Fractured reservoirs have become more attractive for acquisition and development because of recent technological developments and tools for visualizing critical elements of fractured reservoir flow paths. The industry often considers these reservoirs as technically challenging projects that may produce expensive surprises during their development.
Ultimate recovery from these reservoirs is low, sometimes no more than 5-10% of the estimated original oil or gas in place.
The low recoveries arise from:
- Substantial permeability contrast between matrix and fractures.
- The possiblity of highly heterogeneous spatial distribution and connectivity of fractures throughout the reservoir.
Secondary or tertiary recovery processes, design of these processes, well orientation, completion effectiveness, and other engineering considerations are all sensitive to the geometry and properties of fractures that in these reservoirs are on the scale of 10s to 100s of meters.
This scale of fracturing is largely below seismic tool threshold resolution. On the other hand, fracture information derived from wellbore imagery or cores is often highly variable among wells and consequently provides only a local picture of fracture orientations and flow properties around the well.
This local nature and high variability makes it challenging to predict fracture intensity, connectivity, and permeability among wells with much confidence. All these factors are critical for optimizing field performance.
Advances in reservoir characterization approaches, software tools, flow simulators, and visualization technologies have improved modeling of subseismic fracturing among wells and allow this knowledge to be incorporated into reservoir calculations.
Known as discrete fracture network (DFN) modeling, this technology integrates geological, geophysical, and engineering data in a self-consistent and verifiable manner and combines it with rapidly evolving software tools for reservoir modeling and engineering calculations.
Also, because of visualization advances, these tools provide new avenues for field development through a clearer picture of how fracture intensity and connectivity vary throughout the field.
Circle Ridge field in Wyoming's Wind River basin, a mature oil field with historically low recovery, illustrates the integrated approach inherent in DFN modeling, the engineering insights that the application of the DFN visualization of the reservoir-scale fracturing can bring to reservoir plumbing, and the direct linkage of the visualizations and the numerical models underlying them to reservoir development.
Circle Ridge field
Operated by Marathon Oil Co., Circle Ridge field is on the Wind River Reservation and is owned by the Eastern Shoshone and Northern Arapaho tribes. Circle Ridge produces oil from the Paleozoic Phosphoria, Tensleep, and Amsden formations (Fig. 1). These formations in central Wyoming were part of the continental shelf along the eastern side of the Cordillera.
The Amsden is Late Mississippian to Early Pennsylvanian heterogeneous series of marine sandstones, shales, and limestones. Within the Wind River basin, the Amsden varies in thickness from 200 to 400 ft. A conformable contact separates the Tensleep from the Amsden. The lower unit of the Tensleep consists of dolomite and limestone. Shales and fine to medium-grained, cross-bedded sands make up the upper clastic unit.
At Circle Ridge, the Tensleep has a 300-ft gross thickness. An erosional unconformity separates the Pennsylvanian Tensleep from the Permian Phosphoria, formed as the Paleozoic seas withdrew during the Late Pennsylvanian and Early Permian.
The basal unit is primarily a cherty dolomite and thin, hard, brown limestone that was deposited in a shallow marine environment. The upper unit contains red, phosphatic shales and sandstones. Economic oil production most often comes from the Upper and Lower Ervay dolomite sequences in the Phosphoria.
The Laramide Orogeny produced structural deformation in Wyoming characterized by northeasterly crustal shortening and large fault-bounded, basement-involved uplifts that formed intervening basins such as the Wind River basin. Many of the large folds and thrust faults trend northwest, perpendicular to the inferred direction of maximum principal stress. The northwesterly compression also likely produced the natural fractures in the reservoirs.
Circle Ridge field is an asymmetric, doubly plunging fault breached fault-propagation fold. The Red Gully fault separates the reservoirs into an upper thrust sheet termed the overthrust block, and several lower imbricate blocks known collectively as the "subthrust" block. The subthrust block structure is complex because of multiple fault bifurcations that create a stacked series of subthrust reservoir blocks.
