Abstract
Over the last 10 years a series of developments have led to a new understanding of what controls the variations in lithosphere strength, structure and evolution that produce dramatic contrasts between the geological histories of oceans, ancient shields and young orogenic belts. Those developments involve a wide range of observations from a great diversity of geological, geophysical and geochemical disciplines that, none the less, provide a mutually consistent and coherent overall picture. This paper summarizes, in one place, the essential stages in the evolution of the relevant ideas and observations that have led to this situation.
The discovery of plate tectonics in the 1960s concentrated attention on the outer, relatively strong, layer of the Earth called the lithosphere, whose behaviour controls the tectonic features of the Earth's surface. Oceanic lithosphere is produced at mid-ocean-ridge spreading centres and the apparent simplicity of this process, together with the oceans' youth, uniformity and the easily observable relation between water depth and lithosphere age, led quickly to a simple and widely accepted model of its thermal structure (Parsons & Sclater 1977) that has required little modification until very recently. Because the thickness of the oceanic crust varies little from an average value of about 7 km, the rheology of the oceanic lithosphere as a whole is likely to be linked in a simple way to its thermal structure. The continental lithosphere was more difficult to assess. Variations in crustal thickness, together with isostasy, allow no simple relation between height, age and thermal structure on the continents, and so an assessment of continental rheology must be based on possible proxies for strength, the most common of which are earthquake depth distributions and the ability of the lithosphere to support loads, which cause gravity anomalies. In addition, the continental crust is several tens of kilometres thick, and plays a significant part in the overall rheology of the lithosphere. By the early 1980s a view evolved in which the continental lithosphere was thought to consist of a weak lower crust sandwiched between a relatively strong upper crust and strong uppermost mantle (Chen & Molnar 1983). This view, popularly called the ‘jelly-sandwich’ model, remained virtually unquestioned for the next 20 years.
More recently, that view has been challenged in a series of papers beginning with studies by Maggi et al. (2000a, b). Those studies claimed that there was little evidence for the laminated, or ‘jelly-sandwich’ view of continental rheology and that, instead, the long-term strength of the continental lithosphere resided only in its upper part, which was contained wholly within the crust. The effort to explain why that new view might be correct led quickly to a series of developments in related subjects, including: (1) a re-examination of geotherms and thermal structure in both oceans and continents; (2) connections between seismic velocity and temperature; (3) relations between mechanical properties and the metastability of lower crustal rocks in mountain roots; (4) causes of the variations between the ancient continental shields (or ‘cratons’) and younger Phanerozoic orogenic belts; (5) the issue of how the continental cratons were created. The new views originating in the studies of Maggi et al. (2000a, b) have (of course) themselves been challenged; but because the subject has been very fast-moving, debate in publications has not been able to keep pace with the new developments themselves, which makes it difficult for an interested reader to know what is happening. Nor have the different strands of those developments yet been brought together in one place.
The last few years have thus seen a great deal of activity and thought in the area of lithosphere structure, rheology and evolution, which, whatever one's personal views, is undoubtedly important for understanding the geology and evolution of the continents. The aim of this review is to summarize the development of this subject, and to demonstrate that a coherent picture is emerging that reconciles observations from fields as diverse as seismology, gravity, heat flow, rock mechanics, metamorphic petrology and geochemistry. Furthermore, the insights, and agreement, that these widely differing disciplines offer on the same subject produce an overall view that is more robust than any obtained from one of those disciplines alone. As is often the case in Earth Sciences, some of these ideas are not new, and their origins can be found in earlier studies. However, in the interests of a coherent narrative, this review follows a roughly chronological account of their development, as they became relevant or interlinked in the context of the current overall debate about lithosphere structure, rheology and evolution.
Earthquake depth distributions on the continents
It has been known for a long time that, in most places, earthquakes on the continents are confined to the upper half of the crust. This is usually interpreted as a temperature-dependent change from shallow friction-dominated slip on faults to deeper aseismic creep-dominated processes (e.g. Brace & Kohlstedt 1980). An important influence on the formulation of the ‘jelly-sandwich’ or laminated rheological view was the apparent occurrence of rare earthquakes in the uppermost mantle in a few areas, which were thought to indicate a strength contrast between upper mantle and the generally aseismic lower crust (e.g. Chen & Molnar 1983). Such a contrast might reasonably be expected, as creep strength depends mainly on homologous temperature (the ratio of actual temperature to the melting temperature in degrees Kelvin), which could change significantly at the Moho between silica-rich (and possibly wet) lower crustal rocks and the silica-poor (and possibly anhydrous) rocks of the mantle whose melting temperature should be higher (e.g. Mackwell et al. 1998).
