The New Geophysics: stress-forecasting as an alternative to earthquake prediction - a paradigm shift

Transcription

1 24/ Crampin&Gao(2009b) TITLE PAGE The New Geophysics: stress-forecasting as an alternative to earthquake prediction - a paradigm shift Stuart Crampin 1,2,* and Yuan Gao 3 1 School of GeoSciences, University of Edinburgh, Edinburgh EH9 3FW, Scotland UK. 2 now at British Geological Survey, Edinburgh EH9 3LA, Scotland UK. scrampin@ed.ac.uk 3 Institute of Earthquake Science, China Earthquake Administration, Beijing, China. gaoyuan@seis.ac.cn *Corresponding author. Received: 24 th September, Intended for publication in: PAGEOPH Page heading: The New Geophysics and GEMS Address for correspondence: Stuart Crampin School of GeoSciences University of Edinburgh Grant Institute West Mains Road tel: +44 (0) Edinburgh EH9 3JW fax: +44 (0) Scotland UK. scrampin@ed.ac.uk Papers and prints at 1

2 The New Geophysics: stress-forecasting as an alternative to earthquake prediction - a paradigm shift STUART CRAMPIN, 1,2,* and YUAN GAO, 3 1 School of GeoSciences, University of Edinburgh, Edinburgh EH9 3FW, Scotland UK. 2 now at British Geological Survey, Edinburgh EH9 3LA, Scotland UK. scrampin@ed.ac.uk 3 Institute of Earthquake Science, China Earthquake Administration, Beijing, China. gaoyuan@seis.ac.cn * Corresponding author Abstract - This paper suggests that the paradigm shift, before earthquakes can be predicted, is to neglect the impending earthquake source zone, on which most previous prediction techniques have (ineffectually) concentrated, and use shear-wave splitting to monitor stress-accumulation in the rock mass away from the eventual epicentre. Since in situ rock is weak to shear stress, the energy released by large earthquakes must necessarily accumulate over substantial volumes of rock around the impending earthquake source. A new understanding of fluid-rock deformation, a New Geophysics, suggests that such stressaccumulation can be monitored by analysing the effects on shear-wave splitting (seismic birefringence) caused by changes of stress on the geometry of the stress-aligned fluidsaturated microcracks pervading almost all in situ rocks. Observations of shear-wave splitting suggest that large earthquakes occur when all or most of the microcracks in the stressed volume approach fracture-criticality and failure by fracturing. This allows the progress towards fracture-criticality of the highly-compliant fluid-saturated microcrack geometry in most crustal rocks to be monitored and the time and magnitude of an impending larger earthquake stress-forecast. Logarithms of the duration of the stress-accumulation are proportional (self-similar) to the magnitude of the impending event. Knowledge that a large earthquake is approaching allows other precursory data to be interpreted realistically and the location of the fault-plane estimated. Using seismic swarms as the source of shear-waves to monitor the stressed rock, characteristic temporal variations in shear-wave splitting (the inferred stress-accumulation) have been seen in retrospect before some 14 earthquakes worldwide, and on one occasion, the time, magnitude, and fault break of an M 5 earthquake in SW Iceland were successfully stress-forecast in a comparatively narrow time/magnitude window. Unfortunately, the scarcity of suitable seismic swarms for the source of shear-waves, and the irregularity of swarm activity, mean that reliable routine stress-forecasting schemes cannot use seismic swarms as the source of shear-waves. Reliable stress-forecasting requires the precision and controlled-source stability of cross-hole seismics in borehole Stress-Monitoring Sites (SMSs). This paper reviews the evidence for a paradigm shift from earthquake prediction to stress-forecasting and its implications for the future development of SMSs. An appendix proposes GEMS, a Global Earthquake Monitoring System for stress-forecasting all damaging earthquakes worldwide. Key words: Critical-systems, earthquake stress-forecasting; Global Earthquake Monitoring System (GEMS), New Geophysics, shear-wave splitting, stress-accumulation. 2

3 Even a very small effect sometimes requires profound changes in our ideas (Feynman, 1995) 1. Introduction During the 1996, Royal Astronomical Society, Joint Association for Geophysics, Discussion Meeting in London on Assessment of schemes for earthquake prediction (Evans, 1997), Geller (1997) in a comprehensive analysis of past research found no evidence for successful short-term earthquake prediction (see also Geller et al., 1997). Similarly Kagan (1997) found no evidence for successful long-term prediction, and Leary (1997) demonstrated that well-logs showed that in situ rocks are non-linear critical-systems with the implication that earthquakes are inherently unpredictable. Geller (1997) summarised the current understanding: The idea that there must be empirical identifiable precursors before large earthquakes is intuitively appealing, but studies over the last 120 years have failed to support it. Efforts to make longer-term deterministic forecasts, as discussed... by Kagan (1997), have also been notably unsuccessful.... Recent results in non-linear dynamics are consistent with the idea that earthquakes are inherently (or effectively) unpredictable due to highly sensitive non-linear dependence on the initial conditions. Geller concludes We are overdue for a paradigm shift: it appears that the occurrence of individual earthquakes is unpredictable. Some ten years later, this conclusion still holds. Jordan (2006) writes: No practicable methodology has been demonstrated for reliable prediction of large earthquakes on time scales of decades or less. We were still waiting for Geller s paradigm shift. Crampin et al. (2008a) suggest that, the paradigm shift is to ignore the essentially unpredictable earthquake source, but instead use changes in shear-wave splitting to monitor the effects of stress on microcrack geometry in the rock mass at substantial distances from the source. This is based on the recognition that in situ rock is weak to shear stress, as witness the low stress (typically ~2 MPa) released by earthquakes, independent of the earthquake magnitude. Consequently, the enormous strain-energy released by large earthquakes must necessarily accumulate over substantial volumes of rock, probably hundreds of thousands to millions of cubic kilometres before the largest earthquakes to substantial distances from the eventual epicentre. We shall refer to this phenomenon as stress-accumulation. Crampin et al. (2008a) suggest that the required paradigm shift is to neglect the earthquake source zone and use the new understanding of low-level pre-fracturing deformation, the New Geophysics, to monitor small near-negligible changes to microcrack geometry during stress-accumulation in rocks surrounding the source zone (Crampin, 1999, 2006; Crampin et al., 2008a). Using small earthquakes as the source of shear-waves, such stress-accumulation has been recognised before some 15 earthquakes worldwide (Crampin and Peacock, 2008) where the time and magnitude, but not necessarily location, of impending earthquakes could have been estimated if the shear-wave splitting had been routinely monitored. On one occasion, when the stress-accumulation was recognised before the earthquake occurred, and the time and magnitude of a M 5 earthquake in SW Iceland were successfully estimated in real time in a comparatively narrow time-magnitude window (Crampin et al., 1999a). On this occasion, when it was known that a large earthquake was approaching, persistent low-level seismicity correctly identified the impending fault plane. We call this procedure stress-forecasting. We conclude that, with appropriate observations, we have a paradigm shift and earthquakes can be stress-forecast. These various new phenomena are reviewed and the future of stress-forecasting earthquakes discussed. This review involves ideas and concepts that may be unfamiliar to some readers. In order to simplify the main text, and make the arguments more complete without extensive reading of references, wider discussions of key phenomena are included in Appendices A to F (where Appendix F presents GEMS, the opportunity to stress-forecast all damaging earthquakes worldwide). This paper expands and justifies the arguments in Crampin et al. (2008a). 3

4 Note that the reported variations in shear-wave splitting are not earthquake precursors in the conventional sense. The variations are monitoring the effects of the accumulation of stress on the microcrack geometry pervading almost all in situ rocks. Initially, when stress begins to accumulate, from the movement of tectonic plates, say, the impending source zone is not identified. It is only when the effects of the increasing stress on microcrack geometry approach fracture-criticality and microcracks coalesce onto the eventual fault plane the stress starts to relax that the impending fault-plane is identified. The other distinction from conventional earthquake precursors is that the effects of stressaccumulation are progressive, with increasing likelihood of accuracy in magnitude, timing, and hopefully fault-plane location, with increasing time. 2. Azimuthally-varying shear-wave splitting In effectively anisotropic rocks, shear-waves split into two approximately orthogonally polarised waves that travel at different velocities and write characteristic easily-identifiable signatures into three-component seismograms which can be measured in cross-sections of the particle motion known as polarisation diagrams (PDs) or hodograms (Crampin, 1981). Azimuthally-varying shear-wave splitting is widely observed in almost all igneous, metamorphic, and sedimentary rocks, in almost all geological and tectonic regimes throughout the Earth s crust (reviewed by Crampin and Peacock, 2005, 2008). Such azimuthally-varying shear-wave splitting has been widely discussed elsewhere and will not be repeated here. However, several key phenomena relevant to this paper are summarised in Appendix A. PDs are discussed in Section A1.2. (FIGURE 1 HERE) Crampin and Peacock (2008) show that the observed azimuthally-varying shear-wave splitting in the Earth s crust is invariably caused by distributions of parallel vertical stressaligned fluid-saturated microcracks in almost all igneous, metamorphic, and sedimentary rocks in the Earth s crust. The discussion is summarised in Appendix B. Figure 1 shows a schematic illustration of shear-wave splitting in stress-aligned parallel vertical microcracks. The microcracks are grain-boundary cracks in crystalline rocks, and aligned pores and porethroats in granular sedimentary rocks. For propagation within ~35º of the vertical, the faster split shear-wave is polarised approximately parallel to the strike of the vertical microcracks which are themselves aligned approximately perpendicular to the direction of minimum horizontal stress (Crampin, 1999), similar to the opening of hydraulic fractures in seismic exploration (Hubbert and Rubey, 1959). Widespread observations of azimuthally-varying shear-wave splitting in the Earth s crust show shear-wave velocity anisotropy (SWVA) ranging from a minimum of ~1.5% to a maximum of ~4.5% in all ostensibly-intact in situ rocks (Crampin, 1994, 1996, 1999; Winterstein, 1996; Crampin and Peacock, 2005, 2008). Higher values of SWVA may be found in near-surface rocks weakened by stress-release and weathering anomalies, and in heavily fractured rocks at depth in the crust (Mueller, 1991; Crampin, 1994). Crack density can be written: = N a 3 /V; (1) where N is the number of cracks of radius a in volume V (Hudson, 1980, 1981). Since the percentage of SWVA is approximately equal to crack density x 100 (for a Poisson s ratio of 0.25), crack distributions for the observed range of SWVA (~1.5% to ~4.5%) can be illustrated schematically as uniform distributions of parallel vertical penny-shaped microcracks in Figure 2 (Crampin, 1994, 1999). Crack density in (1) is dimensionless, but since microcracks are likely to be dominated by grain-boundary cracks, the average crack size is expected to be less than average grain size (0.2 mm to 2 mm, say). Note that the 4

5 elastic or physical (or philosophical) reason why the percentage of SWVA is approximately crack density x 100, is not understood (John A Hudson, private communication). (FIGURE 2, HERE) For crack densities (SWVA ~4.5%), the microcrack distributions in Figure 2 are comparatively sparse with much solid (uncracked) matrix rock in the interstices between cracks so that the microcracked rock has some shear strength. However, as crack densities approach = 0.1 (SWVA = 10%), Figure 2 shows that microcracks become so closely spaced that shear-strength is lost and any disturbance is likely to cause the cracked rock to fail by fracturing. (At = 0.1, each crack is within a crack radius of eight other cracks in a uniform three-dimensional distribution.) The value of fracture-criticality, when fracturing and failure is likely, can be associated with the percolation threshold, which for parallel cracks is approximately = (SWVA 5.5%) (Crampin and Zatsepin, 1997) and is consistent with the fracture-criticality in Figure 2. The deformation of such compliant microcracks has been quantified by Zatsepin and Crampin (1997) and Crampin and Zatsepin (1997) in the Anisotropic Poro-Elastic (APE) model of the deformation (evolution) of fluid-saturated microcracked rock (Section A2.1). The mechanism of deformation is fluid movement by along pressure gradients between neighbouring microcracks at different orientations to the stress field. APE-modelling shows that for low-levels of deformation, below those at which rocks fail by fracturing, any change of stress modifies the internal geometry of microcrack distributions. (FIGURE 3, HERE) Figure 3 is a schematic (but quantitatively accurate) illustration of the effects of increasing low-level stress on dimensionless distributions of vertical microcracks modelled by APE. Hexagons are elastically isotropic, so the initial configuration at zero stress (top-left diagram) is a (limited) selection of randomly-aligned microcracks where, since there is no differential stress, the cracks have similar aspect-ratios. There is no effective elastic anisotropy in the top two diagrams, until stress reaches a critical value (normalised to one) when cracks normal to the increasing stress first begin to close (bottom left). Anisotropy then immediately jumps from zero to about the 1.5% minimum SWVA actually observed in the crust (Figure 2). With increasing stress, cracks normal to the stress continue to close and begin to line up parallel to the direction of maximum horizontal stress by differentially increasing crack aspect-ratios, as in Figure 3, bottom right. Note that increasing stress tends to lower crack density (there are fewer cracks bottom right, than top-left). Note also that APE and Figure 3 are modelling pre-fracturing deformation at levels of stress well below those at which fracturing (and earthquakes) occur. The principal consistent effect of increasing stress on the microcrack distributions in Figure 3 is an increase in the average microcrack aspect-ratio. This can be recognised in the field by analysing shear-wave splitting for increases in the average time-delays in Band-1 of the shear-wave window. (Shear-wave window and Band-1 are defined in Sections A1.1 and A1.5, respectively.) Variations in shear-wave time-delays showing implied increases in aspect-ratios are the major observable for monitoring changes of stress, as in the stressaccumulation before earthquakes (Crampin, 1999). 3. Monitoring stress-accumulation before earthquakes Shear-wave time-delays can be measured very accurately (Section A1.3) so that shearwave splitting can monitor very small changes to microcrack geometry during stressaccumulation (Crampin and Chastin, 2003; Crampin and Peacock, 2005, 2008; Crampin, 2006). Table A1 (Appendix A) shows that APE modelling matches a large range of different phenomena referring to cracks, stress, and shear-wave splitting. Changes in microcrack geometry in pre-fracturing deformation can be monitored by corresponding changes in shearwave splitting. 5