The field has a steeply dipping southwestern limb and a shallowly dipping northeasterly limb. Moreover, the overthrust block is broader to the southeast and narrows to the northwest. Faulting rather than folding towards the northwest end accommodates some space problems.
The field is shallow at the structural crest, with the Phosphoria reservoir being only a few hundred feet below the surface.
Generic models exist for fracture development in folded rock, but these may be overly simplistic due to the asymmetry, doubly plunging ends, and complex fault imbrications of the field. As a result, a model for Circle Ridge fracture orientations and intensity must be based on the specific structural development rather than on some generic archetype.
In terms of reservoir properties, intergranular matrix porosity dominates oil storage while fractures dominate permeability. Water from the large Tensleep-Phosphoria aquifer goes through natural fractures to the producing wells, leaving much oil in the matrix between fractures largely in place.
After nearly 50 years of oil production, Circle Ridge had only produced 3.5 million bbl from the Amsden, 16.3 million bbl from the Tensleep, and 7.9 million bbl from the Phosphoria. This is a small portion of the estimated original oil-in-place.
Recovering additional oil has been a challenge because of the lack of a means to predict fracture development and flow response of the fracture networks throughout the Tensleep and Phosphoria.
Building a 3D DFN visualization and computational model, based upon the structural history of the field, was a way to capture additional oil.
DFN modeling
DFN modeling originated in the 1980s to predict flow and transport radionuclides through fractured crystalline rocks in efforts to develop technology for designing high-level nuclear waste repositories. This characterization and modeling approach entered the petroleum industry in the early 1990s and over the past few years has emerged as an important evaluation tool.
It has proven useful in various types of fractured reservoirs as well as in reservoirs where matrix permeabilities may be significant, but fracture permeability dominates.
A DFN model explicitly incorporates all fractures that play a role in the variations in reservoir permeability at the scale of the flow process involved.
The model represents small fractures numerically as single polygons, while larger fractures, such as fault surfaces, form as a mesh of smaller planar polygons. Each polygon has properties associated with it, such as permeability, compressibility, roughness, or aperture.
One can convert these DFN models into finite-element meshes for solving and visualizing fluid flow and mass transport within the fracture system. The models, thus, are a quantitative reservoir engineering tool for assessing well locations and completions, as well as improved oil recovery processes.
The models can determine fracture geometry and properties in several ways. Early DFN models often interpolated fracture properties between well control through statistical algorithms.
In the 1990s, more geologically realistic ways of generating DFN models became tenable because of advances in computational power These ways include conditioning the intensity and orientation of the fractures to geophysical imagery, such as azimuthal variations in amplitude-vs.-offset data sets derived from processing 3D seismic.
Geologically conditioned DFN models are diverse and can relate fracturing to such parameters as shale content, matrix porosity, structural curvature, proximity to faults, lithology, bed thickness, and other factors that influence fracture development.
The DFN approach to characterizing fractured reservoirs offers the following two advantages that makes it particularly well suited for improving recovery:
1. The fractures created for the reservoir model are derived from geological, geophysical, and engineering data. Information relating to the natural fracture system can come from multiple sources, such as wellbore image logs, cores, production history, various types of well tests, outcrop analogs, multiwell tracer tests, structural history, flow logs, geophysical imagery, and qualitative considerations such as tectonic setting.
The models can use all these different data types to build the reservoir model. This produces a model that reflects the local fracturing pattern in the reservoir as best determined from available quantitative and qualitative data. It also produces a model that responds quantitatively and qualitatively like the field.
Another advantage is that integration of these different data types into a single model often highlights apparent inconsistencies that help identify assumptions that may not be valid or suggest new additional interpretations
2. DFN numerical models are easily processed to yield data that can be directly input into numerical reservoir simulators or other types of more specialized flow and transport simulation software.
Model development, refinement, and application benefit from and require a 3D visualization of the fracture model and calculation results. Because the DFN model consists of finite polygons in 3D space, rather than of a stacked series of cells or space-filling grid, various techniques and software visualize these models and the modeling process.