However, Maggi et al. (2000a, b) re-examined these rare upper mantle earthquakes and concluded that they were instead in the lower crust. The main reason this re-evaluation led to revised conclusions was an improvement in data quality and quantity since the compilation of Chen & Molnar (1983). In particular, new seismological methods (called receiver function analysis), which detect P-to-S wave conversions at the Moho beneath a seismometer, have led to much improved and more widespread determinations of crustal thickness; and more earthquakes, combined with improved body-waveform inversion programs, have led to clearer patterns of well-determined earthquake depths.
The pattern seen by Maggi et al. (2000a, b), updated since then, is shown in Figure 1. The top row of histograms shows three places where well-determined earthquake depths are confined to the upper continental crust; Moho depths are about 40–50 km in Iran, 25–35 km in the Aegean region and deeper than 60 km in Tibet. Similar patterns are typical of many continental areas, such as California, Nevada, Italy, New Zealand, and the Andes. A different pattern is seen in the lower row of histograms. In East Africa, earthquakes occur throughout the crust, but not deeper than the Moho (e.g. Foster & Jackson 1998; Brazier et al. 2005), with lower crustal earthquakes more common in the southern part of the rift, adjacent to or within the Archaean shields of Tanzania, Zambia and Zimbabwe. The India histogram contains earthquakes that occurred within material that originated as part of India, either in the foreland south of the Himalayan front, or beneath the Himalaya itself. These again occur throughout the crust, which varies in thickness from 35–40 km in peninsular India to more than 50 km beneath the Himalaya. (There is more to be said about this region, which is discussed below.) The Tien Shan also shows earthquakes throughout the crust, but not beneath the Moho; a pattern that is also seen in well-recorded local seismicity (Xu et al. 2006). The Mongolia–Baikal region is another that contains earthquakes in the lower crust, but with no convincing evidence of mantle seismicity (Emmerson et al. 2006). The geological context common to the lower row of histograms is that they all contain earthquakes in, or adjacent to, ancient Precambrian shields.
Histograms of earthquake centroid depths constrained by body-wave seismogram modelling for events larger than mb c. 5.3. This figure updates figure 1 of Maggi et al. (2000a), with additional data from our own work, and that of Emmerson et al. (2006) for the Mongolia–Baikal region. The India histogram distinguishes earthquakes south of the Ganges basin (white) from those within and north of the Ganges basin, but within the Indian shield (in red). In the Tibet histogram the white bars are earthquakes on the frontal Himalayan thrust system. Green bars show Moho depths, which are typically 40–50 km in Iran, 25–35 km in the Aegean and deeper than 60 km in Tibet; in all three regions earthquakes are confined to the upper crust. In parts of Africa, India, the Tien Shan and Mongolia–Baikal earthquakes occur throughout the crust, but not substantially in the mantle (the deepest earthquakes are also where the Moho is deepest). A few earthquakes at depths of 70–90 km beneath southern and NW Tibet (Fig. 8) are omitted here, and are considered separately.
Clearly, no single seismicity distribution characterizes all the regions in Figure 1. The important variation noticed by Maggi et al. (2000a, b) is the contrast between the young Phanerozoic orogenic belts, where earthquakes are restricted to the upper crust, and those adjacent to or within ancient shields, in which the whole crust may be seismogenic. The issue of whether, in some of the latter cases, some seismicity also crosses the Moho into the uppermost mantle is one we return to below; but in all those cases the lower crust is certainly seismogenic. Mantle earthquakes are clearly very rare in the continents, and there is no sign of the bimodal distribution, with an aseismic lower crust above a seismogenic mantle, that is a feature of the generic Chen & Molnar (1983) model. Instead, earthquakes on the continents are contained within a single layer, of seismogenic thickness Ts. (An anomaly, which is only apparent, is in the Himalaya, discussed below.) This was the starting point of this entire story, from which it was clear that the widely accepted ‘jelly-sandwich’ model for the continents needed revision.