6 It is expected that regional stress builds up largely due to the movement of the continental plates. Due to the weakness of crustal rock to shear stress, stress-accumulation is very pervasive in the heterogeneous crust. If stress builds up over a small volume, the rate of accumulation will be relatively rapid, but the eventual earthquake will be relatively small. If stress builds up over a larger volume, the rate of accumulation will be slower, but the eventual earthquake will be larger. It is found that the logarithm of the duration of stressaccumulation is proportional to the magnitude of the eventual earthquake. Table 1 lists earthquakes and volcanic eruptions where changes in time-delays have been recognised in retrospect, with just one real time stress-forecast. These are all the known examples where swarms of small earthquakes are recorded within the shear-wave window by seismic recorders. All such data sets have shown variations whenever there are large or larger earthquakes or volcanic eruptions nearby. (TABLE 1, HERE) From the substantial distances at which the effects are observed in Tables 1 and A1 the approach to fracture-criticality is not confined to the source zone but is spread over a very extensive volume surrounding the impending epicentre. This allows the approach to fracture-criticality, when rocks fail by fracturing, to be recognised at substantial distances from the eventual epicentre. The time of the impending earthquake can then be estimated from the time the projected crack deformation (shear-wave time-delays) reaches levels of fracture criticality as indicated by previous earthquakes in the neighbouring region A successful stress-forecast Figure 4 shows variations in time-delays before a series of earthquakes, including the first successful stress-forecast earthquake, and two volcanic eruptions. The figure shows four years of time-delays (normalised to ms/km) at Station BJA in SW Iceland in two bands of the shear-wave window (Crampin et al., 1999a). The lower diagram shows timedelays in Band-1 ray-path directions sensitive to changes in crack aspect-ratios and hence to stress-accumulation (Section A1.5). The upper diagram is time-delays in Band-2 directions which are sensitive to changes in crack density, which do not appear to have a consistent response to increases in stress, and no distinctive patterns of variations have been identified in Band-2 directions. (Band-2 is also defined in Section A1.5.) The irregular curves through the time-delays are nine-point moving averages visually summarising the variations and are not used in analysis (except for identifying the start of increases in stress). The straight lines in Band-1 are least-squares fits to increasing time-delays. The least-squares lines begin one point before a minimum in the moving nine-point averages and end at a larger earthquake or eruption. (Note that the cause of the large scatter in time-delays is understood but the scatter cannot be reduced or eliminated, see Section A2.3). (FIGURE 4, HERE) The first least-squares increase in Band-1 of Figure 4 marks stress accumulation before the large Gjàlp, Vatnajökull, volcanic (fissure) eruption of October, This is followed by an approximate two year (~2 ms/km/year) decrease in average time-delays with the duration marked by a horizontal line in Band-1. This is interpreted as stress relaxation as the crust (and mantle) surrounding the Mid-Atlantic Ridge responds to the stress changes following the magma injection of the Gjàlp eruption (Volti and Crampin, 2003a, 2003b). This two-year decrease in time-delays is also visible at the three other stations in Iceland where shear-wave splitting was monitored at that time. Towards the end of the two year decrease, we had recognised, with hindsight, four increases in time-delays before earthquakes with magnitudes between M 3.8 to M 5.1 (Table 1a). Each earthquake occurred as the least-squares lines of increasing time-delays reached levels of fracture-criticality varying from about 14 ms/km in 1996 to about 10 ms/km in 1998 (allowing for the average ~2 ms/km/year decrease). The logarithm of the duration of each increase is approximately proportional, and the rate of increase inversely proportional, to the magnitude of the eventual earthquake (Figure 5a). In October, 1998, it was recognised that the slope of the current increase was similar to the slope of a previous increase before a M 6

7 5.1 earthquake some five months earlier, and a preliminary stress-forecast was issued to Iceland. On 10 th November 1998, we made the final (successful) stress-forecast. Text of exchange of messages (in italics) between University of Edinburgh (EU) and Iceland Meteorological Office (IMO) (Crampin et al., 1999a): 10 th Nov EU to IMO: the last plot is already very close to 10 ms/km. This means that an event could occur any time between now (M 5) and end of February (M 6). Three days later: 13 th Nov IMO to EU: there was a magnitude 5 earthquake just near BJA, preliminary epicenter 2 km west of BJA this morning GMT. In response to a less specific stress-forecast a few days earlier, Ragnar Stefánsson, IMO, correctly inferred that the impending earthquake would be on the fault-plane of the M 5.1 earthquake some five months earlier where low-level seismic activity was continuing. This allowed the time, magnitude, and fault-plane of the M 5 earthquake to be correctly stressforecast (Crampin et al., 1999a). (FIGURE 5, HERE) 3.2. Characteristic temporal variations Characteristic long-term temporal increases in shear-wave time-delays, monitoring stress-accumulation, have now been observed, in retrospect, before 14 earthquakes worldwide listed in Table 1a, with the one real-time stress-forecast. Figure 5a shows that logarithms of the durations of the increasing time-delays (the durations of the least-squares line increases) are proportional (self-similar) to the magnitude of the impending earthquakes. The slope of the least-squares line is thought to depend on the rate of strain increase due to movement of tectonic plates. The two outlying points: No. 15, the 1999, Ms 7.7, Chi-Chi Earthquake, Taiwan, in the highly-active Japan/ Philippine subduction zone has the fastest rate of stress accumulation; and No. 13, the 1986, Ms 6, North Palm Springs Earthquake, Southern California, in the strike-slip movement along the North American Plate, has a slower rate of stress accumulation. Note that the several different magnitude scales plotted in Figure 5 may also introduce scatter. In addition to stress-accumulation, Gao and Crampin (2004) recognised that, whenever there are sufficient shear-wave source events, the observed stress-accumulation increases in time-delays abruptly begin to decrease shortly before fracture-criticality is reached and the earthquake occurs. This stress-relaxation is interpreted as the effects of microcracks coalescing (clustering) onto the eventual earthquake fault-plane. Similar stress-increases and stress-relaxation behaviour is also observed in rock-physics stress-cells (Gao and Crampin, 2003). These short-term decreases are observed before nine of the 15 earthquakes (Table 1b), whenever there are sufficient shear-wave source events to display the variations. There are no contrary observations. Figure 5b shows that the logarithms of the durations of crack coalescence are also proportional (self-similar) to the earthquake magnitudes. In our current understanding of the phenomena, this scheme for stress-forecasting time and magnitude of impending earthquakes does not provide information about earthquake location, but knowing a large earthquake is approaching allows other precursory phenomena to be interpreted realistically. This happened with the first successful stress-forecast (previous section) where continuing small-scale seismicity correctly identified the impending fault-plane (Crampin et al., 1999a). We suggest (Crampin et al., 2008a) that monitoring stress-accumulation and the approach to fracture criticality of the rock mass is a paradigm shift that avoids the inherent unpredictability of the earthquake source. Monitoring failure of in situ rock which, in appropriate circumstances, allows the time and magnitude (but not necessarily the location) of the impending earthquake to be estimated in what we call stress-forecasting (Crampin et al., 1999a; Crampin and Peacock, 2005, 2008). Note that similar characteristic changes in shear-wave splitting are also observed before volcanic eruptions (Miller and Savage, 2001; Volti and Crampin, 2003b; Bianco et al., 2006; Crampin and Peacock, 2008), where again 7

8 stress-induced failure of surface rocks is required before magma can erupt at the surface, so that crack distributions again need to approach fracture-criticality before an eruption can occur. 4. The New Geophysics The reason that the almost parameter-less highly-constrained APE model of fluid-rock interactions (Figure 3) matches the wide variety of observations of stress, cracks, and shearwave splitting listed in Table A1 in the highly-complex heterogeneous Earth requires justification (Crampin and Chastin, 2003). The reason for this universality is that the distributions of fluid-saturated microcracks are so closely-spaced (Figure 2) that they are critical-systems verging on failure by fracturing and earthquakes. Verging on criticality is a characteristic of the New Physics (or in the case of the Earth, the New Geophysics) of critical-systems (Davis, 1989; Bruce and Wallace, 1989; Crampin, 2006). Recognition of the New Geophysics of critical-systems of fluid-saturated microcracks imposes fundamentally new properties on conventional subcritical geophysics, including: calculability and predictability (behaviour can be specifically calculated/predicted with APE); even controllability, where if effects can be predicted, the behaviour can be controlled by feedback; universality (APE-modelling applies to all or almost all in situ rock); extreme sensitivity (the highly sensitive non-linear dependence on the initial conditions, Geller, 1997, and hence the deterministic unpredictability of the earthquake source); and deterministic chaos (implied, for example, by the self-organized criticality of the Gutenberg- Richter plots of logarithm of frequency and earthquake magnitude). These new properties seem to be observed in the crust (Crampin and Chastin, 2003; Crampin and Peacock, 2005, 2008; Crampin, 2006). Note that the reason the New Geophysics was not recognised earlier is that the key observable is temporal changes in shear-wave splitting, and the full understanding of shear-wave splitting is only now becoming recognised (Crampin and Peacock, 2008). Appendix C discusses the New Geophysics. We suggest that the required paradigm shift is a result of this New Geophysics, where the new understanding of fluid-rock deformation, both demonstrates the unpredictability of the earthquake source and allows the conditions for stress-accumulation and earthquake occurrence to be monitored by shear-wave splitting. 5. Stress-Monitoring Sites (SMSs) We have shown that stress-forecasting appears to be the paradigm shift for predicting earthquakes, where the critical procedure is monitoring the effects of stress-accumulation on compliant microcrack geometry by analysing shear-wave splitting away from the impending epicentre (see Table 1), and ignoring the seeming unpredictability of the earthquake source zone. This was demonstrated using swarms of small earthquakes as the source of shearwaves. Although the principles of stress-forecasting can be demonstrated using seismic swarms as the source of shear-waves, seismic swarms are far too scarce and far too irregular (Wu et al., 2006) to be used for routine reliable forecasting of all earthquakes. A monitoring system using a controlled-source of shear-waves is required that can be sited where necessary and is deep enough to be below the near-surface stress-release and weathering heterogeneities and below the critical depth so that microcracks have uniform orientations perpendicular to the direction of minimum horizontal stress. This critical depth, where the vertical stress, V, equals the minimum horizontal stress, h, is typically between 500 m and 1000 m (Crampin, 1999). Monitoring the effects of increasing crack aspect-ratios and stress-forecasting earthquakes requires time-lapse observations of shear-wave splitting along ray paths in Band-1 directions sensitive to changes in crack aspect-ratios (Crampin, 1999, 2001, 2006; Crampin and Peacock, 2005, 2008). 8

9 Figure 6 shows the optimum configuration for a borehole Stress-Monitoring Site (SMS). A fixed Downhole Orbital Vibrator (DOV) (Leary and Walter, 2005a, 2005b) in a sourcewell radiates SV-waves downwards in Band-1 directions to three-component geophones in two receiver-wells offset 300 m with azimuths 30º either side of the direction of minimum horizontal stress. The DOV is an eccentric cam rotated within a cylindrical canister hanging freely in a fluid-filled borehole that exerts a radial force on the borehole wall, generating both P-waves and shear-waves. Note that shear-wave transmission from the DOV to the bore-hole receivers needs to in a homogeneous rock without significant horizontal interfaces. Interfaces are likely to reflect and refract the shear-wave ray-path and prevent or seriously attenuate shear-wave signals reaching the receivers. (FIGURE 6, HERE) Note also that the geometrical configuration in Figure 6 replaces an earlier version, where a deeper DOV radiates shear-waves upwards to shallower geophones (Crampin, 2001). Geophones are less constrained by temperature than the electrical motors of the DOV, and it is preferable to have the DOV source at the shallower cooler ends of the ray paths, as illustrated in Figure The prototype SMS at Húsavík, Northern Iceland The prototype SMS was installed in existing boreholes drilled for geothermal purposes, adjacent to the Húsavík-Flatey Fault (HFF) in Northern Iceland. The HFF is a transform fault of the Mid-Atlantic Ridge (MAR), which runs onshore near the fishing port of Húsavík. Funded by the European Commission, the SMS of the SMSITES Project transmitted P-, SV-, and SH-waves horizontally from a DOV at ~500 m-depth between two boreholes, offset 315 m, parallel and approximately 100 m south of the surface break of the HFF (Crampin et al., 2003). Also recorded were variations in water-level in a well on the Island of Flatey immediately above the HFF, and variations in NS and EW Global Positioning System (GPS) displacements spanning the HFF. (FIGURE 7, HERE) In what was intended to be a calibration test of the DOV, the DOV was pulsed (seismic chirps) every s and stacked after 100 sweeps, with only minor interruptions, for 13 days continuously in August 2001, and recorded the spectacular sensitivity in Figure 7 (Crampin et al., 2003). The azimuthal direction of the offset, parallel to the fault break, is approximately in a stress symmetry direction so that shear-waves split into SV- and SHpolarised waves. Although the seismic ray paths were not in optimal source-to-geophone geometry for maximal sensitivity (no Band-1 data), the seismic arrivals, P-, SV, SH-wave, and shear-wave anisotropy, travel-time anomalies, showed exceptional sensitivity. The onset of the seismic variations also coincided with the onset of a one metre fall in level in the water well on Flatey lasting five days, and with North-South (7 mm) and East-West (4 mm) Global Positioning System (GPS) displacements across and parallel to HFF. All seven geophysical measurements coincided with minor seismic swarm activity lasting some 60 hours on the Grímsey Lineament, where the Grímsey Lineament is another MAR transform fault. The total energy release of the warms was equivalent to one M ~3.5 earthquake. The swarm activity was offshore, approximately 70 km NNW of the SMS (Crampin et al., 2003). The observations in Figure 7 are described in more detail in Appendix D and listed in Table D1. This prototype SMS recorded shear-waves in only one horizontal direction. Since the direction was in a plane of vertical anisotropic symmetry, shear-waves split into purely SV and SH-wave polarisations, and carried minimal anisotropic information. In particular, the seismic measurements at ~500 m-depth are probably not deep enough to reach the parallel vertical microcrack orientations expected below the critical depth where V = h. Consequently, the seismic rays are believed to be propagating in a symmetry direction parallel to the strike of the HFF and within near-surface stress release and weathering anomalies. This means there is little constraint on interpretation and little information about the mechanism of the stress changes can be extracted. This was not the optimal geometry 9

10 most sensitive to changes behaviour causing the anomalies. Never-the-less, Figure 7 does confirm that SMSs have the great sensitivity required for monitoring low-level changes in rock mass conditions and stress-forecasting impending earthquakes The sensitivity of SMSs for monitoring stress-accumulation before earthquakes There were 106 earthquakes in the two-and-a-half-day seismic swarm on the Grímsey Lineament correlating with the other measurements in Figure 7 and Table D1. The largest magnitude in the swarm was M 2.8, making a total energy release approximately equivalent to one M ~3.5 earthquake. A M 3.5 earthquake is a comparatively small event with a total conventional source diameter of at most a few hundred metres. Thus the seismic measurements in Figure 7, at distances of ~70 km, were recorded at hundreds of times the likely source dimensions and well beyond that expected in conventional sub-critical geophysics. This demonstrates the extreme sensitivity expected from the crack-critical nature of the crust. Analysis of the sensitivity suggests that stress variations before a M 3.5 event have a surface extent with a radius greater than 70 km = ~15,000 km 2. Since an increase of n in magnitude increases energy by a factor of (10 n ) 3/2 (Choy and Boatwright, 1995), a M 5 earthquake will have approximately 180 times more energy than the M ~3.5 earthquake. Assuming that stress-accumulation before crustal earthquakes is largely confined to the crust, if the energy of an M 5 earthquake is distributed with the same energy density as the equivalent M ~3.5 earthquake in Iceland, the radius of the surface area is ~950 km from the earthquake. Thus conservatively, the effects before a M 5 earthquake would be visible on a seismic network on a grid spacing of conservatively 1200 km where the entire surface is less than 950 km from a SMS. Appendix E suggests possible applications of SMSs, and Appendix F presents GEMS: the opportunity for stress-forecasting all damaging earthquakes worldwide. 6. Conclusions The New Geophysics implies that in situ rock is so weak to shear-stress that the enormous strain-energies released by large earthquakes must necessarily accumulate over enormous volumes of rock. This process is referred to as stress-accumulation. Table 1 suggests that the effects on microcrack geometry of the build-up of stress, and stress relaxation as the earthquake occurs, can be observed at substantial distances from the eventual epicentre by monitoring shear-wave splitting. The observations listed in Table 1 and Table A1 indicate that there are a large number of phenomena consistent with these ideas. There are no known contrary observations. Pervasive stress-accumulation clearly has major implications for earthquake prediction research, yet apart from the current investigations, reviewed in this paper, the effects of extensive stress-accumulation appear to have been largely ignored in previous earthquake research. We suggest that the reason for this is that, before the New Geophysics was recognised, there was no easy way to assess stress changes in the deep interior of the crust. Changes in stress can only be easily recognised now that stress-induced changes to microcrack geometry can be monitored with shear-wave splitting (Crampin, 1994, 2006; Crampin and Chastin, 2003; Crampin and Peacock, 2005, 2008). It is interesting to note that the cause of the inherent unpredictability of earthquakes (Geller, 1997; Leary, 1997), the New Geophysics, also provides the universality and calculability that allows the geometry of microcracks to be monitored by shear-wave splitting and approach of fracture-criticality and earthquakes stress-forecast. We suggest that it may be unique amongst critical-systems that the approach of singularities and critical-points within the interior of a critical-system can be monitored by comparatively non-intrusive techniques such seismic shear-wave splitting. Although it has been sometimes been claimed that small earthquakes can be triggered by the passage of seismic waves, the effect of 10

