izontal well lengths, but which hori- zon and direction are best? .... was a good opportunity to make a. 708 ... kinematic cement is the âbad choles- terolâ of natural ...
Coordinated by M. Ray Thomasson and Lee Lawyer
GEOLOGIC
COLUMN
New directions in fracture characterization STEVE LAUBACH, RANDY MARRETT, and JON OLSON, The University of Texas at Austin, U.S.
Using rotary sidewall cores from a
deep natural gas exploration well, a small E&P company recently measured natural fracture intensity, orientation, and openness in four separate potential target horizons. Results pinpointed one zone having high fracture intensity and open fractures. Another zone with similar fracture intensity was identified as having closed fractures. Along with information on fracture strike, results were used to evaluate the risk of stimulating the well versus drilling a horizontal lateral. Historical production data from the play suggest that natural fractures are key to outstanding production, yet conventional well logs provided little usable information on natural fracture attributes in the four potential completion targets. Why? The vertical well did not intersect visible fractures. Sidewall cores also lack fractures visible to the unaided eye. So how were the measurements used in the fracture evaluation obtained? The methods used in this well to diagnose natural fractures represent innovative use of microfracture and diagenesis data as surrogates for the large fractures that are so difficult to sample. The University of Texas at Austin is conducting the research in cooperation with an industry consortium. Many questions remain, but the potential is clear for these new directions in fracture characterization to have application in exploration and development. One application that will be tested this year in a study supported by the U.S. Department of Energy (DOE) is to use these corebased fracture characterization methods to calibrate seismic fracture detection.
Macrofractures
Figure 1. East Texas microfracture strikes (red rose diagram) from one depth compared with orientations of large fractures in well (*) and regionally in same formation (gray rose diagram). Dispersion in strike of large fractures reflects shifts in regional trends that are challenging to document by using sparse conventional fracture data alone.
An important target. Petroleum in fractured rocks is a growing target of exploration and development, making finding new methods to successfully predict, characterize, and simulate reservoir-scale natural fractures an important challenge facing the upstream oil and gas industry over Editor’s Note: The Geologic Column, which appears monthly in TLE, is (1) produced cooperatively by the SEG Interpretation Committee and the AAPG Geophysical Integration Committee and (2) coordinated by M. Ray Thomasson and Lee Lawyer. 704
THE LEADING EDGE
JULY 2000
Figure 2. Fracture-size distribution prediction, based on scaling patterns of microfractures (green squares). Predictions were confirmed in this case by using data on large fractures (red circles) from a core cut perpendicular to fractures. Data of this type can map fracture-intensity patterns from any core. JULY 2000
THE LEADING EDGE
0000
the next decade. According to recent estimates, the U.S. domestic target in such reservoirs could be hundreds to thousands of trillions of cubic feet; a multibillion-barrel domestic and international oil resource target also exists. Particularly in deep and nonconventional plays and in carbonate rocks, encountering fractures that provide effective flow pathways is key to economic hydrocarbon producibility. Apart from exploration, fractures are increasingly viewed as having a significant role in successful secondary recovery of resources, even in reservoirs that do not fit the production profile of a classic “fractured reservoir.” Clearly, natural fracture characterization is of increasing concern as the industry ventures into more complex, deeper, and unconventional reservoirs. Yet in many cases fractures are difficult or impossible to characterize adequately by using currently available technology. Why has effective fracture network characterization been so elusive? Sampling challenges. Characterizing fracture systems in the subsurface poses many challenges, but one of the most fundamental issues stems from the fact that although large fractures account for most fluid flow, wells typically intersect few large fractures. Fractures are simply hard to sample. This results from two characteristics of fractures and well. (1) Both are commonly nearly vertical, so well trajectories nearly parallel the fractures and (2) the diameter of a well is much smaller than the spacing between large fractures. As a consequence, cores and well logs—the most widely used fracture-detection methods—often yield little or no useful information about fractures. For example, sampling meaningful fracture abundance patterns or obtaining reliable data on shifts in fracture strike within a given unit or from one unit to the next requires representative data from each horizon of interest. Except in rare cases, incomplete sampling precludes this. Consequently, reservoirs that contain fractures have been intractable to describe and interpret effectively, posing serious challenges for exploration, risk assessment, development, and accurate reservoir simulation and management. Horizontal drilling has made this sampling and characterization issue more acute. The ability to drill in one horizon increases the number of potential targets, drilling directions, and hor706
