Rock Fractures and Fluid Flow: Contemporary Understanding and Applications (1996)

This report describes the tools that are currently available for characterizing fracture systems and for understanding, predicting, and controlling fluid flow and chemical transport in fractures. In many applications, fractures have been problematic. Their locations are often a mystery, but their effects on flow can be dramatic. Practitioners frequently ignore fractures or do not account for them adequately because the tools for dealing with them are too expensive, not widely available, or poorly understood. However, in the past few decades, researchers have made many advances in analyzing and characterizing fracture systems, in understanding fluid flow in fractured media, and in accounting for the importance of fracture systems in engineering design. New conceptual, physical, and numerical tools are now available to solve many problems associated with fracture systems.

In this report the committee describes these tools and the scientific theories behind them, gives the essential components of application, assesses their utility and limitations, and suggests topics for further work. This report is for students, educators, researchers, practitioners, and public officials.

Chapter 1 reviews fracture problems encountered in a variety of applications. Subsequent chapters address three key issues that are commonly encountered in these applications:

• How can fractures that are significant hydraulic conductors be identified, located, and characterized? (Chapters 2, 3, 4, and 5)

• How do fluid flow and chemical transport occur in fracture systems? (Chapters 3, 5, and 6)

• How can changes to fracture systems be predicted and controlled? (Chapter 7)

Fractures that are significant for fluid flow are open and connected to other fractures over distances comparable to the spatial and temporal scales of the problem of interest. In some cases there may be myriad fractures that are not significant and only a few that dominate flow behavior. In other cases only fractures having certain orientations (e.g., normal to the least principal stress) may be important. Fractures that conduct a significant amount of fluid may be a particular component of a fracture network, but it may be necessary to understand the entire fracture system to understand why that component is important (e.g., extension features that connect segments of a fault).

Fluid flow and transport take place in the void space created by these hydraulically conductive fractures. The geometry of the void space and the relationship between fracture permeability and matrix permeability determine how flow takes place. If the fractures are poorly interconnected and the matrix rock is relatively impermeable, an analysis of flow may need to account for the network of discrete fractures and could well ignore the matrix. On the other hand, if the matrix is permeable and the fractures are regular and highly interconnected, an analysis of flow could treat the fracture network as an equivalent continuum with fluid storage provided by the matrix. A major factor in analysis of flow and transport is the appropriate representation of the features that control flow.

Understanding and controlling changes to the fracture system are tantamount to understanding and controlling changes in fracture void space. Any combination of changes in stress, fluid pressure, temperature, fluid composition, or chemical conditions that commonly occur in engineering activities can change the geometry of the void space. For example, a decrease in fluid pressure increases the effective stress, which decreases the void space and the permeability of the fratures. Similarly, precipitation of minerals in a fracture (e.g., vein formation) decreases void space and, therefore, permeability.

What are the conceptual, physical, and numerical tools that help address these three key questions? Tools from geology, geomechanics, geophysics, geochemistry, and hydrogeology were primarily developed to solve fracture problems related to petroleum reservoirs, mines, and nuclear waste isolation. They have further application in a variety of reservoir, environmental, and construction endeavors. The tools are in various stages of development and use. This report collects and evaluates the scattered information about them.

In bringing these tools together to study fractures, the power of an interdisciplinary approach is evident. The fundamental reason that fracture problems are difficult to solve is that fracture systems tend to be extremely complex and heterogeneous. In any one discipline there are excellent, well-developed tools for handling homogeneous systems and some tools for handling anisotropic or simple composite systems, but each of these tools has severe limitations when applied to complex heterogeneous systems. The most effective way to solve problems in heterogeneous media is to apply a variety of tools so that the particular

strength of each method compensates for the weaknesses of the other methods. This approach may have applications in other endeavors involving analysis of behavior in the earth.

This chapter summarizes the key tools and shows how they address the three key questions given previously. A number of recommendations that follow from the discussion are given here for completeness. These are italicized. These recommendations are summarized in the Executive Summary of this report.

The tools to address this first key question come primarily from:

• Geology and fracture mechanics (Chapter 2)

• Laboratory and field studies of fracture properties (Chapters 2 and 3)

• Geophysics (Chapters 3 and 4)

• Hydraulic and tracer testing (Chapter 5)

Tools from geology and geomechanics can help address the following questions: What kinds of fractures tend to form in a given environment? What patterns do they commonly assume? Which fractures are more likely to conduct fluid? Understanding of fracture properties and tools from geophysics can help answer the question: How can hydraulically conductive fractures be detected at depths where they are not visible? Hydraulic and tracer testing can provide proof that a feature is hydrologically important by measuring flow and transport properties directly.

Each of these methods has limitations. The genesis of many fracture systems and the resulting fracture geometries are extremely complex and difficult to understand, especially at depths where fractures are not directly accessible. Geophysical methods can detect the mechanical or electrical anomalies produced by fractures, but these geophysical properties do not have a one-to-one correlation with hydrological properties. Well tests measure hydrological properties, but many different distributions of heterogeneity can give the same hydrological response. Moreover, well tests may not be practical, especially on very large scales, when producing wells must be shut down, or in contaminant environments. Used together, these tools can provide an understanding of the likely locations and the hydrological properties of major flow paths.

Quantitative calculations of fracturing based on principles of fracture mechanics (i.e., the study of the mechanics of fracture formation) explain many

fracture patterns observed in the field. Central to the use of fracture mechanics is an understanding of the role of heterogeneities in both stress and material properties. Flaws and other heterogeneities in the rock mass concentrate stress, so fractures begin to propagate. The new fractures perturb the local state of stress and enhance material property heterogeneity, which, in turn, affects the propagation of other fractures. This process creates fracture systems on every scale of observation. The fracture systems that result sometimes have recurrent, recognizable patterns or ''themes." These recognizable patterns can be used to infer the nature of subsurface fracture systems and their control on fluid flow.

Examples of fracture patterns that result in hydraulically significant flow paths include:

• Highly connected, regionally extensive networks of open fractures, such as cooling fractures that form polygonal networks.

• Clusters of open joints or fracture zones.

• Open fractures that occur as parts of faults, such as extension fractures that occur between steps in shear zones.

In the absence of subsequent fracture closure or filling, equivalent porous medium behavior (see Chapter 6) might be more likely in a system of regular cooling joints than in a concentrated system of joint clusters, fracture zones, or tension fractures. Such inferences can form the framework for characterization and modeling of flow behavior. However, many (if not most) fracture system patterns cannot yet be explained or categorized. Furthermore, the nature of fluid flow has not yet been confirmed in many of the patterns that have been described. Knowledge of the processes responsible for producing distinctive fracture patterns that are likely in various lithological and tectonic environments can help us understand systems of fracture flow. These relationships can guide hydrological investigations and should be a focus for further research.

The relationships between fractures at different scales are not yet understood. Although the mechanisms of fracture formation are similar at different scales, lithological and structural heterogeneities may be dissimilar over large ranges of scale. Scale effects in fracture patterns warrant more thorough investigation. This should entail an analysis of scaling relationships as a function of stress history, structure, and lithology at a number of different sites.