Visualization tools are an essential adjunct to DFN computational techniques because fracture network connectivity structures and major flow path development are not readily apparent from a list of numbers.
Reservoir characterization
Building and applying a DFN model for a fractured reservoir is similar to developing a model for matrix properties, although the specific types of data and analyses in each component differ, sometimes substantially.
The process passes through the following three stages (Fig. 2):
- The data analysis phase considers project goals, data availability, and cost of acquiring additional data.
- The DFN model construction phase includes model verification and validation.
- The application phase may follow several paths. These might focus on exploration or production issues, and involve analysis of a full-field DFN model or only a series of local well-scale DFN models.
At Circle Ridge, the goal was to improve recovery in the short-term by drilling or recompleting wells in more-favorable areas or orientations and in the long-term by initiating secondary or tertiary recovery projects. These goals focused the data analysis on developing a predictive model for fracture orientations and intensity that can be applied field-wide throughout the Tensleep and Phosphoria.
The analysis could not rely on geophysical imagery or statistical interpolation between well control for populating the model with fractures between well control because at Circle Ridge:
- No useful geophysical data exist.
- Geostatistical interpolation is uncertain because well coverage in blocks other than the overthrust block and one imbricate of the subthrust block are sparse or nonexistent.
The operator deemed as too costly the acquiring of a 3D seismic survey for attribute processing or obtaining enough individual well-based fracture data to increase geostatistical interpolation certainty. The geologically conditioned DFN model, therefore, involved the reconstruction of the strain history of the reservoirs as they were folded and faulted during the Laramide Orogeny.
The abundant well control on the major faults and key reservoir horizons made a 3D palinspastic reconstruction both technically possible and economically viable.
A palinspastic reconstruction consists of unfolding and unfaulting the rock to restore it to its original horizontal or flat condition. To obtain the best reconstruction, the process needs to exam many algorithms that approximate different folding and faulting styles and movement sequences.
The doubly plunging field anticlinal fold and nonsymmetrical imbrication implies that the reconstruction has no obvious symmetry that would allow a 2D reconstruction. For this reason, the analysis used a 3D reconstruction with Midland Valley Exploration Ltd.'s 3DMove software.
A 3D palinspastic reconstruction requires the building of surfaces, representing faults and boundaries between various units (Fig. 3a). Construction of vertical cross-sections through the structures facilitates the process. The analysis combined three new cross-sections from surface and well data with three existing cross-sections to provide control in critical areas for the reconstruction.
The field has seven major recognized faults. The reconstruction showed that the first major event was the flexural slip folding of the reservoir units. This was followed by several faulting events that started with some faults in the steeply dipping southwestern portion of the field, moved through the subthrust imbricates from the northwestern-most faults progressively southeastward, and then culminated with the youngest movement along the Red Gully fault.
These processes deformed the tetrahedral volume elements of the Tensleep and Phosphoria after each folding and faulting event, making it possible to calculate volumetric and directional strains (Fig. 3b).
In most reservoirs, the primary source for data on fracture orientations and intensity comes from fracture image logs or cores. Outcrops of younger formations at Circle Ridge provided additional insights into fracture pattern variations because they likely were deformed in a similar manner as the underlying Phosphoria and Tensleep.
These 11 outcrops provided data on fracturing in many different structural settings where the amount and orientation of extensional strain was likely to differ (Fig. 4).
Fractures develop in many overthrust settings during the first large strain-producing structural event. Subsequent large events may produce additional fractures, propagate old ones, or move the rock as blocks along the existing fractures.
The largest and first strain-producing event in Circle Ridge was the initial folding of the reservoir units before any appreciable faulting. After the folding occurred at Circle Ridge, faults began to form and the blocks then moved with much less internal strain along these faults.
Comparisons of the strain pattern from the folding with the fracture patterns in outcrop and with fracture patterns interpreted from image logs in three wells suggested that the maximum extensional strain vector produced by the folding is a reliable predictor of fracture orientations and relative intensity in the Tensleep and Phosphoria.
