Subduction zones generate the world’s largest and most destructive earthquakes, most of the world’s tsunamis, and most of the world’s explosive volcanoes. They are also the sites where much of the world’s population is concentrated (the coastal zones) and, over geologic time, where most of the earth’s continental crust and mineral resources are generated. NSF’s MARGINS program includes the Seismogenic Zone Experiment (SEIZE) to study the shallow subduction plate interface that is locked and accumulates elastic strain, periodically released in large or great earthquakes. The scientific rationale for these studies was originally outlined in the SEIZE science plan, based on a workshop held in Hawaii in 1997. The MARGINS program officially began in 1998, and has provided funding for US researchers for focused studies in the Nankai Trough and Central America, complementing research funded by Japan, Germany, Costa Rica, Nicaragua and other nations. It is appropriate to re-evaluate this plan based on these and other data as well as new laboratory and theoretical developments. The SEIZE Science Plan update was carried out in association with the SEIZE 2003 Theoretical Institute in Snowbird, UT, March 2003.
The science plan derived from the 1997 meeting focused on the following questions:
1. What is the physical nature of asperities?
2. What are the temporal relationships among stress, strain and pore fluid composition throughout the earthquake cycle?
3. What controls the updip and downdip limits of the seismogenic zone of subduction thrusts?
4. What is the nature of tsunamigenic earthquake zones?
5. What is the role of large thrust earthquakes in mass flux?
Here we provide some additional scientific background to these questions and discuss the extent to which the MARGINS program has begun to address them (section II), pose some new questions (section III), and revise a plan for obtaining the requisite knowledge (section IV).
Understanding the factors leading to Earth’s largest and most destructive earthquakes is obviously an important goal. While the rupture process as well as factors controlling strain accumulation could be different for smaller (M<6) vs. larger (M>8) events, simple earth-quake scaling relations suggest that, for similar types of focal mechanisms, earthquake moment scales with rupture area over a relatively large range of moment. To a first approximation, we understand why subduction zone earthquakes release the great majority of earth’s seismic energy (they are capable of rupturing large areas) although we do not yet understand the factors that occasionally lead to extremely large rupture areas and truly great (M>9) subduction zone events. While some subduction thrusts produce M~9, others only produce M<7.5. Why? What are the relative roles of fault area, seismic coupling, seismic vs aseismic slip, asperities, type and thickness of subducted sediments, and fluid flow?
Lay and Kanamori (1981) described a model that relates the size of an earthquake to the characteristic size of asperities, defined by these authors as regions that slip by large amounts in an earthquake. In this model, events like the 1960 Chile earthquake rupture large asperities and are characterized by large displacements (maximum displacement in this event probably exceeded 20 m). Regions that typically have smaller earthquakes would be characterized by smaller slip on smaller asperities, limiting rupture area. Unfortunately it has proven difficult to relate asperities to physical features.
There is even some ambiguity in the literature on the use of the term. Do asperities represent regions of high strength (high friction) such that large amounts of locked slip accumulate in the interseismic period, to be released in a subsequent earthquake? Or are they regions of low dynamic friction that are therefore able to slip by larger amounts in an earthquake, promoting rupture over long distances. In this latter view, a high strength region may actually block further propagation of earthquake slip. Are asperities only the source regions for past earthquakes, or are they also regions that are currently locked, to rupture in the future? Are these two definitions in fact the same? In other words, do asperities have enduring significance? If so, are they controlled by specific fault properties or features on the subduction interface? Can these be measured directly or otherwise quantified in any way, e.g., by models? Ultimately, comparing the “patchiness” of locked slip on the subduction interface as measured by interseismic geodetic data to subsequent slip distribution in the next large earthquake (measured seismically or geodetically) may help resolve this problem. “Imaging” the fault surface with active source techniques may also reveal physical features or changes in reflective properties that are diagnostic.
In the ~20 years since the asperity model was first discussed, there has been surprisingly little progress in resolving these issues. Part of the philosophy of the MARGINS-SEIZE program is to accumulate a variety of key data sets in specific focus sites (Central America, Nankai Trough) to better attack such problems.
What controls maximum earthquake size in subduction zones? Why do subduction zones occasionally generate the largest known (M>9) events? A simple empirical model incorporating only plate convergence rate and lithospheric age has been proposed (Ruff and Kanamori, 1980). This model is plausible if one considers only the down-dip width of the seismogenic zone; since earthquake size scales with rupture area, faster convergence rates tend to depress the brittle-ductile transition depth, and young, buoyant lithosphere will tend to subduct at shallower angle, all promoting a larger plate contact area in the seismogenic regime and resulting in larger down-dip width. On the other hand, a key characteristic for the largest subduction zone earthquakes is the along-strike rupture length, and this does not relate in any simple way to plate subduction rate or age.
Once earthquakes are large enough to rupture down to their maximum depth, earthquake moment scales with the along-strike length of the rupture (e.g., Scholz, 1990). Thus, understanding variations in earthquake moment in the subduction environment requires that we understand variations in rupture length. The great 1960 Chile earthquake had the longest recorded rupture length, ~1000 km (Plafker, 1972). Shorter segment ruptures sometimes appear to correlate with certain sea floor bathymetric features such as seamounts or oceanic plateaus (Figure 1), but this does not explain why the rupture zones of rare great earthquakes pass through and connect several such features to achieve very long rupture lengths. At this point we have very little understanding of factors controlling rupture length, although the smoothness of the incoming oceanic plate may play a role, and this could be affected by incoming sediment (see later).
Another factor limiting the applicability of simple models such as those focusing only on plate rate and age is that the rupture length, slip and seismic moment may vary considerably from cycle to cycle along a given trench segment (Thatcher, 1990; Schwartz, 1999). This variability suggests that there may be temporal changes in key properties on the plate interface (Figure 2).