The formation of shear zones is usually studied in the plane of the maximum and minimum principal stresses. However, important hydrological features associated with tension fractures and openings tend to be parallel to an intermediate principal stress. Little is known about the extent and interconnection of fractures in the direction of intermediate principal stress. This is also true for joint clusters. Joints are usually studied in the plane of the maximum and minimum principal stresses, but critical interconnections between these joints can take place in the third dimension or in areas where the orientations of the principal stresses rotate.

Fracture genesis and geometry should be considered a fully three-dimensional problem in order to infer the flow properties in three dimensions.

Geology and fracture mechanics can be applied to help understand fracture patterns and to predict and assess their importance for subsurface flow, but it may be difficult to use these tools to identify the locations of fractures at depth. By understanding the properties of fractures (Chapter 3), it is possible to design tools that detect these fractures remotely. Fractures have relatively more porosity than adjacent rock. Consequently, at least three distinct properties are useful for detection:

• Unfilled fractures are less stiff than surrounding rock, so they represent an elastic anomaly, especially in shear.

• Fluids in fractures can be more or less electrically conductive than the rock matrix, so they represent resistive and electromagnetic anomalies.

• Unfilled fractures are usually more permeable than the surrounding rock matrix, so they concentrate flow.

Rock type and stress history control the geometry of a fracture system, which in turn controls the properties of the fractures. Although stress history may be difficult to determine quantitatively, even a qualitative understanding can be important in understanding the fracture void geometry that is likely to be present. Fractures that have formed because of contractive cooling have distinctive surface patterns and consequently should have distinctive void geometry patterns that are different from fractures that have undergone shear, for example. Small-scale laboratory and large-scale in situ research should be undertaken to determine how the character of the fracture void geometry depends on lithology and stress history.

A large number of tools have been developed to quantify fracture surface and pore space geometry. These include profilometer and optical scanning instruments and a variety of casting techniques. It is now possible to quantify the spatial relationships of asperity height and aperture distributions and to relate these geometric properties to a variety of fracture flow, mechanical, and geophysical properties.

A number of recent studies have quantified the geometry of laboratory fracture samples as well as their hydrological and geophysical properties. These studies greatly benefit the design of methods for detecting conductive fractures and the interpretation of field data. It is not always easy to extrapolate such measurements to field scales, but laboratory studies can provide an understanding of trends in behavior and a framework for interpretating of field data.

Some theoretical analyses suggest a correlation between the elastic properties of fractures and permeability. Fractures that are less stiff may also tend to be more

permeable. The elastic properties of a fracture are a function of the distribution of contact areas in the fracture. More open and permeable fractures have smaller contact areas and are less stiff. Because fractures tend to be less stiff than surrounding rock, they concentrate strain. Fractures thus create a "displacement discontinuity." The displacement discontinuity concept has been used to analyze the propagation of seismic waves across fractures. This has led to a model that provides quantitative relationships between fracture spacing and stiffness and between velocity and attenuation of seismic waves. This model has opened significant new possibilities for the design and interpretation of seismic surveys for detecting fractures. There is also a correlation between electrical and hydraulic conductance, although electrical conductance is not as sensitive to fracture opening as permeability is. Electrical conductance is proportional to the cross-sectional area of the fracture opening (i.e., the local aperture), whereas fluid conductance is locally proportional to the aperture cubed. Thus, fluid flow is much more sensitive to aperture variation than electrical conductivity. Consequently, electrical conductivity measurements cannot be used alone to quantify hydraulic conductivity, either for single fractures or the bulk medium.

An important area of research impacting the interpretation of geophysical surveys is the relationship between seismic and electrical properties of fractures and hydraulic conductance (volumetric flow per unit gradient). More theoretical work should be done to reduce the description of each phenomenon to a minimum number of free parameters, in order to identify quantitative relationships among the various properties. Additionally, the properties of fractures in a network should be related to the bulk properties of the rock.

The relationship between fracture stiffness and wave propagation is particularly interesting for shear waves (S waves). A shear wave propagating through a fracture set splits into two modes. The component vibrating normal to the fracture planes is significantly delayed compared to the component vibrating parallel to the planes. This phenomenon is called shear-wave splitting. Attenuation is larger for waves polarized subnormal to the fracture compared to waves polarized subparallel to the fracture. Research to determine how the properties of fractures affect shear-wave propagation in fractured rock should be undertaken to improve fracture detection.

Theoretical and laboratory studies of the correlation between hydrological and geophysical properties are useful, but it is essential to examine whether these relationships hold in situ. Two approaches hold promise: one is to make in situ measurements of elastic stiffness and electrical and hydraulic conductances of fractures and to use these measurements to develop empirical relationships. However, unique relationships would only be expected to hold under comparable geological conditions. Isolating significant numbers of fractures in a given geological setting might be extremely difficult. Alternatively, one could examine the hydrologic and geophysical properties of a single isolated fracture under different stress regimes (e.g., by changing the fluid pressure). Significant changes in pres-

sure should change void space volumes and contact areas in the fracture and, consequently, stiffness and electrical and fluid conductivities. In situ studies of the geophysical and hydrological properties of fractures should be undertaken to relate the results of seismic and electromagnetic surveys to the hydrological behavior of rocks.

Geophysical methods utilize the anomalous properties of fractures to locate and characterize them remotely. For example, single fractures have different elastic and electrical properties than the rock matrix. Fracture sets cause anisotropic behavior in the rock. Either of these conditions can be detected with geophysical methods. However, the line between detecting a fracture as a heterogeneity or as an anisotropy can become somewhat blurred, especially when the matrix rock is inherently anisotropic. This distinction also depends on the resolving power of the measurement technique used.

Surface reflection surveys utilizing polarized S waves can identify highly fractured rocks at depth by locating regions of high horizontal anisotropy. Similar surveys utilizing P waves fail to detect rocks with vertical fractures because at significant depths the waves propagate nearly parallel to the fractures and thus are largely unaffected by them. Acquisition and interpretation techniques should be more fully developed to take advantage of shear-wave properties of fractures.

Cross-hole and surface-to-borehole tomographic surveys use either seismic or electromagnetic waves to image the velocity and attenuation properties of rock. These methods can be used to identify hydrologically important features with sizes over several hundred meters if the rock is not highly attenuating. Inversion of travel time or attenuation data can resolve anomalous features on the order of a meter or less. The results depend significantly on survey geometry and wavelength. For example, two-sided tomography (e.g., between two parallel boreholes) is much less accurate than three-sided tomography. Efforts should be made to develop borehole shear-wave sources for tomographic imaging.

One of the important practical limitations of tomographic techniques for characterizing fracture hydrology is that tomography requires at least two boreholes that lie in the same plane. In many cases only one borehole is available for imaging the rock. Work should be undertaken to develop borehole seismic reflection methods that can map fractures at an appreciable distance from a single borehole, similar to what can now be achieved with borehole radar.