Gravity and effective elastic thickness
The lithosphere can support loads by bending elastically, and the consequent lateral variations in density cause gravity anomalies that can be observed at the surface. A measure of its elastic strength is revealed by the wavelength over which this bending and support occurs, and is commonly expressed in terms of an ‘effective elastic thickness’, Te, which is the thickness of a uniform elastic beam that is usually used to model the observations. (An analogy is that a thick plank bends on a larger wavelength than a thin plank.) Estimates of Te are then obtained by modelling the relation between gravity and topography, either in the 2D frequency domain (gravity v. wavelength), or as profiles in the space domain (gravity v. distance). In both cases, what is obtained is simply an estimate of the effective elastic thickness, not the depth to the top or bottom of the layer that is actually providing the support.
In the oceans, such analysis has been common for some time (e.g. McKenzie & Bowin 1976; Watts 2001) and yielded results that were unsurprising. Te apparently increased with lithospheric age, giving values that approximately corresponded to the 450 °C isotherm, whereas the maximum depth of intraplate oceanic earthquakes (Ts), which also increases with age, corresponded to higher temperatures of about 700–750 °C (Chen & Molnar 1983; Wiens & Stein 1983). Because the time scale for accumulation and release of seismic energy in the earthquake cycle is shorter than that involved in supporting uncompensated topography for millions of years, Ts is expected to be greater than Te, as was observed. Furthermore, the homologous and actual temperatures at which these two manifestations of elastic strain accumulation ceased are reasonably within the expectations of experimental rock mechanics.
However, in some areas of the continents, similar techniques apparently gave estimates of Te that were very much greater than in the oldest parts of the oceans, with published values up to 130 km in some Precambrian shields. Such large values posed two difficulties. First, they required Te on the continents to be much greater than Ts, unlike in the oceans, as the deepest earthquakes in shields are shallower than 50 km (Fig. 1). Second, they required long-term elastic strength at mantle depths corresponding to temperatures of 800–1000 °C, which is double that at which elastic behaviour ceases in the oceanic mantle, and at temperatures that experiments suggest should be well with the ductile creep regime. There was clearly a problem here, which began to be examined at roughly the same time as the earthquake depth distributions.
McKenzie & Fairhead (1997) first questioned the conventional practice of Te estimation on the continents, and their investigations were followed by others that addressed particular issues with the gravity–topography analysis, or with Ts–Te relations, or with different regions: those of Jackson (2002), McKenzie (2003), Bayasgalan et al. (2005), Emmerson et al. (2006) and Crosby (2007). The principal conclusions of this group are that (1) the very large values of Te found by earlier studies are only upper bounds, and may be much greater than the true values, and (2) where Te is well resolved, it is everywhere less than Ts, as it is in the oceans, and nowhere do the data require Te > Ts. The emphasis in this last statement is important, as, in some places, the nature of the problem or data prevents a well-resolved upper limit to Te, although a lower limit usually can be resolved, and is less than Ts.
The analysis of this group has, of course, been challenged. The issues involved in the 2D frequency-domain analysis have been comprehensively addressed by McKenzie (2003) and Crosby (2007), and are too technical for the arguments to be rehearsed in detail here. The issues related to profile modelling in the space domain have received less attention, and can be illustrated by the flexure of the Indian shield south of the Himalaya (Fig. 2). McKenzie & Fairhead (1997) and Jackson (2002) estimated Te to be about 30–40 km, but with an unconstrained upper bound, whereas Burov & Watts (2006) showed an apparently well-defined value of 70 km, a value similar to that found also by Hetenyi et al. (2006). It is indeed mystifying to many readers that the same problem (and data) should yield such widely differing and incompatible results.
(a) Curves showing misfit of observed to calculated gravity anomalies across the Himalayan foreland basin as a function of Te. The curve labelled McKenzie is from Jackson (2002) and used free-air gravity; the curve from Burov & Watts (2006) used Bouguer gravity. The vertical axis has no units and has been scaled to show the Te positions of the minima in each curve. (b) The reason for the disagreement in (a) is because the position where the flexed Indian plate is broken was specified by Burov & Watts (2006), but not by McKenzie & Fairhead (1997), Maggi et al. (2000a) or Jackson (2002). The break is defined as the location on the profile where the bending moment is zero. The plot shows the effect of fixing the location of the break for the north Indian average profiles. The actual values for the minima are Te=78km when the break is at +100 km, and Te = 39km when it is at −100 km. When the position of the break is not fixed, the best-fitting value of Te is 50 km, with the break at −50 km. (The origin is the location of the gravity minimum on the stacked profiles.) Clearly, the trade-offs between Te, break-point location and bending moment produce a broad minimum whose true upper bound is not well resolved.
