A review of shear wave splitting in the crack‐critical crust

September 8, 2003 15:14 Geophysical Journal International gji˙2037 Geophys. J. Int. (2003) 155, 221–240 A review of shear wave splitting in the crack-critical crust Stuart Crampin∗ and Sebastien Chastin Shear-Wave Analysis Group, Department of Geology and Geophysics, University of Edinburgh, West Mains, Edinburgh EH9 3JW. E-mails: scrampin@ed.ac.uk; schastin@glg.ed.ac.uk Accepted 2003 May 14. Received 2003 May 8; in original form 2002 September 23 SUMMARY Over the last 15 years, it has become established that crack-induced stress-aligned shear wave splitting, with azimuthal anisotropy, is an inherent characteristic of almost all rocks in the crust. This means that most in situ rocks are pervaded by fluid-saturated microcracks and consequently are highly compliant. The evolution of such stress-aligned fluid-saturated grainboundary cracks and pore throats in response to changing conditions can be calculated, in some cases with great accuracy, using anisotropic poro-elasticity (APE). APE is tightly constrained with no free parameters, yet dynamic modelling with APE currently matches a wide range of phenomena concerning anisotropy, stress, shear waves and cracks. In particular, APE has allowed the anisotropic response of a reservoir to injection to be calculated (predicted with hindsight), and the time and magnitude of an earthquake to be correctly stress-forecast. The reason for this calculability and predictability is that the microcracks in the crust are so closely spaced that they form critical systems. This crack-critical crust leads to a new style of geophysics that has profound implications for almost all aspects of pre-fracturing deformation of the crust and for solid-earth geophysics and geology. We review past, present and speculate about the future of shear wave splitting in the crackcritical crust. Shear wave splitting is seen to be a dynamic measure of the deformation of the rock mass. There is some good news and some bad news for conventional geophysics. Many accepted phenomena are no longer valid at high spatial and temporal resolution. A major effect is that the detailed crack geometry changes with time and varies from place to place in response to very small previously negligible changes. However, at least in some circumstances, the behaviour of the rock in the highly complex inhomogeneous Earth may be calculated and the response predicted, opening the way to possible control by feedback. The need is to devise ways to exploit these new opportunities in the crack-critical crust. Recent observations from the SMSITES Project at Húsavı́k in Northern Iceland, gathered while this review was being written, display the extraordinarily sensitivity of in situ rock to small changes at great distances. The effects are far too large to occur in a conventional elastic brittle crust, and their presence confirms the highly compliant nature of the crack-critical crust. Key words: anisotropy, compliance, crack-critical crust, fracture criticality, shear wave splitting. N O T AT I O N APE Band-1 Band-2 CW Anisotropic poro-elasticity: a model for the evolution of fluid-saturated microcracked rock (defined in Section 2.3.) Double-leafed solid angle of directions making angles 15◦ –45◦ either side of the average plane of stress-aligned vertical microcrack distributions (Section 2.3). Solid angle of directions making angles 15◦ to the average plane of stress-aligned vertical microcrack distributions (Section 2.3). Clockwise vibrations of the DOV (q.v.) source (Section 3.5). ∗ Also at: Edinburgh Anisotropy Project, British Geological Survey, Murchison House, West Mains Road, Edinburgh EH9 3LA.  C 2003 RAS 221 September 8, 2003 222 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin CCW DOV EDA RVSP SMS SOC SWVA SWTD TIH TIV VSP σV, σ H, σh sV , s H , sh Counter-clockwise vibrations of the DOV (q.v.) source (Section 3.5). Downhole orbital vibrator: a borehole shear wave source, previously known as the COV, the Conoco downhole orbital vibrator (Section 3.5). Extensive dilatancy anisotropy: the distributions of stress-aligned fluid-saturated microcracks in almost all in situ rocks in the crust (Section 2.3). Reverse vertical seismic profile: seismic exploration source/receiver geometry where borehole source is recorded by surface geophones (Section 2.2). Stress-monitoring site: a source/receiver geometry using crosshole seismics to monitor stress-induced changes to microcrack geometry (Section 3.5). Self-organized criticality: a type of self-organizing system verging on criticality (Section 4.1). Shear wave velocity anisotropy: percentage of shear wave velocity anisotropy (Section 2.3). Time delay between split shear waves (Section 2.3). Transversely isotropic anisotropic symmetry (hexagonal symmetry) with a horizontal axis of cylindrical symmetry (Section 2.2). Transversely isotropic anisotropic symmetry (hexagonal symmetry) with a vertical axis of cylindrical symmetry (Section 2.1). Vertical seismic profiles: seismic exploration source/receiver geometry where surface source is recorded by borehole geophones (Section 2.2). Principal axes of vertical stress, and maximum and minimum horizontal stress (Section 2.3). Principal axes of differential vertical stress, and maximum and minimum horizontal stress, where s V = (σ V −σ h ), s H = (σ H −σ h ) and s h = 0 (Section 2.3). 1 I N T RO D U C T I O N Every 10 years or so we write a review of shear wave splitting and seismic anisotropy (Bamford & Crampin 1977; Crampin et al. 1984a; Crampin & Lovell 1991). These last 10 years have been particularly rewarding, and it looks as if at last we are beginning to understand what shear wave splitting means. It appears to monitor the low-level deformation or evolution of fluid-saturated rocks under changing conditions. We can model, and even predict, the effects of known changes in the complicated heterogeneous crust. This is an important advance, but the reason for this predictability is even more important. The fluid-saturated cracks in the crust are so closely spaced that they form critical systems, verging on criticality and failure. This opens a new window into the behaviour of the crust, which appears to have implications for all solid-earth geophysics. Criticality is well recognized in other areas of geophysics and rock physics, for example Turcotte (1992) and Sornette (2000), but they do not identify the mechanism of fluid-saturated stress-aligned microcracks nor the diagnostic of shear wave splitting. The attitude of many geophysicists to shear wave splitting with azimuthal anisotropy, aligned with the stress field, is somewhat ambiguous. Clearly, recognized in many, perhaps most, rocks in the crust, shear wave splitting is observed, records are processed, modelled and interpreted as indicating propagation through some form of parallel vertical cracks or fractures. In the exploration industry for example, these effects are taken as indicating stress directions and the directions of hydraulic fractures so that the optimum directions of water floods can be chosen. However, the petrological significance of the phenomenon is then usually ignored and its implications for the nature of the rock mass forgotten. Since more detailed examination is not pressing, shear wave splitting, once identified as an isolated phenomenon, is accepted without question and with little interest in what it actually implies. This paper argues that shear wave splitting provides a window into a new understanding of rock deformation in a crust pervaded by fluid-saturated stressaligned grain-boundary cracks and pore throats. Observations of shear wave splitting indicate that microcracking is so extensive and pervasive that almost all rocks may be thought of as critical systems close to levels of fracture criticality, which, when the stress field is modified appropriately, lead to fracturing, faulting and earthquakes. It can be shown that the parameters that control shear wave splitting also control low-level (pre-fracturing) deformation and that the evolution of fluid-saturated cracks under changing conditions can be modelled by anisotropic poro-elasticity (APE). APE modelling matches a huge range of phenomena. On those few occasions when the observational data at depth is accurate the match of APE is exact. More often with less accurate data, the match can only be approximate. Nevertheless, the match means that shear wave splitting can be used to monitor, model and in appropriate circumstances predict the response of deep in situ rocks to changing conditions. This calculability and predictability has numerous potential applications ranging from industrial hydrocarbon exploration and production to a huge variety of natural hazard monitoring, management and mitigation. Shear wave splitting (seismic birefringence) is the most diagnostic, informative and easily observable evidence of azimuthal seismic anisotropy (Crampin 1981, 1985, 1994). Azimuthal anisotropy is now recognized as being characteristic of shear wave propagation through the fluid-saturated stress-aligned grain-boundary cracks and pore throats in almost all in situ sedimentary, igneous and metamorphic rocks (Crampin 1994, 1996; Winterstein 1996). This implies that most rocks in the crust are pervaded by distributions of highly compliant stress-aligned microcracks. Consequently, any interpretation of in situ rock, which does not allow for the presence of such easily deformed stress-aligned fluid-saturated microcracks is at best incomplete, possibly flawed and at worst significantly in error. The deformation or evolution of fluid-saturated cracks under changing conditions has been modelled with APE (Zatsepin & Crampin 1997). Since APE modelling broadly agrees with all relevant observations of cracks, stress, shear waves and shear wave splitting (there are no known contradictions, Crampin (1999a), we claim that we are making some progress towards understanding in situ rock deformation (Crampin & Zatsepin 1997). APE has been approximately calibrated in laboratory stress cells (Zatsepin & Crampin 1996; Crampin et al. 1997, 1999b). In situ rock, however, is remote and difficult to access and, currently, there is only one case study where the effects of APE have been effectively calibrated at depth (Angerer et al. 2000, 2002). These various results confirm that APE is at least a good approximation to pre-fracturing  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust rock deformation and suggest that the deformation of in situ rock can be calculated and, in appropriate circumstances, predicted (Crampin et al. 1999a; Angerer et al. 2000, 2002). These various ideas have stimulated the design of stress-monitoring sites (SMSs) using stateof-the-art borehole instrumentation to monitor shear wave splitting between boreholes in a convenient stress-oriented configuration (Crampin 2001). SMSs are designed to recognize the build up of stress before earthquakes and volcanic eruptions and estimate (stress forecast) the time and magnitude of impending large earthquakes. The implications of why the response of the immensely complicated heterogeneous crust, below 500–1000 m, say, should be calculable, monitorable with shear wave splitting, and in some cases predictable, is remarkable. The fluid-saturated cracks in the crust are so closely spaced and so near fracture criticality at the percolation threshold that the cracks behave as critical systems (Crampin 1998, 1999a; Crampin & Chastin 2001). Fracture criticality is the level of cracking when shear strength is lost and the rock necessarily fractures (Crampin 1994; Crampin & Zatsepin 1997). Note that a similar situation probably applies to the upper mantle, where the cracks would be films of liquid melt in grainboundary cracks (Crampin 1995, 2003; Blackman & Kendall 1997; Mainprice 1997). Critical systems of cracks lead to a new style of geophysics, which has profound implications for the detailed high-resolution properties of in situ rock (Crampin 1999a). However, if current levels of detail and resolution are satisfactory, then the crack-critical nature of the crust can be ignored, but if greater detail and higher resolution are required, then the effects of the crack-critical crust influence almost every observation and measurement. In particular, there are likely to be profound effects on reservoir characterization and oil recovery (Crampin 1999b; Crampin & Chastin 2001), and on stress forecasting earthquakes and other geological hazards. The future of shear wave splitting and seismic anisotropy is a rapidly expanding and developing field with many implications and applications. Note that this is perforce somewhat of a personalized account because there are few other papers relevant to the development we are presenting. The review is necessary to set the parameters of a complicated multifaceted rapidly evolving development. 2 S H E A R - WAV E S P L I T T I N G Shear wave splitting (seismic birefringence), where shear waves split into typically two approximately orthogonal fixed polarizations with different velocities, is characteristic of propagation in media with some form of elastic anisotropy (Crampin 1981). Such splitting writes easily recognizable signatures into the three-component particle motion of shear wave arrivals (visible in particle motion diagrams or hodograms), so that shear wave splitting is the key diagnostic phenomenon for investigating seismic anisotropy. Note that shear wave splitting is controlled by small (typically less than 5 per cent) differences in the velocities of the two polarized shear waves. The power and sensitivity of shear wave splitting is that by rotating seismograms into the preferred polarizations, or by plotting polarization diagrams or hodograms, the time delay between the two split shear waves can usually be measured with much greater accuracy than most second-order measurements. Thus shear wave splitting opens a window into the analysis of small variations of the rock mass that probably no other geophysical measurement can match, and is the underlying reason why shear wave splitting is the key to monitoring the crack-critical crust discussed in Section 4, below.  C 2003 RAS, GJI, 155, 221–240 223 2.1 TIV anisotropy will not be discussed TIV anisotropy is transverse isotropy (hexagonal anisotropic symmetry) with a vertical axis of symmetry where the shear waves split into strictly SH and SV polarizations (Crampin 1986). Such symmetry is characteristic of finely layered horizontal sedimentary strata due to the interactions of reflections and transmissions through thin layers. It is also characteristic of many shales, clays and mudstones, where the anisotropy is caused by horizontal intergranular platelets of mica and other minerals. Such TIV anisotropy has vertical and horizontal move-out velocities that may differ by 30 per cent and cause severe problems in migration and in establishing well-ties in industrial exploration seismics. These technical problems can usually be accommodated by processing (Tsvankin 2001), although, like almost all examples of azimuthal anisotropy, the behaviour in in situ rock has only recently being effectively calibrated (Winterstein & De 2001). Thus, TIV anisotropy is comparatively well understood and probably has few surprises left. However, almost the only real geophysical information it carries is that the rock was laid down in some sort of sedimentary process in some sort of fluid, and that gravity is vertical. We shall refer to TIV anisotropy only in passing. 