Tomography is practically limited to two dimensions, whereas hydrological flow paths in fractured rock are rarely confined to two dimensions. Surface reflection seismic methods that can create three-dimensional images are now relatively common in the petroleum industry. The technique is very useful but expensive because it requires a large number of source-receiver pairs. Efforts

should be made to reduce the cost of three-dimensional seismic methods and to adapt them to shallower and smaller-scale applications.

Images produced by geophysical tomography reflect variations in seismic or electrical properties, not hydrological properties. In some cases it is possible to qualitatively deduce the density and orientation of fractures from these geophysical properties. Quantitative interpretations of hydrological properties from these geophysical properties are extremely difficult, but not impossible, to develop. These interpretations will always be site specific. The primary difficulty is that interconnection of void space controls the hydrological-conductance properties of the rock, whereas interconnection has little effect on most geophysical properties. In fractured rock the relationship between interconnected porosity and total porosity is complex.

Probably the most useful approach for detecting hydrologically important fractures is to identify the particular geophysical signatures of particular geological settings. Case studies can be used to calibrate the methods and identify appropriate signatures. Application of several methods at one site greatly improves the possibility of identifying the signatures of hydrologically important fractures. Once the signatures are identified, the locations of similar features at other sites may be easier to detect.

An example discussed in Chapter 4 involved the laboratory analysis of core samples. It was shown that the matrix rock near a hydrologically active shear zone had a higher seismic velocity than the surrounding rock, possibly caused by deformation or alteration by circulating fluids. The fractures in the shear zone delayed the seismic waves, but the higher velocity of the rock near the shear zone nearly offset this delay. The result was a weak-velocity anomaly. However, fractures in the shear zone increased attenuation. The geophysical signature of this hydrologically important shear zone was a weak-velocity anomaly coupled with a high attenuation anomaly.

Another example discussed in Chapter 4 showed that the radar velocity anomaly was displaced in space with respect to the attenuation anomaly. The velocity anomaly may have been due to fluid-filled porosity, whereas the attenuation anomaly may have been due to clay minerals in the alteration zone around the fractures.

Field studies based on multiple measurements that identify the geophysical signatures of known, hydrologically significant fractures should be undertaken. Learning to interpret and catalog such signatures will greatly improve the use of geophysics in locating hydrologically important fracture systems.

Geophysical methods are very good at detecting changes, so one of the more promising approaches for locating hydrologically active fractures is to image the rock mass before and after changing some aspect of the flow system. The images produced before and after the change are differenced, which cancels out the natural heterogeneities, leaving only the anomalies caused by the hydrologically activated part of the system. A very successful example of this approach, discussed

in Chapter 4, used radar tomography before and after the injection of a highly conductive saline solution. The difference image showed the location of saline water in the rock, not unlike a barium x-ray used in medical applications. The analogy for seismic tomography might be measurements before and after the injection of gas, because a small-percentage decrease in saturation has a large effect on seismic properties. This type of imaging also has important application in remote monitoring of underground processes. Difference tomography is an important and promising area of research. Efforts should be made to develop these techniques for investigation of fracture systems.

Borehole logging techniques used in conjunction with other types of information have been very successful in characterizing fracture flow systems. A large variety of borehole logs are specifically designed for fracture systems. These are readily available and can provide multiple measurements of physical properties along the borehole. However, these logs usually sample only a small volume of the drilling-disturbed rock surrounding the borehole. Most logs are, at best, indirect measures of hydrological properties.

Borehole logging devices to map fractures and detect the rather discrete inflow (or outflow) zones that occur in fracture-dominated systems include a variety of imaging devices and extremely sensitive flowmeters. Development and commercialization of borehole logging devices and techniques for fractured rock should continue because they hold promise for relatively inexpensive, quick, and effective characterization tools.

In many cases the state of the art in geophysical imaging is far more advanced than the state of practice. Efforts should be made to develop and promote state-of-the-art geophysical technologies for routine use in the field. These should include radar, electromagnetic, and seismic methods.

Geophysical investigations can help locate hydrologically important fractures. These investigations are vital for designing well testing programs. In classical porous medium hydrology, stratigraphy determines the location and geometry of permeable units (i.e., "hydrostratigraphic units"). In fracture systems the permeable units (i.e., fracture zones or swarms) may have extremely complex geometries. Effective testing strategies must account for these geometries a priori. For example, to measure the properties of a fracture zone, it is very helpful to know its location, thickness, and orientation. Packers set above and below the fracture zone in both the pumping and monitoring wells isolate the hydraulic response of the zone.

The identification of hydraulically significant fracture zones in boreholes is not necessarily straightforward. Two fractures (or fracture zones) encountered in a borehole may have nearly identical geophysical signatures, but only one may conduct fluid. A single-hole hydraulic test that indicates a large hydraulic

conductivity may be bracketing an important fracture zone, or it may simply be testing a spurious single fracture that is connected to a fracture zone at some distance from the borehole. If the test indicates a small hydraulic conductivity, the test may not bracket the fracture zone, or it may simply be testing a low-conductivity part of the zone. Consequently, the identification of major fracture features from borehole data can be problematic.

A combination of borehole measurements (e.g., transmissivity, fracture frequency, electrical conductivity, acoustic velocity) can be used to identify important fracture zones. At Stripa, for example, a fracture zone index (FZI) was defined as a weighted combination of borehole measurements (see Chapter 4). The weights were a function of the correlation matrix of all the measured variables. The FZI was a much better indicator of hydrologically important fracture zones than single measurements. Fracture zone indices based on multiple geological, geophysical, and hydrological measurements should be developed and tested at several sites in order to determine the general utility of this approach.

Hydraulic tests are valuable tools for diagnosing certain aspects of fracture geometry. Test data are usually interpreted by specifying a conceptual model for the hydrological system in the vicinity of the well and then determining the parameters (e.g., size permeability or storativity of the specific features) that give the best match to the observed hydraulic data. A good match between the data and the model results (arrived at with realistic parameters) lends support to the underlying conceptual model. For example, model solutions exist for the case of a single fracture of large areal extent intersecting a well. Flow geometries in fractured rocks are usually more complex than this simple case; consequently, interpretation of hydraulic tests in fractured systems can be challenging. In addition, fluid flow is described by the diffusion equation, which means that the signal decays rapidly and may not be affected by heterogeneities. Further, large-scale tests can take a long time to complete. Diffusive behavior limits the ability of hydraulic tests to diagnose flow system geometry.

Hydraulic tests permit inferences about the geometry of the fracture network beyond the wellbore by comparing the test results to specific model geometries. For example, a well that intersects a large, highly conductive fracture has a test response like a constant-head boundary. If the rock has a fracture or if other fractures intersect the fracture being tested, the test results will match the mathematical solution for leakage into a high-transmissivity "aquifer." Likewise, a fracture termination will match the solution for an impermeable boundary. Such solutions are widely available and well known. Application of these solutions to fracture systems only requires relating the geometry underlying the previously derived mathematical expressions to the geometry of the fracture system.