The reason why these workers obtained such different results is that they made different assumptions about where the flexed plate is broken, and hence where the bending moment on the end of the plate is zero. Figure 2b shows the value of the misfit between the theoretical gravity profile for a flexed plate and the mean of the observed stacked profiles from northern India, illustrated by Jackson (2002). The continuous line shows the misfit when the position of the break is varied to minimize the misfit for each value of Te. This approach is that of McKenzie (2003), and uses stacked profiles. It yields more stable results for Te than does that of McKenzie & Fairhead (1997), who fitted individual profiles. In contrast, Burov & Watts (2006) fixed the position of the break and then found the best-fitting value of Te. The misfit curves for three chosen positions of the break are shown in Figure 2b as dotted and dashed lines. Clearly, these curves must be tangential to the continuous curve, and must be above it everywhere. The values of Te obtained from the minima of these three curves vary from 39 to 78 km, depending on the position chosen for the break. This effect is responsible for the disagreement. Unfortunately, the location of the break is not well constrained by the geophysical observations. Even if it were, the value of the bending moment on the break would still be unknown. For these reasons, we prefer to allow the location of the break to vary when calculating the misfit, even though the resulting minimum is shallow and the best-fitting value of Te is poorly constrained.
If the re-examination of continental Te values is correct, and Te ≤ Ts everywhere on the continents, then the obvious implication is that there is a single strong layer within the continental lithosphere, and that the strength resides in the seismogenic layer, which is either the upper crust or the whole crust but not, to any significant extent, the mantle (Maggi et al. 2000a). Counter-arguments have been examined by Maggi et al. (2000a) and Jackson et al. (2002). In this revised analysis, the continental situation then resembles that within the oceans, where a single strong layer with Te < Ts has always been the conventional view; but a single generic view does not suit everywhere on the continents. Where earthquakes occur throughout the crust, a case can be made that the long-term strength of the lower crust is greater than that of the mantle. Where earthquakes are restricted to the upper crust, the relative strengths of the lower crust and mantle are generally unknown; but both appear weak compared with the upper crust.
Geotherms and mantle earthquakes
This new view of the continents produced a complication. In the oceans, there is no doubt that the maximum values of Ts and Te, which clearly exceed the typical crustal thickness of c. 7 km, require the mantle to be a source of long-term strength. By contrast, there was no evidence for long-term mantle strength in the continents. This was a puzzle because oceanic earthquakes were thought to occur in the mantle to temperatures of 700–750 °C, whereas Moho temperatures in shields were thought to be as low as 300–500 °C (e.g. Artemieva & Mooney 2001). If the lower crust was seismogenic in some shields (Fig. 1) why wasn't the mantle as well, as it was presumably cold enough to be so? This conundrum led Maggi et al. (2000a) and Jackson (2002) to appeal to the presence (in the continents) or absence (in the oceans) of small quantities of water in the mantle as an additional factor affecting its strength. Although water certainly can affect creep strength (Mackwell et al. 1998), and is an important part of this story (discussed below), it turns out that there is a much simpler explanation for the presence or absence of mantle seismicity: the accepted temperature estimates at depth in both the oceans and continents were incorrect.
Various developments over the last decade have allowed geotherm modelling in the shields to become better understood. Observations in granulite-facies rocks, particularly in the Canadian shield by Jaupart, Mareschal and their co-workers (e.g. Mareschal & Jaupart 2004), showed that heat production in the lower crust was both higher and more evenly distributed than was previously thought. In the mantle, the temperature dependence of thermal parameters, particularly conductivity, which halves over the temperature range of the lithosphere, is more important than was thought, although the effect has been known for some time (e.g. Denlinger 1992; Xu et al. 2004). McKenzie et al. (2005) took account of these effects and fitted geotherms to pressure–temperature estimates from nodules in kimberlites to produce temperature profiles that are also in a steady state with respect to the underlying convecting asthenosphere beneath shields. Some examples are shown in Figure 3. Two features are noteworthy in these profiles: the parabolic shape of the geotherm in the crust, related to the heat generation within it, and the concave-upward shape of the geotherm in the mantle, caused by the decrease in conductivity with depth in the presence of a constant heat flow (i.e. no internal heat production in the mantle). The best example is the Jericho region in northern Canada, where all the nodules come from the same kimberlite pipe (Kopylova et al. 1998). This geotherm also predicts a heat flow from the mantle and from the surface that is compatible with the observations of Mareschal & Jaupart (2004).
