Integrated Reservoir Management: Decisions, Data, and Discipline—Episode Two

Episode 2: Why Some Timely Decisions Do Not Produce Results

The first episode in this series argued that delay is never free. Every week between a correct reservoir diagnosis and a committed field action carries a reservoir cost that is real but rarely quantified. But acting at the right time solves only half the problem.

Two further failure modes sit between a timely decision and a field outcome. The first is whether the information driving the decision deserves the confidence placed in it. The second is whether the decision, once made, ever reaches the field. In practice, both failures often occur in the same asset, on the same decision, in sequence.

Consider a case that many reservoir engineers will recognize. A plateau extension is sanctioned. The model is matched, the forecast looks reasonable, and the team is aligned.

Then, 14 months later, the field is underperforming.

Water arrives earlier than expected. Pressure support is less uniform than the model predicted. The team does not react immediately because the forecast range still provides cover. The field is still, just, within the stated uncertainty bounds.

The model was not obviously wrong. It was wrong in a way the uncertainty representation was never designed to catch. The fracture contribution driving early water encroachment was not a parameter sensitivity. It was a geological concept uncertainty that never appeared in the P10 to P90 range because it was never tested as a distinct scenario. The history match had been achieved without it. The forecast had not been stress-tested against it. And the decision had been made as though the range presented was the range that mattered.

The reason this happens is structural, not accidental. History matching is an ill-posed problem. As Subbey, Christie, and Sambridge (2004) demonstrated, multiple materially different combinations of geological parameters and flow properties can reproduce the same production history.

The reservoir is underdetermined relative to the parameters used to describe it. A single history-matched model therefore presents one member of a large family of equally plausible models as though it were confirmed truth.

Sensitivity runs around that model sample the neighborhood of one solution, not the full space consistent with the data. The P10 to P90 range most teams present has not been derived from an ensemble of models representing distinct geological concepts. It has been derived from parameter variations within a single concept. The uncertainty that actually matters, geological concept uncertainty, is structurally absent from the stated range.

This problem is more severe in carbonate reservoirs than in most sandstones. In a dual-porosity carbonate system, the matrix-fracture transfer function is poorly constrained by production data alone.

Masalmeh et al. (2012), working on a giant Middle Eastern carbonate reservoir, demonstrated that high-resolution static and dynamic model integration was essential to avoid history matches that were numerically consistent but geologically implausible, and that plausibility constraints materially narrowed the uncertainty on EOR performance prediction.

Published experience from onshore Abu Dhabi fields documents the field consequence of this pattern: sector models matched to field-level production data have repeatedly reproduced cumulative behavior well while missing preferential flow conduits, high-permeability streaks identified in core but not modeled explicitly, that drove earlier-than-expected water encroachment once the plateau began. The history match did not need them. The forecast did.

This compounds a broader industry condition.

In SPE 195914, Bratvold and Abellan, drawing on the Norwegian Petroleum Directorate database of 56 field development projects sanctioned between 1995 and 2015, found that production forecasts at final investment decision were both systematically optimistic and systematically overconfident. Actual outcomes fell outside the stated P10 to P90 range far more frequently than an honest 80% confidence interval predicts. The underlying causes, anchoring on a single geological concept, running sensitivities around a single matched model, and institutional pressure to present a range that supports the sanction case, are structural features of how development assets are evaluated across the industry.

Three disciplines address this.

First, challenge the uncertainty range itself. Was it derived from models representing distinct geological concepts, or from parameter variations within a single matched model? If the latter, the range understates the true forecast uncertainty and the decision resting on it is less well-supported than it appears.

Second, track forecast accuracy over time. A lookback process, comparing stated uncertainty ranges with actual outcomes at 12 months, is among the most effective changes a team can make to forecast quality. It is not technically demanding. It is organizationally uncomfortable, which is precisely why it is not done regularly.

Third, match the confidence requirement to the decision at hand. A bounded operating decision, a surveillance adjustment, and a major capital commitment do not all demand the same level of proof. Treating them as if they do leads either to under-scrutiny of large decisions or paralysis on small ones.