11 disturbances due to the passage of seismic waves has not been has not yet been identified in shear-wave splitting. Table 1 indicates that, whenever there is a suitable seismic swarm nearby, the time, magnitude, and in optimal circumstances fault-plane, of an impending larger earthquake can be stress-forecast. Unfortunately, seismic swarms are far too scarce and irregular for reliable routine stress-forecasting, and controlled-source borehole Stress-Monitoring Sites (SMSs) are required. The prototype SMS in Iceland, although not in optimal source-to-receiver geometry, recorded exceptional sensitivity in Figure 7 and Table D1 confirming the science, technology, and sensitivity of SMSs for stress-forecasting the time, magnitude, and in optimal circumstances fault-plane, of impending earthquakes. Evidence suggests that a single three-borehole SMS could stress-forecast all M 5 (that is all damaging earthquakes) within 950 km of the SMS (Section 5.2). This appears to be the necessary paradigm shift for stress-forecasting earthquakes. The time, magnitude, and epicentre of impeding earthquakes cannot be predicted by investigating the unpredictable source zone, but as we have shown above, by monitoring the pervasive stress-accumulation in the crust, the approach to fracture-criticality can be recognised (possibly at considerable distance from the eventual epicentre), so that the time, magnitude and, at least in some circumstances, fault-plane can be stress-forecast. Appendix E discusses possible applications of SMSs. Our confidence that we can stress-forecast earthquakes after 120 years of failure to predict earthquakes (Geller, 1997; Geller et al., 1997; Kagan, 1997; Jordan, 2006) requires some justification. Our confidence is based on the paradigm shift (Crampin et al., 2008a) that the stress-accumulation before all impending large earthquakes modifies microcrack geometry throughout substantial volumes of in situ rock around the impending epicentre, and that this can be monitored by analysing shear-wave splitting. There a large number of observations in Table 1 that support this New Geophysics in the cited references and appendices, with no contrary indications. Monitoring the low-level deformation of the rock mass and the approach of stress-accumulation to fracture-criticality and failure, avoids the inherent unpredictability of the earthquake source. Except for hydraulic fractures, all large fractures begin as the coalescence of microcracks. Consequently, our confidence in the paradigm shift is also because we are monitoring what is probably the underlying mechanism of failure of in situ rock in all earthquakes (and probably all volcanic eruptions): the necessary coalescence of microcracks into fault-planes, fissures, and fractures (Crampin and Peacock, 2008). This means that the techniques are likely to have generality and applicability (the New Geophysics universality) and not be overly dependent on local geological and tectonic circumstances. The outline in Appendix F for GEMS, a Global Earthquake Monitoring System, for a global network of Stress-Monitoring Sites to forecast all damaging earthquakes worldwide, would be expensive. After 120 years of failure in earthquake prediction, progress requires a large-scale team with a large-scale approach, perhaps somewhat analogous to the Large Hadron Collider required for progress in high-energy physics. GEMS would provide the data for the holistic long-range worldwide approach as advocated by Evison (2001). Acknowledgements We acknowledge the valuable contributions and gratefully thank all our numerous coauthors in the cited papers for their collaboration in the ideas contained in this review. GY was partly supported by National Natural Science Foundation of China project This work is published with the approval of the Executive Director of the British Geological Survey (NERC). List of Appendices 11

12 Appendix A: A summary of azimuthally-varying shear-wave splitting in stress-aligned fluid-saturated microcracks Appendix B: The cause of azimuthally-varying shear-wave splitting Appendix C: The New Geophysics - a new understanding of fluid-rock interaction/ evolution Appendix D: The Prototype SMS at Húsavík, Iceland: the SMSITES Project Appendix E: The potential of SMSs for stress-forecasting earthquakes and volcanic eruptions Appendix F: GEMS: the opportunity for stress-forecasting all damaging earthquakes worldwide 12

13 Appendix A A summary of azimuthally-varying shear-wave splitting in stress-aligned fluid-saturated microcracks Shear-wave splitting (seismic birefringence) in the Earth s crust has been extensively observed and discussed elsewhere, with recent reviews by Crampin and Chastin (2003) and Crampin and Peacock (2005, 2008). Here, we present a brief summary of details relevant to this review. On propagating through anisotropic rock, shear-waves split into two approximately orthogonal polarisations which travel at different velocities and write characteristic signatures into the seismic wave-trains (Crampin, 1981). These signatures are most easily identified in three-orthogonal cross-sections of the particle motion known as polarisation diagrams, PDs, particle-motion diagrams, or hodograms (see Section A1.2, below). Azimuthally-varying shear-wave splitting is widely observed in the Earth s crust where the horizontal projection of the faster split shear-wave is generally aligned approximately in the direction of maximum horizontal compressional stress (Crampin, 1994, 1996, 1999; Winterstein, 1996). Although in the past there has been controversy as to the cause of azimuthally-varying shear-wave splitting, the review by Crampin and Peacock (2008) demonstrates that, although in principle shear-wave splitting can be caused by a range of phenomena, the parallel stress-aligned polarisations observed at surface receivers are invariably caused by the fluid-saturated stress-aligned microcracks in most rocks of the crust (these arguments are summarised in Appendix B). There are no confirmed exceptions. Below the critical-depth where increasing vertical stress, V, equals the minimum horizontal stress, h, cracks are generally aligned vertically, approximately parallel to the direction of maximum horizontal stress, although there are approximately-orthogonal exceptions (Section A1.4). Parallel vertical microcracks result in transverse isotropy (hexagonal anisotropic symmetry) with a horizontal axis of symmetry (TIH- or HTI-anisotropy), or a minor variation thereof. Note the caveat azimuthally-varying shear-wave splitting. Horizontally-stratified structures display azimuthally-invariant shear-wave splitting caused by: periodic thin layers (Levin, 1979, 1980); horizontal platelets of mica in shales and clays (Kaarsberg, 1959, 1968); and horizontal layering in the lower crust and mantle (Dziewonski and Anderson, 1981); etc. These phenomena lead to transverse isotropy with a vertical axis of symmetry (TIV- or VTI-anisotropy), where all shear-waves split into strictly SH- or SV-wave polarisations, and there is no azimuthal variation. TIV-anisotropy is commonly observed in sedimentary basins by exploration seismologists. This paper refers only to the more widely observed TIH-anisotropy. Note that sloping layers may also show azimuthal variations in shear-wave splitting but will not show the parallelism of TIH-anisotropy (Tsvankin, 2001). In this appendix, we summarise a number of crucial features of shear-waves and shearwave splitting in both conventional geophysics, and what has been called the New Geophysics where shear-wave splitting shows that fluid-saturated microcracks in most in situ rocks are so closely spaced they are critical-systems (Crampin, 1994, 2006), with all the fundamentally new properties that that implies. The New Geophysics is discussed in Appendix C. A1. Shear-wave splitting in conventional geophysics A1.1. The shear-wave window The shear-wave window is the solid angle of incidence at a (horizontal) free-surface with angular radius sin -1 Vs/Vp (approximately 35º, for a Poisson s ratio of 0.25). For ray paths incident within the window, the apparent horizontal velocity of shear waves is too high for energy to be lost to S-to-P conversions at the free-surface, and the recorded signals at the 13

14 surface repeat the incident waveforms (but with double the amplitude). For shear-wave arrivals outside the window, the energy lost to S-to-P conversions is so great that incident SV-waves cannot be directly observed. These and associated disturbances (Crampin, 1990) mean that interpretable shear-waves and shear-wave splitting are necessarily confined to incidence within the shear-wave window (Nuttli, 1961; Evans, 1984; Booth and Crampin, 1985). Note that horizontally-polarised SH-waves are not affected by the shear-wave window, and disturbances caused by shear-wave windows only apply to arrivals with an SVcomponent of motion. Low-velocity near-surface layers due to weathering and stress-release phenomena frequently result in ray paths curving upwards towards the free surface. As a result, the effective shear-wave window can frequently be extended to straight-line source-to-receiver incidence angles of 45º or 50º. In practice, this does not cause problems, as anomalous polarisations are easily recognised in polar maps of recorded polarisations. As the shear-wave window is controlled by the angle of incidence at the free-surface, the effects of the window are extremely sensitive to topography. Irregular surface (and subsurface) topography severely distorts observations of shear-waves and shear-wave splitting. Since small earthquakes are typically beneath hilly or mountainous terrains, selecting suitably flat surface locations for seismic stations recording shear-wave splitting above small earthquakes is frequently difficult. The effects of irregular topography typically lead to scattered polarisations (Crampin, 1990), which are easily and frequently misinterpreted (Crampin and Peacock, 2008). A1.2. Identifying shear-wave splitting in Polarisation Diagrams (PDs) The human eye has little ability to recognise polarisation phenomena in three-component wiggly-line seismograms (Crampin and Peacock, 2008). However, shear-wave splitting typically writes characteristic easily-recognisable signatures into three mutuallyperpendicular cross-sections of three-component particle motion, known as polarisationdiagrams (PDs, particle-motion diagrams, or hodograms). As long as the seismograms contain sufficiently high frequencies (or high-frequency pass-bands are set high enough), split shear-wave arrivals result in abrupt easily-recognised angular changes in particle motion direction. For nearly vertically propagating shear waves, PDs of the horizontal plane are particularly significant. Accurate time-delays between split shear-waves can be measured by counting samples in horizontal PDs between these abrupt changes. The azimuth of polarisation is the direction of particle motion during the time-delay (Crampin, 1999). The accuracy of the measurement is limited by the sampling rate. Unfortunately, although accurate, plotting and visually analysing and measuring PDs is tedious and time consuming. Indeed, some authors have used the tedium of accurately measuring PDs as an excuse ( cannot afford the time!) for using less-accurate automatic techniques for measuring time-delays. Fully automatic techniques have been attempted but are not wholly successful without very rigorous data rejection (typically eliminating between 40% and 70% of the data) (Crampin and Gao, 2006). Gao et al. (2006) have developed a user-friendly semi-automatic technique, the Shear-Wave Analysis System, SWAS, which switches between screen images of various seismogram displays and PDs. Figure A1 shows screen images of seismograms and PDs of the SWAS semi-automatic measurement of timedelays and polarisations of a M 0.55 earthquake recorded at Station BJA in SW Iceland. Measuring shear-wave splitting parameters with SWAS appears to be successful for more than 80% of shear-wave splitting arrivals within the shear-wave window in Iceland (Gao et al., 2006; Hao et al., 2008). (FIGURE A1, HERE) Note that polarisations of split shear-waves travelling at the group velocity along ray paths are only strictly orthogonal in particular symmetry directions (Crampin, 1981). Since the polarisations of the two split shear-waves, in horizontal PDs, say, are horizontal projections of the particle motion, the observed angular change in PDs may be greater or 14

15 smaller than the true angular difference depending on the polarisations of the shear-waves and the angles of incidence of the ray paths. A1.3. Resolution of measurements of shear-wave velocity anisotropy (SWVA) Shear-wave splitting is the result of small second-order differences in the travel times and polarisations of two shear-waves travelling along similar ray paths, which is difficult to measure in three-component seismograms with arbitrary orientations. Analysis of PDs provides highly accurate measurements of time-delays and polarisations. More conventional time-series analysis can be made by rotating three-component seismograms into the two anisotropic shear-wave polarisations, as in Figure A1. Arrivals are then (approximately) separated into isolated signals and the arrival times and polarisations of SWVA can be measured with the similar first-order accuracy as that of P-wave arrivals on the vertical component. Second-order quantities read to first-order accuracies typically lead to much higher resolution than is generally available in most seismic measurements. Note that it is only possible to separate split shear-wave arrivals into independent noninterfering wave trains by rotation of coordinate axes when the polarisations are strictly orthogonal. Since split shear-wave polarisations in rays travelling at the group velocity are seldom strictly orthogonal (see previous section), the separation is usually only approximate and may lead to difficulties in measuring particularly second arrivals of the slower split shear-wave. In all cases, counting samples in PDs, either manually, or automatically in the semi-automatic SWAS technique, as illustrated in Figure A1 (Gao et al., 2006), is likely to lead to the most accurate measurements. A1.4. Occasional orthogonal polarisations Although polarisations of the faster split shear-waves observed within the shear-wave window at the free surface are usually more-or-less parallel to the direction of maximum horizontal stress, occasionally approximately orthogonal polarisations are observed. There are four principal causes. Firstly, if the source, such as a small earthquake, radiates shear-waves in a particular direction only with polarisations parallel to the slower split shear-wave, then only slower waves will be excited and the faster split shear-waves will be excluded. This can result in 90º-flips with polarisations orthogonal to the direction of maximum horizontal stress. Secondly, on propagation through a uniform distribution of parallel vertical microcracks, there is a line singularity at an angle of about 30º to the normal to the crack plane, where the two shear-wave velocity sheets intersect and the faster and slower split shear-waves exchange polarisations leading to 90º-flips in polarisations of the faster split shear-wave (Crampin and Yedlin, 1981; Crampin, 1991, 1999). Thirdly, the behaviour of shear-waves travelling along ray paths near point- and linesingularities can lead to 90º-flips in shear-wave polarisations. Propagation of body-waves along ray paths is at the group velocity, and this may lead to abrupt 90º-flips for very small changes in direction of propagation (Crampin and Yedlin, 1981; Crampin, 1991). Fourthly, in the presence of critically-high pore-fluid pressures, local microcrack orientations become orthogonal to orientations in normally pressured cracks. This occurs: on all seismically-active faults which always appear to be permeated by high pore-fluid pressures leading to the pronounced ±80% scatter in shear-wave time-delays (Crampin et al., 2002, 2004); in critically-high-pressure fluid-injections in exploration geophysics (Angerer et al., 2002); and in critically-high-pressure hydrocarbon reservoirs (Crampin et al., 1996). These various phenomena all result in 90º-flips in shear-wave polarisations, where the faster and slower split shear-waves exchange polarisations (Crampin et al., 2002). This phenomenon is a feature of the New Geophysics (see Section A2.2 and Appendix C.) A1.5. Monitoring small changes of stress and Band-1 and Band-2 geometry Peacock et al. (1988) recognised intuitively that small increases of low-level stress would cause stress-aligned microcracks to swell (increase in aspect-ratio perpendicular to the 15

