Reservoir Characterization Model

A reservoir characterization model is a three-dimensional digital representation of a specific volume of the subsurface that integrates all available geological, geophysical, and engineering data to describe the spatial distribution of reservoir properties. The model assigns values for porosity, permeability, net pay thickness, fluid saturations, and rock type to a grid of cells that together represent the reservoir volume. Geologists build the model from well log interpretation, core analysis, seismic attributes, and production history. The reservoir characterization model is the foundation for fluid flow simulation: a reservoir simulator runs on top of the characterization model, applying equations of fluid flow through porous media to forecast production rates, recovery factors, and the economics of various development strategies.

Key Takeaways

Reservoir characterization models have two main types. A static model (also called a geological model or geomodel) describes the reservoir's rock properties as they exist today, without simulating flow. A dynamic model (reservoir simulation model) takes the static model and adds fluid properties, pressure-volume-temperature data, relative permeability curves, and a production history to simulate flow. Static models are the starting point; dynamic models are built from them by adding the fluid and flow framework.
Geostatistical methods fill the space between wells with property values that honour the well data but reflect the geological variability of the reservoir. Sequential Gaussian simulation, kriging, and object-based modelling (for channels, lobes, and reefs) are common methods. Each realisation is one possible version of the reservoir that is equally consistent with the data. Running hundreds of realisations gives a probabilistic distribution of properties across the model volume, which translates directly into P10, P50, and P90 reserves estimates.
Upscaling is the process of coarsening a fine geological grid (which may have millions of cells at centimetre to metre scale) to a coarser simulation grid (which might have tens of thousands of cells) that a flow simulator can run in reasonable time. Permeability upscaling is the most critical step: the effective permeability of a coarse cell is not the simple average of the fine-scale permeabilities within it but a harmonic, geometric, or tensor average that reflects how flow paths connect through the volume. Errors in upscaling can produce simulation results that do not match production history even when the fine-scale model is geologically correct.
History matching is the process of adjusting model parameters until the simulator reproduces the observed production history (oil rate, gas rate, water cut, wellhead pressure) from existing wells. Parameters adjusted during history matching include permeability multipliers, fault transmissibility, relative permeability endpoints, and aquifer strength. A model that history matches well gives more confidence in its forecasts, but history matching is not unique: many different parameter combinations can reproduce the same history, and a model that matches the past may still forecast future performance poorly if it matches for the wrong geological reasons.
In the Western Canada Sedimentary Basin, reservoir characterization models are built routinely for major plays: Duvernay shale, Montney tight gas, Viking sandstone, and Devonian carbonate reef pools. For a Montney well pad with 10 horizontal wells, the model integrates microseismic data from hydraulic fractures, LWD logs from each lateral, core data from one or two wells, and 3D seismic attributes to map porosity, total organic carbon, and natural fracture intensity across the drainage volume of the pad.

What a Reservoir Characterization Model Is Built From

Think of a reservoir characterization model as a 3D jigsaw puzzle with millions of pieces. Each cell in the model grid is one piece, and the goal is to assign the right property values to every cell, not just the ones a well has drilled through. The challenge is that wells sample only a tiny fraction of the total reservoir volume. A vertical well with a 20-centimetre borehole in a 10-kilometre by 10-kilometre reservoir samples about one-millionth of the area. Everything between wells is inferred.

The data inputs to a characterization model come from four main sources. Well logs (gamma ray, resistivity, porosity, sonic) give rock type and property values at the wellbore at vertical resolution of 15 centimetres to 1 metre. Core samples give direct measurements of porosity, permeability, and rock type at centimetre scale but are available only where core was cut. Seismic data gives 3D spatial coverage of the entire reservoir volume but at vertical resolution of 10 to 30 metres. Production data (rates, pressures, water cuts) gives an integrated signal of average reservoir performance over the entire drainage volume of each well, but cannot be decomposed into individual rock properties without modelling.

The art of reservoir characterization is combining these data types, each with different scales and different uncertainties, into a model that is internally consistent and honours all the data simultaneously.

Fast Facts

Reservoir characterization modelling became a commercial discipline in the late 1970s and early 1980s as computer hardware advanced enough to store and manipulate three-dimensional grids. Early models used simple interpolation between wells (kriging). The introduction of geostatistical simulation methods in the 1990s, developed primarily at Stanford University's GSLIB project, allowed modellers to generate multiple realisations that preserved geological heterogeneity and provided probabilistic reserves distributions. Software platforms including Petrel (Schlumberger/SLB), IRAP RMS (Emerson/Roxar), and Jewel Suite became industry standards for building static models. Coupled with dynamic simulators (Eclipse, tNavigator, CMG), the static-dynamic workflow is now standard practice for any reservoir with more than a few wells.