2.2 Shear wave splitting with azimuthal anisotropy This paper refers specifically to shear wave splitting varying azimuthally, sometimes misleadingly called TIH anisotropy: transverse isotropy (hexagonal anisotropic symmetry) with a horizontal axis of symmetry. It is misleading because strict TIH (caused by wholly parallel vertical cracks) is exceedingly uncommon. Almost all distributions involve significant elements of non-parallel cracks, although TIH symmetry is frequently a good first approximation. Typically, below a critical depth of 500–1000 m, the polarizations of the faster split shear wave are approximately parallel (within 20◦ ) to the direction of maximum horizontal stress. Such near-parallel polarizations are observed routinely above small earthquakes and in a wide range of exploration configurations: reflection surveys, vertical seismic profiles (VSPs), reverse vertical seismic profiles (RVSPs), cross-hole seismics, etc. Table 1 lists evidence supporting stress-aligned fluid-saturated grain-boundary cracks and pore throats as the source of azimuthal anisotropy. Note that in general, the vertical stress, σ V , is zero at the free surface but increases with depth, and a critical depth is reached when σ V equals the minimum horizontal stress, σ h . Below this depth, cracks open normal to the minimum stress, which is typically horizontal so that the cracks are usually vertical striking approximately parallel to the maximum horizontal stress, σ H (Crampin 1990) and gives the characteristic stress-parallel shear wave polarizations. Above this depth, crack distributions are controlled by stress-release and lithologic phenomena, and may be very disturbed. Below this critical depth, the polarization of the faster shear wave is parallel to the direction of maximum horizontal stress in a broad band across the centre of the shear wave window. Such approximately parallel polarizations are the characteristic, most diagnostic, feature of observations of shear waves in most types of rock (Crampin 1994). (The shear wave window is the cone of ray paths with angles of incidence to the free surface less than 35◦ –45◦ (the actual angle depending on details of near-surface structure). Outside this window, shear waves are seriously disturbed and are contaminated by S-to-P conversions. Only within the window can be the waveforms of the incident shear wave be directly observed at a free surface (Booth & Crampin 1985).) The polarizations are September 8, 2003 224 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin Table 1. Summary of the basis for interpretating shear wave splitting with azimuthal anisotropy as the result of propagation through stress-aligned fluid-saturated grain-boundary cracks and pore throats. Observations of SWVA∗ Interpretation (1) Fast shear wave polarizations within 20◦ of direction of maximum horizontal stress for nearly vertical propagation in most† crustal rocks below 500–1000 m depth [1] (a) Only transverse isotropy (hexagonal symmetry) with horizontal symmetry axis (TIH anisotropy), or a minor perturbation thereof, has this property [2] (b) The only common phenomenon possessing such symmetry is stress-aligned fluid-saturated cracks (specifically microcracks) [2] (2) Observed with similar parameters (minimum and maximum SWVA of ∼1.5 per cent and ∼4.5 per cent, respectively) in almost all sedimentary, igneous, and metamorphic rocks [1] (a) Parameters of anisotropy close to APE§ theory in model of crack evolution (minimum SWVA of ∼1 per cent to maximum at the percolation threshold of 5.5 per cent [3]) (b) Only crack distributions consistent with all observations are fluid-saturated grain-boundary cracks and pore throats [4] (3) Temporal variations in time delays indicate increasing crack aspect ratios as stress builds up before earthquakes [4, 5, 6]. Temporal variations indicate both small cracks and fluid-saturations supporting interpretation 2b, above (4) Shear wave splitting occasionally seen in heavily fractured beds [7, 8]. Large aligned fractures also cause shear wave splitting [7, 8, 9] (5) A large range (15+) of different static and temporal observations of cracks, stress, and shear waves can be matched by APE§ modelling. It is fundamental to the success of APE modelling of the evolution of a crack rock mass (see Table 2, below) that the modelled cracks are highly compliant and this necessitates distributions of micro as opposed to macrocracks. ∗ Shear wave velocity anisotropy; only common exceptions are shales, clays, mudstones, which may have several tens of per cent of lithologically induced transverse isotropy with a vertical symmetry (TIV anisotropy), and oolites and coccoliths which have a highly constrained microstructure and may have very little SWVA. § Model of anisotropic poro-elasticity [10]. Percolation threshold is the theoretical crack density at which cracking is so extensive that statistically through-going fractures exist [3]. [1] Crampin (1994); [2] Crampin (1981); Wild & Crampin (1991); [3] Crampin & Zatsepin (1997); [4] Crampin (1999a); [5] Crampin et al. (1999a); [6] Volti & Crampin (2003a,b); [7] Mueller (1991, 1992); [8] Meadows & Winterstein (1994); [9] Li et al. (1993); [10] Zatsepin & Crampin (1997). † The only approximately parallel because microcracks have a range of orientations with crack normals averaged about the direction of minimum compressional stress so that the anisotropic symmetry is only approximately TIH (Crampin & Zatsepin 1997). See also the discussion of the crack-critical crust in Section 4, below. 10 years ago, Crampin & Lovell (1991) published a review of the first decade since stress-aligned shear wave splitting was first positively identified in records within the shear wave window above small earthquakes (Crampin et al. 1980). Crampin & Lovell tried to answer three questions concerning shear wave splitting: what does it mean, what use can we make of it and what should we do next? Substantial progress has been made in all three questions, reviewed below, but the questions are still relevant. We believe we are just beginning to be able to answer them, and the answers are exciting, although they are not always the answers that were originally expected. 2.3 What does shear wave splitting mean? Crampin (1994, 1996) and Winterstein (1996) reviewed all the then available (∼80) examples of shear wave splitting recorded above small earthquakes and in exploration surveys, which were principally VSPs. With the exception of the TIV anisotropy mentioned above, almost all sedimentary, igneous, and metamorphic rocks below 500–1000 m display azimuthal stress-aligned shear wave splitting. There is a minimum shear wave velocity anisotropy (SWVA) of about 1.5 per cent and a maximum in ordinary unspecified rock of about 4.5 per cent. Higher values of SWVA are found nearer the surface and in heavily fractured rocks and areas of high heat flow (Crampin 1994). The polarizations of the faster waves are subparallel to the direction of maximum horizontal stress. Since the only phenomenon common to all rocks where such symmetry is observed is stressaligned cracks, this strongly suggests that the shear wave splitting observed in most rocks is caused by stress-aligned microcracks (Table 1)—the extensive dilatancy anisotropy (EDA) of Crampin et al. (1984b). Fig. 1 shows the classic schematic illustration of shear wave splitting through distributions of microcracks aligned perpendicular to the direction of minimum horizontal stress. For nearly vertical propagation, the polarization of the faster split shear waves are parallel to the strike of the cracks, which is approximately in the direction of maximum horizontal stress. Since, the percentage SWVA of parallel cracks is approximately equal to 100 times the crack density (Crampin 1994), the values of observed SWVA indicate crack densities of 0.015 ≤ ε ≤ 0.045 in ostensibly intact rocks, where the crack density is ε = Na3 /v, and N is the number of cracks of radius a in volume v. Since grains in most  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust Figure 1. Schematic illustration of shear wave splitting in distributions of stress-aligned fluid-saturated parallel vertical cracks aligned normal to the direction of minimum horizontal stress, where for nearly vertical propagation the polarization of the faster split shear wave is parallel to the strike of the cracks, parallel to the direction of maximum horizontal stress. Such parallel vertical crack orientations are typically found below the critical depth, usually between 500 and 1000 m, where the vertical stress is greater than the minimum horizontal stress. 225 rocks usually have a comparatively restricted range of dimensions, grain-boundary cracks have a similarly restricted range. Assuming uniform equally sized penny-shaped cracks, Fig. 2 gives a schematic illustration of distributions of cracks that have the observed range of SWVA. Crampin (1994) suggested there is a fracture criticality limit, 0.045 ≤ ε ≤ 0.1, separating ostensibly intact rock (ε ≤ 0.045) from rock that is disaggregating at the free surface (0.1 ≤ ε). The fracture criticality limit in stressed fluid-saturated rock is now shown to be associated with the percolation threshold at about ε ≈ 0.055 (Crampin & Zatsepin 1997). This indicates a very narrow range of crack size. There is a difference in radius of less than a factor of 2 (∼1.8) between cracks in ostensibly intact rocks having the minimum crack density usually observed below 500–1000 m (1.5–4.5 per cent SWVA) and rocks that are desegregating at the free surface (≥10 per cent). Such crack distributions in intact rocks typically lead to normalized time delays of less than about 8–10 ms km−1 in most in situ rocks. Since the grains in any particular rock usually have similar sizes, the dimensions of grain-boundary cracks and pore throats will also tend to have a narrow range. This means that the uniform microcrack distributions in Fig. 2 are not too unrealistic. Note that the mathematical derivation of SWVA is sensitive to the ray path through the cracks, the velocities and Poisson’s ratio of the rock matrix, and to the properties of the pore fluid, as well as to the crack density (Crampin 1993). Consequently, SWVA is a variable quantity and does not refer to a fixed geometry of cracks but depends on matrix and pore-fluid properties. For example, the value of SWVA is observed to be higher in areas of high heat flow (Crampin 1994), although the reasons for this are not fully understood. The variation of SWVA in three dimensions is also sensitive to the crack aspect ratio and hence to principal axes of stress and the crack geometry. Fluid-saturated microcracks are highly compliant, and the evolution of fluid-saturated grain-boundary cracks and pore throats under changing conditions can be modelled by APE (Crampin & Zatsepin 1997; Zatsepin & Crampin 1997). The driving mechanism for evolution (deformation) is fluid migration by flow or diffusion along pressure gradients between cracks at different orientations to the stress field. APE models fully 3-D distributions of cracks. Fig. 3 is a schematic but numerically accurate illustration of the effect of the APE deformation mechanism on random distributions of vertical cracks as the maximum horizontal stress is marginally increased. Hexagons have isotropic elastic symmetry so the two (solid) hexagons of cracks, at zero differential stress (top left), are Figure 2. Schematic interpretation of observed percentages of SWVA below the critical depth, interpreted as uniform (dimensionless) distributions of equally sized parallel penny-shaped parallel cracks with the given percentage of SWVA, where ε is crack density (approximately equal to a hundredth of the percentage anisotropy) and a is crack radius. (After Crampin 1994.)  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 226 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin Figure 3. Schematic illustration of the evolution of crack aspect ratios in an initially random distribution of vertical cracks (solid lines) for four values of increasing maximum horizontal differential stress, normalized to the critical value at which cracks first begin to close. Pore-fluid mass is preserved and aspect ratios are correct for a porosity of φ = 5 per cent. (After Crampin & Zatsepin 1995.) a small selection of randomly oriented vertical cracks without any horizontal anisotropy. We define differential stress, s i , as the stress components, σ i , less the minimum stress which in this case, below the critical depth, is σ h , hence s V , s H , s h = σ V −σ h , σ H −σ h , 0, respectively. As differential stress increases, stress-induced pressure gradients cause fluid to move, by flow or diffusion, between adjacent cracks at different orientations to the stress field (top right), but until the stress is sufficient to close normally oriented cracks, there is negligible anisotropy. At the critical stress (normalized to s H = 1 in the figure), cracks begin to close (bottom left) and the level of SWVA jumps from zero to a value close to the 1.5 per cent SWVA minimum observed in the crust (Crampin 1994). As differential stress continues to increase, the cracks increase in aspect ratio and begin to line up. It can be shown that the percolation threshold when the mechanism in Fig. 3 leads to through-going cracks at about 5.5 per cent SWVA (Crampin & Zatsepin 1997), which can be identified with the fracture criticality of Crampin (1994). (Since stress-aligned shear wave splitting is almost always observed in in situ rocks, this implies that the critical horizontal stress is almost always exceeded in rocks in the crust.) APE theory shows that the parameters that control low-level deformation before fracturing occurs are exactly those that control shear wave splitting (Crampin & Zatsepin 1997). Note that the principal effects of marginal changes of stress and pore-fluid pressure, below the level at which fracturing occurs, are modifications to crack aspect ratios. This has recently been confirmed by very different theoretical techniques (Hudson 2000). APE theory is highly constrained with no free parameters yet matches a wide range of phenomena, some of which are listed in