Temperature profiles through the lithosphere based on pressure–temperature estimates from mantle nodules at Jericho (NW Canada) and Udachnaya (Siberia), calculated using the method described by McKenzie et al. (2005). Pressures and temperatures estimated from nodules were calculated from the expressions of $Finnerty & Boyd (1987). The lithosphere consists of the crust (pink), a rigid mechanical boundary layer (dark green) and a thermal boundary layer (light green). Also shown are lines indicating the onset of melting (solidus) for peridotite and the position of the graphite–diamond transformation. The estimated temperatures at the Moho are c. 500 °C at Jericho and c. 630 °C at Udachnaya.
For the purposes of this review, the importance of the profiles in Figure 3 is the Moho temperature estimates, which are c. 500 °C at Jericho and c. 630 °C at Udachnaya in Siberia. Both are higher than previous estimates, which were mostly obtained by trying to fit a single straight line from the surface through the nodule data. The most important single influence on the Moho temperature is the crustal thickness and its increased heat production. Most Moho temperatures in shields were probably underestimated in the past for these reasons. Better Moho determinations show that shields commonly do not conform to the relationship between elevation and crustal thickness seen in young lithosphere, but have thicker crust; leading to underestimates of crustal thickness in the past. Dramatic examples include Archaean parts of Finland, where regions effectively at sea level have crustal thicknesses up to 60 km (Alinaghi et al. 2003), and parts of the Kaapvaal shield system, where low-lying Proterozoic areas, such as the Namaqua–Natal and Limpopo belts, have crustal thicknesses of 45–50 km (Nair et al. 2006).
The dramatic influence of a temperature-dependent mantle conductivity, without which the nodule data are not properly fitted in geotherms, led McKenzie et al. (2005), following Denlinger (1992), to re-examine the temperature structure of cooling oceanic lithosphere. The essential observation here was the well-known age–depth relationship, interpreted by Parsons & Sclater (1977) using constant mantle conductivity to produce their widely accepted plate-cooling model. Denlinger (1992) and McKenzie et al. (2005) remodelled the age–depth data using temperature-dependent thermal parameters; an exercise that does not produce an improved fit to the observations, but does change the plate thickness in the oldest part of the oceans (from about 125 to 105 km) and the depth to particular isotherms, generally depressing them as the temperature gradient must increase in the lower lithosphere, as in Figure 3. When the depths of intraplate oceanic earthquakes are examined as a function of age, Denlinger (1992) and McKenzie et al. (2005) showed that they are restricted to mantle colder than about 600 °C (Fig. 4), rather than the 700–750 °C estimated in earlier studies that used the Parsons & Sclater (1977) cooling model (Chen & Molnar 1983; Wiens & Stein 1983). Emmerson & McKenzie (2007) extended this analysis to recalculate the temperature within subducting oceanic slabs, showing that there, too, earthquakes were restricted to material colder than about 600 °C, even when the earthquakes are deeper than 600 km.
Isotherms and depths of intraplate earthquakes in a cooling oceanic plate, adapted from McKenzie et al. (2005). Black contours show isotherms every 100 °C calculated using temperature-dependent thermal parameters and a plate thickness of 106 km; the 600 °C and 1000 °C contours are emphasized by thicker lines. The grey shaded area is the region of the 600–750 °C temperature range calculated in the older model of Parsons & Sclater (1977), which uses constant thermal parameters and a plate thickness of 125 km. In the new model of McKenzie et al. (2005), the limiting temperature for earthquakes is about 600 °C. In the older model, it is about 750 °C. The sole exception is an earthquake in 1964 on the outer rise of the Chile trench in lithosphere about 45 Ma old, reported by Chinn & Isacks (1983), whose depth we are unable to verify.
This result immediately suggested an explanation for the lack of continental mantle earthquakes: if the mantle is seismogenic only when it is colder than 600 °C, it is probably too hot under the continents. New geotherms calculated in the shields showed that those which were seismically active, such as Siberia in Figure 3b, generally had Moho temperatures above 600 °C. Mohos in younger Phanerozoic regions are likely to be much hotter still. The Jericho region in Canada (Fig. 3a) has a Moho temperature of only about 500 °C, mainly because of its relatively thin crust, but is not deforming and has no natural earthquakes that are large enough to constrain focal depths accurately.
