Oil & Gas Reserves Estimation: P10 / P50 / P90
The platform computes recoverable reserves through four independent methods and aggregates them into a single probabilistic range. Every number has a traceable source — from the state reserves register to a neural network trained on 3,445 fields.
Four independent approaches to reserves estimation
Each method delivers its own estimate. The platform combines them into an ensemble and shows the uncertainty range.
State reserves register
GKZ-approved C1+C2 reserves. Top-priority source.
Nearby analogs
Comparison with 3–5 fields within a 50–100 km radius matched on lithology and conditions.
Neural network
ML model trained on 3,445 fields. Forecast from 25 geological parameters (gradient-boost).
Volumetric method
The classic: S × h × m × β × η. Fallback when no other sources are available.
When probabilistic reserves are required
P10/P50/P90 is not just "three numbers" — it is the language you need for an investor, a bank or a regulator.
Geologists
Pre-screening ahead of a GKZ defense. Fast "what-if" when parameters change.
Finance teams
NPV range across reserve scenarios. A reserves base for project financing.
M&A teams
Independent reserves opinion for deal due diligence.
Strategy
Portfolio ranking by reserves and the efficiency of bringing them on stream.
How the volumetric method integrates with the ML model
The volumetric formula returns a "deterministic" number. The ML model adjusts it based on the similarity of the geological setting to the training set.
- Oil-bearing areaFrom reservoir outlines via seismic data or analogs.
- Oil-saturated thicknessWell logs or correlations with neighboring fields.
- Porosity and oil saturationCore, well logs or modeled values for the reservoir-rock type.
- Recovery factor (RF)From analogs and the ML model — accounting for the development plan and fluid properties.
FAQ on reserves estimation
These are probabilistic estimates of recoverable reserves. P90 is the conservative case: there is a 90% probability that actual reserves will be at least this value. P50 is the most likely estimate (the median). P10 is the upside case: there is a 10% probability reserves could be higher.
This format is the SPE-PRMS industry standard for all investment decisions.
At minimum — site coordinates and reservoir-rock type (clastic or carbonate). That is enough for the platform to pick analogs, locate state reserves register entries and run the neural network. The more of your own data you upload (core, well logs, well tests) — the tighter the P10–P90 range becomes.
The platform produces an estimate methodologically close to GKZ approaches — but the final defense remains the responsibility of your geologist. AVP AI accelerates preparation: pre-screening, comparison with analogs and a full audit trail for every parameter.
The training set consists of fields across Russia and the CIS with known actual reserves on the state register and measured geological parameters. The architecture is gradient boosting over 25 features. Model quality is evaluated on a hold-out set using MAE.
Estimate the reserves of your reservoir right now
Provide the coordinates and the reservoir-rock type. Get a probabilistic reserves estimate with the source of every figure laid out. No lengthy sign-offs — a single run.