16 direction of minimum compressional stress). This was confirmed theoretically and numerically (Crampin and Zatsepin, 1997) with the APE model of fluid-rock deformation (Figure 3, and Section A2.1). Such changes in aspect-ratio are for stress changes well-below levels of stress at which fracturing occurs. These small increases in stress in pre-fracturing deformation can be monitored by observing increases in the average time-delays in Band-1 directions of the shear-wave window (Crampin, 1999). Band-1 of the shear-wave window is the two-leafed solid angle making angles ±(15º to 45º) either side of the vertical plane of the stress-aligned parallel vertical microcracks (Crampin, 1999). The logarithms of the durations of the increase in time-delays are proportional (self-similar) to the magnitudes of the impending events (Figure 5a). Figure A2 is a schematic illustration of the source-receiver geometry of Band-1 and Band-2 directions in the shear-wave window. (FIGURE A2, HERE) However, not only is there an increase in Band-1 time-delays, interpreted as monitoring stress-accumulation before earthquakes, but also immediately before the impending earthquake there is an abrupt change from increasing to decreasing aspect-ratios, interpreted as a stress-relaxation as microcracks begin to coalesce onto the eventual fault-plane (Gao and Crampin, 2004). The logarithms of the duration of the decreases are also (separately) proportional to the magnitudes of the impending events (Figure 5b). Note that Band-2 is the central band of the shear-wave window, ±15º either side of the plane of the cracks (Figure A2). Time-delays in Band-2 vary with crack density, but crack density does not have a simple relationship to changes of stress, and may decrease with increasing stress as in Figure 3 (Crampin, 1999). A2. Shear-wave splitting in the New Geophysics Figure 2 demonstrates that distributions of fluid-saturated stress-aligned microcracks in the Earth s crust are so closely spaced they verge on fracture-criticality and failure by fracturing and earthquakes, and hence are critical-systems. Critical-systems impose fundamentally different properties on the conventional brittle-elastic crust, including calculability, predictability, universality, and extreme sensitivity to initial conditions leading to deterministic chaos (Crampin and Chastin, 2003; Crampin and Peacock, 2005). These new properties are subtle but fundamental, and amount to a New Geophysics with a new understanding of fluid-rock deformation, as discussed in Appendix C. A2.1. The Anisotropic Poro-Elastic (APE) model of deformation/evolution of stress-aligned fluid-saturated microcracks Observations of azimuthally-varying shear-wave splitting show that most igneous, metamorphic, and sedimentary rocks in the Earth s crust are pervaded by distributions of stress-aligned fluid-saturated microcracks (Crampin, 1994, 1996, 1999; Winterstein, 1996; Volti and Crampin, 2003a, 2003b; Crampin and Peacock, 2005, 2008). These microcracks are the most compliant elements of in situ rock, and Zatsepin and Crampin (1997) and Crampin and Zatsepin (1997) developed the Anisotropic Poro-Elastic (APE) model of deformation/evolution of stress-aligned fluid-saturated microcracks under changing conditions. Fluid-saturated microcracks respond to changes of stress (and other phenomena) by modifying microcrack geometry at low-levels of stress, well below levels at which failure by fracturing occurs. The mechanism for such pre-fracturing deformation is fluid movement by flow or dispersion along pressure gradients between neighbouring grain-boundary cracks in crystalline rocks and aligned pores and pore-throats in sedimentary rocks at different orientations to the stress field. The behaviour for small increases in stress is illustrated schematically but quantitatively in Figure 3 (Crampin and Zatsepin, 1995, 1997). APE-modelling is dimensionless, but in the crust the mechanism involves micro-scale (<2 mm, say) movements of fluid around grain boundaries and does not require large-scale fluid motion. Consequently, the response of the microcracks to changing stress is almost 16

17 immediate (Crampin and Zatsepin, 1997) as is observed in both the field (Angerer et al., 2002; Volti and Crampin, 2003b; Gao and Crampin, 2004), and laboratory (Gao and Crampin, 2003). Numerical modelling with APE approximately matches some twenty different phenomena (Table A1) referring to cracks, stress, and shear-wave splitting. The effects on shear-wave splitting have been observed in millions of records along individual source-toreceiver ray paths in exploration seismology (Helbig and Thomsen, 2005) and above small earthquakes (Crampin and Chastin, 2003; Crampin and Peacock, 2005, 2008). (TABLE A1, HERE) Note however that the match of APE to observations is only approximate because the detailed behaviour of rocks at depth can seldom if ever be directly assessed in situ. For example, measurements in boreholes are likely to be severely disturbed by anomalies caused by stress release, temperature changes, and drill-fluid invasion. Observations of Holmes et al. (2000) show directly that stress-release anomalies extend to at least three times the borehole diameter. Other observations suggest the anomalies extend to much greater distances (Crampin et al., 2003). The most accurately calibrated test of APE is where Angerer et al. (2002) used actual injected pressures in APE to match almost exactly the behaviour of shear-wave splitting in multi-component reflection profiles before and after both critically high- and low-pressure CO 2 -injections in a carbonate reservoir. Angerer et al. (2002) matched variations in time-delays and 90º-flips in shear-wave polarisations. Note also that an underlying assumption of APE is that microcracks are so closely spaced in all in situ crustal rocks that they verge on failure by fracturing and consequently are critical-systems, with all the implications implied by the New Geophysics (Appendix C). A º-flips above large seismically active faults Rose diagrams of shear-wave polarisations are usually polarised approximately parallel to the maximum horizontal stress, which is typically between 60º and 90º to the strike of seismically-active faults. Fault-parallel shear-wave polarisations have been observed at two faults: at two locations above the San Andreas Fault (SAF), California, by Peacock et al. (1988) and Liu et al. (1997); and at three stations above the Húsavík-Flatey Fault (HFF), Iceland (Crampin et al., 2003), where HFF is a major transform fault of the Mid-Atlantic Ridge running onshore in Northern Iceland. Using APE to model the behaviour, Crampin et al. (2002, 2004) showed that the most likely explanation of the orthogonal polarisations is that critically-high pore-fluid pressures on all seismically-active faults interact locally with the tectonic stress (normally at a high angle to the fault plane) and reorient microcracks parallel to the fault in the immediate vicinity of the fault plane. If normally pressurised microcracks result in positive, say, shearwave time-delays, the re-oriented microcracks result in negative delays and lead to the observed 90º-flips in shear-wave polarisations. SAF and HFF are major faults traversing a large part of the crust. Consequently, ray paths through the normally pressured microcracks, with positive time-delays, above the fault plane are too short to reverse the negative time-delays established by 90º-flips in the immediate fault zones, and 90º-flips are observed at the surface. A2.3. The scatter in shear-wave splitting time-delays above small seismically-active faults Observations of shear-wave splitting time-delays in the shear-wave window above small earthquakes invariably display a large ±80% scatter, as in Figure 4. It contrast, time-delays in seismic reflection profiles and VSPs in exploration seismic using controlled seismic sources vary smoothly and display negligible scatter (Li and Crampin, 1993; Yardley and Crampin, 1993; and many other examples). Thus the scatter appears to be associated only with earthquake source zones. The scatter is far too large and too consistently present to be caused by conventional geophysical effects, such as inadequate knowledge of geological 17

18 structure and earthquake locations, inaccurate measurements, and complicated anisotropic structure above the earthquake or beneath the station, etc. (Volti and Crampin, 2003a). The previous section showed that critically-high pore-fluid pressures on large seismically-active faults caused 90º-flips in shear-wave splitting at surface receivers above the fault (Crampin et al., 2002). Crampin et al. (2004), again using APE modelling, showed that high pore-fluid pressures on smaller faults, with longer normally-pressured ray paths above the fault to surface receivers, would cause the positive time-delays to generally exceed the 90º-flipped negative time-delays in the immediate vicinity of the fault. Consequently, positive time-delays are generally observed at the surface, but small differences in the ratio of positive to negative path lengths can easily produce the observed ±80% scatter in timedelays (Crampin et al., 2004) as seen in Figure 4. Note that high pore-fluid pressures are expected on seismically-active faults to relieve friction and override asperities (Sibson, 1981, 1990). Thus the universally-observed scatter in shear-wave time-delays demonstrates explicitly that high pore-fluid pressures are present on virtually all seismically-active faults. A2.4. Temporal versus spatial variations in shear-wave splitting As usual with observations of shear-wave splitting and seismic anisotropy above small earthquakes, it is difficult to attribute the source of the variations to particular segments of the ray paths, when geological structures, seismic velocities, and source locations are all subject to significant uncertainties. Consequently, interpretations in terms of temporal and spatial variations are usually ambiguous. For example, Crampin and Gao (2005) recognised temporal variations of both stressaccumulation increases and crack-coalescent decreases in measurements of shear-wave splitting by Liu et al. (2004) before the 1999 Chi-Chi Earthquake, Taiwan. This was despite the claims of Liu et al. (2004) that there were no temporal variations (Crampin and Peacock, 2008). The inherent ambiguity means that the assertion by Liu et al. (2005) that the characteristic variations in shear-wave time-delays, as recognised by Crampin and Gao (2005), are caused by spatial variations rather than temporal variations is difficult to specifically refute without detailed structural information which is never available. However, the variations in Band-1 time-delays before the Chi-Chi Earthquake identified by Crampin and Gao (2005) have similar behaviour and similar self-similar relationships with earthquake magnitudes as those observed before the other earthquakes listed in Table 1 and Figure 5. We suggest it is more consistent to attribute the variations in stressaccumulation and crack coalescence to temporal effects, which appear to be a universal feature before all larger earthquakes (as listed before 15 earthquakes in Table 1), than it is to suppose that the foci of all seismic swarms have (unexplained) spatial patterns of migration of foci displaying similar variations of time-delays at arbitrarily located seismic stations (Crampin and Peacock, 2008). In addition, Crampin and Peacock (2008) show that azimuthally-aligned shear-wave splitting is invariably caused by distributions of stress-aligned fluid-saturated microcracks. Fluid-saturated microcracks are the most compliant elements of the rock mass (Figure 3, Crampin and Zatsepin, 1997). Consequently, temporal variations in time-delays must be expected during the stress-accumulation before large earthquakes. As usual with shear-wave splitting and seismic anisotropy it is easy to make judgements from a limited data base and inadequate knowledge, and assume they are more widely applicable than is appropriate. Correct interpretation requires an understanding of as many observations as are available. In a review of almost all observations of shear-wave splitting above small earthquakes, Crampin and Peacock (2008) identified 17 fallacies or misunderstandings commonly held about shear-wave splitting. 18

19 Appendix B The cause of azimuthally-varying shear-wave splitting There have been various suggestions that the widely observed shear-wave splitting is the result of: mineral alignments (Aster and Shearer, 1992); foliated schists (Brocher and Christensen, 1989); orientations of magmatic dykes (Gerst and Savage, 2004); joints, fractures, and macro-cracks (Barton, 2007); and shallow near-surface structural anisotropy (Aster and Shearer, 1992; Zhang and Schwartz, 1994; Munson et al., 1995); amongst others. These various causes are discussed (and rejected) by Crampin and Peacock (2008) in a critical review of all observations of azimuthally-varying shear-wave splitting and a discussion of common fallacies in interpretation of shear-wave splitting. Crampin and Peacock (2008) identify seven phenomena (listed in Table B1) suggesting that the source of the azimuthally-varying splitting is invariably stress-aligned fluid-saturated grain boundary cracks in crystalline rocks and aligned pores and pore-throats in granular sedimentary rocks. (TABLE B1, HERE) In particular, three critical features directly indicate that the cause of the azimuthallyvarying shear-wave splitting is unquestionably fluid-saturated stress-aligned microcracks (Crampin and Peacock, 2008). 1) Surface observations of the polarisations of the leading (faster) split shear-wave are approximately parallel (or occasionally orthogonal) to the direction of maximum horizontal stress in the shear-wave window (Peacock et al., 1988; Volti and Crampin, 2003b; Crampin and Peacock, 2005, 2008). The only anisotropic symmetry system with such parallel polarisations in a broad band across the centre of the shear-wave window is hexagonal symmetry (transverse isotropy) oriented with a horizontal axis of cylindrical symmetry (TIHanisotropy) or a minor variation thereof (Crampin, 1981, 1999). 2) The only common phenomenon in igneous, metamorphic, and sedimentary rocks that possesses such TIH-anisotropy is fluid-saturated stress-aligned microcracks (Crampin and Peacock, 2008), which take up parallel vertical orientations, perpendicular to the direction of minimum compressional stress which is typically horizontal (Hubbert and Rubey, 1959). 3) This interpretation is confirmed by observations of temporal variations in time-delays between the split shear-waves before larger earthquakes (listed in Table 1), and during fluidinjection in hydrocarbon reservoirs (Angerer et al., 2002). Only fluid-saturated microcracks have such immediate compliance to small changes in stress (Zatsepin and Crampin, 1997; Crampin and Zatsepin, 1997), so that fluid-saturated microcracks are confirmed as the source of the shear-wave splitting. Occasional orthogonal polarisations are discussed in Section A1.4. Such 90º-flips in shear-wave polarisations are comparatively common, and may be caused by: polarisations of the faster split shear-wave not radiated from the source; observations along ray paths either side of line-singularities in either TIH or TIV anisotropy; by critically-high pore-fluid pressures re-aligning microcracks in hydrocarbon reservoirs (Crampin et al., 1996); and in the neighbourhood of seismically-active fault-planes (Crampin et al., 2002). Although each of the various mechanisms, listed in the first paragraph of this section, could in theory cause variations in azimuthally-varying shear-wave splitting, each mechanism requires very particular geological, tectonic, and geometrical conditions. Worldwide observations of many thousands of azimuthally-varying shear-wave splitting along ray paths above earthquakes and millions of ray paths in seismic exploration all displaying approximately similar polarisations and similar ranges of (normalised) time-delays, strongly suggest a similar source of fluid-saturated microcracks (Crampin and Peacock, 2008). However, despite the incontrovertible evidence for microcrack-induced shear-wave splitting listed in Table B1, controversy still persists (Barton, 2007). As we have said before, it is unwise to make judgements from a limited data base and assume they are more widely applicable than is appropriate. 19

20 Appendix C The New Geophysics - a new understanding of fluid-rock interaction/ evolution Observations of shear-wave splitting throughout the crust show that fluid-saturated stress-aligned microcracks are so closely spaced that they verge on fracture criticality and failure in slippage and earthquakes (Figure 2; Crampin, 1994, 1999; Crampin and Chastin, 2003; Crampin and Peacock, 2005, 2008). Such verging on failure at critical points (also known as singularities, bifurcations, phase-transitions, or, in the case of the Earth, fracturecriticality) is a major defining characteristic of critical-systems (Davis, 1989; Bruce and Wallace, 1989). Critical-systems are part of a New Physics, a New Geophysics, where the elements of complex heterogeneous interactive systems initially interact locally, but as they approach critical points (Bruce and Wallace, 1989), the elements cease to interact locally and heterogeneously, but abruptly display coherent behaviour involving collective organisation over enormous numbers of degrees of freedom. It is one of the miracles of nature that huge assemblages of particles subject to the blind forces of nature, are nevertheless capable of organising themselves into patterns of cooperative activity (Davis, 1989). This is the reason the highly-constrained APE-model of rock mass deformation, illustrated in Figure 3, approximately matches the huge range of phenomena listed in Table A1 (Crampin and Chastin, 2003; Crampin and Peacock, 2005). Note specifically that we are unlikely to be able to explain these observations with arguments based on conventional sub-critical geophysics. Experience based on conventional sub-critical geophysics may not be useful in interpreting or understanding the New Geophysics based on a critical-system of fluidsaturated microcracks. A paradigm shift in understanding is required. Critical-systems and self-organisation are extremely common in complex heterogeneous interactive phenomena ranging from: quantum mechanics; to super-fluidity; traffic clustering on roads; the life cycle of fruit flies; the New York stock exchange; and a huge variety of physical phenomena (Bak, 1996). Thus in claiming that the Earth s crust is a critical-system of stress-aligned fluid-saturated microcracks (Crampin, 1998; Crampin and Chastin, 2003; Crampin et al., 2003; Crampin and Peacock, 2005), we are merely asserting that the Earth behaves like almost all other complex heterogeneous interactive systems. The good fortune for geophysicists is that critical-systems in the body of the rock mass can be monitored by analysing shear-wave splitting. The Gutenberg-Richter relationship is probably the best known example of selforganised criticality in the Earth, where the logarithm of the cumulative number of earthquakes and the earthquake magnitude (proportional to the logarithm of the energy) is linear (self-similar) over eight or nine orders of magnitude. The New Geophysics suggests that the underlying physical basis for the Gutenberg-Richter relationship, previously unidentified, is the interactions of critical-systems of stress-aligned fluid-saturated microcracks distributed throughout most rocks in the crust (and probably upper mantle). (The universality of critical-systems suggests that fluid-saturated cracks are also source of shear-wave splitting in the upper mantle, where the cracks would be intergranular films of hydrolysed melt, Crampin, 2003a). New Geophysics provides a new understanding of fluid-rock interaction with important implications for the way in situ rocks deform. In particular, the crust as a critical-system imposes new properties on conventional sub-critical behaviour including universality, calculability, predictability, and extreme sensitivity to initial conditions leading to deterministic chaos (Crampin and Chastin, 2003; Crampin and Peacock, 2005, 2008). These various properties are only now becoming recognised as shear-wave splitting becomes more widely observed and better understood. New Geophysics appears to have many applications, particularly to industrial geophysics (Crampin, 2003b, 2006), as well as stress-forecasting earthquakes as reviewed in this paper. Perhaps more importantly, New Geophysics provides an improved understanding of all fluid-rock interactions, underlying almost all low-level fluid-rock deformation/evolution in the Earth. 20

