Three-dimensional geologic characterizations of reservoirs are used in waterflood simulation studies to predict oil and gas field recovery efficiencies, quantities of water, oil and gas production over time, and for well optimization and economic studies. Quantitative global characterizations of the static reservoir, for example, reservoir connectivity or permeability heterogeneity, can be used to help explain waterflood performance, but with a wide range in result quality.

In the current study, aspects of dynamic simulation of waterfloods were replaced by static approximations using a quantitative localized approach, in which information about distances and properties within the reservoir is used to help predict waterflood performance. A “ladder” of waterflood flow simulators was considered, ranging from simulators using only static or rock property information, to simulators using static and dynamic information, to state-of-the-art dynamic simulators. Predictions based on each simulator type were compared with predictions using a state-of-the-art simulator. The simplest waterflood simulator uses a static model with only permeability and porosity information. These approximations to fluid flow can be improved by including dynamic considerations, such as an approximation of water saturation behind the waterfront (obtained by solving the Buckley-Leverett transport equation) or an approximation to the velocity of the advancing waterfront (obtained by solving the Darcy equation for a single pressure step).

In the cases considered here, comparing swept volume or recovery factor using our simplest static flow simulators with results using a state-of-the-art simulator yielded R² values in the range of 0.5 to 0.65. By better characterizing geometric aspects of the reservoir, R² values of 0.65 to 0.75 were attained. By adding a single pressure step, R² values of 0.8 to 0.85 were calculated. Ranking, or perhaps binning, can be achieved using simulations based on earth models with limited dynamic information. Ultralarge models containing a billion cells can be simulated with little difficulty using this approach.