Speculations on Earthquake Forecasting
A new understanding of the way fluid-saturated microcracked rocks deform suggests that networks of stress-monitoring sites could lead to earthquake forecasting analogous to the way networks of meteorological stations allow weather forecasting and prediction of storms. This new understanding is believed to be a major advance that offers more hope for forecasting earthquakes than was envisioned in the survey Living on an Active Earth: Perspectives on Earthquake Science of the National Research Council's Committee on the Science of Earthquakes, summarized by Tom Jordan in the Opinion for August 2002. This Opinion may also be considered as a response to the plea by John E. Ebel in the Opinion for January 2003, for seismologists to begin forecasting earthquakes.
The linearity and self-similarity of the well known Gutenberg-Richter relationship demonstrate that the Earth behaves as a complex critical system with self-organized criticality, where minor disturbances (small earthquakes or fractures) may cascade into major fractures and large earthquakes. Such criticality is claimed to specifically exclude the possibility of the deterministic prediction of the time, magnitude, and location of impending large earthquakes. Resolving this fundamental difficulty is one of the four complementary perspectives of earthquake research recognized by the NRC report, and as Jordan suggests provides "the opportunity for exciting basic science." We here speculate on some of this new exciting science.
Although as suggested criticality makes the deterministic prediction of the time, magnitude, and location of impending large earthquakes rather unlikely, large earthquakes cannot take place without the accumulation of sufficient stress for the appropriate magnitude energy release. Both observations and the theory of the evolution of fluid-saturated microcracked rock under stress (anisotropic poro-elasticity or APE) demonstrate that in situ rock is weak to shear stress. This means that the accumulation of sufficient stress energy before a large earthquake necessarily takes place over very large volumes of rock: probably tens to hundreds of millions of cubic kilometers before the largest earthquakes, with deformation extending for thousands of kilometers. The APE mechanism for low-level deformation before fracturing takes place (prefracturing deformation) is fluid movement along pressure gradients between the neighboring grain-boundary cracks or low aspect-ratio pores present in almost all rock. The principal effect is that differing pore-pressures modify crack aspect-ratios in systematic ways that can be monitored by analyzing shear-wave splitting along a specific range of solid-angle ray-paths. APE theory suggests and numerous observations appear to confirm that the effects of increases in such low-level stress can be monitored almost anywhere within a very extensive volume surrounding the eventual source zone.
Note that much of the recent published and unpublished work on which these ideas are based can be found at http://www.glg.ed.ac.uk/~scrampin/opinion/. The various results show that APE modeling matches a large range of phenomena referring to cracks, stress, and shear-wave splitting in almost all rocks with only a few well understood exceptions. This suggests that the APE mechanism is at least a good first-order approximation to fluid-rock deformation.
On those comparatively rare occasions when a persistent swarm of small earthquakes allows shear-wave splitting to be analyzed near an impending large (or larger) earthquake, the rate of increasing stress can be monitored by the effects of the changing microcrack geometry on shear-wave splitting. The time of the earthquake can be assessed by the time at which the level of cracking reaches levels of fracture criticality estimated from previous earthquakes, and the magnitude is proportional to the duration of the increase before criticality. We call this stress forecasting. Seen with hindsight in the laboratory and before about a dozen earthquakes worldwide (magnitudes range from mb 1.7 to Ms 6), on one occasion the effects have been used to successfully stress-forecast the time and magnitude of a mb ≈ 5 earthquake in southwest Iceland. On that occasion, knowledge that a larger earthquake was approaching allowed other phenomena (local seismicity) to be used to identify the activated fault. We claim this as the first successfully stress-forecast earthquake, and it also might be the first earthquake deterministically forecast from geophysical as opposed to statistical phenomena. Others have suggested that we were lucky, and perhaps we were, but it was the first time we have ever made such a stress forecast, and we got it spot on.