It may be valuable to compare hydraulic test results to solutions for flow systems that are one, two, and three dimensional. One-dimensional, or linear, flow occurs in an open fracture where the borehole lies in the plane of the fracture (e.g., a vertical fracture intersecting a vertical well). One-dimensional flow can

also occur in a one-dimensional fracture channel. Two-dimensional, or radial, flow occurs in open fractures that are not parallel to the borehole, in highly interconnected networks of fractures confined to a relatively thin stratum, or in two-dimensional fracture zones. Three-dimensional flow occurs in networks of fractures that are highly interconnected throughout a three-dimensional volume of rock or when matrix permeability is comparable to the fracture permeability.

Well test solutions now exist that allow the dimension of flow to take on a range of noninteger flow values between 1 and 3. For example, a two-dimensional fracture zone consisting of poorly interconnected fractures might have a flow dimension between 1 and 2. A flow dimension between 2 and 3 could indicate a poorly connected three-dimensional network. Comparing test data to these solutions can be a powerful diagnostic for system geometry when some additional information about geometry is available. In the absence of any constraints, however, it may be difficult to distinguish between different system geometries. There is a need to relate geometric fractal dimensions for fracture systems to flow dimensions determined by matching noninteger dimension solutions to the well test equation. In this way it may be possible to determine something about the fractal dimension of the fracture system from the well test solution.

The cointerpretation of multiple well tests is a good way to determine critical aspects of fracture flow systems. Cointerpretation forces the system response to be consistent with all the observations simultaneously. Two approaches are useful. In the first, observations are analyzed statistically to find the best-fitting permeability tensor for the medium. In the second, inverse techniques are used to find patterns of heterogeneity that fit the data. Cointerpretation of multiple well tests and well test observations is a promising research area that may provide a practical characterization for many sites. Research in this area should be expanded.

Tracer tests are powerful tools for confirming the existence of connected flow paths. However, meaningful interpretation of the transport properties of the medium require that the geometry and boundary conditions of the flow system be extremely well understood. The only examples of successful tracer test interpretation are for simple fracture geometries, for instance, in cases in which there are dominant fracture zones. The detailed characterization, flow modeling, and subsequent tracer testing of complex fracture systems should have a high research priority. Such tracer tests should include natural gradient tracer tests where possible. This research will allow the development of appropriate tracer testing methodologies and interpretations. There is a need to determine successful testing strategies, including borehole configurations and injection and withdrawal rates.

Tools have been identified that address the first key question—How can fractures that are significant hydraulic conductors be identified, located, and

characterized? Geological investigations can indicate which features are likely to conduct fluid and their likely spatial relationships. Geophysical techniques can detect these features, and well tests can confirm their hydrological role and allow inferences to be made about the fracture network geometry. In crystalline rocks it is now possible to identify important fracture features over scales of several hundred meters. For larger scales of interest, other geological media, or limited resources, it is more difficult. However, it is increasingly possible to make educated hypotheses about the nature of fracture systems.

The ability to identify hydrologically important fractures in crystalline rocks is a result of the number of excellent in situ research facilities in this medium (Chapter 8). Recently, the petroleum industry has begun to develop in situ research facilities in fractured sedimentary units. Characterization research consists of repeated sequences of characterization, prediction, and comparison. In situ research facilities are important because they provide the opportunity to move freely between these stages. Progress in fracture characterization depends to a great extent on the availability of appropriate in situ research sites. These sites are especially useful when a large number of characterization methods can be applied to assess the utility and advantages of each method. Additional in situ facilities should be developed in fractured rocks in a variety of geological environments in order to improve the ability to identify, locate, and characterize hydraulically conductive fractures.

There are three key tools used to understand how flow and transport occur in fracture systems:

• Conceptual modeling (Chapters 2, 3, 4, 5, and 6)

• Mathematical modeling (Chapter 6)

• In situ flow and transport experiments (Chapters 5 and 8)

A conceptual model is a simplified representation of the real system based on a cointerpretation of all the observations. A conceptual model is a hypothesis describing the main features of the geology, hydrology, and site-specific relationships between geological structure and patterns of fluid flow and transport. The conceptual model serves as the basis for mathematical models of flow and transport. There are no standard parameters for these models; each conceptual model defines its own appropriate parameters. Consequently, it is necessary to alternate between in situ testing, interpretation, and definition of a model. Predictions made with the model are compared to measurements, and the model is updated

accordingly. Then the process begins again. Each cycle of prediction, measurement, interpretation, and model update should bring the system representation closer to reality. If the iterative process does not improve the predictions, it is appropriate to question the underlying assumptions. Such an elaborate process is not always possible in practical situations, but it is a fundamental and critical part of successful characterization and predictive modeling.

Building the conceptual model is the most important part of the modeling process. Consider a system with an in situ measurement of transmissivity. The predicted transport velocity for this system will vary by many orders of magnitude depending on whether a three-dimensional porous medium model, two-dimensional parallel-plate model, or one-dimensional channel model is used. The error associated with the choice of conceptual models is much more significant than any measurement error or numerical errors involved in the mathematical simulation.

The amount of detail required in a conceptual model depends on the phenomenon of interest. A simple prediction of flow rate as a function of time is not highly dependent on a detailed representation of the heterogeneity. Thus, problems involving reservoir yield may not require detailed conceptual modeling. Protecting a water supply well from contamination or optimizing a fluid recovery system, on the other hand, requires a more complete understanding of geology and hydrology.

Understanding the geometry of fracture system and how that geometry controls flow is central to conceptual modeling. In the laboratory it is possible to measure the geometry of a portion of a single fracture and quantify the void space that controls flow with a variety of methods. It is possible to control the boundary conditions that produce flow in a fracture and make measurements of flow and transport phenomena. However, if a sample can be brought to the laboratory for study, it usually is not highly fractured. Highly fractured samples do not maintain their integrity and are consequently difficult to collect and study. In many, but not all, cases this is tantamount to saying that only those fractures that are not the primary conduits for flow can be studied in the laboratory. Further, laboratory samples are necessarily small. Consequently, they are not representative, and in most cases it is difficult if not impossible to scale up results obtained in the laboratory to applications in the field.

For these reasons, laboratory studies of fracture flow may be difficult to apply to field cases from a quantitative point of view; in the worst cases, their application may lead to misconceptions about the natural flow systems. Nevertheless, these studies are extremely valuable from the qualitative point of view because they reveal the relationships between fracture geometry and flow and transport properties. Although in situ values of permeability cannot be determined from laboratory studies, such studies do reveal how fracture geometry controls the flow. Laboratory studies are also the primary source of information to link geophysical and hydrological properties.

Recent work has demonstrated that an appropriate conceptual model for flow in a single fracture is that of a two-dimensional porous medium that is sensitive

to stress. This key concept has almost entirely replaced the parallel-plate analogy for fractures. The parallel-plate analogy is useful for predicting changes in permeability that occur as a fracture is opened, such as in hydrofracture operations. However, when fractures have a significant contact area, the parallel-plate analogy is applicable only on the local scale and is not particularly useful for understanding flow and transport on the macroscale. The concept of identifying a single aperture for a fracture (i.e., the parallel-plate analogy) is far too simple for many applications. However, it is impractical to completely define fracture void geometry in most cases. Consequently, simple methods should be developed to describe the geometry of fracture openings to provide sufficient information to determine flow and transport properties.