21 Appendix D The Prototype SMS at Húsavík, Iceland: the SMSITES Project Swarms of small earthquakes are far too scarce and irregular (Wu et al., 2006) to be used for reliable routine stress-forecasting. Controlled-source Stress-Monitoring Sites (SMSs) with the geometry of Figure 6 are the optimum solution for routine stress-forecasting, where the Downhole Orbital Vibrator (DOV) (Leary and Walter, 2005a, 2005b) radiates shearwaves downwards in Band-1 directions to three-component geophones in adjacent boreholes. The depth, X, of the DOV in Figure 6 needs to be greater than the critical depth, typically between 500 m to 1000 m, where the vertical stress, V, equals the minimum horizontal stress, h. Below X m-depth, stresses V H h, and the fluid-saturated microcracks are typically aligned vertically, perpendicular to the direction of minimum horizontal stress, h (Hubbert and Rubey, 1959). Note that the great advantage of the DOV is that sweeps of varying frequency (chirps) are highly repeatable. 12 s-long chirps made up of 4 s to 5 s of increasing frequencies (in our case, to 200 Hz) were followed by 4 s to 5 s of decreasing frequencies at a rate of four chirps each minute, allowing 100-fold stacking. In the SMSITES Project, below, there was no perceptible change in seismic waveforms after a total of ~40,000 chirps. In 2001, the European Commission funded SMSITES Project set up a prototype SMS between two boreholes, drilled for geothermal purposes, parallel and 100 m to the south of the Húsavík-Flatey Fault (HFF) where it runs onshore in Northern Iceland (Crampin et al., 2003). HFF is a transform fault of the Tjörnes Fracture Zone of the Mid-Atlantic Ridge. The DOV was deployed at 456 m-depth and recorded in three-component geophones at 515 m- depth in a borehole offset 315 m. The seismic waves were transmitted almost horizontally and the recording geometry was not optimal for accessing the most sensitive ray path directions. A quasi-continuous 12 hour-survey, 17 th April, 2001, established approximate traveltimes for P-waves (~88 ms), and for SH- and SV-waves (~186.5 ms and ~188.5 ms, respectively). Figure 7 shows observations at the prototype SMS of the SMSITES Project adjacent to the HFF (Crampin et al., 2003). The main recording session, a quasi-continuous 24 hours per day survey, 11th to 24th August, 2001, recorded the travel-time variations in Figure 7, together with North-South and East-West Global Positioning System (GPS) measurements, and water pressures at 30 m-depth in a water well on Flatey Island immediately above the HFF fault break. The advantage of SMSs is that, by differencing repeated shear-wave signals recorded by identical source-receiver geometry (in time-lapse seismics), SMSs recordings are extremely sensitive to nearly negligible changes in stress and other parameters. Table D1 summarises the observations in Figure 7. (TABLE D1, HERE) In Figure 7, a), b), and c) are travel times in ms of: P-waves; SV-waves and SH-waves; and anisotropic time-delays (SV-SH), respectively, propagating nearly horizontally at ~500 m-depth between boreholes 315 m-apart. Shear-waves are split into SV- and SHpolarisations as they propagate in a vertical symmetry direction of the stress field. Also shown are d) NS and EW GPS displacements in mm; and e) pressure measurements in bars at 30 m-depth in a water well on Flatey Island immediately above the HFF (the 'pulse' is a ~1 m-drop in water level lasting five days, and the twice daily ~40 cm oscillations are oceanic tides). All seven observations a) to e) coincide in time with f): a histogram of small scale seismicity on the parallel fault, the Grímsey Lineament, 70 km NNW of the SMS. These various observations are listed in Table D1. The seismicity consisted of 106 earthquakes in 2.5 days where the largest earthquake is M 2.8, where the total energy of the swarm is approximately equivalent to one earthquake of magnitude M ~

22 Preliminary observations in April 2001 suggested that final travel times on 24 th August of both P- and both shear-waves in Figure 7 are close to the background values before there was a stress disturbance. Thus the abrupt increase in travel-times in Figure 7a suggests that P- wave velocities are insensitive to the (implied) stress-accumulation before the M ~3.5 swarm activity, but are sensitive only when cracks begin to open as the earthquake swarm begins. In contrast, as the implied stress-accumulations in Figure 4 suggest, shear-waves velocities are sensitive to stress-accumulation and reached lower values (longer travel-times), probably at fracture-criticality, before the swarm activity begins and stress is relaxed. A M 3.5 earthquake is a comparatively small event with a source zone of a few hundred meters at most. Thus, the seismic observations show remarkable sensitivity to the start of a small disturbance at a distance of several hundred times the likely source diameter. This sensitivity demonstrates that time-lapse seismic SMS recordings have the ability to monitor extremely small changes in rock mass conditions. This is believed to be a unique data set. Variations in GPS measurements and changes in well levels have been observed previously and have been associated with seismicity and earthquakes. The remarkable observations in this data set are that the stacked seismic traveltimes where the highly-repeatable DOV source has allowed time-lapse recordings of seismic travel-times, to accuracies better than ±0.02 ms, correlate with GPS displacements, waterwell levels, and histograms of seismicity. It is interesting that the seven marked variations listed in Table D1 all appear to begin at the start of the seismicity on the Grímsey Lineament before the bulk of the seismic energy is released. With one exception, the various variations are broadly compatible with each other: shear-wave velocities and anisotropy, GPS, and well-level changes appear to be internally consistent, as well as consistent with previous observations of shear-wave splitting, which can be modelled as microcracks opening and closing with seismicity-induced stress changes. The exception is that the P-wave travel-times which, following instantaneous increase at the start of the seismicity, relax linearly over about nine days as cracks close, whereas the shearwaves relax (in classic S-shaped relaxation curves) over about five days. Since the P-waves and shear-wave travel times appear to be measured along similar ray paths, this anomaly may convey important information about how rocks respond to changes of stress in comparatively shallow rocks. Appendix E The potential of SMSs for stress-forecasting earthquakes and volcanic eruptions As yet we have too few observations to determine the overall sensitivity of shear-wave splitting for all magnitude earthquakes in all geological and tectonic regimes. As suggested in Section 5.2 and data listed in Table 1, the inferred sensitivity of SMSs from existing observations in Figures 4 and 7, and elsewhere, suggest that a single three-borehole SMS would be sensitive to stress-accumulation before M 5 earthquakes (that is all damaging earthquakes) within 950 km of the SMS. A SMS would also expect to record stressaccumulation before smaller earthquakes closer to the SMS, and larger earthquakes at greater distances. However, this is based principally on data from Iceland. Iceland has atypical crustal structure (sheeted basalts) and may not be representative of earthquake occurrence elsewhere. However, it can be argued that such critical-system data may well be representative. The universality of critical-systems suggests that effects are pervasive throughout the available space. An example of universality is that observed SWVA is in the narrow range ~1.5% to ~4.5% in all in situ in ostensibly-intact rocks regardless of geology, tectonics, and porosity (Crampin, 1994, 1996, 1999, 2006; etc.). It is suggested that this surprising result for subcritical geophysics can only be explained by the universality of critical-systems. 22

23 The evidence reviewed in this paper and discussed above, suggests that the sensitivity of a single three-borehole SMS with the geometry of Figure 6 would monitor stressaccumulation, and hence stress-forecast the times, magnitudes, and in some circumstances fault-planes, of all damaging earthquakes (M 5) within 950 km of the SMS, as well as smaller earthquakes nearer than 950 km, and larger earthquakes at greater distances. If a large earthquake threatened a large city or other vulnerable site, a SMS gives the opportunity of monitoring an optimising stress-release (and hence reducing hazard) by inducing seismicity by massive hydraulic injection in low-vulnerability locations within the overall stressed volume, such as mountain ranges, deserts, or in some circumstances offshore, if the risk of tsunamis can be discounted. A SMS would allow the effects of the hydraulic fracturing to be monitored and the reduction of stress, and hazard, optimised. Note that lowering the risk of large earthquakes by hydraulic fracturing was discussed and rejected by Raleigh et al. (1976). The danger then was that the fracturing might excite the large event it was designed to inhibit. The difference now is that the SMS would allow the effects to be monitored, and the location of the fracturing moved, if necessary, to achieve the desired relaxation. This is discussed in Crampin et al. (2008b). 23

24 Appendix F GEMS: the opportunity for stress-forecasting all damaging earthquakes worldwide STUART CRAMPIN 1,2, SERGEI ZATSEPIN 3, CHRIS W. A. BROWITT 2, KIYOSHI SUYEHIRO 4, YUAN GAO 5 and LARRY WALTER 6 1 British Geological Survey, Edinburgh, Scotland UK. scrampin@ed.ac.uk. 2 School of GeoSciences, University of Edinburgh, Edinburgh, Scotland UK. cbrowitt@staffmail.ed.ac.uk. 3 formerly at School of GeoSciences, University of Edinburgh, Edinburgh, Scotland UK. 4 JAMSTEC, Yokosuka, Japan. suyehiro@jamstec.go.jp. 5 Institute of Earthquake Science, China Earthquake Administration, Beijing, China. gaoyuan@seis.ac.cn. 6 APEX HiPoint, Englewood, CO80112, USA. lwalter@apexpe.com. Strange as it may seem, we understand the distribution of matter in the interior of the sun far better than we understand the interior of the earth. (Feynman 1995) Abstract - The logical extension of a single or small network of Stress-Monitoring Sites (SMSs), as discussed in Appendices D and E, is GEMS, a Global Earthquake Monitoring System deploying world-wide network of SMSs. GEMS would need a network of, perhaps 200, three-borehole SMSs on a 1200 km-grid in seismic regions and a 2500 km-grid elsewhere, both onshore and offshore. Recent technological and internet developments could control both onshore and offshore SMSs. Not only would GEMS stress-forecast all damaging earthquakes (M 5) worldwide, the range of benefits would include a long-term stress forecasting service analogous to weather forecasting, control for mitigating seismic hazards, and a better understanding of evolution in the solid Earth beneath our feet. 1. Introduction Despite nearly three hundred years of geology and over a century of geophysics, we know remarkably little about how rocks deform a few meters beneath our feet. This is partly because rocks at depth are extraordinarily remote. We cannot access the behaviour directly without destroying in situ conditions by the severe traumas of partial de-stressing, partial cooling, and fluid disruption. This lack of understanding has serious consequences for earthquake hazard. The 1995, M 7.2, Kobe Earthquake in Japan killed 6,000 people and caused an estimated 250 billion dollar damage. Earthquakes such as Kobe; 1999, Izmit, Turkey, 17,000 dead; 2001, Bhuj, India, 40,000 dead; 2003, Bam, Iran, 34,000 dead, occur frequently, indiscriminately, and cause incalculable suffering and loss. These four earthquakes were all close to magnitude M 7. The largest earthquakes, such as the 2004, Mw ~ 9.2, Sumatra-Andaman earthquake, whose tsunami killed over 250,000, released some three orders of magnitude more energy (but fortunately are orders of magnitude less frequent). Currently there is no effective earthquake prediction programme and we are frequently fatally surprised. Recent advances in understanding and modelling fluid-rock interaction (Crampin, 2006) lead to a paradigm shift in forecasting earthquakes (Crampin et al., 2008a, and this paper). Large earthquakes release enormous amounts of stress which, since rock is comparatively weak to shear stress, must accumulate over very large volumes of rock before the earthquake can occur. The paradigm shift is to ignore the earthquake source which is deterministically 24

25 unpredictable (Geller, 1997), but instead analyse shear-wave splitting and monitor stressaccumulation in the rock mass surrounding the impending earthquake. The evidence shows that the approach to fracture-criticality, when the microcrack distributions are so closely spaced they lose shear-strength and fracture in earthquakes, can be monitored at substantial distances from the eventual source zone (Crampin et al., 2008a). A combination of three recent developments provides the opportunity for the paradigm shift to be exploited in GEMS: the Global Earthquake Monitoring System The first is, the recognition that almost all in situ rocks, certainly in the crust, and the seismogenic parts of the mantle are pervaded by self-organised scale-invariant systems of fractures ranging from open fluid-saturated grain-boundary microcracks and preferentially-oriented pore space from sub-millimetre to millimetre scales to plate-boundaries at scales of thousands of kilometres (Crampin and Peacock, 2008; Heffer and Bevan, 1990). In the crust, the fluids are usually water-based salt solutions, but may be hydrocarbons, and in the upper mantle, fluids are likely to be intergranular films of hydrologised melt (Crampin et al., 1986; Crampin, 2003a). Shear-wave splitting shows that the cracks are so closely spaced they verge on fracturing and are critical-systems (Crampin, 2006), and are a New Physics (Davis, 1989), or New Geophysics. The criticality, particularly of small-scale microcracks, is the underlying reason for rocks extreme sensitivity to small disturbances. The second development is that we now know that details of stress-induced low-level deformation of crack distributions can be monitored by shear-wave splitting (seismic birefringence), so that the accumulation of stress before earthquakes can be monitored and the release of stress in earthquakes stress-forecast (Crampin et al., 2008a). Finally, and crucially important, recent advances in borehole instrumentation and technology allow polarized shear-waves to be monitored by repeated crosshole shear-wave transmission measurements at borehole SMSs, which may now be both onshore and offshore (Crampin et al., 2003). 2. Stress-Monitoring Sites (SMSs) The prototype SMS, developed by the European-Commission-funded SMSITES Project in Iceland used wells, previously drilled for geothermal purposes, adjacent to the Húsavík- Flatey Transform Fault of the Mid-Atlantic Ridge where it runs onshore in Northern Iceland (Crampin et al., 2003). The well geometry was not optimum for a SMS and signals were restricted to horizontal propagation at 500 m-depth between wells 315 m-apart. Never-theless, despite non-optimal geometry the records were spectacularly sensitive to small disturbances of stress in Figure 7. In what was intended to be a source calibration test of the borehole source (the Downhole Orbital Vibrator, DOV, Leary and Walter, 2005a, 2005b), the DOV was pulsed every 12 to 20 seconds for 24 hours for 13 days, yielding over 40,000 records at each of four downhole three-component geophones 1 m-apart. Hundred-fold stacking gave travel-time accuracies of ±0.02 ms. Fortuitously, the recordings coincided with a burst of low-level seismicity, 70 km NNW of SMSITES on another transform fault, the Grímsey Lineament, and remarkable anomalies were recorded (Figure 7). The variations in seismic travel-times between the two wells and the shear-wave anisotropy also correlated with NS and EW Global Positioning System (GPS) variations and with changes in water level in a well on the Island of Flatey immediately above the fault. This sensitivity to low-level seismicity (equivalent energy to one M 3.5 earthquake) at 70 km distance at hundreds of times the conventional source dimensions is far greater than would be expected in a conventional sub-critical brittle-elastic crust and is another demonstration of the crack-critical nature of the Earth s crust. These observations confirm the science, technology, and sensitivity of SMSs for monitoring changes of stress in the Earth s crust and stress-forecasting the times and magnitudes of impending earthquakes. Although the experiment was designed to monitor small changes, we were surprised by the sensitivity actually recorded. Well-level changes and GPS 25