Table 2 (Crampin 1999a, 2000). The match of observations to APE modelling in Table 2 can be very accurate, although typically it is difficult to obtain accurate measures of crack behaviour at depth because the high temperatures and pressures make in situ rock essential inaccessible. Note that the crack density in Fig. 3 is lower at s H = 3 than at s H = 0, as increasing stress tends to close cracks and hence lessen the crack density and the associated SWVA. The only unambiguous change as stress increases is the increase in the average crack aspect ratio (as noted above). The major observational effects of changing aspect ratios are comparatively subtle changes in a range of directions in the 3-D variation of time delays between the split shear waves. The effect of increasing the aspect ratio of parallel vertical cracks on shear wave splitting is to increase the average time delay along ray paths in the double-leafed solid angle of directions (referred to as Band-1) making angles 15◦ –45◦ to the plane of the cracks (Crampin 1999a). Time delays in Band-2 (directions within 15◦ of the crack plane) are sensitive only to crack density. Table 2 lists matches of APE modelling in exploration seismology and behaviour before earthquakes (see the next section), as well as with laboratory experiments in stress cells. It is this wide range of agreement of APE modelling with observations that confirms that the shear wave splitting with azimuthal variations observed in the crust is typically caused by stress-aligned fluid-saturated microcracks. Large cracks would be stiff and much less compliant. Note that shear wave splitting is controlled by the crack/pore throat geometry, and is largely independent of the noncrack porosity (equant porosity) (Crampin & Zatsepin 1997). The large number of dynamic effects in Table 2 confirm that fluidsaturated cracks make the rock mass highly compliant. If internal conditions in the rock mass are changed in any way, the crack geometry responds and the response can be monitored directly with shear wave splitting, and modelled or predicted by APE. Certainly large fractures, if they are aligned, will cause significant shear wave splitting. However, the only confirmed observations, to our knowledge, are those of Mueller (1991, 1992) and Li et al. (1993) in the Austin Chalk, and Angerer et al. (2000, 2002) in Vacuum Field, New Mexico (who models a structure of large fixed fractures and compliant microcracks). Large cracks are comparatively stiff, and the reports in Table 2 suggest, and the results of Angerer et al. (2000, 2002) confirm, that the dominant cause of shear wave splitting is propagation through the compliant fluid-saturated grain-boundary cracks and pore throats (microcracks) that pervade most rocks. The phenomena in Table 2 have been discussed in more detail elsewhere (Crampin 1997, 1999a,b, 2000). Table 1 summarizes the evidence for interpreting the cause of azimuthal shear wave splitting as propagation through fluid-saturated microcracks, specifically grain-boundary cracks in low permeability rocks and pore throats in porous rocks. The evidence is overwhelming that the source is microcracks, particularly the extraordinary match of APE modelling to observations in Table 2 and the modelling of Angerer et al. (2000, 2002), and yet the evidence is almost wholly indirect. In situ cracks are remote, essentially inaccessible, and are nearly transparent to everything except seismic shear waves (Crampin 1981, 1999a). Note that recent observations reported in Section 6, below, confirm the extraordinary compliance of the fluidsaturated cracked rock mass and sensitivity of shear waves to nearly negligible changes in stress.  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust 227 Table 2. Match of APE* modelling to observations (Crampin 1999a, updated). Ref. (Obs.) Ref. (APE) [1] [1] [1] [1] [1] [2] [2] [2] [2] [1] [3] [4] [5, 6] [2, 5, 6] Temporal changes in SWVA during production procedures (8) Changes before and after pumping tests (9) Changes before and after high pressure CO2 flood in carbonate reservoir [7] [6, 8] [6] Temporal changes in SWTD¶ before earthquakes (10) Variations of time delays before earthquakes (with hindsight) (11) Successful forecast of time and magnitude of an M = 5 earthquake in SW Iceland [9, 10] [9] [2] [11] § [12] [14] [13] [14] [15] [16] Static effects Field observations of SWVA† (below 500–1000 m depth) (1) SWVA in all rocks independent of porosity and geology (2) Minimum SWVA in intact rock: observed ≈1.5 per cent; APE modelled ≈ 1.0 per cent (3) Maximum SWVA in intact rock: observed ≈4.5 per cent; APE modelled ≈ 5.5 per cent (4) Narrow range of crack density: 0.025 ≤ ε ≤ 0.045 (5) Proximity of fracture criticality (at percolation threshold) ≈5.5 per cent Other field observations (6) Fracture criticality limit specifies crack distributions with a range of dimensions of about nine orders of magnitude (7) π/2 shear wave polarization changes (90◦ flips) in overpressurized reservoirs Dynamic effects Temporal changes in SWTD before volcanic eruption (12) Variations in SWTD for some 5 months before 1996 September 30; Vatnajökull eruption, Iceland, at distances of: 230 km WSW; 170 km SW; and 240 km, N Variations of shear waves in laboratory experiments (13) Variations of SWVA and permeability in uniaxial stress cell (14) Variations of (isotropic) shear wave velocities to changes in confining pressure and pore-fluid pressure for oil-, water-, and gas- (dry) saturations in stress cells of sandstone cores (15) Variations of velocity and attenuation from sonic (transducers) to seismic (resonant bar) frequencies § § anisotropic poro-elasticity; † SWVA, shear wave velocity-anisotropy; § Effects compatible with APE; ¶ SWTD, shear wave time delays; [1] Crampin (1994); [2] Crampin & Zatsepin (1997); [3] Heffer & Bevan (1990); [4] Crampin (1997, 1999a); [5] Crampin et al. (1996); [6] Angerer et al. (2000, 2002); [7] Crampin & Booth (1989); [8] Davis et al. (1997); [9] Crampin et al. (1999a); [10] Booth et al. (1990), Crampin et al. (1990, 1991), Liu et al. (1997), Gao et al. (1998); [11] Volti & Crampin (2003a,b); [12] King et al. (1994); [13] Zatsepin & Crampin (1996); [14] Crampin et al. (1997, 1999b); [15] Sothcott et al. (2000a,b); [16] Chapman et al. (1998, 2000). ∗ APE, 3 S H E A R - WAV E S P L I T T I N G — C U R R E N T POSITION Here we attempt to answer the second question of Crampin & Lovell (1991) ‘what use can we make of it?’. The polarizations of the faster split shear wave are generally assumed to indicate the alignment of cracks and fractures. As noted above, the polarizations are frequently claimed, without proof, to indicate the alignment of large fractures permitting easy oil flow. Unambiguous demonstrations of anisotropy due to large fractures are few (Mueller (1991, 1992) and Li et al. (1993) in the Austin Chalk, Texas, and Davis et al. (1997), Duranti et al. (2000) and Angerer et al. (2000, 2002) in Vacuum Field, New Mexico; Angerer et al. (2000, 2002) analysed shear wave splitting in the presence of both microcracks and large fractures). There is also little direct evidence in other literature that the preferred directions of flow in oil fields are due to large fractures rather than the preferred flow directions through highly permeable stress-aligned microcracks.  C 2003 RAS, GJI, 155, 221–240 The last 2 years have seen several applications of shear wave splitting. 3.1 Predicting the response of a reservoir to known changes during recovery processes Angerer et al. (2000, 2002) analysed, interpreted and modelled with APE (in effect predicted with hindsight) the response of the Vacuum Oil Field, New Mexico, to two CO2 injections resulting in pressure increases of 1000 psi (6.9 MPa) and 200psi (1.38 MPa), respectively. These were Phases VI and VII of the Reservoir Characterization Project (RCP) of Colorado School of Mines (Davis et al. 1997; Duranti et al. 2000). The Phase VI injection of 17 MPa was an overpressure. The part of the Vacuum Oil Field accessed had a flat layer-cake structure. Extensive 4-D 3-C reflection record sections suggested that changes in shear wave splitting were the most diagnostic effects of both CO2 injection pressures (Davis et al. 1997; Duranti September 8, 2003 228 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin Figure 4. (a) Pre-injection waveforms of a multicomponent nearly vertical reflection survey near the centre of the Vacuum Field, New Mexico, carbonate reservoir (Davis et al. 1997). S1, S2 and P are record sections with mutually orthogonal polarizations, where the horizontals S1, and S2, have been rotated into the split shear wave arrivals parallel (S1) and perpendicular (S2) to the direction of maximum horizontal stress, respectively. The left-hand five traces are observed waveforms at neighbouring receivers 17 m apart, and the right-hand three traces are synthetic seismograms modelled by APE to match the shear wave arrivals. Top and bottom of injection zone for shear waves (established by extensive analysis at Colorado School of Mines) are marked by arrows with time delays in ms km−1 . (b) Post-injection waveforms after a high pressure CO2 injection. Again, the left-hand traces are observations and right-hand traces are synthetic seismograms modelled by APE with structure from (a) and an injection pressure of 2500 psi. (After Angerer et al. 2000, 2002). et al. 2000). Angerer et al. (2000, 2002) processed the Phase VI records and determined an initial structure of large fixed faults with an internal microcrack structure with 2 per cent SWVA. Synthetic seismograms, calculated by ANISEIS (Taylor 2000), through the initial macrocracked model reproduced the shear wave splitting arrivals in Fig. 4(a) from reflections from the top and bottom of the San Andres Formation target zone, where the CO2 was injected. The three sections in Figs 4(a) and (b) are the horizontal S1 and S2 polarizations parallel and perpendicular, respectively, to the maximum horizontal stress, and the vertical P-wave polarization. The time delays between the top and bottom reflections of the target zone in Fig. 4(a) are 176 ms for S1 and 178 ms for S2, indicating that S1, polarized parallel to the maximum horizontal compressional stress, is the faster split shear wave in the target zone. Note that  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust other investigations at RCP identified the reflections from the target zone, where the numerous other arrivals are various reflections and reverberations. The five traces in the left-hand side of the record sections in Fig. 4(b) show significant changes following the high-pressure CO2 injection. The three modelled traces to the right are calculated by inserting the specific injection pressure into the APE model of the initial fracture model in Fig. 4(a). The match of observed to modelled arrivals is almost exact so that APE has correctly calculated (in effect predicted) the response of the cracked rock mass to the injection pressure. (Note that the changes to the P-wave arrivals also match, but P-wave arrivals contain little information and are easy to match (Crampin 1985).) APE shows that the difference in the shear wave response is largely caused by the increased aspect ratio as the microcracks expand with the increased pore-fluid pressure (Angerer et al. 2000, 2002). A similarly satisfactory match was also found for the lowerpressure Phase VII CO2 injection (not shown), using the same initial cracked model and inserting the lower pressure in APE. The Phase VII injection was in a similar well, 25 m from the Phase VI injection well, with varying conditions and a lower injection pressure. This means that APE has in effect predicted the response of the microcrack structure to two very different injection pressures. This is the best in situ calibration of APE to date. Fig. 4 shows another characteristic of shear wave splitting. In contrast to Fig. 4(a), where S1 is the faster wave through the San Andres Formation, the orthogonal polarization S2 in Fig. 4(b), is faster with a time delay of 184 ms as opposed to 204 ms for S1. This means that the polarizations of the faster split and slower shear waves have interchanged. Angerer names this phenomenon a ‘90◦ flip’. Such 90◦ flips, together with large values of SWVA, are characteristic of shear wave splitting in overpressurized rocks. In overpressurized hydrocarbon reservoirs, 90◦ flips have previously been observed in the Caucasus oil field (Crampin et al. 1996; Slater 1997). Such 90◦ flips are also characteristic of shear wave splitting immediately above active faults and have been observed at two places on the San Andreas Fault (Crampin et al. 1990, 1991; Liu et al. 1997) and on the Húsavı́k-Flatey Fault (HFF) in Iceland (Section 6.2, below). High pore-fluid pressures are necessary to relieve frictional forces and explain fault slip with no observable frictional heat flow (Sibson 1990; Gudmundsson 1999). Thus, Angerer et al. (2000, 2002) show that, in some circumstances at least, it is possible to predict the response of a reservoir to quantifiable oil-field operations. Since the response can be monitored by analysing shear wave splitting, this means that the response of the reservoir to a known operation, such as a fluid injection, can, in some circumstances, be controlled by adjusting the injection pressure to optimize the rock mass response. This means that the response can be monitored and controlled by feedback long before production rates would have indicated whether the procedures were satisfactory or not. 3.2 Converted phases One of the difficulties of analysing shear waves is the expense of using shear wave vibrator sources onshore and the lack of an efficient shear wave source offshore. It has long been recognized that P–S converted phases are in many cases a cheaper alternative technique for generating shear waves (Garotta & Marechal 1987). Conversions at depth would have the advantage that shear waves would initially have the higher frequencies associated with P waves. Thomsen (1999) calls such converted waves C waves and presents a strong  C 2003 RAS, GJI, 155, 221–240 229 case for their use in studying shear wave anisotropy. The major difficulty as Thomsen recognizes is that: (1) results are highly directiondependent. If the sagittal plane is one of the preferred split shear wave polarizations, the other split shear wave will not be generated, with possibly misleading results; and (2) the fixed P-wave polarizations almost never give the opportunity of giving two shear wave source orientations. This means that C waves cannot give the supporting evidence of two shear wave splitting measurements along the same ray path to provide confirmation of the interpretation. Note that a further cheaper optimum technique for monitoring shear wave anisotropy is suggested in Section 5.1, below. 