Many studies have compared seismic moment release to the full plate convergence rate, and report either “seismic coupling”, the ratio of seismic slip to plate motion (Kanamori, 1977) or the amount of missing slip as the percentage of “aseismic” slip. Although these were initially assumed to reflect physical coupling on the plate interface and be consequences of convergence rate and age of the subducting plate, further study has suggested that such correlations are weak (Pacheco et al., 1993). While there are other difficulties and problems in estimating seismic coupling and locked slip from seismic data (McCaffrey, 1997; Norabuena et al., 1998), the large numbers of such studies that point out differences between the full plate rate and seismic rate, as well as newer geodetic studies suggesting that some seismogenic zones do not appear to be fully locked, requires explanation.
One difficulty is that there are significant trade-offs between geodetically determined “coupling” and a variety of other parameters, including the post-seismic response to past earthquakes, assumed down-dip width, dip, and curvature of the subduction zone, and these need to be fully investigated before we can obtain rigorous, quantitative estimates of the amount of locked slip on the plate interface. Accurate estimates of locked slip, how this varies with depth, along strike, and with time, are not yet available for most subduction zones. Geodetic studies can measure strain accumulation above the seismogenic zone, and have documented fully coupled (locked) seismogenic zones, freely slipping seismogenic zones, and perhaps partially coupled zones. What is not clear is whether partially coupled zones represent regions with uniform properties, or a spatially more heterogeneous plate interface characterized by small, fully locked patches intermixed with freely slipping zones, with the “mixture” poorly resolved by available data. It is also possible that the explanation for “partial coupling” reflects a temporally heterogeneous interface with time-varying properties that are “averaged out” in campaign-style geodetic measurements (this latter problem will presumably be resolved once continuous geodetic monitoring capability is well-established).
A series of seismic reflection studies offshore Costa Rica (von Huene et al., 2000; Ranero et al., 2001; McIntosh et al., 2000) have imaged the locations of subducting seamounts, and these are implicated as source regions for several M=6-7 earthquakes during the past decade (Protti et al., 1995; Bilek et al., 2003). These studies document our ability to tie physical features such as seamounts with earthquake sources, and potentially with geodetically determined locked patches. The regions where seamounts are clearly implicated as seismic sources are also those in which the earthquake magnitude is less than 7.5. Larger earthquakes (Mw>8.0), are often associated with regions of thick trench sediment (Ruff, 1992) and/or a strong forearc (McCaffrey, 1995).
Transient Strain. An important class of phenomenon that was not well understood at the time of the first SEIZE workshop in 1997, but is now becoming more widely appreciated, concerns transient strain events. Several of these events have been imaged with continuous GPS in Japan, Cascadia and other subduction zones in the last few years (the requisite network does not yet exist in Central America). Future SEIZE research will undoubtedly focus on the implications of these time-transient phenomena for seismogenic zone processes.
Slow transient deformation in subduction zones may be quite common, but until recently has been only rarely observed. Slow/silent earthquakes were initially postulated on the basis of borehole strainmeter data (Sacks et al., 1978; Linde et al., 1996), elevation data (e.g, Linde and Silver, 1989), and low frequency seismic observations (e.g., Beroza and Jordan, 1990). The advent of GPS, and reductions in the cost of receiver hardware and the complexity of data reduction, facilitate dense GPS networks and much better recording of this important class of geophysical phenomena, largely outside the frequency band of seismometers. Several transient events related to subduction have now been recorded in Japan (Heki et al., 1997; Hirose et al., 1999; Ozawa et al., 2001), Kamchatka (Burgmann et al., 2001), Mexico (Lowry et al., 2001) and Cascadia (Dragert et al., 2001).
In fact, every subduction zone that has been instrumented with continuous GPS, even relatively sparse networks, have observed these phenomena, suggesting that they are quite common. Below we briefly discuss some of the classes of transient strain phenomena that have been described in the literature. With the exception of viscoelastic relaxation in the lower crust and upper mantle, these phenomena likely involve temporal variations in pore fluid pressure (either as cause, effect or both), reflecting either changes in fluid production (e.g., from metamorphic reactions) or changes in permeability. Variation in pore fluid pressure within or near the plate boundary is particularly relevant to the issue of strain accumulation and release because of its direct control of effective normal stress on the fault plane, as first pointed out by Hubbert and Rubey (1959). Observations of transient fluid flow in “CORKed” ODP holes are increasingly common, lending support to the idea that variations in fluid flow in the subduction environment play a key role in a variety of processes. Obara (2002) has recently reported nonvolcanic tredmor at the down-dip edge of the Japan seismogenic zone and noted its possible relationship to fluid migration in this critical region. Section 2.6 focuses on fluids specifically..
Precursory phenomena (“Fore slip”). A long-standing and controversial question deals with the presence or absence of observable phenomena precursory to seismic failure. Fore-slip, anomalous motion immediately before a major earthquake, has been recognized in a few cases, for example the great 1960 Chile earthquake (Linde and Silver,1989). However, well-documented fore-slip observations are exceedingly rare, and obviously much more difficult to obtain than afterslip observations, since they generally require data collection protocols (e.g., a reliable continuous network) to be in place before the event. The great majority of earthquakes have no clear precursory signals (Geller et al., 1997), implying that such signals are either rare—consistent with the hypothesis that earthquakes are essentially unpredictable—or of low amplitude and/or so close in time to the main earthquake that they are difficult to observe with current geodetic techniques. Perhaps we are not monitoring with the right instruments and/or with the right frequency window.