Structural, Stratigraphic, and Petrophysical Frameworks

Building a reservoir characterization model proceeds in layers. The structural framework comes first: fault surfaces are interpreted from seismic, and the top and base reservoir horizons are mapped. These surfaces define the 3D grid geometry, controlling where the reservoir exists in space and how faults divide it into compartments.

The stratigraphic framework subdivides the reservoir grid into zones and layers that correspond to geological time units. A Viking sandstone model, for example, might divide the reservoir into five zones corresponding to lowstand, transgressive, and highstand systems tract deposits, with each zone further subdivided into layers that preserve the internal vertical variation of properties. Getting the zonation right matters because a simulated fluid front moves through zones differently depending on their permeability contrast.

Petrophysical population fills the grid with property values. Facies (rock types) are modelled first using the well log facies interpretation and seismic facies attributes as constraints. Then porosity and permeability are assigned within each facies type using statistics derived from core analysis and log petrophysics. In a clean fluvial sandstone, porosity might average 18 percent with a standard deviation of 3 percent. In an argillaceous sandstone, porosity might average 12 percent with the same standard deviation. Generating realistic spatial distributions of these properties across the grid, with spatial correlation lengths (variogram ranges) that match the known geological continuity of the facies, is the core work of geostatistical population.

Model Scale and Uncertainty

No reservoir model is right. Every model is a simplification that captures some features of the real subsurface and misses others. The value of a model lies in what it tells you about uncertainty: which reservoir properties matter most to the forecast, and what is the range of outcomes consistent with the available data.

Uncertainty analysis (also called sensitivity analysis or probabilistic modelling) varies key model parameters within their credible ranges and runs the dynamic simulator to generate a probability distribution of outcomes. In a Cardium sandstone pool, the most important uncertainties might be: the net pay thickness distribution (P10 = 4 m, P50 = 7 m, P90 = 12 m), the horizontal permeability (P10 = 5 mD, P50 = 20 mD, P90 = 80 mD), and the water saturation cutoff (P10 = 35%, P50 = 45%, P90 = 55%). Combining these uncertainties through Monte Carlo sampling gives the reserves distribution that drives investment decisions.

A reservoir characterization model is also called a reservoir model, geomodel, static model (for the pre-simulation geological version), or geological model. Related terms include reservoir simulation (the dynamic modelling of fluid flow through a reservoir model using numerical methods; the simulation model is built on top of the static characterization model and adds fluid properties and flow equations to forecast production), geostatistics (statistical methods for spatially distributed data; used to populate reservoir models with property values between wells using variograms, kriging, and stochastic simulation), history matching (the process of adjusting reservoir model parameters until the dynamic simulator reproduces the observed production history; a matched model is the basis for production forecasting), upscaling (the process of coarsening a fine-scale geological model to a coarser simulation grid while preserving the effective flow properties; a critical step between static modelling and dynamic simulation), and net pay (the thickness of reservoir rock that is productive under current economic and technical conditions; net pay is one of the key properties populated in a reservoir characterization model and is controlled by porosity and water saturation cutoffs).

How a Reservoir Model Error Added CAD 140 Million to a Judy Creek Development Cost

An operator held acreage in the Judy Creek area of central Alberta on a block that had produced oil from the Devonian Swan Hills carbonate reef since the 1960s. A new reservoir team was brought in to plan a waterflood expansion to improve recovery from the existing pool. The team built a reservoir characterization model using all available well data, seismic, and the 40-year production history.

The model predicted that the reef crest was well connected across the entire eastern flank, and that a line of injectors placed along the eastern rim could sweep oil efficiently toward the producing wells on the reef interior. The waterflood plan called for 8 new injectors and 12 new producers at a total capital cost of CAD 340 million. The model forecast a recovery factor increase from 28 percent to 41 percent, adding approximately 11 million barrels of incremental oil.

After 18 months of injection, the waterflood was performing significantly below expectations. Water breakthrough was occurring in some producers much earlier than the model predicted, while other producers were seeing no water at all. Pressure data showed that one area of the reef was not communicating with the injectors. The team re-examined the original seismic interpretation and found a set of tight, low-throw faults that had not been included in the structural framework because they were near the limit of seismic resolution. These faults divided the reef into four compartments that communicated only through restricted pathways.

Rebuilding the model with the fault compartments required drilling three additional injectors to access the poorly swept compartments, and shutting in two injectors that had been displacing oil into the wrong compartment. The remediation program cost CAD 140 million above the original budget. The post-mortem identified the root cause as insufficient seismic resolution at the time of the original modelling combined with overconfidence in the connectivity assumption. A subsequent reprocessing of the original 3D seismic with updated algorithms resolved the fault system clearly. The revised model matched the production history and guided the corrected injection strategy, which eventually achieved the planned recovery factor.