Reservoir Simulation: Definition, History Matching, and Development Planning

What Is Reservoir Simulation?

Reservoir simulation is the numerical modelling of fluid flow (oil, gas, and water) through a porous reservoir rock, used to predict reservoir behaviour under different production and injection scenarios and to optimise field development decisions. A reservoir simulator divides the reservoir into thousands to millions of grid cells, each assigned properties such as porosity, permeability, fluid saturations, and relative permeability — then solves the coupled partial differential equations governing multiphase fluid flow, pressure equilibration, and mass conservation across all cells simultaneously, advancing in time steps from initial conditions to forecast conditions. The simulator is calibrated (history matched) against the observed production history — rates, pressures, GOR, and water cut — to ensure the model honours real reservoir behaviour before using it to predict future performance under alternative development strategies. Modern reservoir simulation is central to every major field development decision: infill drilling locations, injection strategy, EOR design, plateau rate setting, and ultimate recovery factor estimation all derive from simulation studies.

Simulator Types and Workflow

The black-oil simulator is the industry workhorse — it models oil, gas, and water as three pseudocomponent phases with pressure-dependent PVT relationships (formation volume factor B_o, B_g, gas-oil ratio R_s, viscosities) but does not track individual hydrocarbon components. Black-oil simulators handle the vast majority of conventional oil field development studies: waterfloods, pattern floods, gas injection, aquifer influx, and natural depletion. Commercial examples include Eclipse (SLB/Schlumberger), CMG IMEX, tNavigator (RFD), and Intersect (SLB). The compositional simulator tracks individual hydrocarbon components (C1, C2, C3, C4, C5, C6+) through an equation-of-state (EOS) — required when phase behaviour is rate- and composition-dependent: gas condensate (retrograde condensate dropout near the wellbore changes fluid composition), CO₂ flooding (CO₂ miscibility with crude depends on composition and pressure), volatile oils (high GOR changes composition significantly during depletion), and LNG production optimisation. Compositional simulation is computationally 5–30× more expensive than black-oil simulation per cell.

The simulation workflow follows four stages: (1) geological model building — a geocellular model integrating seismic, well log, and core data defines the static reservoir properties (porosity, permeability, net pay) in a fine-scale grid (often 50–500 million cells); (2) upscaling — the fine geological model is upscaled to a coarser simulation grid (typically 100K–5M cells) using flow-based or arithmetic averaging techniques that preserve effective flow properties; (3) history matching — the model is calibrated to observed production and pressure data by adjusting uncertain parameters within geological and physical constraints; (4) forecasting — the history-matched model is run under alternative development scenarios (well locations, injection rates, EOR strategies) to identify the optimal development plan. The uncertainty in the forecast is quantified by running multiple history-matched models (P10/P50/P90 scenarios) and mapping the range of outcomes.

Reservoir Simulation Synonyms and Related Terminology

Reservoir simulation is also referred to as:

• Numerical reservoir simulation — emphasises the numerical (finite difference or finite element) solution method, distinguishing it from analytical reservoir models
• Dynamic reservoir model — the simulation model that changes with time (fluid saturations, pressures evolve during the run), contrasted with the static geological model which is time-independent
• Full field model (FFM) — a reservoir simulation model that represents the entire field rather than a single-well or sector model; used for development planning and reserves booking
• History matching — the calibration process that adjusts the simulation model to match observed production data; the most time-consuming step in the simulation workflow

Related terms: Material Balance, Relative Permeability, Decline Curve, Recovery Factor

Frequently Asked Questions About Reservoir Simulation

What are the main sources of uncertainty in a reservoir simulation forecast?

Reservoir simulation forecasts carry significant uncertainty from multiple sources that compound through the workflow. Structural uncertainty — the depth and geometry of the reservoir top, fault locations, and OWC/GWC depths — from the seismic interpretation propagates into grid geometry and initial hydrocarbon volumes. Petrophysical uncertainty — porosity and permeability estimation from well logs, and the upscaling from log to simulation scale — affects initial volumes and flow capacity. Rock-fluid uncertainty — relative permeability and capillary pressure curves from laboratory SCAL measurements that may not represent the full range of reservoir conditions — controls how efficiently fluids are swept and how quickly water breaks through. Fluid property uncertainty — PVT data from unrepresentative samples, or EOS tuning to limited data — affects phase behaviour predictions. Boundary condition uncertainty — aquifer size and connectivity, fault seal integrity, and natural fracture network transmissibility — controls pressure support and drainage compartmentalisation. Each source of uncertainty produces a range of history-match-consistent models that forecast different recoveries — the P10/P50/P90 range (10th, 50th, 90th percentile of forecast outcomes) across an ensemble of models quantifies the combined uncertainty. Fields with high structural complexity, limited well control, and short production history have wide P10/P90 bands; mature fields with decades of production, many wells, and abundant dynamic data converge to tighter forecast ranges.

How is thermal reservoir simulation different from conventional black-oil simulation?

Thermal reservoir simulation solves the energy balance equation in addition to the fluid flow equations — tracking temperature as well as pressure, saturation, and composition throughout the reservoir. This is required when injected fluids (steam, hot water) or in-situ combustion significantly change reservoir fluid viscosities and phase behaviour. Heavy oil viscosity is extremely sensitive to temperature: bitumen viscosity falls from millions of centipoise at reservoir temperature (10–20°C in Athabasca) to hundreds of centipoise at 150°C (steam chamber temperature in SAGD). Thermal simulators (CMG STARS is the industry standard; SLB INTERSECT also handles thermal) model steam chamber growth, heat conduction into adjacent rock, steam condensation and drainage, and the resulting oil mobility increase simultaneously. Geomechanical coupling — stress changes from thermal expansion during steam injection — is increasingly modelled in coupled thermo-geomechanical simulations for well integrity assessment in high-rate SAGD. Thermal simulation is computationally more expensive than black-oil simulation due to the additional energy balance equations and the strong non-linearity of steam properties near saturation.

How is machine learning being integrated with reservoir simulation?

Machine learning (ML) is being integrated with reservoir simulation primarily through proxy modelling (surrogate models) and data-driven history matching. Proxy models are ML algorithms trained on full simulation runs — once trained, the proxy predicts simulation outputs (total production, water breakthrough time, recovery factor) 1,000–10,000× faster than the full simulator. This speed enables uncertainty quantification (10,000 scenarios to map P10/P50/P90), optimisation (finding well placement and injection strategy that maximises NPV), and real-time decision support. Data-driven history matching using ensemble smoother with multiple data assimilation (ESMDA) simultaneously updates hundreds of parameters (permeability multipliers, fault transmissibility) across an ensemble of model realisations. Companies including Equinor, bp, Shell, and TotalEnergies have published results showing ML-assisted history matching reduces the time from months to weeks while producing better uncertainty quantification than manual matching.

Why Reservoir Simulation Matters in Oil and Gas

Reservoir simulation is the primary tool through which billions of dollars of field development capital are allocated — it determines where infill wells are drilled, how much injection water or gas is needed, whether EOR is economic, when facilities should be sized for plateau production, and ultimately how much oil and gas a field will recover. Every major oil field development plan submitted for regulatory approval and investment sanction is supported by a reservoir simulation study. The confidence level in the simulation forecast controls the investment risk accepted by operating companies, joint venture partners, and governments that collect royalties on the produced hydrocarbons. As fields become more complex (deepwater, ultra-HPHT, fractured carbonates, heavy oil), reservoir simulation has evolved from a specialised analysis tool to the central hub of integrated reservoir management — continuously updated with new production, 4D seismic, and well log data to guide operational decisions in real time.