Laboratory studies that take samples to frictional failure have documented precursory failure phenomena, perhaps related to exponential crack growth and precursor dilatation, i.e., crack opening prior to rupture. Better documentation of the laboratory conditions leading to such behavior may lead to improved understanding of when, where and why such behavior does (or does not) occur in subduction zones, and guide appropriate field measurements. Geodetic instrumentation capable of high precision continuous monitoring (e.g., borehole strain and tilt, continuous GPS) needs to be installed, and rigorous data analysis/interpretation protocols need to be in place.
After-slip, with periods of days to a year, may occur after major earthquakes, and may propagate down-dip or up-dip from the main thrust plane, or be confined largely to the location of the main shock. It decays logarithmically, with the majority of motion occurring within the first few hundred days after the main earthquake. Hutton et al., (2001) report measurable motion 3.5 years after the Mw 8.0 1995 Colima-Jalisco earthquake in Mexico. In some cases, cumulative moment from these events approaches or exceeds that released in the main earthquake (e.g., Heki et al., 1997; Burgmann et al., 2001). Consequently after-slip may in part explain the observation that seismic moment released at subduction plate boundaries is less than the total moment expected if all plate motion is accommodated seismically. The strain released as afterslip should be observable in geodetic fault locking studies prior to the earthquakes, along with the coseismic slip, and hence be distinguishable from steady aseismic interseismic slip.
Afterslip observations have a number of important applications, one major one being to test rate- and state-variable friction laws (Marone et al., 1991; Marone, 1998; Hutton et al., 2001). These are now widely believed to be the appropriate friction constitutive law governing the earthquake process. Accurate estimation of decay time, as well as rigorous numerical tests of whether observed decay is in fact logarithmic, and remains logarithmic over the entire observing period, would enable tests of rate/state friction laws with field observations.
Poroelastic deformation can be significant following a large earthquake, as demonstrated for the Landers earthquake (Masterlark and Wang, 2002; Peltzer et al., 1996). Masterlark et al., (2001) predict several centimeters of transient poroelastic deformation following the 1995 (Mw=8) Jalisco-Colima subduction zone earthquake. All too often, the poroelastic deformation is ignored and inadvertently lumped together with either afterslip or viscoelastic relaxation.
Slow slip events can occur on the subduction interface with no clear relation to major earthquakes. Dragert et al., (2001) document an event in the Cascadia subduction zone, where about 2 cm of slip on the plate interface propagated up-dip over a broad area over a period of several weeks, with moment equivalent to a M~6.7 earthquake. Similar events appear to have occurred in a quasi-periodic fashion in the past (Miller et al., 2002). Ozawa et al., (2001) describe an event where up to 20 cm of slip on the subduction interface in the Nankai Trough accumulated over a period of about 1 year. Lowry et al., (2001) document a similar event in southern Mexico on the Middle America trench. Observing more such events, understanding why they occur, and why they do not always lead to instabilities that lead to earthquakes, are critical research areas. To our knowledge, all subduction zones that are well instrumented with GPS have now observed such events, suggesting that they may in fact be quite common and may also contribute to the apparent discrepancy between total (plate motion-related) moment and seismic moment.
Earthquake-stimulated viscoelastic flow. For periods longer than about one year, it is generally believed that deviations from steady, interseismic strain accumulation reflect viscoelastic relaxation in the lower crust and upper mantle, stimulated by coseismic rupture, afterslip, and other stress transfer processes. Such deviations may persist for several years (perhaps several decades, depending on viscosity) after the main earthquake. In theory, upper crustal flow due to viscous response of the lower crust could also be stimulated by the slow slip events described above, but the signal may be small. All such observations of surface deformation can in principle be inverted to obtain information on the rheology of this important region, otherwise inaccessible for study.
Combinations. Of course real earthquakes generate multiple deformational mechanisms, and increasingly our data are adequate to resolve two or three for a given event. Pollitz et al., (1998) and Azua et al., (2002) used a combination of viscoelastic and afterslip deformation mechanisms. Masterlark and Wang (2002) demonstrate that a combination of viscoelastic and poroelastic deformation is required to account for transient postseismic deformation.
One focus of the 1997 SEIZE workshop concerned the factors controlling the depth limits of the seismogenic zone. The up-dip and down-dip limits of rupture in great subduction-thrust earthquakes are important for assessing and understanding seismic and tsunami hazard, and more generally for understanding the physical processes involved in generating earthquakes. The down-dip limit determines the landward extent of the seismic source, important for assessing earthquake hazard at inland localities. An accurate “mapping” of this feature also helps us understand the physical processes that control locking on the plate interface. For example, does plate interaction change from locking to stable sliding at a thermal boundary, at a mineralogical boundary that is thermally controlled, or at a mineralogical boundary controlled by pressure (e.g., transition to phases with lower water content)?
Great earthquakes have variable maximum depth of rupture, ~10-50 km. Potential factors controlling this limit related to properties of the incoming plate include (a) composition; (b) temperature; (c) fault material state change; and (d) fault zone seismic “coupling”. The elastic properties of the upper plate may also play a role, by increasing the total seismogenic contact area. As is the case in continental fault zones there appears to be a thermal limit of about 350°C (e.g., Hyndman and Wang, 1995), in agreement with laboratory data for the maximum temperature for velocity-weakening seismic behaviour in rocks of crustal composition (e.g., Tse and Rice, 1986; Blanpied et al., 1991; 1995). Detailed thermal models show that the downdip limit of great earthquakes and of the interseismic locked zone agrees well with this temperature for hot subduction zones, including SW Japan (Nankai), Cascadia, Mexico, and S. Chile (Hyndman et al., 1995; Hyndman and Wang, 1995; Oleskevich et al., 1999; Currie et al., 2002). However, for cold subduction zones this critical temperature is at great depth and another limit must apply. That second limit may be the forearc mantle, which is inferred to be serpentinized and aseismic (Hyndman et al., 1997). The forearc mantle is usually at 35-45 km depth for continental subduction zones and ~10 km for island arcs. Because of their structure, serpentine minerals are expected to be aseismic, but laboratory data do not yet give a clear story for their aseismic or seismic behaviour. Peacock and Hyndman (1999) argue that the forearc mantle in contact with the thrust should contain significant amounts of talc due to rising silica rich fluids. Talc is a very weak mineral and is unlikely to act seismically. The thrust intersection of the forearc mantle agrees with the downdip seismogenic limit in many subduction zones, but there are discrepancies, for example in N. Japan and the Aleutians, so questions remain.