3.3 Stress forecasting the times and magnitudes of earthquakes It has long been suspected (Crampin 1978; Crampin et al. 1984b) that changes in shear wave splitting would monitor changes of microcrack geometry caused by changes of stress before earthquakes. Peacock et al. (1988) observed comparatively consistent changes in time delays in Band-1 of the shear wave window above small earthquakes, which they suggested indicated stress-induced changes in crack aspect ratios. The hypothesis that the major effect of increasing stress on crack distributions was indeed to increase aspect ratios was confirmed by APE some 10 years later (Crampin & Zatsepin 1997; Zatsepin & Crampin 1997). Crampin et al. (1990, 1991) showed that the observations of Peacock et al. (1988) appeared to be caused by changes before the M s = 6 North Palm Springs earthquake. This technique was difficult to confirm elsewhere as suitably persistent swarms of small earthquakes to use as a source of shear waves are very uncommon. Until 1997, changes in shear wave splitting in Band-1 directions before earthquakes had been observed only on four occasions worldwide: before two earthquakes on the San Andreas Fault in California (Peacock et al. 1988; Crampin et al. 1990, 1991; Liu et al. 1997); one in Arkansas (Booth et al. 1990); and one on Hainan Island, China (Gao et al. 1998). The breakthrough came when shear wave splitting was monitored in the European Commission funded PRENLAB-1 and PRENLAB2 Projects (Stefánsson et al. 2000) using Iceland as a natural laboratory for earthquake prediction studies. Iceland is above a highly seismic offset of the Mid-Atlantic Ridge. Increases in time delays in Band-1 shear wave splitting were observed routinely before earthquakes and before some volcanic eruptions in SW Iceland (Volti & Crampin 2003a,b). Observations before both earthquakes and volcanoes confirm that changes of shear wave splitting are due to the changes of stress along the ray path rather than associated with changes in the immediate source zone of earthquakes. The underlying assumption is that the rock mass is weak to shear stress. Fig. 2 indicates that all in situ crack distributions verge on fracture criticality. This means that the stress released by a large earthquake necessarily accumulates over an enormous volume of rock, probably tens to hundreds of millions of km3 before an M = 8 earthquake. In Iceland to date, changes in shear wave splitting have been observed before earthquakes with magnitudes between M = 3.5 and 5.6 at distances of 14 and 43 km, respectively (Volti & Crampin 2003a,b). (Note that Icelandic magnitude M is approximately equivalent to body-wave magnitude, m b .) The time and magnitude of an M = 5 earthquake was successfully stress forecast from the data in Fig. 5 (Crampin et al. 1999a). This forecast assumed a more or less constant rate of deformation from the movement of the Mid-Atlantic Ridge. If stress accumulates over a small volume, the rate of accumulation will be fast for a comparatively short period of time before fracture criticality is reached and the final earthquake September 8, 2003 230 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin Figure 5. Shear wave splitting time delays for 1996 January 1 to 1999 December 31, at Station BJA, SW Iceland. The middle and top diagrams show the variation of time delays with time for ray paths in Band-1 and Band-2 directions. The time delays in ms are normalized to a 1 km path-length. The vertical lines through the time delay points are error bars based on errors in hypocentral distance. The irregular lines are nine-point moving averages. The straight lines in Band-1 are least-square estimates beginning just before minima of the nine-point average and ending when a larger earthquake or an eruption occurs. The arrows indicate the times of these larger events with magnitudes and epicentral distances indicated. The bottom diagram shows the magnitudes of earthquakes greater than M = 2 within 20 km of the recording station. (After Volti & Crampin 2003a,b.) would be comparatively small. Whereas if the stress accumulates over a larger volume, the rate would be slower and over a longer period of time, but the resulting earthquake would be larger. With this assumption, if relationships between the rate and duration of increase with magnitude can be estimated from previous earthquakes, as they were in Iceland (Crampin et al. 1999a), the time and magnitude of an impending earthquake can be estimated from the time that the increase reaches levels of fracture criticality. We call this process stress forecasting. The magnitude to duration relationship is approximately linear over the small range of magnitudes for which we have data in Iceland. These effects were recognized and the time and magnitude of an M = 5 earthquake were successfully stress forecast (Crampin et al. 1999a). The location of the impending earthquake cannot be estimated directly from shear wave splitting, where effects are seen at over 40 km from an M = 5 earthquake. However, if it is known that a large earthquake is approaching, other precursory activity can be interpreted correctly. Ragnar Stefánsson at the Icelandic Meteorological Office correctly predicted the location of the stress forecast event from the continued seismicity following a previous earthquake forecast (Crampin et al. 1999a). Note that because of the difficulty in quantifying errors in estimating both rates of increase of stress and levels of fracture criticality, stress forecasts are given in terms of a smaller earlier to larger later (SELL) window. Note also that the largest m b = 5.6 (Ms = 6.6) earthquake in SW Iceland since 1996 was not stress forecast because there were insufficient (shear wave source) earthquakes for 7 weeks at the beginning of the increase and the increase was not recognized (Volti & Crampin 2003b). Note also that the large scatter of time delay measurements in Fig. 5 is caused by the crack-critical nature of the distributions of fluid-saturated stress-aligned cracks as discussed in Section 4.3, below. 3.4 Forecasting volcanic eruptions The first implied increase of aspect ratios from increases of time delays in Band-1 in Fig. 5, from 1996 May to September, ends at the time of the Vatnajökull eruption at the beginning of 1996 October. (Note that data are sparse, and the least-squares line is only through nine points. Consequently, the nine-point moving average does not show the same increase as the least-squares line.) Similar implied increases are also seen at Stations GRI, KRI and SAU, Volti & Crampin (2003b). We interpret this as indicating increasing stress prior to the hydraulic fracture of the 10 km long fissure opened by the eruption when the rocks reached fracture criticality. Following the eruption, a least-squares line fit to the time delays in both Band-1 and Band-2 of the shear wave window show a comparatively linear decrease of about 2 ms km−1 yr−1 for about 2 years. Similar behaviour is seen in Bands 1 and 2 at all four seismic stations in Iceland, BJA, GRI, KRI and SAU, where there were reliable measurements of shear wave splitting, at distances up to about 240 km north and 240 km WSW of the Vatnajökull eruption (Volti & Crampin 2003b). This slow decrease is interpreted as the relaxation in stress as the Mid-Atlantic Ridge adjusts to the new stress regime following the eruption. We suggest these observations associated with the eruption confirm that: (1) changes in shear wave splitting are measuring the effects of changes of stress in the rock mass rather than the details of  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust 231 seismogenic zones; (2) shear wave splitting is a sensitive diagnostic of small stress-induced changes to in situ rocks; and (3) the effects of increases of stress may be seen at very large distances. Recently, several volcanic eruptions have disturbed the stress regime in SW Iceland and no earthquakes have been stress forecast since 1998, although the M = 5.6 event (see the end of the last section) might have been forecast had there been sufficient source earthquakes. It is possible that the effects reported for eruptions in Iceland may not be typical of all eruptions. However, similar effects have been observed elsewhere. Miller & Savage (2001) reported 90◦ flips in shear wave polarizations following an eruption on Mount Ruapehu, New Zealand (the monitoring was not continuous). They interpreted these as due to high-pressure magma injections in the comparatively shallow crust. This is consistent with our observations at Vatnajökull. We suggest that continuous observations at Ruapehu would have shown changes in time delays as the fluid pressure increased as we saw at Vatnajökull in Fig. 5. We note that phenomena associated with shear wave splitting are very pervasive, and are probably associated with the universality of critical systems discussed in Section 4. 3.5 Developing stress-monitoring sites The above results indicate that changes in shear wave splitting can monitor the effects of increasing stress before earthquakes and lead to stress forecasting the time and magnitude of impending events. However, reliable routine stress forecasting using earthquakes as the source of shear waves is severely restricted because of the scarcity of suitably persistent swarms of small earthquakes to use as a source of shear waves. Routine stress forecasting away from such swarms requires the controlled source seismology of stress-monitoring site (Crampin et al. 2000; Crampin 2001). SMSs use cross-hole seismics between boreholes to measure shear wave splitting in the same Band1 directions that showed the changes above the small earthquakes in Fig. 5. Fig. 6 shows the optimum SMS geometry. Stress monitoring needs to analyse shear waves below the uppermost 500–1000 m of the crust, below the critical depth where σ V = σ h , so that cracks tend to be parallel and vertical. This also avoids the severe attenuation and scattering usually present in the uppermost layers (Leary & Abercrombie 1994; Leary 1995). The optimum source for radiating split shear waves in SMSs is the downhole orbital vibrator (DOV), which was recently commercialized by Geospace Engineering Resources Inc., Houston. The DOV, previously known as the Conoco Orbital Vibrator or COV (Liu et al. 1993), is an eccentric cam swept both clockwise (CW) and counterclockwise (CCW) to exert a rotating radial pressure on the borehole wall (Cole 1997). The sum and differences of the recorded CW and CCW signals can be combined to simulate radiation from orthogonal point forces and hence simulate radiation of orthogonal shear wave polarizations oriented with respect to horizontal pilotgeophones mounted on the DOV casing (Daley & Cox 2001). This is exactly what is required to routinely monitor shear wave splitting. Preliminary observations from the first SMS are discussed in Section 6. below. 4 S H E A R - WAV E S P L I T T I N G I N A C R A C K - C R I T I C A L C RU S T 4.1 Critical systems The capability of APE to model, calculate, even predict, the response of fluid-saturated microcracked rock to changing conditions in a highly complicated heterogeneous crust is remarkable and requires some explanation. Observations of shear wave splitting indi C 2003 RAS, GJI, 155, 221–240 Figure 6. Specifications for a stress-monitoring site (protected by Patent Application PCT/GBOO/01137), where D is the depth below which the minimum compressional stress is horizontal so that below that depth cracks tend to be aligned vertically. The DOV source operates from a depth of D + 300 m to D + 1000 m in the deeper well and receivers are at a depth D in (optimally at least two) vertical wells at an offset of 300 m. The azimuths of the offsets should be within 45◦ of the direction of minimum horizontal stress, which for this illustration is taken to be north–south. cate that stress-aligned fluid-saturated microcracks are a remarkably pervasive feature with similar parameters in almost all rocks in the crust (Crampin 1994). The calculability is thought to be because the fluid-saturated cracks in the crust are so closely spaced that they are critical systems. Critical systems involve dynamic interactive processes that below criticality perturb only locally. Once systems reach criticality, all members of the critical system influence all other members (Ma 1976; Jensen 1998). The transition temperature of equilibrium thermodynamics is the classical critical system, but critical systems are common in an enormous range of phenomena, including almost all complex systems in nature (Bak 1996). Crampin (1998) and Crampin & Chastin (2001) suggest that stress-aligned fluid-saturated cracks in the Earth’s crust are also interactive critical systems. Similar schemes for the earth have been suggested previously in the self-organized criticality (SOC) of Bak & Tang (1989) and Bak et al. (1988). The advance here is that the microscale mechanism for deformation has been identified as stress-induced fluid movement along pressure gradients between adjacent fluid-saturated grain-boundary cracks and pore throats. This is a quantifiable physical process that can be modelled, monitored and calculated by APE (Crampin & Zatsepin 1997; Crampin 1997, 1999a; Zatsepin & Crampin 1997). The criticality is the reason APE modelling matches the huge range of phenomena in Table 2. Bruce & Wallace (1989) show how different critical systems have similar statistical behaviour at September 8, 2003 232 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin criticality, despite very different subcritical physics. This is known as critical-point universality and implies the self-similar scaling seen in crack distributions (Heffer & Bevan 1990), the well-known Gutenberg–Richter relationship (Kagan 1992), and many other phenomena in the Earth and in rock physics (Crampin & Chastin 2001). Critical systems are typically sensitive (the ‘butterfly’s wings’ sensitivity) to otherwise negligible variations in initial conditions that can lead to order-of-magnitude differences as the systems evolve (Bruce & Wallace 1989). This manipulation of fluid-saturated microcracks makes in situ rock highly compliant (Crampin et al. 2003a), where the behaviour at (fracture) criticality is the occurrence of fracturing and earthquakes. Crampin & Chastin (2001) suggest that the reason for the match of the nearly parameterless APE modelling to deformation in a complex heterogeneous Earth is because the response of such critical media is controlled by the non-linear behaviour near criticality. Consequently, the behaviour can be modelled by critical-point universality of Bruce & Wallace (1989), and is a mean-field theory (Jensen 1998). Crampin & Chastin (2001) note that similarities in critical behaviour can be misleading. It is tempting to use simplistic models for complex Earth processes and make claims for relevance merely because they produce similar statistics and similar self-similarity. Much quoted examples, as in Hergarten (2002), are the slider-block model of Burridge & Knopoff (1967) and the cellular automata of Kadanoff et al. (1989) and Rundle & Klein (1993). Models with SOC are widely available but they are only phenomenological. They reproduce the statistics of SOC but offer little insight into the subcritical physical process, which is the region of interest for geophysics in understanding the behaviour before fracturing or faulting or earthquakes. The perceived similarity with some highly simplified earth mechanism is arbitrary and largely irrelevant to a better understanding of earth processes. The power and relevance of APE modelling is that, because the small-scale physical behaviour is identified and modelled (hopefully) correctly, APE modelling is satisfactory over a very wide range of different phenomena (Table 2). 