26 variations have previously been observed by several authors at substantial distances from earthquake epicentres, but this is the first time that variations in four seismic measurements, vector GPS, and water-well levels have been observed simultaneously before earthquakes. 3. Advances in borehole technology The operation and recording of permanent installations of seismic receivers and energy sources within deep boreholes is now well-established in oil industry seismic surveys. Borehole seismic recorders routinely operate at several kilometres depth at temperatures up to 150ºC. The most significant advance has been in controlling the DOV source and understanding the behaviour and characteristics of the polarised seismic signals it generates (Leary and Walter, 2005a, 2005b). New developments of this seismic source provide the means to reliably control the source, record observations, and process signal measurements by satellite technology both onshore and offshore. This means that the whole shear-wave monitoring operation could be controlled and processed remotely via Internet technology, so that a global network of SMSs (GEMS) could be managed effectively on a continuous real-time basis both onshore and offshore. 4. The GEMS global network of SMSs The concept of Stress-Monitoring Sites is believed to be a significant advance. For the first time there is the opportunity for controlled-source operations to monitor stress-induced changes to microcrack geometry by non-invasive seismic techniques at depth in in situ rock. The power of a single SMS is that it can monitor very subtle changes in behaviour by timelapse techniques. In very quiet conditions, preferably at or below 1000 m-depth, records of the highly-repeatable DOV signals can be differenced to monitor the effects of very small changes in rock mass conditions. The measurements allow exceptional accuracies of ±0.02 ms (±20 s) over 315 m (Crampin et al., 2003). Note that although not specifically addressed by the discussions in this paper, the accuracy of SMSs would also be valuable for investigating the frequency dependence of seismic velocities. Such dispersion is currently of interest to the oil industry as a means of investigating the dimensions of the cracks that cause shear-wave splitting in hydrocarbon reservoirs. The seismic measurements in Figure 7 are clearly not at the limit of their range. Conservative extrapolation suggests that a single SMS would be able to monitor changes induced by M 3.5 earthquakes to 100 km, and correspondingly M 5 to ~1000 km, say, and M 8 earthquakes to the scale of tectonic plates, if not worldwide. This suggests that GEMS, a global network of ~200 SMSs, on a 1200 km-grid in regions of seismicity and 2500 km-grid elsewhere, would be able recognise stress accumulation and stress-forecast the times and magnitudes of all earthquakes with magnitudes greater than M 5, and many M > 4 events, worldwide. In particular, what would be guaranteed is that the accumulation of stress before all damaging earthquakes would always be recognised. No change would indicate no impending large earthquake and hence security. However, if changes were observed, the estimate of the time of occurrence would depend on the rate of the tectonic stress accumulation which may vary from place to place, and possibly from time to time. The suggested GEMS network of a 1200 km-grid in seismic areas and a 2500 km-grid elsewhere would lead to some 200 SMSs, after adjusting distributions for stable and unstable regions. There are large stable areas both onshore, such as the Canadian Shield, and particularly offshore as in oceanic basins, which are believed to be almost completely aseismic and would probably show little variation in stress, although this would be open to confirmation. Note that routine drilling of deep wells offshore is only now becoming 26

27 feasible as the Riser Drillship 'Chikyu' of the Integrated Ocean Drilling Program, IODP, now becomes available. Riser technology allows deeper and more easily re-enterable wells to be drilled offshore. Indeed, networks of borehole seismometers across ocean floors have been proposed to record and analyse earthquake data (Suyehiro, 2002). A 1000 km grid was suggested, filling in the largest gaps in the worldwide network of seismic stations, and would be passive, monitoring earthquakes as they occurred. 5. Lowering the potential for large earthquakes As the accumulation of stress before large earthquakes is so extensive, any changes of stress or, more generally, any large-scale increases of stress, could be recognised at substantial distances from the impending earthquake epicentre. Consequently, if accumulating stress is believed to be threatening a large city or other vulnerable location, in principle, the accumulating stress could be diminished almost anywhere within the larger stressed volume, by inducing small earthquakes, and the potential for city-threatening earthquakes reduced. The most direct way to release stress would be by hydraulic pumping operations in non-vulnerable areas nearby, within 500 to 1000 km, say, of the threatened city. Hydro-fracturing is a routine oil-company operation. Stress release by hydraulic fracturing could be sited in areas of low population and infrastructure such as amongst mountains or deserts, or even offshore, with suitable allowances for tsunamis. However, this is an untested procedure and the effects are currently not known. The great advantage of GEMS would be that the effects of such hydraulic pumping could be monitored so that the results of the tests could be optimised. The intention would be to release stress by exciting small earthquakes in areas within the larger stressed volume where earthquakes would be less destructive. The seismic (acoustic) events as oil reservoirs are depleted demonstrate that this is possible. Such hydraulic fracturing operations would need to be massive, extensive, and very costly. However, the 1995 Kobe earthquake has been estimated as costing $250 billion dollars U.S. Had the accumulation of stress been recognised by GEMS, a premium of 0.5% would provide $1.25 billion for hydraulic fracturing if a city such as Kobe was shown to be threatened by a large earthquake. This would not be a blind investment. The SMSs of GEMS would allow the effects to be monitored and the stress release optimised. If the hydraulic fracturing at one location was not proving effective hydraulic fracturing could be relocated within the stressed volume until an effective relaxation regime had been established. Note that lowering the risk of large earthquakes on specific faults by hydraulic fracturing on the actual fault was suggested many years ago (Raleigh et al., 1976). At that time, the major disadvantage was that such operations might excite the event they are designed to prevent. The advance now is the recognition that the stress accumulation is so extensive that hydraulic-fracture-induced events could be triggered at substantial distances from any vulnerable location, and that any such changes would be monitored by GEMS. 6. A stress-forecasting service Monitoring stress changes and directions at a single SMS is analogous to a single weather station, where the principal measurements are changes in air pressure, and wind speed and direction. The patterns of behaviour can be used to estimate, particularly the stability of the weather, and the likelihood of storms. (One of us finds it a useful guide to look at a barometer each day before stepping into Scottish weather!) The power of weather forecasting comes from networks of such weather stations, where recognising areal and temporal patterns of behaviour allow relatively accurate forecasting. However, weather is 27

28 another critical-system so that weather forecasting has all the uncertainties and sensitivity inherent in critical-system of complicated heterogeneous interactive phenomena. It is anticipated that identifying previously unrecognised patterns of behaviour with GEMS would allow a stress forecasting service analogous to weather forecasting. Such stress-forecasting should provide some predictive capability for the longer-term (we guess at five to ten year) estimation of earthquake scenarios, so that long term preparations for earthquake hazard could be instituted. Currently, such questions are not even raised by the scientific community, because there is no means of acquiring relevant information. Stressforecasting with GEMS would open this new capability. It is worth pointing out that GEMS would provide the data for new investigations of Earth deformation that have not been previously available. Earthquake forecasts would be by investigations of stress and crack evolution, and provide information for stress modelling and some understanding of the phenomena. This is in contrast to probabilistic statistical predictions, which even if they were correct would provide no increased understanding of earthquake occurrence and Earth deformation. 7. GEMS as a new tool for monitoring Earth evolution in the 21 st century Apart from the earth tides, ocean tides, and other astro-geophysical influences, the major driving force of Earth evolution is expected to be the generation, spreading, and subduction of tectonic plates. We do not know, and currently have no means of assessing, the dynamics of plate motion and the way stress is distributed, except by modelling based on inadequate information. The two year relaxation stress implied from the decrease in time-delays following the 1996 Vatnajökull eruption in Iceland (Volti and Crampin, 2003b) suggests that these movements are highly episodic. It is well known that earthquake occurrence is fractal and varies over scales from minutes to millions of years. The reasons for this are likely to be the interaction of the dynamics of the core with movements of the mantle and movements of oceanic plates, but we currently have minimal information. Consequently, there are many comparatively simple questions, such as whether plates are pushed by ridges or pulled by subduction zones, and the underlying reasons behind cycles of greater or lesser seismicity, which need to be answered. But the major questions are what drives the plates and how and why they vary with time. Currently we have no means of acquiring such information which is crucial for understanding the evolution of the Earth. GEMS by monitoring stress deformation over the near-surface of the Earth would provide for the first time the means of investigating the dynamic evolution of the interior of the crack-critical Earth. It is perhaps worth noting that over 70% of the surface of the Earth is water beneath which lie some 50% of all earthquakes (there appear to be more earthquakes onshore than offshore). This means that wholly satellite-based Synthetic Aperture Radar, Global Positioning System, displacement, or other similar measurements which are confined to observations of the solid surface, cannot monitor a substantial proportion of all earthquakes. Only an onshore and offshore borehole-based system such as GEMS can monitor the approach of all earthquakes. 8. Conclusions The effects of changes of stress on shear-wave splitting are comparatively subtle and easily overlooked or misunderstand (Crampin and Peacock, 2005, 2008). Consequently, interpretations of temporal changes in shear-wave splitting are sometimes claimed to be controversial. We suggest that the evidence supporting APE modelling and temporal changes is vast (Table A1) and is confirmed by the unique observations from the SMS in Iceland (Figure 7). The observed sensitivity to remote seismicity is remarkable and marks a new property of the in situ rock mass. Despite not knowing exactly how stress behaves 28

29 before earthquakes, though previous stress monitoring experience provides indications, the New Geophysics, the new sensitivity, and the state-of-the-art technology are all proven attributes, although not necessarily wholly understood. GEMS would have the capability of monitoring stress changes and stress-forecasting all damaging (M 5) earthquakes worldwide. The benefits of Global networks of SMS are summarised in Table F1. GEMS at the suggested grid size of 1200 km in seismic areas, and 2500 km elsewhere, and the establishment of regional processing centres, could stress-forecast the times and magnitudes of all damaging earthquakes, M 5, worldwide. The greatest advantage may be peace of mind. The absence of change would mean there could not be an imminent large earthquake nearby. Secondly, GEMS would provide monitoring for deterministic control for lessening the potential for a large earthquake by mitigation methods such as massive hydraulic fracturing operations. These would be practical advantages for understanding and mitigating earthquake hazard and would place mankind for the first time in some control of damaging earthquakes worldwide. Thirdly, GEMS would provide the data for a stress-forecasting service, similar to the familiar weather forecasting, for the longer-term estimation of stress and earthquake occurrence. Fourthly, GEMS would provide a network of other borehole instrumentation for passive geophysical monitoring where very quiet locations would allow time-lapse monitoring of other geophysical phenomena and open up a whole new range of geophysical investigations. Finally, providing a tool for investigating the dynamic evolution of the Earth on which our lives depend would provide an enormous intellectual stimulus for understanding the Earth in the 21 st century. GEMS, estimated as a five to ten billion U.S. dollar development, is matched in Earth Science only by the scale of oil industry investments. However, multi-billion dollar decisions need to be made. For example, the question of whether new buildings in the New Madrid Seismic Zone, USA, which has occasionally suffered very large earthquakes, should have the same earthquake resistant designs as coastal California, which has many more slightly smaller earthquakes. The argument between Frankel (2003), Project Chief for the U.S. Geological Survey, for seismic hazard maps for different designs and Stein et al. (2003) for the same designs, has multi-billion dollar implications for the cost of new buildings. In the absence of real information, the answers depend on essentially philosophical differences about how to forecast and prepare for future natural hazard about which much is not well understood (Stein et al., 2003). GEMS would eventually (it would need several years to build-up a sufficient data) provide real factual information on which to base such costly decisions. In contrast, a few billion dollar investment in GEMS would, for the first time, place man in some control of earthquake hazards, as well as providing the intellectual stimulus for investigating the dynamic behaviour of the solid Earth on which we are totally dependent every day of our lives. GEMS would provide the basic factual information for informed decisions about the future behaviour of the stressed crack-critical Earth. Although the cost is likely to be too great for current geophysical research monies, the benefits of forecasting all damaging earthquakes worldwide should attract funding from the World Bank, the World Health Organisation, UNESCO, insurance companies, earthquake vulnerable cities, cities vulnerable to volcanic eruptions, and others. 9. Note added on submission As this paper was being prepared for submission, the M 7.8 Sichuan Province, China, Earthquake occurred on 12 th May, 2008, with nearly 70,000 casualties. It is likely that had there been even a single Stress-Monitoring Site installed in China, the stress-accumulation 29

30 before this large earthquake would have been recognised, and the approach of a sizeable earthquake stress-forecast. A single SMS would probably not have indicated the location, but knowing a large earthquake was approaching might have allowed other precursory phenomena to be realistically interpreted to indicate the fault-plane or epicentre, as happened with the successful stress-forecast earthquake in SW Iceland (Crampin et al., 1999). REFERENCES *Angerer, E., Crampin, S., Li, X.-Y., and Davis, T.L. (2002), Processing, modelling, and predicting time-lapse effects of overpressured fluid-injection in a fractured reservoir, Geophys. J. Int. 149, Aster, R.C., and Shearer, P.M. (1992), Initial shear wave particle motions and stress constraints at the Anza Seismic Network, Geophys. J. Int. 108, Bak, P. (1996), How nature works - the science of self-organized criticality (Copernicus, Springer, Berlin). Barton, N. (2007), Rock quality, seismic velocity, attenuation and anisotropy (Taylor and Francis, London). Bianco, F., Scarfi, L., Del Pezzo, E., and Patanè, D. (2006) Shear wave splitting time variation by stress-induced magma uprising at Mt. Etna volcano, Geophys. J. Int. 167, Booth, D.C., and Crampin, S. (1985), Shear-wave polarizations on a curved wavefront at an isotropic free-surface, Geophys. J. R. Astron. Soc. 83, Booth, D.C., Crampin, S., Lovell, J.H., and Chiu, J.-M. (1990), Temporal changes in shear wave splitting during an earthquake swarm in Arkansas, J. Geophys. Res. 95, 11,151-11,164. Brocher, T.M., and Christensen, N.I. (1989), Seismic anisotropy due to preferred mineral orientation observed in shallow crustal rocks in southern Alaska, Geology 18, Bruce, A., and Wallace, D. (1989), Critical point phenomena: universal physics at large length scale, In The New Physics (ed. Davis, P.) (Cambridge Univ. Press) pp Chapman, M., Zatsepin, S.V., and Crampin, S. (1998), Anisotropic dispersion in stresssensitive poroelasticity. 60th Conf., EAGE, Leipzig, Extended Abstracts 1, Chapman, M., Zatsepin, S.V., and Crampin, S. (2000), Incorporating stress-sensitivity into dynamic poro-elasticity, 70th Ann. Int. SEG Mtg., Calgary, Expanded Abstracts 2, Crampin, S. (1981), A review of wave motion in anisotropic and cracked elastic-media, Wave Motion 3, Crampin, S. (1990), The scattering of shear waves in the crust, PAGEOPH 132, Crampin, S. (1991), Effects of point singularities on shear-wave propagation in sedimentary basins, Geophys. J. Int. 107, *Crampin, S. (1994), The fracture criticality of crustal rocks, Geophys. J. Int. 118, *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 Mtg., Dallas, Expanded Abstracts 1, Crampin, S. (1998), Shear-wave splitting in a critical crust: the next step, Rev. Inst. Franc. Pet. 53, *Crampin, S. (1999), Calculable fluid-rock interactions, J. Geol. Soc. 156, *Crampin, S. (2001), Developing stress-monitoring sites using cross-hole seismology to stress-forecast the times and magnitudes of future earthquakes, Tectonophysics 338, Crampin, S. (2003a), Aligned cracks not LPO as the cause of mantle anisotropy, EGS- AGU-EUG, Joint Ass., Nice, 2003, Abstract P