Following the two-dimensional porous medium analogy, it is not surprising that flow in fractures occurs along the paths of least resistance, which are formed by the most efficient connections between large pores. Fluid flow through these pores is effectively ''channelized." The necks in these "channels" control flow. During transport of solutes, the large pores, which contain relatively large volumes of water, dilute the solute and thus delay its movement. Research is needed to determine how the geometry of a fracture gives rise to preferential flow paths and determines the amount of rock surface area that will affect matrix diffusion and reactive transport.

The two-dimensional porous medium model for flow in a fracture is particularly important for two-phase and multicomponent immiscible flow. Theoretical and laboratory studies indicate significant phase interference (the presence of one phase that blocks the flow of other phases) in fractures when the matrix rock is impermeable. Conceptualization of the fracture as a two-dimensional porous medium provides a basis for understanding this phenomenon because the blocked phase does not have the third dimension to use as an alternate route.

In multicomponent flow the nonwetting phases tend to occupy the larger pores. If the fracture surfaces become sealed with impermeable minerals or if the matrix rock is near saturation, transport of the wetting phase in the fracture may be even more channelized than in single-phase flow. Increased channelization implies short fluid flow paths. If the matrix is permeable, however, the wetting fluid, which initially enters the system through fractures, migrates into the smaller pores of the matrix. In this case the matrix controls the flow and the fractures act as barriers. These two conceptual models give radically different values of travel time: the channelized flow model will conduct fluids much more quickly than the permeable matrix flow model. More work is needed to determine which of these mechanisms is dominant under realistic conditions.

Many analyses of two-phase flow in fractures assume that the flowing phase is continuous. However, observations have shown that isolated "blobs" of either phase may move. For example, air bubbles can be entrained in water flow. Another mechanism that may be important is phase displacement. As pressure in the wetting phase increases, isolated portions of the wetting phase may displace

the nonwetting phase occupying the larger, more permeable pores in the fractures. Subsequently, the nonwetting phase displaces the wetting phase again. This suggests that two-phase flow in fractures may be oscillatory in nature. Researchers have observed oscillatory two-phase flow in the laboratory. Additional work should be undertaken to determine if the period of oscillatory behavior increases with the scale of the flow system. Further, the behavior of two-phase flow in fractures may be truly chaotic.

Development of appropriate conceptual models for multiphase flow in fractures is a critical research problem, with applications to many problems of societal interest, for example, the siting of nuclear waste repositories above or even below the water table, predicting the effects of corrosive gas releases from the corrosion of nuclear waste containers, for nonaqueous-phase liquid contamination, and for enhanced oil recovery. New theoretical and laboratory work should be undertaken to relate multiphase flow in fractures to aperture distribution, matrix properties, and stress. There is a critical need for experimentation and observation of two-phase flow in natural systems.

Given insights gained from laboratory tests, field observations provide the basis for developing conceptual models appropriate to the site and problem at hand. All the tools described above under the first key question are brought into play. The identification of fracture patterns, inferences about which fractures are open and connected, characterization of fractures through geophysical imaging, and well testing are used as discussed above. The analyst interprets the data to develop a working description of the fluid flow system. The process should be flexible and interactive. Uncertainties in the model indicate the need for further data collection.

The characterization of fracture systems and the development of conceptual models require access to the field. Conceptual modeling studies should be pursued in a variety of geological settings in order to define the relationships between lithology, stress, structure, fracture style, and permeability. Researchers should integrate these studies for presentation to the earth sciences community in the form of review articles as well as detailed, narrowly focused technical articles.

The best way to develop conceptual models is through examination of case studies; Chapter 8 gives some recent examples. Good case studies are available for fractures in crystalline rocks. The Stripa Project, for example, stands out for its use of geophysics in conceptual modeling. Here, researchers made an integrated interpretation of the flow system from a comprehensive set of geophysical and hydrological measurements. The project documented the process and thus serves as a significant resource. The Underground Research Laboratory case study is a good example of the application of geological understanding to hydrology. Researchers correlated the structure of a major shear zone to measured hydrological properties. This correlation provides a basis for extrapolating hydrological properties through geological analysis. Work at the Grimsel site in Switzerland is a good example of integrating geological mapping and geophysical imaging.

The Mirror Lake example shows that various geophysical and hydrological measurements are needed to characterize flow and transport in fractured rock. The Geysers case study shows how to use the relationship between hydrological activity and monitored acoustic emissions to define a reservoir.

In crystalline rock sites, fracture zones dominate the hydrology. Many of these zones have undergone significant shear deformation, and they are not uniformly permeable. Hydrology dominated by flow in fracture zones may not be a universal property of crystalline rocks, but there is certainly enough evidence to suggest that this is a common occurrence. Case studies for fracture systems in stratified rocks indicate that the hydraulically significant fractures are the classic strata-bound joint systems, with joint clusters forming regions of significant permeability. Compared to the number of research facilities in massive crystalline rocks constructed mainly for nuclear waste storage research, there is a dearth of multidisciplinary experimental facilities in bedded rocks, which are important for many resource and structural problems. Experimental facilities in bedded rocks should be developed. Conceptual models developed at such facilities will aid conceptual modeling process at other sites with similar geological conditions.

Given an appropriate conceptual model, the analyst can choose from a large suite of numerical modeling tools. It is important to note that all of these models use similar mathematical constructs. They differ mainly in their representation of physical heterogeneity. The nature of the geometry that controls flow, the scale of the problem of interest, and the phenomena being modeled dictate the choice of how to discretize the system, that is, the choice of modeling approach.

Numerical model types comprise a spectrum ranging from equivalent continuum models to discrete fracture models. Equivalent continuum models have the advantage of being well known and commonly applied. They tend to be more useful for well-connected fracture systems, for relatively large scales of interest, and where information is needed on average behavior rather than details of the flow paths. These models treat large portions of a flow region as having uniform properties. Such a treatment may fail to adequately represent the flow field if the fractures have highly variable properties or are poorly interconnected. The approach of using a stochastic continuum may overcome many of these problems. In this approach the continuum is discretized into elements and a stochastic process determines the properties of each element. This approach is likely to be useful when fracture connectivity is high, but there is still a significant variability in fracture conductance.

Multiple-continuum models have been used extensively in cases where fractures form the dominant flow paths and the matrix porosities are high enough to provide significant storage. The literature on this topic is extensive, especially the petroleum literature. The interested reader may consult any of several good

reviews of this literature (e.g., Kazemi and Gilman, 1993). These models work best when the fracture network is well connected, for relatively large scales of interest, and where information is needed on average behavior rather than details of the flow path.

At the other end of the spectrum are those discrete fracture models that attempt to include every conductive fracture as an explicit feature in the model. Discrete fracture codes assign continuum properties to subdomains and prescribe how these subdomains interact. The subdomains may represent a single fracture, a channel in a fracture, or a fracture zone, for example. These models are suitable for calculating complex flow paths resulting from the geometry of the interconnected fracture network. A stochastic process places the fractures in the model according to specified statistical distributions describing the location, orientation, extent, and conductive properties of fractures, as well as any spatial relationships between fractures.