4.2 The behaviour of a crack-critical system with SOC A critical system of cracks with SOC has profound implications for classical linear geophysics. Table 3 lists some of the direct implications for conventional geophysics (discussed more fully in Crampin & Chastin 2001). There is much bad news and many conventional assumptions are invalid for analysis at higher temporal and spatial resolution. For example, the self-similar scaling means that: spatial and temporal heterogeneities exist at all scalelengths; Gaussian statistics (averages) are valid only in limited temporal and spatial domains; and it is not possible to reliably extrapolate from place to place or from time to time so that any given measurement may degrade with time. Similarly, the sensitivity of criticality to initial conditions means that there is the possibility of long-range and long-time interactions between hydrocarbon reservoirs and between earthquakes in different regions. There is also very good news. In at least some circumstances, the response to given phenomena can be modelled and predicted, as exemplified by the APE modelling of Angerer et al. (2000, 2002) and the successfully stress forecast earthquake of Crampin et al. (1999a). If the response of the rock mass to known changes can be predicted, in some circumstances it may be possible to control the response by feedback (item b4 in Table 3). Note that these advantages are only valid if natural processes are given time to evolve. Jensen (1998) shows that SOC behaviour may be expected only in ‘slowly driven, interaction-dominated threshold systems’. Consequently, one essential ingredient to SOC is that the system evolves slowly from a marginally stable state of metastability toward a threshold. In the Earth, the critical state at the threshold is fracture criticality at the percolation threshold, when fracturing, faulting and earthquakes occur. When the threshold is reached there is fracturing and the system relaxes to another metastable state. The slow drive is necessary in order for the intrinsic properties of the system to have sufficient time to control the dynamics (a continuous stream of sand on a sand pile would not have the discrete avalanches characteristic of SOC). The reader is referred to Crampin & Chastin (2001) and references therein for further discussion. 4.3 The scatter in measurements of time delays One of the remarkable features of the measurements of time delays above small earthquakes is the exceptionally large scatter of the data points, which still show temporal variations when averaged by leastsquares lines (Crampin et al. 1999a; Volti & Crampin 2003a,b). The temporal variations have geophysical significance (for stress forecasting earthquakes, Crampin et al. (1999a) even though the scatter Table 3. Direct implications of distributions of cracks in the crust being critical systems with self-organized criticality, SOC (after Crampin & Chastin 2001). (a) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) The bad news Spatial and temporal heterogeneities exist at all scalelengths Gaussian statistics (averages) are only appropriate in limited situations Inability to extrapolate reliably from place to place Inability to extrapolate reliably from time to time Hence the expectation that any measurement may degrade with time Possibility of long-range interactions between hydrocarbon reservoirs Possibility of long-range interactions between earthquakes in different regions Possibility of long-term interactions in hydrocarbon reservoirs Possibility of long-term interactions between earthquakes in different regions Behaviour of crustal deformation may not correspond to or be explicable by conventional geophysics (b) (1) The good news when the rock mass is responding to slow changes Current configuration of crack geometry within the deep interior of the rock mass or reservoir can be monitored with shear wave splitting Current configuration of cracks in rock mass or reservoir can be evaluated by APE Response of rock mass or reservoir to known changes can be calculated by APE Response of rock mass or reservoir can be controlled by feedback by repeating b1, b2 and b3, above (2) (3) (4)  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust is greater than the variation. Similar scatter is observed in almost all observations of time delays above small earthquakes (Peacock et al. 1988; Crampin et al. 1990, 1991; Liu et al. 1997 and further examples in Volti & Crampin 2003a,b). The sensitivity of shear wave splitting (Section 2) means that the scatter is likely to be variations of small second-order quantities. However, the scatter is much too large to be due to errors in location, errors in local structure, or to misinterpretations or misidentifications (Volti & Crampin 2003a,b). Consequently, the scatter appears to be a fundamental feature of all measurements, reflecting what is presumably a fundamental property of the behaviour of stressed, fluid-saturated, microcracked rock (Volti & Crampin 2003a,b). (The only exceptions where significantly less scatter are observed occur when the earthquakes are used as the shear wave source signals are above isolated swarms (Booth et al. 1990; Gao et al. 1998) typically less than about 1 or 2 km in diameter (Crampin 1991). The reduction in scatter may be because the source-to-receiver ray paths are more restricted so that one source of variation is more limited.) The suggested behaviour is that because the pervading fluidsaturated cracks are a critical system, the effects, the distributions of increased aspect ratios in Fig. 3, are necessarily heterogeneous and cluster in time and space (Volti & Crampin 2003a,b). The deformation mechanism in Fig. 3 is very sensitive to even nearly negligible changes of stress (Crampin et al. 2003a). The variations in earth and ocean tides are well known to cause changes in water well levels (presumably also due to the opening and closing of stress-aligned cracks). So even when stress is slowly accumulating before an earthquake, the stress tensor is continually disturbed by the variations of tides. Thus the proximity to criticality implies that there is a continuous temporal and spatial adjustment or re-arrangement of the clusters of distributions of larger aspect ratios. (Such variations of clustering are typified by Ising models of critical behaviour (see, for example, Bruce & Wallace 1989).) This means that shear waves, particularly from varying earthquake foci, are likely to propagate along slightly different paths as clusters vary in time and space, and in particular pass through crack distributions with significantly different crack aspect ratios. Consequently, the splitting displays a range of shear wave time delays. These clusters of different crack parameters also cause the variations in shear wave polarizations, which are typically observed to vary by 15◦ –20◦ about the mean value (Peacock et al. 1988; Crampin et al. 1990, 1991; Liu et al. 1997; Volti & Crampin 2003a,b). As stress increases, the clusters of increased aspect ratios get larger and their separation, the correlation length ξ , the distance between the clusters of larger aspect ratios, also increases. As stress builds up, ξ increases until the correlation length spans the particular structural region as fracture criticality at the percolation threshold is reached, and the most prominent line of weakness, usually a pre-existing fault, slips and the impending large earthquake occurs. Although difficult to quantify, with too much scatter and not enough data, there is some indication that the scatter of time delays increases, in Band-1, as the impending earthquake approaches. This would be expected as the characteristic length, ξ , of the disturbance increases and the potential for greater scattering also increased. Note that the recent reports from the SMSITES Project in Section 6.3, below, show that the scatter is indeed the result of criticality. The scatter is caused by the 90◦ flips in shear wave polarizations in the high pore-fluid pressures on all seismically active fault planes.  C 2003 RAS, GJI, 155, 221–240 233 5 S H E A R - WAV E S P L I T T I N G — T H E FUTURE Here we try to update the answers to the last question of Crampin and Lovell, with specific implications: what should we do next? Some of the direct implications of crack-critical systems with SOC are listed in Table 3 and their practical implications are listed in Table 4. These various results have profound implications for much of conventional reservoir characterization and hydrocarbon recovery, and much other solid-earth geophysics. The implications are mostly detailed, largely but not exclusively, nearly negligible effects, which can only be monitored by shear wave splitting. Consequently, if conventional understanding and accuracy are satisfactory and sufficient, the effects of the crack-critical system in the Earth’s crust and mantle may be largely neglected. However, if better, more detailed, spatial and temporal resolution is required, then the implications of crack-critical systems need to be exploited. If we wish to obtain more than the (typically) 30–40 per cent of the oil in place usually recovered, or if we wish to predict the time and magnitude of earthquakes, then we need to take advantage of the good news in Table 3 and exploit the practical implications in Table 4. Note that Crampin & Chastin (2001) do not claim that the lists in Tables 3 and 4 are inclusive: merely that there is some, often indirect, evidence for each listed item, so that Tables 3 and 4 are valid in the present state of our understanding. There will certainly be further bad news, good news and practical implications in the future, when we understand the crack-critical crust rather better than we do now. The next four sections list four applications directly leading from the recognition of crack-critical systems. There are a wide variety of other possibilities. 5.1 Monitoring hydrocarbon recovery with single-well imaging The observation that many reservoir parameters display self-similar distributions, such as crack distributions (Heffer & Bevan 1990), and the 1/f -noise characteristic of well-logs (Bak et al. 1987; Leary 1991; Bean 1996), means that the larger the data sample is the larger the possible variation. Consequently, Gaussian averages are no longer valid (Table 3) (except in particular limited circumstances). Similarly, critical systems imply that detailed measurements no longer have temporal and spatial stability, but vary with time and place with temporal and spatial heterogeneity. This means that the whole basis of conventional reservoir characterization is no longer well-founded. To improve performance we need to devise new strategies. Crampin (1998) and Crampin & Chastin (2001) suggest (Table 3) that measurements are valid only at the time and place they are taken. Additionally, all surface-based measurements, source and/or receiver, are limited in resolution by passage through the absorption and scattering in near-surface structures (the uppermost 500– 1000 m, at least) (Leary & Abercrombie 1994; Leary 1995). This means that passage through near-surface layers is typically limited to frequencies less than typically 80 Hz for P waves and 20 Hz for shear waves. Consequently, to obtain measurements with sufficient resolution to improve on current practice, in hydrocarbon production, for example, measurements need to be made at the time (during production) and place (the particular location in the reservoir) they are required. One way, possibly the only way, to measure such properties is by a single- or dual-well imaging time-lapse configuration (Crampin September 8, 2003 234 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin Table 4. Some practical implications of critical crack systems with SOC for fluid–rock interactions within the Earth’s crust (after Crampin & Chastin 2001). (a) (1) (2) (3) (4) (5) (6) (7) (8) (b) (1) (2) (3) (4) (5) (6) (7) (c) (1) (2) (3) (4) (d) (1) General implications Fluid-saturated crack distributions are highly compliant and crack geometry responds to small nearly negligible changes of stress, pressure and other phenomena Since fluid–rock properties vary with time, and vary from place to place, measured fluid–rock properties are only strictly valid at the place and time they are measured. Hence, the need for measurements with single-well imaging if accurate specifications are required Since fluid–rock interactions have a dominant effect on almost all physical and chemical behaviour within the crust and mantle (see a2, above), these various effects apply to a huge range of geophysical phenomena, particularly those associated with any deformation, including almost all processes during hydrocarbon recovery Behaviour of stress-aligned fluid-saturated crack distributions appears to be remarkably uniform (within certain limits) even in very heterogeneous structures Pre-fracturing deformation of any given fluid–rock configuration can be monitored by observations of shear wave splitting Pre-fracturing deformation can be modelled by anisotropic poro-elasticity (APE) Response of fluid–rock systems to known changes can be calculated by APE Response to calculated changes (a6, above) can be monitored by shear wave splitting (a5, above), and the response controlled by adjusting changes to optimized response Specific implications Implications for hydrocarbon exploration and production Reservoir properties may change from place to place Reservoir properties may change with time, even without production processes Relevant properties need to be measured at the place and time they are needed Response to known changes can be calculated and predicted (Angerer et al. 2000, 2002) Response of a reservoir can be controlled, in the sense of b4, Table 3 Possibility of long-range and long-time correlations across and between reservoirs (Heffer et al. 1995) There is a limit to the resolution of any measurement Implications for earthquake