Analysis showed that the orientation of the maximum extensional strain was frequently perpendicular to the primary extensional fracture set and parallel to secondary fracture set (Fig. 5).
In addition, comparing the extensional strain magnitude at the three wells in the strain field with the relative fracture intensity showed a good qualitative correspondence between them. Fig. 3c shows the DFN model based on the extensional strain from the overthrust and adjacent subthrust Tensleep blocks.
Golder Associates Inc.'s Fracman software created this model.
Fracture development
The DFN model at this stage realistically models fracture geometry but has no fracture fluid flow properties such as fracture permeability, compressibility, and pore volume. It also has not been tested to determine whether it can predict reservoir flow behavior.
Calculating these fracture flow values and validating the model require a comparison of the model's connectivity structure with the pattern of connectivity deduced from well production behavior or direct numerical simulation of single well or multiwell tests.
For calibration and comparison, the work at Circle Ridge included a bromide tracer test and a nitrogen injection test.
To derive fluid flow properties for the fractures in the DFN model, the analysis included simulating the transient pressure behavior of an injection well by making a local, well-scale DFN model, converting it to a finite element mesh, and then adjusting the fracture flow properties until the calculations produced an adequate pressure-vs.-time match (Fig. 6).
The DFN model consisted of two sets, both orthogonal to bedding and each other. The dominant set was orthogonal to the direction of extensional strain from the folding event. Color-coding the discrete fractures according to the values produced a display of the pressure field or flow field as a function of time.
The DFN transient flow simulation primarily serves as a means to calibrate the fracture flow parameters. It also serves, however, as a validation of the fracture geometry in the sense that one only has to adjust the fluid flow and fracture storage properties to match transient well tests.
Visualization of the pressure field makes it easy to identify well connected or poorly connected portions of the reservoir and whether flow in the fracture system is channelized along specific corridors or is more diffuse.
The breakthrough patterns in the two tracer experiments generally coincided with the fracture orientations and highly connected pathways that formed as corridors of high strain. The high-strain corridors correspond to hinges that developed during the folding of the field.
Generally the two zone types found were:
- Those that are subhorizontal, having formed when the rock was upwarped and flattened along the thrust.
- Those that are subparallel to dip, which formed to accommodate the doubly plunging fold on either end of the field.
In the Tensleep overthrust-block nitrogen injection test, for example, nitrogen injected at the top of the structure moved along fracture zones associated with regions of high flexural strain. The test showed the most rapid breakthroughs and the largest surface pressure responses in wells that are either very close to the injection well or lie along the fracture corridors.
Wells further away or not on fracture corridors associated with flexure zones showed little surface pressure response or rapid breakthrough. Of note also was the magnitude of surface pressure response that in some cases was affected by the timing of gas breakthrough relative to the end of gas injection.
Production logs and high-resolution injection profiles also proved invaluable in verifying the DFN model. These logs provided the information for determining which fractures indicated by the image logs controlled flow in a particular well. Often this information leads to a different interpretation than would be reached without the flow logs.
Individual well tests and tracer experiments in both the overthrust and subthrust blocks, thus validated the generation of the DFN model. The next step generated a DFN model for each block.
Enhanced recovery
DFN models can be used for engineering decisions in several different ways. Some are local and short-term, such as what is the best direction to drill a well to maximize the number of fractures intersected, while other questions may involve full-field reservoir simulation.
Because fracture orientations are not random and intensity is variable, connectivity in the fracture network varies through the reservoir units. In some locations, the fracture intensity is low enough that isolated fracture networks develop, forming fracture compartments.
This use of DFN models makes it possible to locate regions in the reservoir that have higher fracture intensity, calculate the tributary drainage volume around a well, predict how many fractures a well of a particular orientation and location might intersect, and assess whether interference among wells is likely (Fig. 7).
Knowledge of the locations of untapped, oil filled, fracture compartments can generate economic infill locations. The increased understanding of the overall field fracture matrix geometry can also greatly aid placement decisions for both gas and water injectors.
