from pessimistic to optimistic, allowing the decision-maker to assess the probability of achieving project threshold returns, the expected value of the investment, and the downside exposure; the correlation structure between uncertain parameters (if porosity is high, is net pay also likely to be high, or are they independent?) must be correctly specified in the Monte Carlo model for the output distribution to accurately represent the geological reality.
• Subsurface uncertainty quantification using multiple geological scenarios (discrete scenario modeling) provides an alternative to purely statistical Monte Carlo approaches when the uncertainty is structural rather than continuous — for example, if there is genuine geological uncertainty about whether a formation is a channel sand or a sheet sand (two distinct depositional concepts that would have very different continuity, connectivity, and recovery factor), representing this as a continuous distribution of permeability values underestimates the bimodal nature of the uncertainty; the discrete scenario approach explicitly builds two or more geological models (one for each plausible scenario), assigns probabilities to each scenario (based on geological analogs, seismic evidence, and expert judgment), simulates production forecasts from each model, and combines the scenario results weighted by probability to produce the expected value and range; this approach is particularly effective for frontier exploration appraisal where the geological concept itself is the primary uncertainty rather than the specific values of rock properties within a known concept.
• Uncertainty in production forecasting accumulates through the chain of technical parameters — porosity uncertainty, net pay uncertainty, permeability uncertainty, recovery factor uncertainty, and facilities capacity uncertainty each contribute to the total production forecast range, and when all uncertainties are combined, the P10/P90 ratio of the production forecast may be 3:1 or more even for relatively mature field developments where subsurface data is extensive; managing this cumulative uncertainty requires understanding which parameter contributes most to the total variance (the tornado chart or sensitivity analysis that identifies the swing parameter — the one whose uncertainty range has the most impact on the output), so that additional data acquisition or engineering work can be targeted at the most important source of uncertainty rather than applied uniformly across all parameters; in most reservoir engineering applications, permeability (and particularly the spatial distribution of permeability heterogeneity) is the dominant source of production forecast uncertainty, which explains why injection well data, production logging, and pressure transient analysis are consistently the highest-value data types for reducing forecast uncertainty in mature field development decisions.
• Decision-making under uncertainty requires an explicit framework that prevents the cognitive biases that lead to systematically poor decisions — the most common cognitive biases in petroleum engineering include overconfidence (probability distributions that are too narrow, representing the analyst's confidence in their model rather than the true geological variability), anchoring (adjusting estimates from an initial reference case rather than building distributions from first principles), and base rate neglect (ignoring the historical success rate of similar plays when assessing prospect probability of geological success); SPEE (Society of Petroleum Evaluation Engineers), SPE-PRMS (Petroleum Resources Management System), and various regulatory bodies have established frameworks for unbiased reserve and resource estimation that include independent technical review, probabilistic reporting, and comparison of estimates against outcomes as a calibration check; companies with mature uncertainty management cultures track the consistency between their P90 estimates and the eventual outcomes of the decisions made based on those estimates, using the historical performance to calibrate future estimates and identify systematic biases in their technical teams.