A mature oilfield is the most data-rich asset an operator will ever own, and often the most misunderstood. Decades of production, pressure and well data sit in the archive, and the simulation model has been history-matched so many times that everyone trusts it. That trust is the problem. Late in life, the barrels that remain are precisely the ones the model was never built to see.
A good match is not a good forecast
Early in my career I worked on automatic history matching: letting optimisation algorithms adjust a model until it reproduced four decades of field history. The lesson I took from it was not about the algorithm. It was that many quite different models reproduce the same history equally well, and then forecast very different futures. The match constrains the past; it barely constrains the next well.
If your infill decision rests on one history-matched model, you have not reduced the uncertainty. You have hidden it.
Where the remaining oil hides
- Compartments the seismic could not resolve. Small faults and baffles that leave pockets of oil behind producers, invisible until a well is drilled into them.
- Layers the waterflood bypassed. High-permeability streaks take the water; the tighter layers above and below keep their oil.
- Sands assumed to pinch out. Reservoir bodies interpreted as ending where the data ended, rather than where the geology did.
Each of these leaves a signature in the production data long before it appears in the model: anomalous pressure support, a well that waters out too early or too late, a producer whose decline does not fit its neighbours. Reading those signatures across hundreds of wells is pattern work, and it is exactly where open-source tools and automated screening have changed what one engineer can do in a week.
From many models to one plan
On a large mature field with seven owners, the technical work was only half the job. Every partner had its own model, its own view of the remaining potential and its own appetite for risk. What unlocked the infill campaign was a single plan tested across the full range of uncertainty: an ensemble of models rather than a favourite one, with each candidate well judged on how it performed across all of them.
A well that is excellent in the optimistic case and a dry hole in the pessimistic one is a gamble. A well that is good in nearly every case is a decision. Ranking by robustness rather than by the best case gave the partners something they could all sign, and the campaign added more than ten million barrels to projected recovery.
What I would tell an operator now
- Treat the archive as the asset. Clean, catalogue and re-read the well and pressure data before touching the model.
- Run the ensemble, not the base case. Compute is cheap; a wrong well is not.
- Rank wells by robustness. The best well in the base case is rarely the best decision.
- Benchmark the forecast. Compare the predicted uplift with what similar campaigns actually delivered. See benchmark before you commit.