To stress-forecast large earthquakes using seismic swarms as a source of shear-wave signals requires suitable source-network geometry to monitor shear waves along the particular solid-angle directions most sensitive to low-level changes of stress. Such stress-forecasting also requires sufficiently persistent swarm activity to provide reliable shear-wave source signals throughout the build-up of stress. There were no source earthquakes for the initial seven weeks of the build-up of stress for the largest earthquake for some decades in southwest Iceland. As a result the increase was not recognized in time and we failed to forecast the earthquake. It is clear that reliable and routine stress-forecasting needs a dependable shear-wave source, and controlled-source measurements are required. The suggested option is stress-monitoring sites (SMSs) using cross-hole seismics between 1-km- to 2-km-deep boreholes so that shear-wave splitting can be monitored along appropriate ray-path directions. The wells need to be deep enough to be below the stress-release and weathering anomalies in the uppermost few hundred meters.
The first (preliminary) SMS was developed in the (European commission-funded) SMSITES Project on the onshore continuation of the Húsavík-Flatey Fault of the Tjörnes fracture zone of the Mid-Atlantic Ridge at Húsavík in northern Iceland. We used a source at 500-m depth to record horizontally propagating seismic waves between boreholes. Recording continuously for two weeks, we were fortunate to record well observed anomalies in P, SH, and SV waves correlating with small-scale seismicity at 70 km and also correlating with a well-level change and north-south and east-west GPS measurements.
This unique data set shows the detailed response of crustal rock to a comparatively minor disturbance at a considerable distance. The data have not yet been fully interpreted and contain several puzzles, including the varying durations and forms of the anomalies, that are likely to yield crucial information about how rocks deform. What they do clearly demonstrate is the sensitivity of crustal rock to small disturbances. A mb 4 earthquake with the same energy release as the small-scale seismic activity at 70 km is a comparatively small earthquake, yet its effects are observed at a distance of several hundred times its likely source diameter. Such sensitivity is far too great to be explained by a conventional brittle-elastic crust, where stress is proportional to strain, but is expected in a compliant crack-critical (CCC) crust verging on criticality as implied by the success of the APE mechanism of deformation. The butterfly's wings in Brazil causing the tornado in California is the classic analogy for the sensitivity of criticality. Thus, although the geometry of the SMS is not optimal for monitoring stress-induced changes by shear-wave splitting, the SMSITES experiment has recorded a unique data set. These data verify that the crust is highly compliant and go some way to confirm both the science and technology of SMSs for stress-forecasting the times and magnitudes of impending large earthquakes in a CCC crust.
It is worth noting that stress (force) and strain (displacement), although strongly interacting and in many respects dependent, have different behaviors before earthquakes. Strain can be released by aseismic slip and slow earthquakes (and by APE-type fluid-rock deformation), whereas stress is released when crack distributions are so strained that they reach fracture criticality, when rocks fracture and earthquakes occur. This means that, for example, GPS measurements of displacement (strain) cannot invariably identify the approach of earthquakes, as strain may be released aseismically. The value of observations of shear-wave splitting at stress-monitoring sites is that they directly monitor the approach of crack distributions to fracture criticality when earthquakes occur.
Networks of Stress-monitoring Sites
Current experience and practice suggests that single SMSs can monitor changes of stress and can stress-forecast the times and magnitudes, but not necessarily the locations, of impending large earthquakes at distances of several hundred kilometers from the SMS. This is analogous to a single barometric weather station, where a rapid drop of air pressure indicates the likelihood of nearby storms. Networks of (usually satellite-linked) weather stations, where the principal measurements are air pressure and wind direction and speed, allow systematic patterns and variations of weather to be recognized so that weather can be forecast over larger areas (with all the inherent uncertainties of critical systems).
It is suggested that networks of SMSs, measuring stress directions and implied changes of stress, would be new tools for monitoring patterns of in situ stress not previously available. Such networks would provide unique information and allow changing patterns of stress to be examined for the first time. A grid spacing of 200 km to 300 km, say, would probably allow the times and magnitudes of M 5 or greater earthquakes to be stress-forecast within the network, although the lower magnitude range would need to be estimated in practice for the particular network and region. It is likely that such mapping of the spatial distribution of stress changes would provide greater insight into the process of stress accumulation before earthquakes than any of the more limited measurements of stress currently available.