31 *Crampin, S. (2003b), The New Geophysics: shear-wave splitting provides a window into the crack-critical rock mass, The Leading Edge 22, *Crampin, S. (2006), The New Geophysics: a new understanding of fluid-rock deformation, In Eurock 2006: Multiphysics coupling and long term behaviour in rock mechanics (eds. Van Cotthem, A., Charlier, R., Thimus, J.-F., and Tshibangu, J.-P.) (Taylor and Francis, London) pp Crampin, S., and Booth, D.C. (1989), Shear-wave splitting showing hydraulic dilatation of pre-existing joints in granite, Sci. Drilling 1, *Crampin, S., and Chastin, S. (2003), A review of shear-wave splitting in the crack-critical crust, Geophys. J. Int. 155, *Crampin, S., and Gao, Y. (2005), Comment on "Systematic Analysis of Shear-Wave Splitting in the Aftershock Zone of the 1999 Chi-Chi, Taiwan, Earthquake: Shallow Crustal Anisotropy and Lack of Precursory Changes." by Liu, Teng, Ben-Zion, Bull. Seism. Soc. Am. 95, *Crampin, S., and Gao, Y. (2006), A review of techniques for measuring shear-wave splitting above small earthquakes, Phys. Earth. Planet. Inter. 159, *Crampin, S., and Peacock, S. (2005), A review of shear-wave splitting in the compliant crack-critical anisotropic Earth, Wave Motion 41, *Crampin, S., and Peacock, S. (2008), A review of our current understanding of shear-wave splitting and common fallacies in interpretation, Wave Motion 45, Crampin, S., and Yedlin, M. (1981), Shear-wave singularities of wave propagation in anisotropic media, J. Geophys. 49, Crampin, S., and 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 Mtg, Houston, Expanded Abstracts *Crampin, S., and Zatsepin, S.V. (1997), Modelling the compliance of crustal rock: II - response to temporal changes before earthquakes, Geophys. J. Int. 129, Crampin, S., Booth, D.C., Evans, R., Peacock, S., and 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., and 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.C., Shearer, P.M., Berger, J., J. Geophys. Res. 96, Crampin, S., Zatsepin, S.V., Slater, C., and Brodov, L.Y. (1996), Abnormal shear-wave polarizations as indicators of pressures and over pressures, 58th Conf., EAGE, Amsterdam, Extended Abstracts X038. Crampin, S., Rowlands, H.J., Zatsepin, S.V., Smart, B.J., Edlmann, K., and Crawford, B. (1997), Predicting the response to effective stress of cores with different pore fluids, 59th Conf., EAGE, Geneva, Extended Abstracts 2, CO22. *Crampin, S., Volti, T., and 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., and Somerville, J.McL. (1999b), APE-modelling of fluid/rock deformation of sandstone cores in laboratory stress-cells, 61th Conf., EAGE, Helsinki, Extended Abstracts 1, *Crampin, S., Volti, T., Chastin, S., Gudmundsson, A., and 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., and Gao, Y. (2003), Shear-wave splitting in a critical crust: III - preliminary report of multi-variable measurements in active tectonics, J. Appl. Geophys. 54, *Crampin, S., Peacock, S., Gao, Y., and Chastin. S. (2004), The scatter of time-delays in shear-wave splitting above small earthquakes, Geophys. J. Int. 156,

32 *Crampin, S., Gao, Y., and Peacock, S. (2008a), Stress-forecasting (not predicting) earthquakes: A paradigm shift?, Geology 36, *Crampin, S., Zatsepin, S.V., Browitt, C.W.A., Suyehiro, K., Gao, Y., and Walter, L. (2008b), GEMS: the opportunity for forecasting all damaging earthquakes worldwide, In Appendix F, Crampin and Gao (2009), Proc. Evison Symp. PAGEOPH, this issue. Davis, P. (1989) The New Physics: a synthesis, In The New Physics (ed. Davis, P.)(Cambridge University Press) pp Davis, T.L., Benson, R.D. Roche, S.L., and Talley, D. (1997), 4-D 3-C seismology and dynamic reservoir characterization - a geophysical renaissance, 67th Ann. Mtg, SEG, Dallas, Expanded Abstracts 1, ; see also , Dziewonski, A.M., and Anderson, D.L. (1981), Preliminary Reference Earth Model (PREM), Phys. Earth Planet. Inter. 25, Evans, R. (1984), Effects of the free surface on shear wavetrains, Geophys. J. R. Astron. Soc. 76, Evans, R. (1997), Assessment of schemes for earthquake prediction: Editor s introduction, Geophys. J. Int. 131, Evison, F.F. (2001), Long-range synoptic earthquake forecasting: an aim for the millennium, Tectonophysics 338, Feynman, R.P. (1995), Six easy pieces (Addison-Wesley, New York). Frankel, A.D. (2003), Comment of 'Should Memphis build on California's earthquakes', by S., Stein, S., Tomasello, J., and Newman, A., and Reply, EOS 84, 29, Gao, Y., and Crampin, S. (2003), Temporal variation of shear-wave splitting in field and laboratory in China, J. Appl. Geophys 54, Gao, Y., and Crampin, S. (2004), Observations of stress relaxation before earthquakes, Geophys. J. Int. 157, *Gao, Y., and Crampin, S. (2006), A further stress-forecast earthquake (with hindsight), where migration of source earthquakes causes anomalies in shear-wave polarisations, Tectonophysics 426, Gao, Y., Wang, P., Zheng, S., Wang, M., Chen, Y., and Zhou, H. (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, *Gao, Y., Hao, P., and Crampin, S. (2006), SWAS: a Shear-Wave Analysis System for semi-automatic measurement of shear-wave splitting above small earthquakes, Phys. Earth. Planet. Inter. 159, Geller, R.J. (1997), Earthquake prediction: A critical review, Geophys. J. Int. 131, Gerst, A., and Savage, M. (2004), Seismic anisotropy beneath Ruapehu volcano: A possible eruption forecasting tool, Science 306, *Hao, P., Gao, Y., and Crampin, S. (2008), An Expert System for measuring shear-wave splitting above small earthquakes, Comp. Geosci. 34, Heffer, K.J., and Bevan, T.G. (1990), Scaling relationships in natural fractures: data, theory, and application, Soc. Pet. Eng., SPE Helbig, K., and Thomsen, L. (2005), 75-plus years of anisotropy in exploration and reservoir seismics: A historical review of concepts and methods, Geophysics 70, 9ND-23ND. Holmes, G.M., Crampin, S., and Young, R.P. (2000), Seismic anisotropy in granite at the Underground Research Laboratory, Manitoba, Geophys. Prosp. 48, Hubbert, M.K., and Rubey, W.W. (1959), Role of fluid pressure in mechanics of overthrust faulting, 1, Mechanics of fluid filled porous solids and its application to overthrust faulting, Geol. Soc. Am. Bull. 70, Hudson, J.A. (1980), Overall properties of a cracked solid, Math. Proc. Camb. Phil. Soc. 88, Hudson, J.A. (1981), Wave speeds and attenuation of elastic waves in material containing cracks, Geophys. J. R. Astron. Soc. 64, Jordan, T.H. (2006), < 32

33 Kaarsberg, E.A. (1959), Introductory studies of natural and artificial argillaceous aggregates by sound propagation and X-ray diffraction methods, J. Geol. 67, Kaarsberg, E.A. (1968), Elasticity studies of isotropic and anisotropic rock, Trans. Soc. Min. Eng. 241, Kagan, Y.Y. (1997), Are earthquakes predictable? Geophys. J. Int. 131, King, M.S., Chaudhry, N.A., and Ahmed, S. (1994), Experimental ultrasonic velocities and permeability of sandstones with aligned cracks, 61th Conf. EAGE., Vienna, Extended Abstracts P113. Leary, P.C. (1997), Rock as a critical-point system and the inherent implausibility of reliable earthquake prediction, Geophys. J. Int. 131, Leary, P.C., and Walter, L.A. (2005a), Physical model for the downhole orbital vibrator (DOV) I. Acoustic and borehole seismic radiation, Geophys. J. Int. 163, Leary, P.C., and Walter, L.A. (2005b), Physical model for the downhole orbital vibrator (DOV) II. Rotary correlation wavelets and time-lapse seismics, Geophys. J. Int. 163, Levin, F.K. (1979), Seismic velocities in transversely isotropic media I, Geophysics 44, Levin, F.K. (1980), Seismic velocities in transversely isotropic media II, Geophysics 45, Li, X.-Y., and Crampin, S. (1993), Linear-transform techniques for processing shear-wave splitting in four-component seismic data, Geophysics 58, Liu, Y., Crampin, S., and Main, I. (1997), Shear-wave anisotropy: spatial and temporal variations in time delays at Parkfield, Central California, Geophys. J. Int. 130, Liu, Y., Teng, T.-L., and Ben-Zion, Y. (2004). Systematic analysis of shear-wave splitting in the aftershock zone of the 1999 Chi-Chi, Taiwan, Earthquake: shallow crustal anisotropy and lack of precursory changes, Bull. Seism. Soc. Am. 94, Liu, Y., Teng, T.-L., and Ben-Zion, Y. (2005), Reply to Comment on Systematic Analysis of Shear-Wave Splitting in the Aftershock Zone of the 1999 Chi-Chi, Taiwan, Earthquake: Shallow Crustal Anisotropy and Lack of Precursory Changes, by Yungfeng Liu, Ta-Liang Teng, and Yehuda Ben-Zion, by Stuart Crampin and Yuan Gao, Bull. Seism. Soc. Am. 95, Miller, V., and Savage, M. (2001), Changes in seismic anisotropy after volcanic eruptions: evidence from Mount Ruapehu, Science 293, Mueller, M.C. (1991), Prediction of lateral variability in fracture intensity using multicomponent shear-wave surface seismic as a precursor to horizontal drilling in the Austin Chalk, Geo. J. Int. 107, Munson, C.G., Thurber, C.H., Li, Y., and Okuba, P.G. (1995), Crustal shear wave anisotropy in southern Hawaii: Spatial and temporal analysis, J. Geophys. Res. 100, 20,367-20,377. Nuttli, O. (1961), The effect of the earth s surface on the S wave particle motion, Bull. Seism. Soc. Am. 51, Peacock, S., Crampin, S., Booth, D.C., and Fletcher, J.B. (1988), Shear-wave splitting in the Anza seismic gap, Southern California: temporal variations as possible precursors, J. Geophys. Res. 93, Raleigh, C.B., J. D. Healy, J.D., and Bredehoeft, J.D. (1976), An experiment in earthquake control of Rangely, Colorado, Science 191, Sibson, R.H. (1981), Controls on low-stress hydro-fracture dilatancy in thrust, wrench and normal faults terrains, Nature 289, Sibson, R.H. (1990), Rupture nucleation on unfavorably oriented faults, Bull. Seism. Soc. Am. 80, Sothcott, J., O'Hara, S.G., Khazanehdari, J., and McCann, C. (2000a), From sonic to ultrasonic - the acoustic properties of reservoir sandstones, 61th Conf. Eur. Ass. Geophys. Eng., Glasgow, Extended Abstracts D

34 Sothcott, J., McCann, C., and 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. Mtg Soc. Explor. Geophys., Calgary, Expanded Abstracts 2, Stein, S., Tomasello, J., and Newman, A. (2003), Should Memphis build on California's earthquakes, EOS 84, 29, 177 and Suyehiro, K. (2002), Illuminating Earth's mantle and core: a new challenge for ODO, In Achievements and opportunities of Scientific Ocean Drilling (ed. Becker, K.), Joides J., Spec. Issue 28, *Volti, T., and 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, D.A.) Geol. Soc. Lond., Spec. Publ. 212, *Volti, T., and 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, D.A.) Geol. Soc. Lond., Spec. Publ. 212, *Winterstein, D. (1996), Anisotropists Digest 147, <anisotropists@sep.stanford.edu>. *Wu, J., Crampin, S., Gao, Y., Hao, P., Volti, T., and Chen, Y.-T. (2006), Smaller source earthquakes and improved measuring techniques allow the largest earthquakes in Iceland to be stress-forecast (with hindsight), Geophys. J. Int. 166, Yardley, G.S., and Crampin, S. (1993), Shear-wave anisotropy in the Austin Chalk, Texas, from multi-offset VSP data case studies, Can. J. Expl. Geophys. 29, Zatsepin, S.V., and Crampin, S. (1996), Stress-induced coupling between anisotropic permeability and shear-wave splitting, 58 th Conf. EAGE, Amsterdam, 1996, Extended Abstracts CO30. *Zatsepin, S.V., and Crampin, S. (1997), Modelling the compliance of crustal rock: I - response of shear-wave splitting to differential stress, Geophys. J. Int. 129, Zhang, Z., and Schwartz, S.Y. (1994), Seismic anisotropy in the shallow crust of the Loma Prieta segment of the San Andreas fault system, J. Geophys. Res. 99, *Papers/preprints available at 34

35 Table 1 Epicentral distance, magnitude, and duration of changes in shear-wave splitting time-delays observed before earthquakes and volcanic eruptions. a) Observations of inferred stress-accumulation before earthquakes. No. Location Approximate epicentral distance (km) Magnitude Approximate Duration (days) Ref. 1 Swarm at BRE, N Iceland 7 M* Swarm at BRE, N Iceland 7 M SW Iceland 10 M Dongfang, Hainan, China 9 M L Enola Swarm, Arkansas 3 M L SW Iceland 14 M Parkfield, California 14 M L SW Iceland 10, 43 M , SW Iceland 10, 43 M , Grímsey Lineament, Iceland 50, 92, 96 M , 163, SW Iceland 2, 36 M , (successfully stress-forecast) 12 Shidian, Yunnan, China 35 Ms N Palm Springs, California 33 Ms , 8, 9 14 SW Iceland 3, 46 Ms 6.6/M , Chi-Chi earthquake, Taiwan 55 Ms b) Observations of inferred stress relaxation/ crack coalescence before earthquakes. 1 Swarm at BRE, N Iceland 7 M Swarm at BRE, N Iceland 7 M Enola Swarm, Arkansas 3 M L Grímsey Lineament, Iceland 50 M SW Iceland 2 M (successfully stress-forecast) 6 Shidian, Yunnan, China 35 Ms N Palm Springs, California 33 Ms , 8, 9 8 SW Iceland 3, 46 Ms 6.6/M , Chi-Chi earthquake, Taiwan 55 Ms c) Observations of changes before volcanic eruptions 1 Gjàlp, Vatnajökull, Iceland (stress-accumulation) 240,240,230 large, fissure eruption Mount Etna, Sicily 1, 5 Small (stress-accumulation and 90º-flips) 3 Mount Ruapehu, New Zealand (90º flips ) 2-15 Small - 13, 14 d) Observations of swarm at the Stress-Monitoring Site, Húsavík, Iceland 1 Grímsey Lineament ~70 M Gao and Crampin (2004); 2. Volti and crampin (2003b); 3. Gao et al. (1998); 4. Booth et al. (1990); 5. Liu et al. (1997); 6. Gao and crampin (2006); 7. Crampin et al. (1999a); 8. Peacock et al. (1988); 9. Crampin et al. (1990,1991); 10. Wu et al. (2006); 11. Crampin and Gao (2005); 12. Bianco et al. (2006); 13. Miller and savage ((2001); 14. Gerst and savage (2004); 15. Crampin et al. (2003). *Iceland seismic catalogue magnitude M m b Observed at more than one seismic station Older Iceland magnitude scale, now changed to M 4.9 for compatibility with other listed values As interpreted by Crampin and Peacock (2008) 35