THE LEADING EDGE
JULY 2000
Figure 3. Are fractures open or closed? This well comparison illustrates fracture quality predictions according to observations of mineral-sealed microfractures and diagenesis. Predictions of preserved fracture porosity (green or orange squares) match observed porosity patterns in large fractures (green or red circles). Although microfractures having apertures less than 0.01 mm are closed because they are below the emergent threshold, they can be used to identify proportions of postkinematic cement in the rock matrix (degradation). Thus, predictions of both closed and open macrofractures are possible. izontal well lengths, but which horizon and direction are best? Is the bed fracture prone? Are fractures open? How far is it advisable to drill? In most cases it is far too expensive to characterize fractures in each potential target with horizontal wells. Because a single horizontal lateral tests so few beds, information from horizontal wells may not provide all the answers needed for designing a horizontal-drilling program. The implication for seismic tech-
niques that rely on fracture detection without direct sampling is clear. Too often, despite acquisition of expensive log and core data, wells may fail to provide the critical direct evidence of fracture attributes that is the ground truth needed to calibrate. The development of all remote fracture-detection methods is impaired by this fundamental data limitation. Surrogates. We think that the predicament created by these sampling JULY 2000
THE LEADING EDGE
0000
challenges can now be overcome. Microfracture proxies for large fractures are the surrogates that can provide complete, reliable, bed-by-bed evidence of fracture attributes. Microfractures are invariably abundant (i.e., thousands in a typical thin section). By using new laboratory imaging devices, natural microfractures can be distinguished from those created by drilling and handling. For some attributes, such as fracture porosity, these surrogates currently provide qualitative data, as we explain below. For other attributes, such as orientation and size distribution, quantitative measurements are now possible. Because small samples, such as oriented sidewall cores, can be used for this type of fracture analysis, it is possible to collect relatively inexpensive fracture information from many more horizons within a given well than would be practical by conventional methods. Figures 1-3 show examples of the new surrogate-based assessment methods. These examples show successful predictions of fracture orientation, size distribution, and openness in wells in which special data sets were available to independently validate these predictions. Orientation. Fracture strike can frequently be obtained from large fractures in full-diameter oriented cores or well logs. But in many instances data are lacking from the zone of interest, or there is ambiguity about whether fractures visible on image logs are natural or drilling induced. Figure 1 shows how microfractures can complement conventional data sets: Microfracture strikes frequently match nearby large fractures. Because of their abundance, microfractures can identify regional and local shifts in fracture strike that may be vague even in the most complete conventional data sets. Scaling. Scaling is an application of fracture population statistics that allows quantitative predictions of macrofracture attributes. Predictions are based on microfracture observations. Consistent scaling patterns of fracture apertures suggest that large and small fractures are commonly merely different-size fractions of the same fracture population. In Figure 2, an example from west Texas, because the spatial frequency of fractures having apertures smaller than 1µ to nearly 1 cm follows a single relation, the microfractures pro708
THE LEADING EDGE
JULY 2000
a)
b)
c)
d)
Figure 4. Fracture patterns derived from geomechanical modeling, showing effects of differing subcritical crack index (n values) on resultant fracture spacing and clustering. Models show fracture traces (red lines) in map view and have identical boundary and initial conditions. vide accurate means of predicting the abundance of large fractures. Note that the regression line in this example is only fit to the microfractures, yet the large fractures are accurately predicted. This measure of fracture intensity can be thought of as an estimate of “average fracture spacing” (for fractures of a specified size), although averages are misleading where fractures are clustered in swarms. Used with care, rigorous scaling analysis is a potentially powerful tool for reservoir management (for example, by specifying optimal horizontal well length) and a link to fractured reservoir simulation and seismic response. Which fractures are conductive? Some methods based on surrogates do not rely solely on observation of microfractures. An example is identification of areas having fractures
that are open and capable of conducting fluid flow. Conventional data from core and logs rarely provide this information unless fractures intersect the well. How can open and closed fractures be distinguished where the only fractures in the sample are cement-filled microfractures? For many subsurface fracture systems, it is surprisingly easy to assess whether large fractures are open or not without sampling them. All that is required is a sample from the vicinity of the fracture (i.e., the same bed) and an appreciation of the two ways diagenesis closes fractures. Figure 3 illustrates a case in which sealed microfractures and diagenesis information from two cored wells were successfully used to predict fracture quality. Both wells have fulldiameter core and contain large fractures that have been variably filled with natural mineral cements, so this was a good opportunity to make a JULY 2000