In a general way we understand the up-dip limit for subduction zone seismicity: porosity losses and mineral dehydration reactions in response to increasing pressure and temperature initially release large amounts of water from subducted sediment and oceanic crust, probably maintaining high pore fluid pressures on the fault interface and limiting frictional build-up of shear stress and strain. At some point, however, the rate of fluid loss decreases, increasing effective stress and potentially allowing build-up of elastic strain and seismogenic behavior (e.g., Moore and Saffer, 2001).
At the beginning of the SEIZE initiative in the mid-1990’s there was considerable focus on the transition from smectite to illite clays as a possible control on the up-dip seismic limit. This transition is known to coincide with the 100°C-150°C isotherm, which in many subduction zones approximately marks the up-dip seismic limit. Further heat flow measurements at a number of subduction zones (including the Nankai Trough and Costa Rica margin) and modeling of the estimated temperatures as a function of landward distance and depth have led to a refined picture of thermal state of the seismogenic zone. While there are still large uncertainties in thrust temperatures and estimates of the updip limits, results are consistent with a thermal control of the updip limit at 100°-150°C. However, new laboratory data do not support a simple model of smectite-illite representing the transition between velocity strengthening and velocity weakening. Experimental work, in part funded by the MARGINS-SEIZE program, suggests that illite remains velocity strengthening and thus is unlikely to cause a transition to seismogenesis (Saffer and Marone, 2003). Some other at least partly temperature-controlled process downdip must be sought. One option is rising pore pressure downdip as a consequence of temperature-controlled diagenetic processes that reduce permeability.
Exhumed accretionary prisms clearly show a range of diagenetic-metamorphic processes above 150°C that could also trigger velocity weakening behavior. Most promising among these is quartz mobility expressed as pressure solution, cementation, veining, and coatings of shear surfaces (Fisher and Byrne, 1990; Moore and Saffer, 2001; Ujiie, 2002). Because quartz is velocity weakening (e.g., Blanpied, 1995), its introduction along shear surfaces could also induce seismic behavior (see Figure 3). Drilling into the seismogenic zone will obviously be important to answering the question of which material (or materials) control onset of seismogenic behavior.
It is also possible that individual subduction zones may have different dominant processes controlling this important transition, depending on the nature of subducted material and other specific properties (one SEIZE does not fit all)!. For example, the onset of seismogenic behavior could be both temperature-dependent (cementation/pressure solution) and stress-dependent (compaction). Both of these properties depend on the materials in the subduction interface plus fluid pressure distribution, and it seems likely that this could lead to a highly variable depth limit (Figure 4). Different regions may become “seismogenic” at different depths in the system and may do so in patches or more uniformly depending on the specific margin. It is also possible that topographic features on the down-going plate impart important perturbations to the local state of stress, permeability, and distribution of materials and fluid pressures within the subduction fault.
Accurate recording of the locations, sizes and other characteristics of moderate and smaller magnitude earthquakes that “illuminate” the plate interface between great events is another way to investigate up-and down-dip limits. Unfortunately, given subduction zone geometry, most such earthquakes tend to be poorly located, or their locations have systematic errors. This reflects the fact that most subduction zone earthquakes can only be characterized on the basis of teleseismic data. Even if local seismic arrays are available, they are usually “one-sided” (i.e., sited only landward of the trench). Improving the accuracy of event location requires simultaneous recording by seismometers on land, over the down-dip portion, as well as on the sea floor, around the up-dip portion, at an array of azimuths and distances. This is obviously a technical and logistical challenge, and until recently there have been few such measurements. Several observation programs relevant to this problem have now begun under the auspices of the MARGINS program, in both Central America and Japan. For example, seismometer deployments occurred near the Osa Peninsula in southern Costa Rica in November-December 1999, and off the Nicoya Peninsula in northern Costa Rica in January-June 2000, for comparison to geodetic results (Dixon et al., 2001; DeShon et al., 2003). Each experiment included simultaneous deployments of IRIS/PASCAL broadband seismometers on land, and state of the art ocean bottom seismographs (OBS) offshore. The Nicoya deployment involved 20 PASCAL stations deployed for 18 months, and 14 OBS for 6 overlapping months, with thousands of events recorded. Initial results from these deployments (Newman et al., 2002; DeShon et al., 2003) suggest spatial variability in the updip limit of this seismogenic zone.
Nankai has also had careful land and OBS deployments conducted. A key finding is that there is a very low background level of microseismicity on the plate interface. Why this region is so different compared to Central America, and whether it is a stable or time-transient feature, is not known.