Models should be used to find out how fragile the base case is, not primarily to produce it. The most dangerous forecast is not the uncertain one. It is the one that looks certain enough to stop being questioned.

Getting the confidence right is necessary. It is not sufficient.

Even when a decision is timely and well-supported, it can fail at the next step. In one asset where the author was involved, a heterogeneous carbonate reservoir was brought on stream under an Early Production Scheme (EPS) with a deliberate and planned development strategy: produce for approximately 2 years to generate early cashflow and gather reservoir data, then initiate a water-injection line-drive scheme to provide pressure maintenance ahead of full field development. The subsurface case for that sequencing was sound, and the injection timing was not an afterthought. It was built into the plan.

Injection did not start in year 2. It started in year 5. The 3-year slip accumulated through a combination of facilities construction delays, budget-cycle constraints, and the organizational complexity of coordinating a large injection infrastructure program across multiple functions.

Each individual delay had a reason. None of those reasons included a visible, running estimate of what the reservoir was paying while the organization worked through them. By the time injection started, reservoir pressure in the EPS area had fallen to approximately 75% of initial pressure. The displacement efficiency consequence was estimated internally at approximately 3% OOIP from the EPS area deferred beyond the lease duration. Not lost in the ground but moved outside the contractual window within which the field could recover it.

What makes this case instructive is that the original decision was correct. The EPS-then-injection sequence was a legitimate strategy and the cashflow case for early production was real. What the process did not consistently surface was the reservoir cost of the implementation slip, in the same economic units as the investment under discussion.

Thakur, in SPE 20748, identified this failure 3 decades ago: the most common reason reservoir management programs fail is not their technical quality but the gap between developing a plan and implementing it. Capital decision processes are rigorous. The reservoir cost of time is less consistently part of the conversation. When it is, the trade-off between cashflow timing and recovery efficiency can be made explicitly. When it is not, it is made by default.

Closing this gap requires three things to be true simultaneously.

The reservoir cost of delay needs to be expressed in the same economic units as the project cost, not buried in a technical appendix but stated as a headline in the decision submission: every month of delay in this program costs an estimated $X million in discounted recovery.

The implementation timeline needs to be tracked against what the reservoir is paying, not just against the process milestone, because accountability for moving a submission forward typically exists while accountability for the reservoir cost of the time it takes rarely does. And time-sensitive recommendations need an explicit expiry: if a pressure-maintenance case is built on specific reservoir conditions, state that it needs formal revisitation if sanction has not occurred within a defined window.

The most damaging failure mode is a recommendation sitting in the system, technically agreed but economically deteriorating, while the organization believes the decision has effectively already been made. The practical lesson across both failure modes is the same. Good technical work is necessary. Calibrated confidence is necessary. Neither is sufficient without a decision process that keeps the reservoir cost of time visible to everyone making the implementation decision.

The final episode of this series will bring all three disciplines together and propose how to build a system that gets better at decisions over time rather than repeating the same ones.

For Further Reading

SPE 84148 Applied Reservoir Management Principles With Case Histories by J.N. Ezekwe.

SPE 161485 Developing High-Resolution Static and Dynamic Models for Waterflood History Matching and EOR Evaluation of a Middle Eastern Carbonate Reservoir by S.K. Masalmeh, L. Wei, H. Hillgartner, R. Al-Mjeni, and C. Blom.

SPE 20748 Implementation of a Reservoir Management Program by G.C. Thakur.

Prediction Under Uncertainty in Reservoir Modeling by S. Subbey, M. Christie, and M. Sambridge. Journal of Petroleum Science and Engineering (2004).

SPE 195914 Production Forecasting: Optimistic and Overconfident, Over and Over Again by R.B. Bratvold, E. Mohus, D. Petutschnig, and E. Bickel.

The Importance of Special Core Analysis in Modeling Remaining Oil Saturation in Carbonate Fields by S.K. Masalmeh and X.D. Jing. International Symposium of the Society of Core Analysts.