The discrete approach may require significant data collection efforts. Data requirements can be difficult to meet, particularly if individual fracture conductances are required. Conductance distribution measurements have been attempted with hydraulic tests using short test intervals to isolate single fractures. However, it is not usually possible to isolate single fractures in situ or to determine how the conductance measurement is affected by other fractures that are connected to the fractures being measured. Consequently, different fracture conductance distributions can be inferred from a single set of flow test data.

If a fracture can be isolated in a well test, it is not usually exposed in an excavation or outcrop where the trace length can be measured. Consequently, it is extremely difficult to find statistical correlations between fracture size and conductivity. One way to overcome this difficulty is to generate a random series of fracture systems that match the observed distributions of fracture trace length, orientation, and transmissivity. These, in turn, provide for a range of possible models.

In the past, the discrete approach was limited by the amount of computational power required to account for every fracture. However, this limitation is disappearing rapidly; presently available computers now have the necessary computing power for practical problems. Models with hundreds of thousands of fracture elements have been run in workstation environments. Further advances are anticipated with parallel computers. Research into efficient numerical methods continues to be important for practical application of fracture models.

One way to reduce the computational requirements is to change the way in which a "discrete" feature is interpreted. For example, on a scale of tens of meters, the model may include every fracture that is larger than about 10 cm. On a scale of hundreds or thousands of meters, however, fractures can be lumped into discrete zones that are interpreted as single features. This "lumping" or "equivalencing" produces a model that has both discrete and continuum properties. The more features that are lumped together and described in terms of average

properties, the closer the model is to an equivalent continuum model. A rationale for lumping features of the fracture network when modeling either fluid flow or the transport of reactive solutes should be developed.

An attractive approach to fracture system modeling incorporates the hierarchical nature of fracture systems. This involves the generation of fractures according to stochastic rules inferred from observations made at outcrops. For example, some methods describe the geometry of the random smaller fractures with respect to larger fractures that are either random or deterministic. The construction of the model network can mimic the inferred genesis of the natural fracture system. This involves the creation of random flaws that grow at each iteration according to rules inferred from fracture mechanics and calibrated by observations from outcrops. The generation of fracture models based on fracture genesis incorporates physical processes into the fracture pattern. Consequently, these methods provide a physical basis for extrapolating fracture patterns to locations where there are no measurements. Mathematical descriptions that are based on physical properties of fracture genesis provide an additional level of confidence in model predictions. The development of hierarchical fracture models based on physical processes should be a priority for research.

It is also possible to create hierarchical fracture systems through inverse methods. In this approach a general arrangement of fluid conductors is first defined based on a conceptual model. Then the particular arrangement and strength of the fracture conductors are optimized such that the model's results match certain observations, for example, interference or tracer test results. These methods are particularly useful when the pattern of conductors determines the hydraulic behavior. For example, fracture systems can be modeled as simple, partially filled lattices that lie on planes representing major fracture zones. These are called equivalent discontinuum models. The system is fully described by specifying the presence or absence of each lattice element. However, a rather large number of parameters must be optimized in this process. Consequently, another issue that arises from the use of inverse analysis is the nonuniqueness of resulting models. A large number of inverse solutions can be found, and these can be examined to identify common elements.

Recently, iterated function systems (IFSs) have been used to represent networks of conductors. IFSs use a small number of parameters to control an iterative process that determines a geometric object called an attractor or a fractal. If the attractor defines the variations in hydraulic conductance, the parameters of the iterated function system can be optimized so that the system reproduces the observed test results. Using an IFS, a complex fracture flow system can be fully described by using tens of parameters rather than thousands. Research to find simple but sufficient mathematical descriptions of fracture networks and to determine what constraints the data actually place on the model should be a priority for inverse modeling research.

Little work has been done to understand the relationship between fracture connectivity and matrix permeability. When matrix permeability is low, fracture connectivity controls flow. When matrix permeability is comparable to fracture permeability, flow can pass from one fracture to the next through the matrix and fracture connectivity is not important. As long as fracture connectivity is an important factor in the models, continuum assumptions are suspect. Parameter studies should be conducted to see how the relationship between fracture connectivity and fracture network permeability is affected by matrix permeability.

On large (regional) scales it may be more difficult to locate the hydrological features and to test the hydraulic behavior of the fractured rocks. One of the most promising methods for inferring hydrological behavior at large scales is to use natural geochemical tracers. The use of natural geochemical tracers to constrain large-scale flow models in fracture systems deserves more attention. These efforts should include research to understand geochemical processes and boundary conditions.

Discrete or hierarchical models are useful when it is necessary to model the locations of flow paths and transport velocities. In many fracture systems, flow paths are highly irregular compared to those predicted by continuum models; consequently, models that account for heterogeneity are at a distinct advantage. However, the development of appropriate conceptual models for transport in fractured rock is not yet complete. It is difficult to characterize solute distribution, especially when transport occurs in narrow pathways in fracture planes. It is also difficult to design and run effective tracer tests in fractured rocks. Such research is ongoing at a number of sites, including the Mirror Lake site operated by the U.S. Geological Survey in New Hampshire and the Underground Research Laboratory operated by Atomic Energy of Canada Limited in Manitoba. Similar work is planned at the Hard Rock Laboratory, operated by the Swedish Nuclear Fuel and Waste Management Company in Sweden.

The modeling of two-phase flow, heat transfer, and chemical transport in fracture networks is almost completely confined to single- and dual-porosity continuum models. These modeling efforts are hampered by a poor understanding of two-phase flow parameters as well as the mechanisms governing flow behavior. Recently, fracture network approaches have been used to investigate the behavior of networks under two-phase conditions. These studies utilized what are essentially theoretical percolation models. As mentioned above, these modeling efforts are hampered by a poor understanding of the appropriate physics to use in the conceptual model. Models of two-phase flow in fracture networks can help reveal how the geometry of a network controls flow. Such theoretical research on the nature of two-phase flow in fracture network should be continued.

One of the key problems in the prediction of reactive solute transport is a poor understanding of the geometry that controls reactions between the rock and the transported species. The problem of reactive transport in fracture systems is largely unsolved. Theoretical investigations and experimental work should be

supported to link hydrogeological and geochemical processes in a transport model for reactive solutes that accounts for the unique properties of fluid pathways in fractured rocks. Efforts should also be directed toward determining the effective contact area between a reactive solute and the network of fractures forming connected pathways.

In situ experimentation is important for determining how flow and transport occur in a fracture system. Field-based flow and solute migration experiments should be carried out at a variety of scales, ranging from several tens of meters to hundreds of meters in a variety of host media with a range of styles in fracturing. Field experiments can test the appropriateness of conceptual models. The design of these field-based experiments should address data requirements for evaluating modeling approaches and capabilities for both flow and transport. This effort should also address the practicality of collecting appropriate data in routine hydrogeological practice.