In January-February 2000, a geodetic network in Central America that was first measured in 1994 (Lundgren et al., 1999) was resurveyed to improve the accuracy of the existing GPS site velocities, and new sites were installed in Costa Rica and Nicaragua. An accurate and well-sampled regional surface velocity field is being defined as these new sites are re-occupied, and interpreted in the context of the new earthquake data described above. Preliminary results suggest that a large patch beneath the Osa Peninsula is fully locked and accumulating strain at or near the plate convergence rate, and also experiences back-arc shortening on the Panama fold and thrust belt at rates of ~1-2 cm/yr. In the better sampled Nicoya region, patches of essentially fully locked areas alternate with regions of lower coupling (Figure 5). However, the resolution of these data is not yet sufficient to determine if this “patchiness” in fact represents adjacent fully locked vs. freely slipping regions or true “partial coupling” and how these patches correlate with plate boundary microseismicity
Costa Rica is one of the few places where the land is close enough to the trench (Osa and Nicoya Peninsulas) that terrestrial geodetic data can provide constraints on the updip limit of the seismogenic zone. Generally geodetic data only constrain the downdip limit of the locked zone and great earthquake coseismic rupture zone. This is one of the reasons Costa Rica was chosen for focused MARGINS/SEIZE studies.
For the much better instrumented Japan subduction zone, a precise GPS interseismic velocity field has been available for several years, and indicates strain accumulation consistent with a fully locked seismogenic zone in most areas. The dense network of continuous GPS stations has also documented several transient strain events (e.g., Heki et al., 1997; Ozawa et al., 2001), and elucidated important tectonic aspects (Heki and Miyazaki, 2001; Miyazaki and Heki, 2001).
Tsunami earthquakes are defined as events that because of their slow (possibly very slow) rupture speed, do not effectively radiate seismic energy, but do actively excite tsunami waves. At this point, not much is known about the mechanical processes that allow rupture to proceed at such slow speeds, however there is mounting evidence that there is a significant decrease in rigidity in and around the seismogenic updip limit (Bilek and Lay, 1999). Currently it is not understood if the decrease in rigidity is controlled by subducted sediments or if other factors, including fluid flow, play a significant role. Changes in the rupture mechanics of shallow earthquakes may also influence tsunami earthquake generation. For some of these earthquakes the rupture upward through the normally inactive up-dip limit, all the way to the sea floor, may contribute further to the generation of tsunami waves.
Understanding why some earthquakes occasionally rupture through this normally aseismic region is an important question and may relate to rate- and state-variable friction. Some models have proposed rupture propagating along splay faults in the accretionary wedge. Others have suggested that heterogeneous distribution of asperities and rate- and state friction variations may also be important for producing the large amounts of shallow slip observed during tsunami earthquakes (Bilek and Lay, 2002). Obviously, anything we learn about how tsunami earthquakes are generated and propagated would play an important role in mitigating tsunami hazards.
Some slip seems to be too slow even for tsunamis and is only detected in geodetic data. We can therefore consider three speed classes for slip on subduction thrust faults:
(a) Fast; generates seismic energy,
(b) Intermediate; generates large tsunamis but small or no earthquake (tsunami earthquake);
(c) Slow; only seen in geodetic data, often shallower or deeper than seismic zone; may be related to deep slow slip events.
Understanding controls on rupture speed, and understanding why some earthquakes and slow slip events apparently rupture through the normally aseismic up-dip and/or down-dip limits are important questions and may ultimately bear on our understanding of the physical processes on the plate interface.
Role of subducted sediment. Regional stress and earthquake studies suggest that subduction thrust faults are weak. Sediment subduction may play an important role, providing a mechanism for bringing large amounts of water to the plate interface. However, there is little consensus on the nature of this role and even less hard data. For example, subduction of large amounts of sediment could generate a large, relatively uniform region of lowered effective friction coefficient, perhaps facilitating rupture over large areas (more sediment equals larger maximum magnitude earthquakes). On the other hand, very low effective friction coefficient could reduce the tendency for strain accumulation and seismic rupture (Pacheco et al., 1993) (more sediment equals less seismicity).
The stress drop in subduction earthquakes is inferred to be close to complete. This might reflect:
(a) high pore pressure that reduces normal stress;
(b) inherently weak fault gouge, or;
(c) something unique to the dynamic rupture process.
Mass Transfer. Transfer of mass in subduction zones can occur as solid mass or as dissolved mass carried by fluid flow. In the solid phase, transfer of material from one plate to the other is a fundamental part of the subduction process. In the upper part of subduction zones, including the seismogenic zone, this can take the form of the addition of material from the subducting plate to the base of the overriding plate by underplating, with consequent uplift. Alternatively, removal of material from the base of the overriding plate by a number of processes leads to tectonic erosion, manifested by subsidence. We do not know whether major thrust earthquakes are part of the mechanism of either of these processes, or whether earthquakes arise purely from slip between the two plates with no material transfer. The association of areas of rupture with regions of the forearc known to exhibit underplating or tectonic erosion suggest that large thrust earthquakes are involved in either one or perhaps both of these processes. This issue can be resolved by comparing:
1) seismic reflection images of the basal detachment;
2) the earthquake or microearthquake-determined locations of the detachment, and;
3) changes in shape of the sea floor above the zone of underplating or tectonic erosion.
Thermal Models. Numerical thermal models give us our best temperature estimates for the subduction thrust. New subduction zone specific thermal models using finite element methods have greatly improved our temperature estimates over older generic models using finite difference. However, many uncertainties remain, including (a) the isotherms intersect the thrust at a fairly shallow angle which limits thrust temperature resolution, (b) frictional heating, although concluded to be small, remains an uncertainty, (c) transient effects, (d) updip temperatures can still be improved by more detailed description of the thrust dip profile, including how much and where sediment is scraped off, and the seafloor profile at the model location (and how has this varied with time). Detailed heat flow measurements are needed, because comparison of predicted with observed surface heat flow is one of the more important tests of such models. Other parameters needed are the thermal properties of the forearc, its radioactive heat generation, and the thermal state of the incoming oceanic crust and overlying sediments.