36 Table A1 Summary of match of APE*-modelling to observations (after Crampin and Chastin, 2003, updated) STATIC EFFECTS Field observations of SWVA (below m depth) 1 SWVA observed in all rocks independent of porosity and geology Minimum SWVA in ostensibly intact rock: observed ~1.5%; APE modelled ~1.0% Maximum SWVA in ostensibly intact rock: observed ~4.5%; APE modelled ~5.5% Narrow range of observed crack density: Proximity of intact rocks to fracture-criticality at crack density = (SWVA ~5.5%), where fracture-criticality can be associated with the percolation threshold. Other field observations 6 Fracture-criticality specifies crack distributions with a range of dimensions of about nine orders of magnitude. Ref. (Obs) Ref. (APE) º-flips in shear-wave polarisations in critically high-pressured reservoirs. 5, 6 2, 5, º-flips in shear-wave polarisations immediately above major tectonic faults. 7, 8, High fluid-pressures on all seismically faults cause ±80% scatter in time-delays. 7, DYNAMIC EFFECTS Temporal changes in SWVA during production procedures 10 Changes in SWVA before and after hydraulic pumping tests Changes in SWVA before CO 2 -injections (CO 2 -sequestration) in carbonate reservoir. 6, 14 6 Temporal changes in SWTD before earthquakes 12 Characteristic changes in SWTDs in retrospect before 14 earthquakes (M 1.7 to Ms 7.7). 7, 11, Successful stress-forecast of time, magnitude, and fault-plane of M 5 earthquake Temporal changes in SWVA before and after volcanic eruptions 14 Characteristic changes observed in retrospect at 240 km along several azimuths before Gjàlp eruption, Iceland and 2 ms/km/year decrease interpreted as Mid-Atlantic Ridge adjustment to magma injection Similar characteristic changes seen above eruptions on Mount Etna, Sicily, as those seen 17 2 before earthquakes, including 90º-flips in shear-wave polarisations º-flips in shear-wave polarisations observed above Mount Ruapehu, New Zealand. 18, 19 2, 10 Changes in SWVA before and after failure in laboratory stress cells 17 Similar changes to those seen before earthquakes before failure in laboratory stress cells. 20 Changes in seismic travel times at prototype SMS in Northern Iceland 18 Prototype SMS records effects of swarm seismicity equivalent to one M 3.5 earthquake at 70 km-distance on borehole observations of P-, SV-, SH-, and SV-SH travel times at 500 m- depth over 315 m-offset, NS and EW GPS measurements perpendicular and parallel to the Húsavík-Flatey Transform Fault, and water-well levels in a well above the fault. Variations of shear-wave behaviour in laboratory experiments 9 19 Ultrasonic variations of SWVA and permeability in uniaxial stress cell Ultrasonic variations of (isotropic) shear-wave velocities for changes in confining pressure and pore-fluid for oil-, water-, gas- (dry) in sandstone cores in stress cells Cont. 36

37 21 Ultrasonic variations of velocity and attenuation from sonic (transducers) to seismic (resonant bar) frequencies. 1. Crampin (1994); 4. Crampin (1997, 1999); 7. Peacock et al. (1988); 10. Crampin et al. (2002); 13. Crampin and Booth (1989); 16. Crampin et al. (1999a); 19. Gerst and Savage (2004); 22. Zatsepin and Crampin (1996); 25. Chapman et al. (1998, 2000). *Anisotropic poro-elasticity Effects compatible with APE 2. Crampin and Zatsepin (1997); 5. Crampin et al. (1996); 8. Liu et al. (1997); 11. Volti and Crampin (2003b); 14. Davis (1997); 17. Bianco et al. (2006); 20. Gao and Crampin (2003); 23. Crampin et al. (1997, 1999b); Heffer and Bevan (1990); 6. Angerer et al. (2002); 9. Crampin et al. (2003); 12. Crampin et al. (2004); 15. Crampin and Peacock (2008); 18. Miller and Savage (2001); 21. King et al. (1994); 24. Sothcott et al. (2000a, 2000b); Shear-wave velocity anisotropy Shear-wave time-delays 37

38 Table B1 Summary of the principal reasons why stress-aligned shear-wave splitting is strongly indicative of fluid-saturated microcracks (after Crampin and Peacock, 2008) Reason for interpretation as fluid-saturated stress-aligned microcracks 1 The only anisotropic symmetry system leading to parallel polarisations within the central band of the shear-wave window is hexagonal symmetry (transverse isotropy) with a horizontal axis of cylindrical symmetry (TIH-anisotropy). 2 The only common geological phenomena having TIH-anisotropic symmetry in all rock types are fluid-saturated stress-aligned microcracks. 3 Fluid-saturated microcracks are the most compliant elements of in situ rocks and changes inhear-wave splitting have been observed in retrospect before (currently) some 15 earthquakes (Table 1). No other source of anisotropy has such immediate compliance. 4 The underlying rationale of the anisotropic poro-elastic (APE) model of the evolution of fluidsaturated rocks, which approximately matches a very large range of phenomena (Table A1), is the existence of closely-spaced distributions of stress-aligned fluid-saturated microcracks prevalent in almost all rocks. 5 Temporal variations of shear-wave splitting following both high- and low-level CO 2 -injections in a carbonate reservoir were exactly matched by APE, hence tending to confirm the crackinduced origin of shear-wave splitting. 6 Highly-diagnostic 90º-flips in shear-wave polarisations caused by critically high pore-fluid pressures are seen in high-pressurised oil reservoirs, and above major seismically active fault planes, again tending to confirm crack-induced shear-wave splitting. 7 Observations of highly scattered shear-wave time-delays above all seismically-active faults are the results of 90º-flips in polarisations caused by critically high pore-fluid pressures on all seismically active fault planes, again tending to confirm crack-induced shear-wave splitting. Ref. 1, 2, 3, 4, 5, 6 1, 2, 3, 4, 5, 6 3, 4, 5, 9, 10 2, 3, 4, 7, 8, 11 4, 9, 11, 12 5, 7, 8, 9, 12, 13 3, 5, 8, Crampin (1981); 4. Zatsepin and Crampin (1997); 7. Crampin et al. (2002); 10. Gao and Crampin (2004); 13. Wu et al. (2006). 2. Crampin (1994); 5. Volti and Crampin (2003a, 2003b); 8. Crampin et al. (2004); 11. Crampin and Zatsepin (1997); 3. Crampin and Chastin (2003); 6. Crampin (1999); 9. Crampin et al. (1999a); 12. Angerer et al. (2002); 38

39 Table D1 Summary of variations in Figure 7 at the SMS at Húsavík, August, 2001 (after Crampin et al., 2003) Figure No. Nature of phenomenon Approximate size or amplitude Approximate duration (days) Seismic variation (%) 7a P-wave travel-time: instantaneous increase followed by linear decrease. 7b* SH-wave travel-time: approximately constant level followed by S-shaped decrease to a constant level. 7b* SV-wave travel-time: approximate constant level followed by S-shaped decrease to a constant level. 7c* SV SH travel-time anisotropy: scattered values followed by irregular increase to a constant level. 5 ms ms 4, ms 3.5, ms 4, d North-South GPS displacements: instantaneous increase followed by exponential decrease. 7 mm 11-7d* East-West GPS displacements: impulse followed by increase to permanent displacement, appropriate for movement on approximately EW fault. 3 mm, 4 mm 4, 9-7e* Pressure in bars at 30 m-depth in water-well on Flatey: tidal variations throughout, with pulse drop in level. Equivalent to 40 cm ocean-tides, 1 m temporary drop in water level ~0.5, 5-7f Half-day histogram of seismicity within 100 km of SMS. Initial burst of activity is on Grímsey Lineament, ~70 km from SMS. Initial activity has 106 events, M 2.8, with energy equivalent M ~ *Variation includes two phases with both amplitudes and both durations are listed as appropriate 39

40 Table F1 The benefits of GEMS. 1) Provide data to stress-forecast of times and magnitudes of all damaging earthquakes worldwide with magnitudes greater or equal to M 5 (and many greater than M 4). 2) Provide facilities to monitor the effects of massive hydraulic fracturing operations to optimise stress release to mitigate earthquake hazards threatening vulnerable locations. 3) Provide data for a stress-forecasting service, analogous to weather forecasting, which would give longer-term estimates of earthquake occurrence and hazard, as well as Earth evolution. 4) Provide a network of deep boreholes for passive monitoring of broadband seismics, gravity, resistivity, magnetism, etc., in exceptionally-quiet environments for time-lapse monitoring of the dynamics of Earth evolution. 5) Provide a new controlled-source tool for monitoring the evolution of the crack-critical Earth to stimulate geoscience at the beginning of the 21st century. 40

41 FIGURE CAPTIONS Figure 1 Schematic illustration of seismic shear-wave splitting in stress-aligned fluid-saturated microcracks aligned normal to the direction of minimum horizontal stress. Such parallel vertical microcrack orientations are typically found below the critical depth where the (increasing) vertical stress, V, equals the minimum horizontal stress, h. This is usually between 500 m- and 1000 m-depth. For near vertical propagation, the polarisation of the faster split shear-wave is typically parallel to the direction of maximum horizontal stress, H. Figure 2 Cross-sections of uniform dimensionless distributions of parallel stress-aligned fluidsaturated penny-shaped microcracks for observed percentages of shear-wave velocity anisotropy, crack density, and crack radius a (after Crampin, 1994). Figure 3 Schematic but quantitatively-accurate illustration of the Anisotropic Poro-Elastic (APE) dimensionless model of the evolution of aspect-ratios of stress-aligned fluid-saturated microcracks under changes of four values of increasing stress, s h. Pore-fluid volume is preserved and aspect-ratios are chosen to give a porosity of 5%. The mechanism of deformation is fluid-movement along pressure gradients between neighbouring microcracks at different orientations to the stress field. Figure 4 The first successfully stress-forecast earthquake. The figure shows time-delays (normalised to ms/km) at Station BJA in SW Iceland for four years (Volti and Crampin, 2003a, 2003b). The irregular curves are nine-point moving averages summarising variations in the normalised time-delays. The lower diagram is time-delays in Band-1 ray-path directions sensitive to crack aspect-ratios, and hence to stress-accumulation. The straight lines in Band-1 are least-square fits to increasing time-delays, starting one point before a minimum of the moving average and ending at a larger earthquake, or eruption, when the time-delays approach levels of fracture-criticality determined from previous events. The upper diagram is time-delays in Band-2 which are sensitive to crack density (after Crampin et al., 1999a). The large ±80% scatter in both Band-1 and Band-2 is caused by critically-high pore-fluid pressures, which cannot be reduced or eliminated (Section A2.2). The large observational scatter dominates the statistics of the regression lines and is irrelevant to the relative merits of the individual least-squares fits. Figure 5 Earthquake magnitude plotted against (a) logarithm of the duration of stress-accumulation and (b) logarithm of the duration of crack coalescence. Numbers refer to numbered events in Table 1a and 1b. Events with open durations (Table 1a, nos 1, 2, 5, and 7) are not plotted. When different stations give different values for the same event, values from the nearer station are plotted. Figure 6 Optimum ray-path geometry for stress-monitoring sites. The downhole orbital vibrator (DOV) in the source borehole at X m-depth radiates shear-waves downwards in Band-1 directions to three-component geophones from X m- to X m-depth in two boreholes at 300 m-offset in azimuthal directions 30º either side of the direction of minimum horizontal stress. X needs to be greater than the critical depth where increasing vertical stress, V, equals the minimum horizontal stress, h. 41

42 Figure 7 Variations at the SMSITES SMS from August 8-24, 2001 between two boreholes parallel and offset ~100 m South of the ~EW surface break of the Húsavík-Flatey Fault (HFF) where the fault runs onshore in Northern Iceland. Travel times in ms at ~500 m-depth over 315 m of: (a) P-waves (black dots); (b) SV-waves (green crosses) and SH-waves (blue crosses); (c) SV SH (red circles); (d) GPS displacements in mm North-South across the HFF (blue circles) and East-West parallel to the HFF (red crosses); (e) pressure at 33 m-depth in water well on Flatey Island in bars showing ocean tides and an anomalous ~1 m-drop in water level (blue); and (f) twelve-hourly histogram of seismicity within 100 km of SMS, Húsavík (black) (after Crampin et al., 2003). Figure A1. SWAS semi-automatic measurement of small (M 0.55) earthquake recorded within the shearwave window at Station BJA in SW Iceland measured by the (after Gao et al., 2006). (a) Screen images of seismograms, from top: EW-, NS-, Vertical-, and rotated horizontal Fast- and Slow-polarisations with second marks. 0.1 s time intervals are marked for the S- wave polarisation diagrams (PDs) in (b) and the fast and slow shear-wave arrivals are marked by vertical bars. (b) Screen images of mutually-orthogonal PDs from the top are: sagittal section; horizontal view from source (Up, Down, Left, and Right from the source), and horizontal section (Towards, Away, Left, and Right from the source). Each column of PDs shows particle motion in the 0.1s (10 sampling points) time-intervals marked in the seismograms spanning the P-wave and S-wave arrivals in a) and b). The x N marked above each column indicates relative scaling. SWAS/ES fast and slow shear-wave picks are marked by circles in the horizontal PDs in-time interval X. At the bottom is an enlarged horizontal polarisationdiagram spanning the selected fast and slow picks. The template shows current selection of row, column, first point, and last point of the time-delay, and directions for visual adjustment. The line marks the average polarisation between the two circled points. Small ticks on the PDs mark time-series samples, which are outward facing for clockwise rotation, and inward facing for anti-clockwise rotation. Figure A2. Ray-path geometry for Band-1 and Band-2 directions. ABCD is a section of the crack plane of parallel vertical cracks passing through the recorder position S on a horizontal freesurface. Band-1 directions to the horizontal free-surface, where time-delays are sensitive to crack aspect-ratios. are those within the solid angle EFGHS subtending 15º to 45º to the crack plane (within the effective shear-wave window at 45º, Booth and Crampin, 1985). Band-2 directions to the horizontal free-surface, where time-delays are dominated by crack density, are those within the solid angle ADEHGS subtending 0º to 15º to the crack plane. Both Band-1 and Band-2 directions include the equivalent solid angles reflected on the other side of the imaged crack plane. 42

43 h Therearefluid-filedmicrocracksin amostalrocks. H O V Thesecracksarethemost mobile ementsoftherockmasand H mediatelyshowthefectsof any O changestres. Thenewunderstandingofrock deformationisthathedetailed efectsofchangingstrescanbe monitoredbyseismicshear-wave spliting. V Figure 1. Figure 2. 43

44 Figure 3. I I I Figure 4. 44

45 a) b) Figure 5. 45

46 N30ºW 0º N30ºE R 300m X km DOV 45º 45º R kmm 17º 17º R R Figure 6. 46

47 Figure 7. 47