The puzzle that led Maggi et al. (2000a) and Jackson (2002) to consider the additional effect of water on mantle rheology thus has a much simpler solution, which is that the mantle appears to be seismogenic if it is colder than c. 600 °C, in both the oceans and continents. No other effect appears to be necessary. The mantle beneath most continental regions is likely to be hotter than 600 °C, explaining why it is both aseismic and does not contribute to long-term elastic strength. There are two circumstances that might lead to the continental mantle being colder than 600 °C, and seismically active, if it is deforming. One is in an Archaean shield with relatively thin crust, as at Jericho (Fig. 3a). This will be relevant for our discussion below of northern India. The second is if continental mantle is cooled by an underplating, subhorizontal subducting oceanic slab, as beneath certain parts of the Andes, where Emmerson (2007) has indeed found evidence for one (possibly two) earthquakes in the Andean continental mantle. However, water has not ceased to be important in this topic. If earthquakes occur throughout the crust in some shield areas, such as parts of east Africa and Siberia, and yet the mantle is aseismic because the Moho is at 600 °C or more, those earthquakes must occur in crust that is substantially hotter than the usual c. 350 ± 100 °C estimated at the base of the seismogenic layer in normal continental areas. The occurrence of earthquakes in crustal material at temperatures of c. 500 °C almost certainly requires the mineralogy to be dry (e.g. Mackwell et al. 1998, Jackson et al. 2004), and anhydrous granulite-facies rocks, probably of mafic composition, may well be characteristic of the lower crust in many shields (Rudnick & Fountain 1995). We return to this issue below.
Lithosphere temperature and velocity structure from surface waves
The next development in this story involved a technological advance, made possible by high-quality measurements of ground motion by modern broadband digital seismometers. Analysis of fundamental and higher-mode surface waves can now provide estimates of shear-wave velocity as a function of depth, Vs(z), over much of the continents, with a horizontal resolution of 200–500 km and a vertical resolution of 25–50 km (e.g. Debayle & Kennett 2000; Ritsema & van Heijst 2000; Priestley & Debayle 2003). Priestley & McKenzie (2006) used the observed spatial variation of Vs(z) in the Pacific Ocean, which has excellent surface-wave coverage, to obtain a relationship between Vs(z) and lithosphere age (Fig. 5a). As the oceanic lithosphere temperature structure as a function of age was already known (Fig. 4), they were then able to construct empirical relations between the temperature (T, in °C), the S-wave velocity (Vs), and the depth (z) in the lithosphere (Fig. 5b). Their empirical relation in Figure 5b shows the dramatic decrease in Vs at temperatures above about 1200 °C that is also observed experimentally (e.g. Gribb & Cooper 1998; Jackson et al. 2002). It is this effect that is responsible for the ‘low-velocity zone’ at the base of plates, not the presence of partial melt.
(a) Shear-wave velocity structure beneath the Pacific Ocean, obtained from surface-wave tomography by Priestley & McKenzie (2006), averaged as a function of age. Noteworthy features are the cooling of the lithosphere away from the ridge, and the almost constant velocity at its base, which corresponds to a temperature variation of less than c. 20 °C. (b) The three black lines show Vs at 50 km depth beneath a cooling oceanic plate as a function of temperature, with 1σ upper and lower error bounds, obtained from (a) using the thermal model of McKenzie et al. (2005). The dashed line is the fit of the parameterized expression derived by Priestley & McKenzie (2006) to those data at 50 km depth. The red lines show the observed Vs (continuous line) and parameterized fit (dashed line) at 75 km. The rapid change in Vs above 1200 °C should be noted.