geophysics Deterministic prediction of time, magnitude, and place of large earthquakes is likely to be impossible (Geller 1997; Kagan 1997; Leary 1997) With sufficient source seismicity (Crampin et al. 1999a), or appropriate crosshole SMS observations (Crampin 2001; Crampin et al. 2003a), times and magnitudes of future large earthquakes can be stress forecast. Other information may than indicate location (Crampin et al. 1999a) In presence of sufficient source seismicity, or appropriate crosshole SMS observations, times of future volcanic eruptions can be stress forecast (Volti & Crampin 2003a,b) There is the possibility of long-range and long-time correlations between earthquakes Implications for rock physics Much of the behaviour in stress-cells in the rock-physics laboratory can be modelled and predicted by APE (Crampin et al. 1997, 1999b; Chapman et al. 1998, 2000) et al. 1993; Peveraro et al. 1994; Crampin 1999b). In such imaging, a string of three-component geophones in a producing well would be inserted behind casing, or behind tubulars, to record signals from a source (for example the DOV) pulsed in-line in the same well (single-well imaging) or in an adjacent well (dual-well imaging). The signals recorded within a production zone will be scattered from the internal structure within the reservoir and, except in fortuitous circumstances there will be no reflections from approximately planar reflectors to interpret. It is suggested that the only way to monitor the detailed fluid–fluid, fluid–rock interactions during hydrocarbon production, which are likely to be dominated by the behaviour and statistics of the crack-critical systems discussed in this paper, is by single-well or dual-well time-lapse imaging: analysing the changes in scattering induced by movements of fluid–fluid fronts within the producing reservoir. Appropriate instrumentation for strings of receivers and borehole shear wave sources in a single- or dual-well configuration has recently become available. We suggest that many of the instrumental developments for well imaging have been solved, and single- or dual-well deployments would be much cheaper alter- natives to 4-D three-component reflection profiles, as well as providing high-frequency signals and resolution in the appropriate parts of the producing reservoir. 5.2 Likelihood of greater hydrocarbon recovery at slower production rates The crack-critical crust has many implications for the behaviour of the reservoir. To take advantage of the good news in Table 3(b), a crucial requirement is that any induced change during production procedures must be sufficiently slow to leave time for natural stress-relaxation phenomena to occur: for example, fluids must be allowed to percolate unforced. Calculable APE behaviour depends on SOC, and a necessary requirement for SOC is that the process should be driven slowly (Jensen 1998). Fast changes, which do not allow stress relaxation, would deliver the bad news in Table 3(a) (the tendency for the reservoir to behave chaotically, irregularly and unpredictably) without any of the mitigating advantages (regularity, calculability and predictability). This is typically the current  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust situation where maximum production rates are sought, and the system is overdriven so only a relatively small proportion of oil in any reservoir can be extracted. Such overdriven fluid–fluid and fluid– rock interactions are irregular and unpredictable, and surface-based seismics do not have sufficient resolution to monitor the progress in detail, hence the need for the single- or dual-well imaging of the previous section. There is likely to be true chaos in fast driven processes, not merely calculable deterministic chaos. This suggests that producing reservoirs are likely to behave in more regular ways (and are more likely to achieve their expected production targets), if production rates are sufficiently slow to allow stress relaxation as production proceeds. There are many examples of overmature reservoirs where nodding donkeys continue to produce acceptable oil-to-water percentages at very low production rates. This review suggests that such fields may well produce a larger proportion of their initial reserves at slow production rates than fields with more aggressive production strategies. It is also suggested that to take advantage of a slower higher production requires slower production to begin from the outset of production. It seems unlikely that a slower regime would be effective once a field has been rapidly produced, and water breakthrough, and other probably irreversible processes had disturbed the initial fluid–rock structures. Note that we do not know just how slow this slower production needs to be to take advantage of the good news in Table 3(b). It may be that even small (possibly marginal) decreases in production rates could lead to substantially higher overall production. The highpressure CO2 injection in Angerer et al. (2000, 2002) was matched by APE modelling some 2 weeks after the injection had been completed, which since APE depends on subcritical physics, presumably means it had reached an approximately steady state. Thus ‘slow’ in this case may mean a comparatively short period of time (days or weeks rather than months or years). However, it is expected that optimum rates will vary from reservoir to reservoir and from field to field. To optimize production rates while maximizing oil production requires new production strategies while using single-well imaging to monitor the progress of fluid movements within the reservoir during production. To the best of our knowledge, such strategies have not yet been attempted. 5.3 Stress forecasting (not predicting) earthquakes with stress-monitoring sites Mankind has been seeking ways to predict the magnitude, time and place of earthquakes for over 120 years (Milne 1880) with a singular lack of success. Much of the research has been directed at seeking some form of precursor, and a very large number of possibly precursory phenomena have been identified before earthquakes. However, no particular precursor has been consistently observed, and when a precursor has been identified, it typically bears no quantifiable relationship with the magnitude, time or place of the particular earthquake with which it appears to be associated. It is now recognized that the earth is so complex and heterogeneous that the magnitude, time, and place of earthquakes are unlikely to be predicted deterministically (Geller 1997; Kagan 1997; Leary 1997, amongst many others). The hypothesis that makes stress forecasting the time and magnitude of earthquakes possible is that rock is so weak to shear stress (see the discussion of Fig. 2), that the necessary stress increase before an earthquake builds up over an extensive volume of rock. Eventually a large volume of rock approaches fracture criticality and the earthquake occurs at the weakest point which will typically  C 2003 RAS, GJI, 155, 221–240 235 be on an existing fault plane. The evidence presented here suggests that the stress build-up, and the approach to criticality, can be recognized at substantial distances from the eventual epicentre. The duration and rate of the increase are proportional and inversely proportion, respectively, to the magnitude of the impending event, which occurs when the increase reaches fracture criticality (Volti & Crampin 2003a,b). This led to the successful stress forecast of the time and magnitude of an M = 5 earthquake using swarm earthquakes as the source of shear waves (Crampin et al. 1999a). Local seismicity correctly indicated the location. SMSs use controlled-source crosshole seismics (Fig. 6) to monitor the state of the in situ crack distributions and their progress towards fracture criticality without the need for persistent swarms of source earthquakes (Crampin 2001). In principle, controlled-source seismology is capable of great precision in measuring shear wave splitting (Li & Crampin 1992), so we can expect the source of the scatter of normalized time delays in Fig. 5 to be resolved, and the SELL windows to be much more tightly defined. 5.4 Other applications of shear wave splitting There are two main types of application for shear wave splitting. Those involving: (1) static effects—the measurement and interpretation of the initial state of the rock mass; and (2) dynamic effects—the measurement of changes in the rock mass by time-lapse techniques. (a) Static effects. It has been recognized for many years that the polarizations of the faster split shear waves are aligned approximately parallel to the direction of maximum horizontal stress and hence parallel to the maximum horizontal permeability in the rock matrix through which the shear waves pass (Crampin 1981; Alford 1986; Zatsepin & Crampin 1996). This means that the orientations of hydraulic fractures, directions of water floods and other hydrocarbon production strategies can be optimized by analysing shear wave splitting. This strategy appears to be recognized by the oil industry, although not always adopted, as there are alternative techniques for obtaining similar estimates. (b) Dynamic effects—monitoring toxic-waste and nuclear-waste repositories. The case studies in Sections 3.1, 3.3 and 6.1, below, show that shear wave splitting is sensitive to otherwise negligible changes in rock mass conditions. This means that leakage or other site instabilities in toxic-waste and nuclear-waste repositories could be detected by monitoring shear wave splitting with cross-hole timelapse techniques. The optimum technique would be to set up borehole instruments to bracket the repository with ray paths from a DOV source to three-component geophones. CW and CCW sweeps of the DOV would again generate orthogonal shear wave polarizations. Note that the ray paths should be far enough from the repository to avoid interface waves and to separate the most informative waves, the direct shear wave arrivals, from reflections and refractions from side walls of the repository. (c) Dynamic effects—monitoring slope and other near-surface instability. It is suggested that the APE mechanism for pre-fracturing deformation (fluid migration along pressure gradients between neighbouring microcracks), while proven at depth, is also valid for near-surface deformation. The items in Table 2 suggest that APE deformation is based on a fundamental relationship between the evolution of fluid-saturated microcracks and stress, which will be present in all in situ materials, even poorly consolidated soils. It is expected that fracturing or failure only occurs when fracture distributions reach fracture criticality. This means that monitoring shear wave splitting in appropriate near-surface geometries will also September 8, 2003 236 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin allow the approach of fracture criticality and local instability to be recognized. A number of small-scale shear wave sources have been developed, ranging from horizontal pistons, to a variety of oriented weight drops and swinging hammers. These could be used to monitor the near-surface microcrack structures for hillside and dam site stability testing, as well as stability of foundations for buildings, dams and tunnels. There are many other examples of local instability problems where analysing shear wave splitting would allow the approach to fracture criticality and fracturing to be monitored. 6 O B S E RVAT I O N S F R O M T H E F I R S T STRESS-MONITORING SITE: C O N F I R M AT I O N O F T H E C R A C K - C R I T I C A L C RU S T The first stress-monitoring site (Section 3.5) was developed in the European Commission funded SMSITES Project at Húsavı́k in Northern Iceland, where the Húsavı́k-Flatey Fault of the Tjörnes Fracture Zone of the Mid-Atlantic Ridge runs onshore (Crampin et al. 2000, 2003a; Crampin 2001, www.smsites.org; see also www.glg.ed.ac.uk/∼scrampin/opinion). The SMSITES Project has recently yielded four significant results, which provide strong suppport for the arguments in this review. (1) A data set of eight simultaneous measurements showing great compliance and sensitivity of in situ rock to small remote disturbances (Crampin et al. 2003a). (2) Evidence that seismically active fault planes are pervaded by critically high pore-fluid pressures (Crampin et al. 2002). (3) Explanation of the ±80 per cent scatter (as in Fig. 5) observed in measured time delays above small earthquakes (Crampin et al. 2003b). (4) Increases in time delays before large earthquakes show precursory decreases immediately before earthquakes occur (Gao & Crampin 2003). 6.1 Eight simultaneous measures displaying stress sensitivity to small disturbances A detailed calibration of the recording system at SMSITES, in which the highly repeatable downhole orbital vibrator borehole source was pulsed repeatedly every 12–20 s for 13 days, happened to coincide with the start of a 2.5 day burst of low-level seismicity at 70 km distance (Crampin et al. 2003a). We recorded seismic traveltimes propagating horizontally at 500 m depth between boreholes 315 m apart. Stacked records yielded a resolution of ±20 µs. The boreholes are approximately parallel and about 100 m south of the major WNW to ESE striking Húsavı́k-Flatey Fault, a transform fault of the MidAtlantic Ridge. The following traveltime anomalies in were recorded (Crampin et al. 2003a). (1) P-wave traveltimes showed initially an abrupt 5 ms (6 per cent) increase which then decayed linearly back to the original level over 10 days. (2) SH-wave traveltimes showed a 2 ms (2 per cent) reduction over 4 days in an ‘S-shaped’ relaxation curve typical of measurements of a phenomena relaxing after some disturbance. (3) SV -wave traveltimes where 2 ms later than SH waves, but showed a similar 2 ms (2 per cent) ‘S-shaped’ relaxation curve over 4 days. (4) SV –SH traveltime anisotropy showed a 0.2 ms (10 per cent) increase in the ∼2 ms time difference between the split shear waves over 6 days. (5) A continuously monitored water well on the island of Flatey immediately above the HFF showed an abrupt 1 m reduction in level for 5 days superimposed on 40 cm ocean tide oscillations. (6) Global Positioning System (GPS) measurements in a NS direction across the WNW to ESE trending HFF showed an abrupt 7 mm extension, which decreased to the original level over about 11 days. (7) GPS measurements across HFF in an EW direction showed a similar 3 mm right-lateral extension which decreased to the initial position after 4 days, and then increased to a permanent 4 mm EW extension in a displacement characteristic of the dextral HFF. (8) The start of all these anomalies coincided with a 2.5 day burst of low-level swarm-type seismicity on the Grı́msey Lineament, a parallel transform fault 70 km from SMSITES. The release of stress at these small earthquakes is thought to be the driving mechanism for all the anomalies. This data set is probably a unique imaging of the effects of low-level deformation on the rock mass. We were fortunate to record them. Continuous well level measurements at Flatey showed that the 1 m 5 day ‘pulse’ in water level was the only such pulse in 15 months of records. The data set has not yet been fully interpreted and will be discussed more fully elsewhere. In brief, some measurements appear to be compatible. Splitting into SH and SV waves, rather than other shear wave polarizations, is expected because propagation parallel to the fault is a symmetry direction. The 1 m drop in water level is approximately what is expected from a 7 mm NS GPS extension over microcracks and pores in a 200 m thick sandstone at the top of which the seismic waves are propagating. However, there are puzzling features: in particular, the different relaxation times of the various measurements. Note the earthquakes mark abrupt releases of stress on the Grı́msey Lineament. Such sensitivity to small disturbances at 70 km distance, the energy release of which is probably equivalent to one M = 4 earthquake, is not expected in a conventional brittle elastic crust. Thus the sensitivity is a direct confirmation that the crust of the Earth is a crack-critical system verging on fracture criticality and failure. Note that SMSs monitor changes within the interior of the rock mass. In particular, shear wave splitting monitors the approach of the rock mass to fracture criticality when rocks are so heavily fractured that shear-strength is lost and fracturing, faulting, and earthquakes occur. In contrast, GPS measurements monitor surface displacements. These are necessarily consistent with the overall movements of tectonic blocks, as Michel et al. (2001) demonstrated in their analysis of GPS measurements before and after an earthquake in Sumatra, However, GPS measurements do not monitor the approach of fracture criticality and earthquakes. 