Projection method
How the pathways to 2040 are produced: a physical driver model fitted to each facility's reported record, run forward under five operating scenarios, with a band built from the model's own hold-out error. Thresholds are tested against the pathways, never built into them.
- Horizon
- 2026 to 2040
- Scenarios
- 5
- Threshold tests
- 4
- Band draws
- 240
The driver model
Each facility's activity is projected from curves fitted to its own complete reported years, then history-matched so the first projected year reproduces the latest reported production rate exactly.
- Oil
- Arps hyperbolic decline from the facility's observed peak.
- Gas
- A saturating gas-oil ratio times oil, capped at the facility's demonstrated handling capacity.
- Water
- Logistic water cut.
- Fuel gas
- Linear in gas produced.
- Reinjection
- Held at the latest reported gas balance.
- Export
- The residual, capped at demonstrated export capacity.
- Flaring
- Not fitted: it is the policy lever. The base year is the mean of the last three complete reported years, the same value the threshold tests use.
Validation
The model is refitted on complete years to the end of 2023 and asked to predict the years after. The error is computed from the stored record each time this page is built.
Bias -5.6% · 6 facility-years predicted
Bias -14.5% · 6 facility-years predicted
The bias is negative: the model under-predicts the withheld years. That is the expected signature of decline curves on fields receiving infill drilling, which adds production the curve fitted to earlier years cannot anticipate. It is stated here rather than corrected away, because the correction would require knowing the drilling programme.
The band
Each projected year carries an interval from 240 draws over two sources of uncertainty.
- Driver error
- Log-normal on oil and gas, with a spread equal to the hold-out error and widened with the square root of the horizon, capped at 45%. A decline fit degrades the further out it runs.
- Flare combustion efficiency
- Triangular: 94–98–99.5% for an unrepaired flare, 99–99.5–99.9% once a repair programme reaches it. The sensitivity analysis identifies it as the dominant unmeasured term.
- Reproducibility
- The generator is seeded and consumed in a fixed order, so the band is reproduced exactly from the stored seed.
Scenarios and thresholds
The five scenarios vary the two things the sector controls — where associated gas goes and whether a facility programme is deployed — plus an offtake-failure downside. Four thresholds are then evaluated against each pathway in 2030.
An earlier design solved each scenario to its target. It collapsed: the abatement stack and natural decline met the methane targets without any reduction in flaring, two scenarios became identical, and the finding disappeared. Keeping thresholds external to the scenarios is what lets a pathway fail a test it was not designed to meet — and what gives a pass its meaning.
The four thresholds are screening tests, not statutory compliance determinations: the World Bank Zero Routine Flaring proxy, the Global Methane Pledge reduction, an EU methane-intensity screen and a carbon-intensity benchmark.