THE LEADING EDGE
0000
“blind” test. By taking samples from areas of the core that lack large fractures, the test identified the overall difference between the wells (mostly open versus mostly closed fractures) and pinpointed shifts in fracture porosity preservation within each well. How does it work? We find in many rocks that there is a diagenesis event contemporaneous with fracturing, and large aperture fractures have mineral bridges and abundant preserved porosity. (In other words, the cements that are contemporaneous with fracturing do not fill in the large fractures.) Fractures with apertures smaller than a characteristic size are completely filled. The fracture aperture size of the transition from filled to partly open we term the emergent threshold. Empirical evidence from many formations suggests that the emergent threshold varies with cement mineralogy. For instance, quartz-rich rocks tend to have a transition to open fractures at a much smaller aperture than carbonate rocks. Figure 2 shows a frequency distribution of fracture apertures from core data. Only fractures having apertures larger than ~0.1 mm have significant preserved porosity; smaller fractures are mostly closed, and they probably closed soon after they formed. This structural and diagenetic phenomenon of scale-dependent porosity preservation above a certain threshold size implies that the main culprit that causes large fractures to close is later cement that precipitates after fractures cease opening. At that stage, fractures are merely another variety of pore that can be filled in, and predicting fracture quality boils down to identifying and measuring the volume of cement available to clog the fracture system, the post-fracture-opening or “postkinematic” cement. This postkinematic cement is the “bad cholesterol” of natural fracture systems, and where it is prevalent, flow in the fracture system is impeded. As we mentioned in the discussion of sampling, making meaningful observations of flow-scale natural fractures in drilling operations is challenging. Therefore, it is unlikely that many large fractures will be sampled that preserve the relationship between diagenesis and fracture opening. However, the pore space of the rock can serve as a proxy for the fracture relationships. To detect postkinematic cement, all that is required is evidence of the timing of fracture formation and 710
THE LEADING EDGE
JULY 2000
of contemporaneous cement precipitation. This information can usually be gleaned from microstructure observations and conventional petrographic measurements. Tiny microfractures generally lack postkinematic cement because they sealed shortly after formation, but the rock’s pore space is a surrogate, or proxy, for the porosity of the large fractures. Where the pore space contains a high proportion of cement that postdates fracture opening, the quality of large fractures is likely to have been damaged (degraded). In the example in Figure 3, this bad-cholesterol parameter (degradation) successfully identified the well having open, conductive natural fractures (the producer) from the well having closed, postkinematic cementfilled fractures (the dry hole). Note that the two wells are indistinguishable by using conventional plug porosity (and conventional porosity log) data. Among other applications, this is a useful approach for mapping fracture quality and identifying fracture “sweet spots,” for calibrating well logs, and for testing models or remotedetection techniques that predict fracture attributes. Predictions ahead of bit. Appreciation of the close relationship between fracturing and diagenetic processes can also yield more meaningful fracture predictions ahead of the bit. Geomechanical models of fracture growth are powerful tools for predicting fracture attributes, such as clustering and connectivity, which are challenging to measure for subsurface fracture arrays. Our models are most sensitive to assumptions about strain and the mechanical properties of the rock at the time of fracturing. Strain can often be estimated from structure maps (e.g., curvature analysis) or, if a subsurface rock sample is available, can be measured from microfracture-size-population data. Thus, the critical unknown parameters for modeling are paleorock properties, which undoubtedly reflect diagenesis. Where natural fractures grew slowly in the chemically reactive subsurface environment, the key rock property is called the subcritical crack index, a measure of crack velocity in rock. We have recently developed methods to measure this rock property in sedimentary rocks in the laboratory. The importance of subcritical crack growth is that it exerts fundamental control on fracture pattern geometry. For a given strain event, the subcritical crack index for a particular bed will control the resulting fracture pattern in