Exhumed faults. Study of exhumed subduction thrusts should be a key aspect of the SEIZE program. They provide the requisite “ground truth” for many of the ideas discussed here, and are also a necessary first step for drilling because they provide information on the material properties likely to be encountered at depth.
Seismic reflection, downdip limit. The subduction thrust reflection image generally is thin and sharp where seismogenic, but becomes a thick reflection band deeper in areas of “slow slip”; the concept of a “plastic shear zone” has been proposed (e.g., Nedimovic and Hyndman, in press). This change in width from the shallow brittle part of the fault zone to a deeper wide shear zone where there is more plastic deformation is well recognized in exhumed continental fault zones. The change in subduction thrust reflection image may therefore provide an indicator of the seismic-aseismic boundary.
Seismic reflection, updip limit. The updip aseismic zone has been shown to have variable positive and negative polarity reflections. These have been interpreted in terms of variable pore pressure
Fluids play a key role in faulting and earthquake mechanics (e.g., Hickman et al., 1995; Raleigh et al., 1976) and in all five questions noted in Section 1.2. Fluid pressure likely controls a wide range of faulting characteristics, from fault strength to rupture propagation, in both subduction zones and continental settings (e.g., Rice, 1992; Johnson and McEvilly, 1995). In subduction zones, elevated pore pressures (approaching lithostatic values) result from a combination of rapid compaction and low permeability typical of marine sediments. Studies have shown that pore pressure affects decollement strength, structural development, and taper angle (e.g., Hubbert and Rubey, 1959; Davis et al., 1983) because it controls effective stress. In addition, pore pressure has been postulated to influence the position of the shallow limit of seismogenic faulting behavior, through its control on effective stress and consolidation state (e.g., Moore and Saffer, 2001; Scholz, 1998). Pore pressure is thought to affect earthquake source duration (e.g., Bilek and Lay, 1998) and the time-dependence of aftershock activity. The compaction state and fluid flow patterns associated with subsurface pore pressure are also important for interpretation of chemical data as a constraint on fluid mass fluxes (e.g., Bekins et al., 1995; Saffer and Bekins, 1998).
Geologic observations of exhumed subduction zones document a close connection between fluid over-pressure and faulting (e.g., Sibson, 1990). Slow-slip events in the down-dip region of the Cascadia subduction zone may be related to cycling of metamorphic reactions (Dragert et al., 2002). These reactions release water, possibly in conjunction with low matrix permeability that limits drainage, resulting in elevated pore pressure and low effective stress, allowing fault slip and opening high-permeability pathways.
Direct measurement of fluid pressure and permeability within active fault zones is challenging, but important to the SEIZE initiative. We need to understand the distribution of pore pressure, dewatering from compaction and metamorphic dehydration reactions, permeability, and porosity—both along strike and down-dip—on subduction plate boundaries. Quantitative models linking transient changes in pore pressure, fluid production, permeability, and porosity to transient strain events are just beginning to be developed (e.g., Saffer and Bekins, 2002, Revil and Cathles 2002). A unique aspect of the MARGINS-SEIZE program is the ability to link such processes and models through a unified field program.
Some examples of key issues related to pore pressure and seismogenic faulting for the five original SEIZE questions are discussed below.
Fluid pressure, through its influence on effective stress, is known to profoundly affect fault strength (e.g., Hubbert and Rubey, 1959) and the sliding stability of faults (e.g., Scholz, 1998). Fluid pressure distribution within fault zones is an important parameter controlling the rupture and aftershock sequence in some earthquakes (e.g., Bosl and Nur, 2002; Miller et al., 1996). Thus, the spatial variability of pore fluid pressure within fault zones is a potentially significant factor in controlling the distribution of fault strength and slip behavior.
Strain, pore pressure, and fluid flow are intimately related at subduction zones where hydrologic process are highly active (e.g., Sibson, 1990). Fluid pressure is thought to be related to earthquake triggering, post-faulting evolution of permeability and effective stress, and volumetric strain, in addition to the spectrum of transient strain behavior discussed above. One question that can be addressed at both the Nankai and Costa Rica margins is whether the physical properties, chemistry, and state of the fault zone change with time throughout the earthquake cycle (inter, co- and immediate post- seismic periods). To address this issue requires sampling and instrumentation of long-term chemical and hydrological sensors in boreholes and at the surface (Figure 7). Physical properties and state variables such as permeability, seismic velocity, fluid pressure, stress, temperature, and fluid chemistry and flow rates are important parameters.
Information gained from pore water chemistry, cementation, and pressure/temperature monitoring could have profound implications for our understanding of the temporal relationships of key parameters (stress, pore pressure and permeability) and mineral reactions that occur at depth and that affect fault mechanical behavior and the nature and magnitude of fluid/chemical fluxes through subduction systems. Two hydrologic hypotheses can be simultaneously addressed during long-term monitoring of the hydrogeological system along faults. These concern the depth from which hydrologically active faults have their fluids sourced and how the movement of these relate to the state of stress, fluid chemistry, and dynamics of faulting.
Hypothesis 1 suggests a deep connection with chemical dehydration reactions that effectively drives fluid pressure and fluid transmission up major faults. Such transmission may occur transiently and can be coupled to rupture. Fluids transmitted up the fault from depth will be progressively modified by shallow water sources and reactions.
In Hypothesis, 2 up-dip flow is limited and fluid pressures within the fault are set by local conditions. Under these conditions pore pressure in the fault can respond to stress changes and pore elastic effects imposed during the earthquake cycle, but fluid movement and fault dilation is limited and does not necessarily drive instabilities and rupture by a fault valving mechanism. Likewise, the importance of high-amplitude seismic “bright spots” that can occur along the faults and their relationship to fluid flow and fluid pressure are unknown but may provide a means to design experiment that will allow us to study along fault fluid transmission scenarios. In Figure 7, we show for Hypothesis 1 and 2 different types of patchiness that might indicate different types of fluid transmission indicators. There may also be little to no fluid concentration and dilation on the fault plane, in which case any geophysically observed patchiness may related to other factors such are the wall rock properties and lower velocity gouge distributions.