Such networks of course would be extremely expensive to set up and maintain. Up to now research into predicting earthquakes has traditionally been done on a shoestring. Not surprisingly, shoestrings do not stretch very far. Most investigations of complex critical systems require expensive instrumentation: networks of weather stations and particle accelerators, to name but two. It would be indeed be surprising if investigating the detailed behavior of the immensely complicated heterogeneous crack-critical Earth could be done on the cheap.
The New Understanding
This new understanding of the behavior of the compliant crack-critical (CCC) crust where prefracturing deformation can be monitored with shear-wave splitting may be considered as a new geophysics. Analogous to what is sometimes called the new physics, this geophysics is new because it recognizes an extremely compliant crust far beyond the compliance expected by conventional geophysics. It is also new because it opens the possibility of exerting some control over Earth processes. The new geophysics has disadvantages (limits to the resolution of conventional geophysical interpretation) but also has important benefits. In particular, prefracturing deformation can be calculated by APE, effects of known changes predicted, and in some circumstances possibly even controlled by feedback. Thus, this new geophysics presents an opportunity for a wholly new approach to earthquake hazards (and incidentally monitoring hydrocarbon production).
One enormous advantage of this approach is that it directly monitors the approach to fracture criticality that triggers the earthquake and monitors deformation before failure, before faulting, and before earthquakes occur. That is, it monitors the behavior of the largely unperturbed rock mass where shear-wave splitting is sensitive to microcrack geometry rather than large-scale closed, or partially closed, faults and fractures. The behavior is calculable because the CCC crust verges on criticality. Once fracture criticality is reached and failure and earthquakes occur there is deterministic chaos. As is sometimes attempted, the effects at any earthquake source may in some circumstances be calculated or modeled by one-off calculations for specific fault and fracture geometries and other particular local and regional conditions. In general the results are likely to be highly sensitive to detailed initial conditions (the "butterfly effect") with no universality, no generality, and no repeatability. It is the "predictable" behavior of the intact rock mass verging on criticality before the earthquake occurs that is the key for stress-forecasting earthquakes, and this is what networks of stress-monitoring sites could provide.
Final Speculations: The Possibility of Mitigating the Potential for Large Earthquakes
As suggested, the weakness of fluid-saturated microcracked rock implies that the stress accumulation before large earthquakes takes place over very large volumes of rock. In principle, the overall increase in stress could be reduced by the release of accumulated stress almost anywhere within the larger stressed volume. Such reductions could be made by a program of hydraulic fracturing operations, or possibly controlled explosions at depth. Hydraulic fractures would not necessarily have to take place along known active faults, unlike the controlled stress release advocated in the 1960's and 1970's. By monitoring the effects of stress-release operations by the network of SMSs, observers would be in control and could optimize the effects and hence utilize the results. In particular, if a network of SMSs indicated that stress was increasing in an area that threatened a major city or other vulnerable location, the overall stress accumulation could be mitigated by hydraulic fracturing operations in a nonvulnerable area (perhaps offshore) within the larger stressed volume. The potential for a large earthquake could be reduced. Further development and testing of the theories discussed in this paper should reveal whether such seeming far-fetched concepts are in fact viable.
This work was partially supported by the European Commission SMSITES and PREPARED Projects, contract numbers EVR1-CT1999-40002 and EVG1-CT2002-00073. Yuan Gao was supported partly by China MOST under Contracts 2001BA601B02 and NSFC Project 40274011, and partly by the UK Royal Society Fellowship Program.
Stuart Crampin, Yuan Gao*, and Sebastien Chastin
*also at Center for Analysis and Prediction
To send a letter to the editor regarding this opinion or to write your own opinion, contact the SRL editor by e-mail.
Posted: 22 May 2003