It is not straightforward to compare the results of flow and transport experiments to stochastic model predictions. The stochastic model produces multiple predictions and a distribution of results that can be quite broad. Field experiments are usually very expensive; frequently, only one experiment is carried out, producing only a single realization of the flow system. It is difficult to evaluate the model prediction using this single realization. Measurement on a small scale exacerbates the problem. For example, at Stripa, prediction of total inflow to six parallel boreholes in a 1-m radius was much better than prediction of flow into a single borehole because the six-borehole measurements represented an average over a larger scale. In addition, it is not always possible to measure the quantities predicted by the model. Conversely, the measurements frequently do not constrain those critical aspects of the model that affect the predictions. In other words, it is very difficult to "validate" a model.

Improved methodologies for formulating conceptual models of flow and transport in fractured media are under development. Efforts to translate these models into quantitative simulation tools should include estimates of the uncertainties in the predictions. Modelers and experimentalists should devise useful, appropriate alternatives, such as the process of "confidence building," to strict model validation. Where possible, field investigations should be designed for multiple repetitions of the same simple experiment, as this may serve to build confidence better than one complex, all-encompassing experiment. Efforts should be made to focus model and field comparisons on predicted quantities that have small variances.

Conceptual modeling, numerical modeling, and in situ testing can be used to address the question: How do fluid flow and chemical transport occur in

fracture systems? Several conceptual models have now been developed for crystalline rocks, and several examples are available. However, additional work should be undertaken to develop appropriate conceptual models for transport and multiphase flow. Conceptual models should have a strong physical basis to provide a reasonable rationale for extrapolation and scaling up. More investigations should be undertaken in bedded sedimentary units. A wide variety of numerical models are available, and progress with these is noteworthy. Fracture flow and transport models should be developed and made more practical and usable. A poor understanding of the physical basis for complex phenomena in fractured rock is primarily responsible for inhibiting their modeling, and research should be directed at this problem. As with the first key question, progress depends on the availability of experimental facilities, and such facilities should be supported.

The role that fractures play in fluid flow and transport can change quickly and dramatically. Changes in effective stress, shear deformation, or temperature gradients, as well as fracture leaching or filling, can change the void space that controls fluid flow. Relatively small changes in void space can be significant because fractures behave like two-dimensional porous media: when the fluid flow in the fracture is blocked, the third dimension is not available as an alternate pathway. Conversely, opening a constriction can dramatically increase flow. Predicting and attempting to control changes in fracture systems are often the most critical parts of an endeavor involving fracture flow.

Change occurs through physical or chemical processes that alter the pore space of a fracture. For example, a fracture may deform in response to changes in effective stress caused by changes in boundary conditions, fluid pressure, loading, or temperature. Fluids may undergo phase or component changes, where one phase or component blocks the flow of the other. Solids may be added to the fractures by injection of hydrofracture proppant or grout. Mineral precipitates may fill the void spaces. Leaching of the fracture walls by corrosive fluids may occur. These processes are further complicated by coupling between mechanical, hydraulic, chemical, and temperature effects.

Problems involving changes to fracture systems are usually addressed through:

• Laboratory studies of coupled behavior in single fractures (Chapter 3)

• In situ testing and procedures (Chapter 7)

• Mathematical models of coupled phenomena (Chapter 7)

There have been relatively extensive laboratory studies of the interaction of effective stress and flow. The same is not true for the relationship between shear stress and flow. Very little laboratory work has been done to study the way in which fractures become filled or leached.

The parallel-plate analogy traditionally used to describe fracture geometry implies that fracture conductance is proportional to the "hydraulic" aperture cubed (the cubic law). The cubic law works well for fractures whose sides are not in contact. However, it breaks down as the fracture sides come into contact. In this case the amount and distribution of the contact area control the flow. As the contact area increases, the aperture available for flow decreases faster than predicted by the cubic law. Numerical models that relate fracture void geometry, deformation, and flow are available and are generally good at matching experimental results. However, generalized relationships between fracture stress, closure, and flow properties are not yet available. Research should be undertaken to relate specific changes in flow properties and stress to different lithologies and fracture histories. Understanding gained in the laboratory should be correlated to behavior observed in the field. Part of this work should be focused on studies relating the local stress heterogeneities to fracture permeability and flow.

The relationship between shearing and fracture conductance is poorly understood. It must be considered in three spatial directions: (1) in the plane of the fracture in the direction of the shear, (2) in the plane of the fracture perpendicular to the shear; and (3) perpendicular to the plane of the fracture.

Conductance in the direction of shear could increase owing to dilation accompanying deformation, but continued shear or high normal stresses can produce an impermeable gouge in the fracture plane. In the plane of the fracture and in the direction perpendicular to shear, permeability can increase for two reasons. First, dilation of the fracture may create channels perpendicular to the direction of shear. Second, small fractures may form parallel to the intermediate principal stress. The direction perpendicular to the plane of the fracture is important if fluid storage in the rock matrix is of interest. Here, dilation should have little effect, but gouge would cause a significant decrease in communication between the fracture and the matrix. The effect of secondary fracturing may offset the effect of gouge production. Our understanding of these phenomena is incomplete. A better understanding of the effect of shear on fracture permeability should be the focus for new research because of its significance for engineering projects, as well as for understanding the nature of fracture conductance in natural shear zones.

Little laboratory work has focused on precipitation, sedimentation, and leaching in fractures. Fluid velocities in large pore spaces of fractures are likely to be lower, which may encourage precipitation. On the other hand, dissolution may occur preferentially at the asperity contacts, owing to higher contact pressures.

Precipitation that takes place preferentially in the large pores would tend to even out the aperture size. Thus, filling the large pores would have little effect on the conductance and a large effect on storage. Consequently, precipitate-laden fluids could continue to flow through and fill the fractures. If precipitation occurs evenly throughout the fractures, the necks will be closed first, reducing fracture permeability and leaving isolated void spaces. These fractures may appear as geophysical anomalies but would not be significant fluid conductors. In general, the opening of fractures increases fluid conductance and lowers the pressure, which tends to encourage precipitation, thus closing the fractures, decreasing the conductance and increasing the pressure. Further study of vein formation may provide insight into problems involving chemical precipitation in fractures as well as the nature of relict flow systems. Laboratory and theoretical studies should determine how and when fractures become filled or leached and how these changes affect fluid flow and transport. Part of this effort should involve examination of the rheology of grouts and their flow behavior in complex fractures.

The importance of couplings between mechanical, hydraulic, chemical, and temperature effects in fracture systems should be evaluated in the field, and techniques should be developed to predict their effects on fractures. It is especially important to understand the couplings between these processes for projects with long design lives because the effects of changes in fracture systems may be hard to predict based on short-term testing and evaluation schemes.