Direct observations (e.g., Raleigh et al., 1976; Bosl and Nur, 1992; Johnson and McEvilly, 1995), field study of exhumed faults, and theoretical models (e.g., Sleep and Blanpied, 1992) all indicate that fluid pressure affects earthquake nucleation. An additional complexity arises because both permeability and porosity are known to depend on effective stress, and thus vary as fluid pulses propagate (e.g., Rice, 1992). In subduction zones these types of behavior may provide destabilization and fault-triggering mechanisms, perhaps “powered” by nearby earthquakes that affect permeability (e.g., Miller et al., 1996) or by chemical reactions (in which case fluids could be characterized by distinct chemical and isotopic signals (Hypothesis 1 in Figure 7).
Fault movement itself may change the surrounding permeability structure, pore pressure distributions and thus flow rates and pressure distribution, giving rise to post-earthquake transients (e.g., Sibson, 1990) or triggering phenomena (e.g., Bosl and Nur, 1992). Regional changes in the permeability of shallow aquifer systems may be monitored in wells and fault controlled seeps on land, and with long-term pressure and temperature measurements in boreholes and surface flux meters offshore.
A number of recent studies have looked at the static stress changes associated with earthquakes and stress triggering of nearby faults. While syn-faulting stress transfer is often quite small, additional postseismic poroelastic and viscoelastic stress relaxation may enhance static stress and transient pore pressure effects far from the earthquake. The temporal distribution of Landers earthquake aftershocks suggest that a combination of coseismic strain and subsequent pore pressure decay (and resulting poro-elastic stress changes) resulted in significant stress change out to distances of ~50 km from the fault (Bosl and Nur, 2002). These should be measurable both in bore holes and potentially at the seabed.
In the same manner that it may affect sliding stability of fault regions through its control on effective stress, pore pressure is thought to be one of several factors that mediate the upper transition from aseismic to seismic slip (termed the “Updip Limit” of the seismogenic zone) (e.g., Scholz, 1998; Moore and Saffer, 2001). Increased effective stress at the fault interface increases the tendency for unstable slip; systematically increasing effective stress with depth in subduction zones may result from a combination of increased overburden and declining fluid pressure or fluid production.
As noted above, the processes controlling tsunami earthquake rupture are not well understood. In particular, the properties and state of the plate interface which allow rupture to extend through the normally aseismic region above the updip limit of the seismogenic zone are unknown. Distinct patches of elevated pore fluid pressure, in conjunction with the complex frictional behavior of some clays, have been hypothesized to control this process, by allowing slip to propagate to the seafloor (e.g., Seno, 2002).
Transport of dissolved mass by fluid flow along faults from deep reaction zones has been observed in subduction zones worldwide (e.g., Kastner et al., 1991). Current monitoring studies are based on the hypotheses that pore fluid chemistry along shallow thrusts faults may be used to infer mineralogy, temperature, and reactions occurring at seismogenic depths (Hypothesis 1 in Figure 7). However, it is not always clear that such updip transport occurs in every environment (Hypothesis 2 in Figure 7) and there are important questions that have to be addressed concerning the widespread applicability of large scale updip fluid migration. Even if fluid migration is limited to local scales, the role of fluid mass transport in vein formation may also be important in changing the material properties of the fault. For example, vein formation occurs along faults as solutes are transported upward to lower temperature and pressures. Changes resulting from vein formation such as permeability decreases and increases in mechanical cohesion may be central to seismogenic processes.
Because sampling by drilling and direct measurement of fluid pressure and hydrologic properties is only plausible at shallow depths and over limited spatial extent, addressing these outstanding issues requires an integrated plan including the following:
Theoretical modeling provides a means to test hypotheses related to feedbacks between hydrologic and mechanical processes, identify and quantify important controlling processes, and direct future data collection efforts. Viable models for subduction zones are needed at multiple scales to (a) link fluid flow models with laboratory and field data, (b) extend results from areas that are well-constrained by direct monitoring to unmonitored regions, and (c) couple fluid flow and mechanical (deformation) models.
Fluid production from compaction, hydrocarbon generation, and metamorphic reactions are central aspects of existing hypotheses explaining earthquake and pore water geochemical observations. Better constraints on the distribution of fluid production, as well as any expected chemical and isotopic signatures, are therefore critical. These objectives can be realized through a combination of laboratory studies (e.g., deformation tests, experimental petrology) and theoretical modeling.
Permeability of both fault zones and country rock (matrix) strongly mediate the development of elevated pore pressures. Thus, estimates of permeability at a range of scales–for both faults and matrix–are needed. These estimates can be obtained by direct measurement on core samples (cm scale), hydraulic testing in boreholes and between pairs of boreholes (10’s-100’s of m scale), and inverse modeling (several km scale).
Direct measurement, long-term monitoring. Direct measurement of pore fluid pressure, via long-term monitoring and down-hole measurement in boreholes, is needed to evaluate the magnitude of pore pressure within fault zones and adjacent wall rocks, as a means to test hypotheses for fault weakness, and as a constraint on theoretical models. Long-term monitoring is an integral component of testing hypotheses discussed above relating temporal variation in pore pressure, permeability, and fault slip. Surface measurements of flow at fault-controlled cold seeps, associated with transient changes in fault permeability and pore pressure distribution, may also give further cost effective field indications of slip events (silent or otherwise) in regions away from boreholes (Tryon et al., 2001, 2002). Together with surface measurements of episodic flow that are associated with coseismic volumetric strain events, these may give us a geographically broader array of allied measurement opportunities to track the impact of rupture events over regional scales.