Many other reservoir engineering decisions rely on results from numerical reservoir simulators. One can use DFN models for calculating fracture permeability, porosity, sigma factor, or other quantities that directly reflect the underlying fracture geology, and thus will produce more geologically realistic and useful simulation results. For example, the sigma factor is related to the typical dimensions of a matrix block within the cell.
The typical matrix-block dimensions from the fractures in a specific cell can be computed by the fracture spacing histograms in three directions, followed by the generation of the matrix blocks with a Monte Carlo sampling process (Fig. 8). From the realizations, one can determine the mean or median matrix block shape, volume, and typical spacing or fracture pore volume for that cell.
The fractures in the DFN model within a specific cell can also be used to calculate the directional fracture network permeability in that cell (Fig. 9). Each fracture within the cell is converted into a finite element mesh.
One assigns no flow or other types of boundary conditions to four contiguous faces of the cell, while assigning a constant pressure boundary to the remaining two opposite faces. This calculates, in a similar manner to a jacketed piece of core tested in the lab, the flux of oil or other fluids entering one end and exiting through the opposing face.
The only unknown parameter is the effective grid cell-scale fracture network permeability because fluid properties such as density and viscosity and the surface area of the opposing grid cell faces are known, the pressure gradient is specified, and the flux is calculated. In this manner, one can calculate the effective permeability for all grid cells in all three grid-cell directions directly from the fractures in the DFN model that occur in the cell.
This type of calculation has two important advantages:
- The fracture permeability is derived directly from geology and well test results, thus producing values that may more accurately reflect the actual reservoir flow.
- The results are calculated directly at the scale of the grid cell. As a result, one does not have to upscale the model, which is difficult for conventional reservoirs and problematic for fractured reservoirs.
Future
DFN technology has advanced considerably over the past few years. Commercial and proprietary software now exists to generate DFN models that reflect a wide variety of geological and mechanical processes.
The creation of fractured reservoir models using the geological and mechanical algorithms that have been developed have helped engineers better understand why some wells produce economically while others do not, and where and how to locate new wells to take advantage (or in some cases, avoid) the fractures.
The major technical challenges lie in expanding the capacity of these models to simulate multiphase flow and coupled matrix-fracture flow.
Current DFN flow solvers discretize the fracture networks, but treat the matrix as a storage element. This is because the irregular geometry of fractures can lead to highly irregular matrix blocks between fractures, not the simple sugar cube, slab, or matchstick idealizations common to most existing dual-permeability reservoir simulators.
The discretization of the matrix between the fractures remains a challenging task with much room for improvement.
Work is under way to simulate multiphase flow in fracture systems. Current finite element solvers based on the DFN model only simulate single-phase flow. When multiphase flow is required, the DFN model generates the input for the fracture components of the finite difference dual porosity, dual-permeability simulators.
Nonetheless, DFN models provide useful insights into the behavior of fractured reservoirs and can lead to improved recovery and reduced costs.
The authors
Paul La Pointe is manager of petroleum services for Golder Associates Inc., Redmond, Wash. He has more than 25 years' experience in the characterization of fractured reservoirs systems. La Pointe holds a PhD in engineering (rock mechanics) and an MS in structural geology from the University of Wisconsin.
Jan Hermanson is business area manager for Golder associates AB, Stockholm. He has more than 10 years' experience in designing and developing site characterization models for flow and transport in crystalline rock masses for the disposal of nuclear waste and for characterization and structural geological evolution of fractured sedimentary and crystalline reservoirs.
R. Parney is a geophysicist and hydrogeologist for Golder Associates Inc., Redmond, Wash., with more than 15 years' experience in geophysics and flow problems in fractured reservoirs. Parney has a PhD in hydrogeology from University of British Colombia and an MS and BS in geophysics from University of Calgary.
Mike Dunleavy is a reservoir engineer with Marathon Oil Co. in Cody, Wyo. He has worked 19 years as a petroleum engineer in various locations. Dunleavy holds a BS in petroleum engineering from the University of Wyoming and an MS in environmental engineering from Montana Tech. He is an SPE member.