6.2 Critically high pore-fluid pressures on seismically active fault planes Seismic stations immediately above major faults, the San Andreas fault in California (Peacock et al. 1988; Liu et al. 1997) and now the Húsavı́k-Flatey Fault in northern Iceland (Crampin et al. 2002), display 90◦ flips in shear wave splitting polarizations approximately orthogonal to the direction of regional tectonic stress and result in fault parallel directions. Crampin et al. (2002) model these effects  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust with APE and show that they can be interpreted as the effects of critically high pore-fluid pressures pervading the immediate vicinity of the fault. The suggestion is that, as frequently noted, seismically active fault zones require high pore-fluid pressures to relieve friction and separate asperities before slip can take place. Such 90◦ flips are only visible above major faults because it is only above major faults that more than half of the ray path of a shear wave to seismic stations at the surface is close to the fault in rocks pervaded by high pore-fluid pressures. Most earthquakes occur on small faults where 90◦ flips only occur near the fault, so that the majority of the ray path is in normally pressurized rock with normal not flipped polarizations. 6.3 Explanation of ±80 per cent scatter in measured time delays above small earthquakes The time delays in shear wave splitting recorded in the shear wave window above small earthquakes typically display a scatter of approximately ±80 per cent about the mean (as in Fig. 5). It should be noted that observations of time delays in exploration seismics away from earthquake source zones do not show significant scatter. This makes the scatter difficult to explain by conventional geophysics or error analysis (Section 4.3; Volti & Crampin 2003a). Such small earthquakes are typically generated by slip on comparatively small faults. The suggestion is that although 90◦ flips are present, caused by the high pressures, when propagating close to the fault, the majority of the ray path to the surface is in normally pressurized rocks which do not show flipped polarizations. Since in general, the normally pressurized ray path is longer than the high-pressure segment in the vicinity of the fault, the typical tectonic-stress aligned polarizations are observed at the surface. The time delay at the surface is the difference of the effects of the high-pressurized ‘flipped’ segments and the normally pressurized normally polarized segments of the ray path. Crampin et al. (2003b) model these effects and show that small differences in triaxial stress and pore-fluid pressure can cause a ±80 per cent scatter. Since every earthquake modifies the triaxial stress and pressure fields by some stress release, even small changes in triaxial stress and pore pressure can easily cause the observed ±80 per cent scatter (Crampin et al. 2003b). The widespread universal observations of this ±80 per cent scatter suggests that, not surprisingly, all earthquakes are generated by slip on critically high pressurized fault planes. Although we have identified the source of the scatter, the scatter cannot be eliminated. To calculate the scatter requires extremely detailed knowledge of the geological and geophysical structure, details of the irregularities on the fault plane, and the stress and pore pressure in the earthquake source zone and along the whole of the ray path to the recorder. Clearly, we are never able to model these effects in sufficient detail. 6.4 Stress relaxation in the earthquake source zone before the earthquake occurs We have established in the main text that increasing time delays in Band-1 of the shear wave window monitor the effects of accumulating stress on crack aspect ratios. Although several anomalies had been recognized, previously it had been thought that time delays in Band-1 increased until stress was released at the time of the earthquake (Crampin 1999a). Reappraisal of the data sets shows that whenever there is adequate data to record the phenomenon there is a precursory decrease in time delays before the earthquake occurs. This decrease has only been observed within a few fault diameters  C 2003 RAS, GJI, 155, 221–240 237 of the source and is believed to be caused by some form of stress relaxation in the source zone. The duration of the precursory decrease varies with earthquake magnitude from a few tens of minutes before mb ≈ 2 events in northern Iceland to over 2 months before the 1988 M s = 6 North Palm Springs earthquake in California. The plot of magnitude against log of the time duration is linear. Similar precursory decreases are also seen in samples in laboratory stress cells before fracturing occurs (Gao & Crampin 2003). 6.5 Interpretation Taken together, these four results are a major confirmation and extension of the ideas advanced in this review. In particular, they indicate the relevance and universality of the APE model for the evolution of fluid-saturated microcracked rock to changes of stress in a compliant crack-critical system. The results confirm that in situ rock, in at least the crust of the Earth, is extraordinarily compliant as a result of the critical system of fluid-saturated microcracks, as well as confirming the science and the technology of stress-monitoring sites for stress forecasting the times and magnitudes of large earthquakes. An explanation for the ±80 per cent scatter in time delays above has been sought for some 15 years since the phenomenon was first recognized by Peacock et al. (1988). It is encouraging that such a long-standing problem seems to have been satisfactorily resolved. It appears that we are beginning to understand shear wave splitting. We suggest that these recent advances in shear wave splitting, together with the suggestions in the main text, present several new opportunities for monitoring, calculating and predicting rock mass deformation. 7 C O N C LU S I O N S We have shown how the effective elastic constants of fluid-saturated microcracks are extremely sensitive to fluctuations in the stress field. The response is strongly non-linear and much more sensitive than laboratory tests on small samples. There are four major conclusions of this survey of seismic shear wave anisotropy. (1) Shear wave splitting is highly sensitive to changes to the fluidsaturated stress-aligned grain-boundary cracks and pore throats pervading almost all in situ rocks in the crust (and upper mantle). (2) The detailed internal geometry of this microcracked rock mass is very sensitive to small nearly negligible changes in in situ conditions, which cause observable changes to shear wave splitting. (3) The evolution of stress-aligned fluid-saturated microcracks can be modelled, sometimes with great accuracy, by APE, where the mechanism of deformation is fluid movement along pressure gradients between neighbouring grain-boundary cracks and pore throats at different orientations to the stress field. APE modelling matches a huge range of static and dynamic phenomena. Since APE is highly constrained with no free parameters, the response of the rock mass to known conditions can be calculated and in some circumstances predicted. (4) The underlying reason for the calculability and predictability of a complicated heterogeneous crust is that the rock mass is so heavily cracked that it can be considered as a critical system. This has profound effects on many aspects of conventional solid-earth geophysics, some of which are listed in Tables 3 and 4. In particular, many parameters of the rock mass are controlled by the behaviour of fluid-saturated microcracks, rather than the complicated heterogeneous subcritical rock. September 8, 2003 238 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin Taken together these four conclusions amount to a new understanding of low-level rock deformation, where the deformation of in situ rock in response to changing stress is controlled by the changes in the geometry of the stress-aligned fluid-saturated microcracks. If current usage in resolution and accuracy is sufficient, the above four items above can be ignored. However, if we wish to understand detailed fluid–rock interactions (including shear wave splitting), to extract more oil from a reservoir, or to forecast the time and magnitude of earthquakes, the good news in Table 3 must be exploited, in the ways suggested in Section 6. We suggest this could be a turning point for solid-earth geophysics. The ideas outlined in this review are believed to be a major advance in understanding rock mass pre-fracturing deformation, with some disadvantages but with some important new properties to exploit. This is believed to be a fundamental new understanding of rock mass deformation with totally new applications. This is a geophysical (and geological) renaissance as Davis et al. (1997) once suggested for a small part of this revolution. We suggest that these developments could lead to the practice and techniques of both exploration and earthquake geophysics being significantly different in a few years time. However, the interior of the Earth is remote, toxic, subject to high temperatures and pressures and is remarkably difficult to understand—witness some of the results in this review which are, we suggest, new and surprising, even after half a century of intensive geophysical investigations, including some 20 years of analysis of shear wave splitting. Some of the applications in Sections 5 and 6 are (observation-based) speculations. We should be interested in any information or plans that may add data to help to realize these speculations and advance this revolution in solid-earth geophysics. AC K N OW L E D G M E N T S This paper was partially supported by the European Commission SMSITES and PREPARED Projects, contract numbers EVR1CT1999-40002 and EVG1-CT2002-00073, respectively. The various ideas in this paper have developed over many years in many discussions with many persons too numerous to mention individually. However, we should like to acknowledge our immediate colleagues Yuan Gao, Peter Leary, Xiang-Yang Li, Enru Liu, David Taylor, Theodora Volti and Sergei Zatsepin, and thank them for the invaluable part they have played and are playing in the development of these ideas. REFERENCES Alford, R.M., 1986. Shear data in the presence of azimuthal anisotropy, Dilley, Texas, 56th Ann. Int. SEG Meeting., Houston, 1986, Expanded Abstracts, 476–479. Angerer, E., Crampin, S., Li, X.-Y. & Davis, T.L., 2000. Time-lapse seismic changes in a CO2 injection process in a fractured reservoir, 70th Ann. Int. SEG Meeting., Calgary, 2000, Expanded Abstracts, 2, 1532–1535. Angerer, E., Crampin, S., Li, X.-Y. & Davis, T.L., 2002. Processing, modelling, and predicting time-lapse effects of overpressured fluid-injection in a fractured reservoir, Geophys. J. Int., 149, 267–280. Bak, P., 1996. How Nature Works, Springer-Verlag, New York. Bak, P. & Tang, C., 1989. Earthquakes as self-organised critical phenomenon, J. geophys. Res., 94, 15 635–15 637. Bak, P., Tang, C. & Wiesenfeld, K., 1987. Self-organized criticality: an explanation of 1/f -noise, Phys. Rev. Lett., 59, 381–384. Bak, P., Tang, C. & Wiesenfeld, K., 1988. Self-organized criticality, Phys. Rev. A, 38, 364–374. Bamford, D. & Crampin, S., 1977. Seismic anisotropy—the state of the art, Geophys. J. R. astr. Soc., 49, 1–8. Bean, C., 1996. On the cause of 1/f -power spectral scaling in borehole sonic logs, Geophys. Res. Lett., 23, 3119–3122. Blackman, D.K. & Kendall, J.M., 1997. Sensitivity of teleseismic body waves to mineral texture and melt in the mantle beneath a mid-oceanic ridge, Phil. Trans. R. Soc. Lond. A, 355, 217–231. Booth, D.C. & Crampin, S., 1985. Shear-wave polarizations on a curved wavefront at an isotropic free-surface, Geophys. J. R. astr. Soc., 83, 31– 45. Booth, D.C., Crampin, S., Lovell, J.H. & Chiu, J.