Based on the presented discussion, a number of new questions can be posed, supplementing the questions posed at the beginning of the SEIZE initiative:
1) What controls the overall distribution of seismic energy release during a subduction zones earthquake (up, down, and sideways). Is there one P-T-X condition that defines the onset and down dip limit, or do they vary with the material properties fault geometry, pore pressure, and state of stress in the subduction system?
2) What controls the sometimes heterogeneous distribution of locking patterns on the plate interface and subsequent variations of energy release during earthquakes? Are such “asperities” linked by common physical processes within the fault region or governed by separate, unrelated phenomenon? Can such features be accurately mapped with microearthquakes? With space geodesy? What are the prediction errors associated with typical mechanical models for subduction zone strain accumulation? Do the patches vary in time, and if so, over what time scale? How do these heterogeneous features influence tsunami earthquake generation?
3) What controls the rate of propagation and slip rates of earthquakes and the distribution of fast, slow, tsunamigenic, and silent earthquakes in time and space?
4) What is the nature of temporal changes in strain, fluid pressure and stress during the seismic cycle? Do these change gradually during the seismic cycle or are there transient interseismic phenomena that lead to strain and energy release at various times during the seismic cycle?
5) What are the prediction errors associated with typical mechanical models? For example, Masterlark et al., (2001) demonstrate enormous prediction errors associated with the homogeneous material property assumption in both forward and inverse models of GPS displacements caused by dislocations along a fault. The analysis was performed using finite element models (see attached figure from Masterlark et al., 2001).
We are now four years into the SEIZE study, and significant data collection efforts have been made in Japan and Central America (see Appendix for discussion of focus site selection criteria). The updip limit of a 1999 rupture zone for a M=6.9 earthquake has been carefully mapped off southern Costa Rica. It appears to be limited to depths in excess of 13 km. There is some indication of a shallower limit farther south but that is outside the boundaries of the OBS experiment. Clearly more OBS experiments are needed to fully understand the up-dip limit of the seismogenic zone and to investigate its possible spatial and temporal variations (this may only be useful in Central America, as the rate of microseismicity is quite low in Nankai). One borehole seismometer has been installed offshore Japan, and the initial results are very promising. Similar instrumentation is necessary off Central America.
Much geodetic work has been carried out in both regions and we have significantly better understanding of locked vs. slipping zones. Excellent methodology for studying locked vs. slipping zones along the thrust interface has also been developed. Further geodetic data and modeling are necessary to establish detailed understanding, and to begin to investigate the difficult question of temporal variation. Continuous GPS in Japan has recorded a number of transient events on the plate interface, and these are revolutionizing our previous picture of a “static” (between major earthquakes) plate interface. A similar network is necessary in Costa Rica.
A great deal of seismic reflection data has been obtained, sufficient to image the entire seismogenic zone beneath Nicaragua, Costa Rica and Southeastern Japan. Two three-dimensional reflection surveys have been done off Japan, and one 3-D in Central America.
Based on these and other studies, some general characteristics for the two regions have emerged:
Temporal and spatially complex, with behavior that varies considerably along strike. In particular the basement topography and sediment thickness vary greatly, with “patchy” seamount distribution, possibly related to patchy locking. Characteristic earthquakes tend to be smaller in magnitude (M~7-7.5) with a limited slip distribution, possibly controlled by bathymetry; slow tsunamigenic earthquakes have occured off Nicaragua.
Larger magnitude (M~8-8.5) earthquakes, with relatively uniform fault properties and only small lateral changes; large areas are fully locked most of the time; limited microseismicity.
What has not yet occurred in either region is direct sampling, e.g., by drilling or submersible vehicles. Drilling is planned in both regions, however a number of site characterization studies still need to be carried out. For example, no 3-D seismic reflection surveys exist in the most promising areas for riser drilling in either Japan or Central America. The general properties outlined above for each region will be used as a guide to picking appropriate drilling targets. Particularly for Central America, with its great spatial variability, several drilling targets spanning both cross-strike and along strike directions are appropriate. The proximity of land to the trench in Costa Rica suggests the possibility of joint land-ocean drilling.
The new drilling program (IODP) is progressing well. The riser vessel has been built and an RFP has been submitted for a non-riser vessel, the latter to begin scientific drilling in 2005. Non-riser drilling has been done in both the Nankai Trough and offshore Costa Rica, the latter with two drilling legs off the Nicoya peninsula. Complex drilling proposals have been submitted for use of both riser and non-riser ships for drilling into the seismogenic zone off both Japan and Central America.
The original SEIZE Science Plan envisioned a 10 year program. However, because the funding level of the MARGINS Program has been less than anticipated and started somewhat later than anticipated, we now believe that achieving SEIZE objectives will require a 15 year program, 2000-2014.
The ordering of certain elements of the program is obvious, for example, extensive geophysical surveys (Table 1, item 2) are required prior to deep drilling (Table 1, item 5). Other aspects of the program are strongly interwoven, with results from one potentially triggering further studies in another. The first 5-6 years of SEIZE will focus on developing geological and geophysical background (Table 1, items 1-4) for the candidate seismogenic zones. Monitoring of seismicity, strain, and fluid flow, whether at the surface or in boreholes, is required for the full duration of SEIZE to document and understand the spectrum of transient phenomena (Table 1, items 1 & 5). Modeling and laboratory experiments will be necessary throughout, to guide data acquisition and evaluate results. Riser drilling, beginning in about 2007, will ultimately test predictions of the nature of the seismogenic zone.