terms of fracture clustering as well as the ultimate fracture intensity (Figure 4). Thus, knowledge of the subcritical fracture parameters can be used to help accurately predict the spatial organization of a fracture pattern. This spatial organization will have a primary impact on fluid flow. Ultimately, we think that understanding subcritical crack processes and perfecting measurement techniques will enable us to predict which beds in a particular sequence are most likely to be fractured, and how those fractures are organized. Model results can be compared with empirical observations of spatial organization of fracture arrays. By using empirical links between rock properties and diagenetic state, it may even prove possible to accurately predict fracture attributes using models of diagenetic history. Calibrating seismic. We foresee fracture characterization using seismic methods becoming a tool of choice because it provides one of the few effective ways to sample an exploration target prior to drilling. But there are many hurdles. Some are fundamental and will be challenging ever to overcome (multiple open fracture sets in the same volume, resolution limitations). However, other limitations stem from inadequate calibration with subsurface fracture observations and unexploited opportunities for enhanced data processing. The most common approach to seismic prospecting for fractures is based on the Alford rotation and the concept of shear wave splitting, but other approaches, such as variation in reservoir reflection as a function of sourcereceiver offset and azimuth, have promise. Information about actual subsurface fracture patterns in a given test locality is required to verify these methods. In our view, sufficient evidence now indicates that microfractures and other surrogates can provide the type of bed-by-bed data on fracture attributes needed to calibrate the seismic response, including new S-wave data analysis technology. Open versus filled fractures, fracture strike, and relative fracture intensity are all parameters that are expected to affect seismic response (and fracture permeability). Yet these attributes are currently extremely challenging to document reliably from seismic data alone (can it be done at all?), and then only for large structures such as faults in areas of relatively simple geology. For fracJULY 2000
THE LEADING EDGE
0000
ture-swarm targets in complex stratigraphic and structural settings such as, for example, deeply buried Cretaceous and Tertiary natural gas plays in the western United States, interpretation of the results of remote methods will not be straightforward. To advance to the next generation of quantitative remote fracture detection in undrilled areas, a ground-truth-calibrated seismic method is needed. For fracture attributes, such sample-based fracture calibration would play a role similar to that of vertical seismic profiles for calibration of stratigraphic interpretation, except that calibration will be for fracture attributes deduced from seismic signals. Fracture orientation controls direction of Swave anisotropy, openness determines whether fractures will have a firstorder seismic signature, and size distribution controls magnitude of the seismic signature. Open fracture length may be related to the magnitude of velocity anisotropy for shear waves propagating through fractured rock. Because individual fractures in natural fracture systems vary across a tremendous range of sizes and follow power-law scaling, simple averages of fracture attributes are not useful. However, the scaling parameters of fracture populations can be incor-
0000
THE LEADING EDGE
JULY 2000
porated into quantitative relations of fracture attributes with seismic and fluid-flow parameters. As a by-product, relations between seismic attributes and fracture-related fluid flow can be established. For example, Swave anisotropy should be linearly proportional to fracture porosity, and the cube of anisotropy should be proportional to fracture permeability. These theoretical relationships offer the potential of before-the-bit prediction and mapping of fracture permeability. Testing of this hypothesis will constitute a component of the new research supported by DOE. This research could yield an empirical view of how seismic waves respond to open fractures, as well as conceptual understanding of the physical basis for the calibration. The future. This summary highlights a few new directions in fracture characterization that build on heightened appreciation of the close relationship between fracturing and evolving rock and fluid properties (diagenesis). Research into these relations has uncovered several potentially valuable surrogates for fracture attributes that can be measured even when large fractures have not been sampled. Many questions remain. A mea-
sure of this uncertainty is the outcome of the small company’s exploration well mentioned in the opening paragraph. Despite indications from “surrogates” that in several horizons large, open fractures probably exist in the vicinity of the well, the decision was made to abandon the well. The next several years should show whether the new directions in fracture characterization highlighted here are successful in helping to convert similar wells, and perhaps whole plays, into economic producers. Suggestions for further reading. “A method to detect natural fracture strike in sandstones” by Laubach (AAPG Bulletin, 1997). “Extent of power-law scaling for natural fractures in rock” by R. Marrett et al. (Geology, 1999). “Permeability, porosity, and shear-wave anisotropy from scaling of open fracture populations” by Marrett (in Fractured Reservoirs: Characterization and Modeling, Rocky Mountain Association of E Geologists, 1997). L Corresponding author: Steve.Laubach@ beg.utexas.edu
JULY 2000
THE LEADING EDGE
711