-M., 1990. Temporal changes in shear wave splitting during an earthquake swarm in Arkansas, J. geophys. Res., 95, 11 151–11 164. Bruce, A. & Wallace, D., 1989. Critical point phenomena: universal physics at large length scale, in The New Physics, pp. 236–267, ed. Davis, P., Cambridge Univ. Press, Cambridge. Burridge, R. & Knopoff, L., 1967. Model and theoretical seismicity, Bull. seism. Soc. Am., 57, 341–371. Chapman, M., Zatsepin, S.V. & Crampin, S., 1998. Anisotropic dispersion in stress-sensitive poroelasticity, 60th Conf. EAGE, Leipzig, 1998, Extended Abstracts, 1, 10–10. Chapman, M., Zatsepin, S.V. & Crampin, S., 2000. Incorporating stresssensitivity into dynamic poro-elasticity, 70th Ann. Int. SEG Meeting, Calgary, 2000, Expanded Abstracts, 2, 1536–1539. Cole, J., 1997. Orbital vibrator, a new tool for characterising interwell reservoir space, Leading Edge, 13, 281–283. Crampin, S., 1978. Seismic wave propagation through a cracked solid: polarization as a possible dilatancy diagnostic, Geophys. J. R. astr. Soc., 53, 467–496. Crampin, S., 1981. A review of wave motion in anisotropic and cracked elastic-media, Wave Motion, 3, 343–391. Crampin, S., 1985. Evaluation of anisotropy by shear-wave splitting, Geophysics, 50, 142–152. Crampin, S., 1986. Anisotropy and transverse isotropy, Geophys. Prosp., 34, 94–99. Crampin, S., 1990. Alignment of near-surface inclusions and appropriate crack geometries for geothermal hot-dry-rock experiments, Geophys. Prosp., 38, 621–631. Crampin, S., 1991. An alternative scenario for earthquake prediction experiments, Geophys. J. Int., 107, 185–189. Crampin, S., 1993. A review of the effects of crack geometry on wave propagation through aligned cracks, Can. J. Expl. Geophys., 29, 3–17. Crampin, S., 1994. The fracture criticality of crustal rocks, Geophys. J. Int., 118, 428–438. Crampin, S., 1995. The anisotropy of fluid-saturated rock, Gen. Ass. Int. Union. Geod. Geophys., Boulder, 1995, Abstracts Week A, A166– A167. Crampin, S., 1996. Anisotropists Digest 149 and 150, anisotropists@ sep.stanford.edu. Crampin, S., 1997. Going APE: I—Modeling the inherent anisotropy of intact rock, 67th Ann. Int. SEG Meeting, Dallas, 1997, Expanded Abstracts, 1997, 1, 952–955; see also 956–959, 921–924. Crampin, S., 1998. Shear-wave splitting in a critical crust: the next step, in 8th Int. Workshop on Seismic Anisotropy, Boussens, Vol. 53, 749–763, ed. Rasolofosaon, P., Rev. Inst. Franc. Pet. Crampin, S., 1999a. Calculable fluid–rock interactions, J. Geol. Soc., 156, 501–514. Crampin, S., 1999b. Implications of rock criticality for reservoir characterization, J. Pet. Sci. Eng., 24, 29–48. Crampin, S., 2000. The potential of shear-wave splitting in a stress-sensitive compliant crust: a new understanding of pre-fracturing deformation from time-lapse studies, 70th Ann. Int. SEG Meeting., Calgary, 2000, Expanded Abstracts, 2, 1520–1523. Crampin, S., 2001. Developing stress-monitoring sites using cross-hole seismology to stress-forecast the times and magnitudes of future earthquakes, Tectonophysics, 338, 233–245. Crampin, S., 2003. Aligned cracks not LPO as the cause of mantle anisotropy, EGS-AGU-EUG Joint Ass. Nice, 2003, Abstract EAE03-A-00205.  C 2003 RAS, GJI, 155, 221–240 September 8, 2003 15:14 Geophysical Journal International gji˙2037 Shear wave splitting in the crack-critical crust Crampin, S. & Booth, D.C., 1989. Shear-wave splitting showing hydraulic dilatation of pre-existing joints in granite, Sci. Drilling, Vol. 1, 21–26. Crampin, S. & Chastin, S., 2001. Shear-wave splitting in a critical crust: II— compliant, calculable, controllable fluid–rock interactions, in Anisotropy 2000: Fractures Converted Waves and Case Studies, Proc. 9th Int. Workshop on Seismic Anisotropy, Cape Allen 2000, Vol. 6, pp. 21–48, eds Ikelle, L.T. & Gangi, A., SEG Open File Publ. Crampin, S. & Lovell, J.H., 1991. A decade of shear-wave splitting in the Earth’s crust: what does it mean? what use can we make of it? and what should we do next?, Geophys. J. Int., 107, 387–407. Crampin, S. & Zatsepin, S.V., 1995. Production seismology: the use of shear waves to monitor and model production in a poro-reactive and interactive reservoir, 65th Ann. Int. SEG Meeting, Houston, 1995, Expanded Abstracts, 199–202. Crampin, S. & Zatsepin, S.V., 1997. Modelling the compliance of crustal rock—II. Response to temporal changes before earthquakes, Geophys. J. Int., 129, 495–506. Crampin, S., Evans, R., Üçer, B., Doyle, M., Davis, J.P., Yegorkina, G.V. & Miller, A., 1980. Observations of dilatancy-induced polarization anomalies and earthquake prediction, Nature, 286, 874–877. Crampin, S., Chesnokov, E.M. & Hipkin, R.G., 1984a. Seismic anisotropy— the state of the art II, in Proc. 1st Int. Workshop on Seismic Anisotropy, Suzdahl, 1982, eds Crampin, S., Hipkin, R.G. & Chesnokov, E.M., Geophys. J. R. astro. Soc. 76, 1–16. Crampin, S., Evans, R. & Atkinson, B.K., 1984b. Earthquake prediction: a new physical basis, Geophys. J. R. astr. Soc., 76, 147–156. Crampin, S., Booth, D.C., Evans, R., Peacock, S. & Fletcher, J.B., 1990. Changes in shear wave splitting at Anza near the time of the North Palm Springs Earthquake, J. geophys. Res., 95, 11 197–11 212. Crampin, S., Booth, D.C., Evans, R., Peacock, S. & Fletcher, J.B., 1991. Comment on ‘Quantitative measurements of shear wave polarizations at the Anza seismic network, southern California: implications for shear wave splitting and earthquake prediction.’ by Aster, R., Shearer, P.M. & Berger, J., J. geophys. Res., 96, 6403–6414. Crampin, S., Queen, J.H. & Rizer, W.D., 1993. The need for underground research laboratories in sedimentary basins, 63rd Ann. Int. SEG Meeting., Washington, Expanded Abstracts, 493–496. Crampin, S., Zatsepin, S.V., Slater, C. & Brodov, L.Y., 1996. Abnormal shear-wave polarizations as indicators of pressures and over pressures, 58th Conf. EAGE, Amsterdam, 1996, Extended Abstracts, O38. Crampin, S., Rowlands, H.J., Zatsepin, S.V., Smart, B.J., Edlmann, K. & Crawford, B., 1997. Predicting the response to effective stress of cores with different pore fluids, 59th Conf. EAGE, Geneva, 1997, Extended Abstracts, Vol. 2, CO22. Crampin, S., Volti, T. & Stefánsson, R., 1999a. A successfully stress-forecast earthquake, Geophys. J. Int., 138, F1–F5. Crampin, S., Zatsepin, S.V., Rowlands, H.J., Smart, B.J. & Somerville, J. M., 1999b. APE-modelling of fluid/rock deformation of sandstone cores in laboratory stress-cells, 61th Conf. EAGE, Helsinki, 1999, Extended Abstracts, Vol. 1, 2–08. Crampin, S., Volti, T. & Jackson, P., 2000. Developing a stress-monitoring site (SMS) near Húsavı́k for stress-forecasting the times and magnitudes of future large earthquakes, in Destructive Earthquakes: Understanding Crustal Processes Leading to Destructive Earthquakes, eds Thorkelsson, B. & Yeroyanni, M., Proc. 2nd EU-Japan Workshop on Seismic Risk, June 23–27, 1999, Europ. Comm., Res. Dir. Gen., 136–149. Crampin, S., Volti, T., Chastin, S., Gudmundsson, A. & Stefánsson, R., 2002. Indication of high pore-fluid pressures in a seismically-active fault zone, Geophys. J. Int., 151, F1–F5. Crampin, S., Chastin, S. & Gao, Y., 2003a. Shear-wave splitting in a critical crust: III—preliminary report of multi-variable measurements in active tectonics, in Proc. 10th Int. Workshop on Seismic Anisotropy, Tutzing, 2002, eds Gajewski, D., Vanelle, C. & Psencik, I., J. appl. Geophys., in press. Crampin, S., Peacock, S., Gao, Y. & Chastin, S., 2003b. The scatter of timedelays in shear-wave splitting above small earthquakes, Geophys. J. Int., in press.  C 2003 RAS, GJI, 155, 221–240 239 Daley, T.M. & Cox, D., 2001. Orbital vibrator seismic source for simultaneous P- and S-wave crosswell acquisition, Geophysics, 66, 1471–1480. Davis, T.L., Benson, R.D., Roche, S.L. & Talley, D., 1997. 4-D 3-C seismology and dynamic reservoir characterization—a geophysical renaissance, 67th Ann. Int. SEG Meeting, Dallas, 1997, Expanded Abstracts, 1, 880– 882; see also 883–885, 886–889. Duranti, L., Davis, T.L. & Benson, R.D., 2000. Time-lapse analysis and detection of fluid changes at Vacuum field, New Mexico, 70th Ann. Int. SEG Meeting., Calgary, 2000, Expanded Abstracts, 2, 1528– 1531. Gao, Y. & Crampin, S., 2003. Observations of stress relaxation before earthquakes, Geophys. J. Int., submitted (ref. GD079F). Gao, Y., Wang, P., Zheng, S., Wang, M. & Chen, Y.-T., 1998. Temporal changes in shear-wave splitting at an isolated swarm of small earthquakes in 1992 near Dongfang, Hainan Island, Southern China, Geophys. J. Int., 135, 102–112. Garotta, R. & Marechal, P., 1987. Shear wave polarization survey using converted waves, 57th Ann. Int. SEG Meeting, New Orleans, 1987, Expanded Abstracts, 657–658. Geller, R.J., 1997. Earthquake prediction: a critical review, Geophys. J. Int., 131, 425–450. Gudmundsson, A., 1999. Fluid overpressure and stress drop in fault zones, Geophys. Res. Lett., 26, 115–118. Heffer, K.J. & Bevan, T.G., 1990. Scaling relationships in natural fractures, Soc. Pet. Eng., Paper 20981. Heffer, K.J., Fox, R.J., McGill, C.A. & Koutsabeloulis, N.C., 1995. Novel techniques show links between reservoir flow directionality, Earth stress, fault structure and geomechanical changes in mature waterfloods, Soc. Pet. Eng., Paper 30711. Hergarten, S., 2002. Self-organized Criticality in Earth Systems, SpringerVerlag, Berlin. Hudson, J.A., 2000. The effect of fluid pressure on wave speeds in a cracked solid, Geophys. J. Int., 143, 302–310. Jensen, H.J., 1998. Self-organized Criticality, Cambridge Univ. Press, Cambridge. Kadanoff, L.P., Nagel, S.R., Wu, L. & Zhou, S.M., 1989. Scaling and universality in avalanches, Phys. Rev., A39, 6524–6533. Kagan, Y.Y., 1992. Seismicity: turbulence of solids, Nonlinear Sci. Today, 2, 1–13. Kagan, Y.Y., 1997. Are earthquakes predictable?, Geophys. J. Int., 131, 505– 525. King, M.S., Chaudhry, N.A. & Ahmed, S., 1994. Experimental ultrasonic velocities and permeability of sandstones with aligned cracks, 56th Conf. EAGE, Vienna, 1994, Extended Abstracts, P113. Leary, P., 1991. Deep borehole log evidence for fractal distribution of fractures in crystalline rock, Geophys. J. Int., 107, 615–627. Leary, P.C., 1995. The cause of frequency-dependent seismic absorption in crustal rocks, Geophys. J. Int., 122, 143–151. Leary, P.C., 1997. Rock as a critical-point system and the inherent implausibility of reliable earthquakes prediction, Geophys. J. Int., 131, 451–466. Leary, P. & Abercrombie, R., 1994. Frequency dependent crustal scattering and absorption at 5–160 Hz from coda decay observed at 2.5 km depth, Geophys. Res. Lett., 21, 971–974. Li, X.-Y. & Crampin, S., 1992. Linear-transform techniques for processing shear-wave splitting in four-component seismic data, Geophysics, 58, 240–256. Li, X.-Y., Mueller, M.C. & Crampin, S., 1993. Case studies of shear-wave splitting in reflection surveys in South Texas, Can. J. Expl. Geophys., 29, 189–215. Liu, E., Crampin, S., Queen, J.H. & Rizer, W.D., 1993. Velocity and attenuation anisotropy caused by microcracks and macrofractures in multiazimuthal reverse VSPs, Can. J. Expl. Geophys., 29, 177–188. Liu, Y., Crampin, S. & Main, I., 1997. Shear-wave anisotropy: spatial and temporal variations in time delays at Parkfield, Central California, Geophys. J. Int., 130, 771–785. Ma, S., 1976. Modern theory of critical phenomena, Frontiers in Physics, Benjamin-Cummings, Menlo Park. September 8, 2003 240 15:14 Geophysical Journal International gji˙2037 S. Crampin and S. Chastin Mainprice, D., 1997. Modeling the anisotropic seismic properties of partially molten rocks found at mid-oceanic ridges, Tectonophysics, 279, 161–179. Meadows, M. & Winterstein, D.L., 1994. Seismic detection of a hydraulic fracture from shear-wave VSP data at Lost Hills Field, California, Geophysics, 57, 11–26. Michel, G.W., Becker, M., Reigber, Ch., Tibi, R., Yu, Y.Q. & Zhu, S.Y., 2001. Regional GPS data confirm high strain accumulation prior to the 2000 June 4 Mw = 7.8 earthquake at southeast Sumatra, Geophys. J. Int., 146, 571–582. Miller, V. & Savage, M., 2001. Changes in seismic anisotropy after volcanic eruptions: evidence from Mount Ruapehu, Science, 293, 2231–2235. Milne, J., 1880. Seismic science in Japan, Trans. Seism. Soc. Japan, 1, 3–33. Mueller, M.C., 1991. Prediction of lateral variability in fracture intensity using multicomponent shear-wave surface seismic as a precursor to horizontal drilling in Austin Chalk, Geophys. J. R. aston. Soc., 107, 409–415. Mueller, M.C., 1992. Using shear waves to predict lateral variability in vertical fracture intensity, Leading Edge, 33, 29–35. Peacock, S., Crampin, S., Booth, D.C. & Fletcher, J.B., 1988. Shear-wave splitting in the Anza seismic gap, Southern California: temporal variations as possible precursors, J. geophys. Res., 93, 3339–3356. Peveraro, R.C.A., Leary, P.C. & Crampin, S., 1994. Single-well wideband borehole seismics in the UNIWELL configuration, Proc. Eur. Pet. Conf., Europec. London, 1994, 1, SPE Paper 28854, 491–506. Rundle, J.B. & Klein, W., 1993. Scaling and critical phenomena in a cellular automation slider-block model for earthquakes, J. Stat. Phys., 72, 405– 413. Sibson, R.H., 1990. Rupture nucleation on unfavourably oriented faults, Bull. seism. Soc. Am., 80, 1580–1604. Slater, C.P., 1997. Estimation and modelling of anisotropy in vertical and walkaway seismic profiles at two North Caucasus Oil Fields, PhD dissertation, University of Edinburgh. Sornette, D., 2000. Critical phenomena in natural sciences, Springer. Sothcott, J., O’Hara, S.G., Khazanehdari, J. & McCann, C., 2000a. From sonic to ultrasonic—the acoustic properties of reservoir sandstones, 62nd Conf. EAGE, Glasgow, 2000, Extended Abstracts, D-39. Sothcott, J., McCann, C. & O’Hara, S.G., 2000b. The influence of two different pore fluids on the acoustic properties of reservoir sandstones at sonic and ultrasonic frequencies, 70th Ann. Int. SEG Meeting, Calgary, 2000, Expanded Abstracts, 2, 1883–1886. Stefánsson, R., Gudmundsson, G.B. & Slunga, R., 2000. The PRENLAB2 project, premonitory activity and earthquake nucleation in Iceland, in Destructive Earthquakes: Understanding Crustal Processes Leading to Destructive Earthquakes, eds Thorkelsson, B. & Yeroyanni, M., Proc. 2nd EU–Japan Workshop on Seismic Risk, June 23–27, 1999, Europ. Comm., Res. Dir. Gen., pp. 161–172. Taylor, D.B., 2000. ANISEIS Manual: Version 5.5, Macroc Ltd, 31 Palmerston Place, Edinburgh. Thomsen, L., 1999. Converted-wave reflection seismology over inhomogeneous, anisotropic media, Geophysics, 64, 678–690. Tsvankin, I., 2001. Seismic Signatures and Analysis of Reflection Data in Anisotropic Media, Pergamon, Amsterdam. Turcotte, D.L., 1992. Fractals and Chaos in Geology and Geophysics, Cambridge Univ. Press, Cambridge. Volti, T. & Crampin, S., 2003a. A four-year study of shear-wave splitting in Iceland: 1—Background and preliminary analysis, in New insights into Structural Interpretation and Modelling, ed. Nieuwland, R., Geol. Soc., Spec. Publ., 212, 117–133. Volti, T. & Crampin, S., 2003b. A four-year study of shear-wave splitting in Iceland: 2—Temporal changes before earthquakes and volcanic eruptions, in New insights into structural interpretation and modelling, ed. Nieuwland, R., Geol. Soc., Spec. Publ., 212, 135–149. Wild, P. & Crampin, S., 1991. The range of effects of azimuthal isotropy and EDA-anisotropy in sedimentary basins, Geophys. J. Int., 107, 513– 529. Winterstein, D., 1996. Anisotropists Digest 147, anisotropists@sep. stanford.edu. Winterstein, D. & De, G.S., 2001. VTI documented, Geophysics, 66, 237– 245. Zatsepin, S.V. & Crampin, S., 1996. Stress-induced coupling between anisotropic permeability and shear-wave splitting, 58th Conf. EAGE, Amsterdam, 1996, Extended Abstracts, C030. Zatsepin, S.V. & Crampin, S., 1997. Modelling the compliance of crustal rock—I. Response of shear-wave splitting to differential stress, Geophys. J. Int., 129, 477–494.  C 2003 RAS, GJI, 155, 221–240