There is considerable laboratory data to establish constitutive relationships between effective stress and fracture permeability for single fractures in rock cores (Chapter 3), but field-based data are few in number. Wellbore operations cause stress-related changes in fracture systems. Under fluid extraction, formation pressures decrease and fracture conductances can decline. However, stress changes do not always affect fracture permeability. Stress sensitivity may be absent in diagenetically altered fractures with stiff mineral "bridges" that prop open the fracture. For shallow flow systems, fracture permeability can be entirely independent of the state of stress or flow. At the other extreme, hydraulic fracture studies indicate a strong relationship between permeability, stress, and flow. The borderline between these two regimes seems rather poorly explored. Work should evaluate the practical significance of stress-sensitive fluid flow in fractured rock for a range of field-scale problems. Work should also identify the threshold and scale where the stress regime affects fluid flow.

Pressure decreases near an extraction well, leading to a decline in permeability. It may be possible to overcome these pressure effects through hydrofracture. Hydrofracture creates new fractures to augment the existing permeability. Hydrofracturing increases the permeability by creating fluid conductors that intersect the well. In a sense the hydrofracture simply increases the effective radius of

the well. If the hydrofracture connects the well to an existing fracture system, a significant increase in hydraulic communication will result. However, if the present stress orientation is responsible for the existing fractures, the hydrofractures will likely be parallel to the existing fractures. In this case, hydrofracturing is unlikely to increase fracture connectivity and permeability. Better methods should be developed to predict hydrofracture geometry in complex geologies, especially when the rocks exhibit nonelastic behavior, in order to predict the effect of the hydrofracture on the local fluid flow regime.

Water well drillers are now commonly using hydrofracture technology. Typically, water well drillers do not use proppants and could benefit from proppant technology developed in the petroleum industry. Hydrofracture technology developed in the petroleum industry should be transferred to water supply hydrofracture practice.

Fluid injection is more complex than extraction because chemical reactions between the injected fluid or particulates and the formation can plug the fractures. Consequently, the pressure needed to inject fluid increases, possibly to the point of hydrofracturing the formation. In general, such hydrofractures propagate upward as well as outward. A vertical hydrofracture may be particularly problematic for disposal wells because it can breach the strata chosen to isolate the waste.

Temperature changes create thermal stresses that affect hydraulic conductance. Cold fluids injected into hot rock produce thermal stresses that decrease effective stresses. Placing a heat source in rock (e.g., nuclear waste) may close some fractures and open others. These stress changes, which are caused by pore pressure changes and thermal volume changes, can substantially alter the hydraulic properties of the rock mass. Work should be undertaken to relate temperature changes to effective stress changes and change in flow properties through in situ experiments conducted where fracture systems have been extensively characterized.

Injection of grout into fracture systems from boreholes is a technology that is receiving increasing attention. Grouting helps prevent fluid flow into underground openings and around dams. This technology has been used to isolate previously emplaced toxic waste. The success of a grouting program depends on the characterization of the fracture system. Grouting programs are more likely to be successful if the flow paths are well-characterized beforehand. Discrete fracture flow models developed for a site may be very useful in planning an isolation system, especially if the model can simulate the fluid properties of the grout. Geophysical monitoring of the grouting process provides important information about the progress of the operation and its success. Given the vast amount of waste in the underground and the extreme difficulty associated with its remediation, as well as the large number of landfills situated over fractured rock, grouting as an isolation technique in fractured rock deserves much more attention. Research should couple grouting techniques with the characterization and flow modeling of the fracture system.

Corrosive waste materials or in situ mining operations can leach fracture walls or fillings. Withdrawal of geothermal fluids can leave precipitates in fractures. The coupling between flow and chemistry is very difficult to understand and predict. Fracture characterization and monitoring of fracture systems should be linked to the chemical processes in controlled in situ experiments.

The effects of stress, precipitation, and temperature are coupled. For example, hot, mineral-rich fluids precipitate minerals as they cool. As the fractures become filled with minerals, resistance to flow increases. If there is a constant driving force, an increase in fluid pressure and a consequent decrease in effective stress will result. The increased pressure causes fractures to open or hydrofracture, increasing the conductance, which, in turn, increases the amount of fluid circulation and precipitation. This cyclic behavior has been observed in nature. For example, accretionary wedges associated with plate subduction are the source of ''book quartz," which consists of fractures filled with quartz laminations representing multiple cycles of filling and opening. Further work in understanding how natural coupled processes affect fracture patterns would be useful in applications such as in situ mining and geothermal energy production. Studies of fracture filling should be keyed to fracture characterization as a way of advancing an understanding of large-scale coupled processes.

A fully coupled model of fracture behavior links hydrological, stress, temperature, and chemical effects. It may be possible to link all four of these effects when the geometry and boundary conditions are very simple. For more complex situations it is difficult to incorporate even two of these phenomena in a coupled model.

Hydrogeological simulation models that are linked to mechanical deformation models originate primarily in the field of geotechnical engineering. The models generally use a discrete representation for the fracture network, with separate values assigned to fracture and matrix stiffnesses. These hydromechanical models have been used to examine the nature of the coupling between fluid pressure response and deformation in idealized settings. Workers attempting to fit hydromechanical models to data sets from single-borehole deformation experiments have great difficulty in identifying unique sets of parameters that adequately characterize the field data. Some success has been achieved in duplicating single-fracture experiments. However, because of the multitude of parameters that determine the nonlinear hydromechanical response of a rock mass containing a network of fractures, the prospect of site-specific predictive simulations is not promising in the near future. Field-scale experiments should be conducted to test models that calculate the deformation of a fractured rock mass during fluid injection or withdrawal. Efforts should be made to construct simplified hydromechanical models that predict changes in flow conditions from changes in stress.

Numerical models for calculating the effects of temperature changes relate temperature gradients to changes in stress. Models exist to calculate the opening of a single fracture caused by thermoelastic effects. However, existing coupled stress, flow, and temperature models may not predict long-term behavior, particularly for two-phase flow, because they may not include the right physics or account for complex fracture geometry. In situ experiments, particularly at large scales, should be conducted to develop reliable models for the effects of coupled stress, flow, and temperature.

Numerical models that couple fracture flow, stress, and fracture filling are used to address the deformation of fractured rock masses. For example, deformation of a rock mass caused by well drawdown, or the effects of grouting and drainage on a fractured rock mass underlying a dam, can be calculated with these models. Models that address coupling between flow and leaching or precipitation in fractures are not generally available. Such models should be developed, as they may be necessary for certain contaminant transport or in situ mining problems.

The factors controlling chemical precipitation during reinjection of fluids are incompletely understood. Researchers have modeled silica redistribution for geothermal reinjection problems. Simulation of more complex chemistries, such as carbonate precipitation, is very difficult because of buffering effects. Better methods for predicting permeability reduction owing to chemical precipitation (scaling) in fracture networks should be developed.

Answers to the third key question—How can changes to fracture systems be predicted and controlled?—are critical in a variety of endeavors of societal importance. The effects of changes in effective stress are better understood than changes in shear stress; the effects of temperature changes are better understood than chemical changes. Predicting and controlling these changes requires thorough characterization of the fracture systems. There is much practical experience with coupled fracture problems, but there are few controlled sites where the various processes can be sorted out from the effects of the fracture geometry. Again, research facilities in a variety of rock types should be established. These facilities should be used to develop technology to control and predict changes to fracture systems.