The next stage was to use observations of Vs(z) to estimate geotherms on the continents. A check of this procedure against geotherms already known from kimberlite nodule suites (Fig. 3) showed that the rapid change in temperature gradient at the base of the lithosphere could be well estimated from the seismological data; but with two caveats. First, at depths less than about 100 km, the seismic velocities determined by surface waves may be affected by downward ‘leakage’ from the overlying low-velocity crust, and therefore be poorly estimated. Second, the estimates of temperature T from Vs are most accurate above about 1100 °C, where Vs changes rapidly in response to changes in T (see Fig. 5b). Priestley & McKenzie (2006) then demonstrated that geotherms can be fitted to seismologically derived estimates of T(z) in the same way as they were fitted to the nodule data in Figure 3, to obtain estimates of lithosphere thickness on the continents, provided that only T(z) values exceeding 1100 °C at depths of 125 km or more are used. An & Shi (2006), using a different method from that of Priestley & McKenzie (2006), also mapped Vs(z) obtained from surface waves into T(z) and thence estimated lithosphere thickness variations under China; both studies obtained similar results in that region. Thus we have a seismological method for mapping the lithosphere thickness on the continents, providing it exceeds 125 km, which it does in most shields.
The essential feature of the base of the lithosphere is that it corresponds to a rapid change in temperature gradient over the thickness of the thermal boundary layer, rather than to a particular temperature itself. For this reason, as Priestley & McKenzie (2006) pointed out, it does not correspond to any particular feature in profiles of Vs(z); a fact unappreciated by some seismologists. It is only by fitting geotherms to Vs(z) that the base of the lithosphere becomes apparent.
Mapping the continental cores
With these new techniques, Priestley & McKenzie (2006), Priestley et al. (2006) and McKenzie & Priestley (2007) were able to produce maps of lithosphere thickness over the vast continental areas where it exceeded 125 km; an example from North America is shown in Figure 6. The thicknesses estimated from Vs(z) and from kimberlite nodule mineralogy are independent and agree to within about 25 km. An additional check is provided by the presence of diamonds in kimberlites, which is expected only where the lithosphere thickness exceeds about 140 km.
Lithosphere thickness beneath North America, calculated by fitting geotherms to temperature estimates obtained from shear-wave velocities, by Priestley & McKenzie (2006). The yellow line shows the approximate edge of the North American craton, determined from surface geological observations. Purple dots are the locations of diamond-bearing kimberlites, and yellow–green dots are kimberlites with no diamonds. The adjacent numbers are lithosphere thicknesses in kilometres estimated from mantle nodule geochemistry using the method of McKenzie et al. (2005).
The maps produced by Priestley & McKenzie (2006) confirmed that the lithosphere can reach thicknesses of 250 km or more; a fact already known from nodule data, but necessarily restricted to a few locations. However, there were also some surprises. Many of the regions of very thick lithosphere are associated with known Archaean or Proterozoic cratons, but not all cratons have thick lithosphere, and some of the thickest lithosphere, such as beneath Tibet, is not associated with a known craton at all. Furthermore, the edges of the thick lithosphere, even allowing for the limited horizontal resolution of the surface-wave techniques, correspond only poorly to the geological boundaries of the cratons mapped at the Earth's surface. For example, it appears that Phanerozoic orogenic belts have been thrust several hundreds of kilometres beyond the edges and towards the interior of the Scandinavian, North American and Siberian cratons. Thick lithosphere underlies the West Siberian Basin, a site of several kilometres of Mesozoic and Tertiary subsidence, and appears to be continuous between the Scandinavian and Siberian shields, underlying also the Urals. The NE China craton currently has thin lithosphere, although Ordovician mantle nodules show that it was thick in the Palaeozoic. For these reasons, and because the ages of these regions of thick lithosphere cannot be determined from seismology, Priestley & McKenzie (2006) referred to these regions as continental ‘cores’ rather than ‘cratons’.
The dramatic variations in lithosphere thickness between the cores and other regions are first-order features of the continents that have had an enormous influence on their evolution. This influence is achieved through the compositional characteristics of cores (and some cratons) that affect their buoyancy and rheology. It is thought that the constant thickness of old oceanic lithosphere is achieved through the boundary layer becoming unstable and convectively detaching, once it has thickened enough (Parsons & McKenzie 1978). Abnormally thick lithospheric mantle on the continents must be stabilized somehow, almost certainly by reducing its density by melt extraction (O'Hara 1975; Jordan 1979). The effects of such a process are shown in Figure 7. Increased melting of the mantle preferentially removes garnet (the heaviest mineral component) and reduces the iron content of the residual mantle. This reduces the density significantly, although having little effect on the seismic velocity (and it is because Vs is controlled primarily by temperature and not composition that it can be used to map lithosphere thickness). The depleted mantle lithosphere is thus rendered relatively buoyant, allowing it to achieve thicknesses of 200 km or more without delaminating; far exceeding the maximum thickness of c. 100 km seen in the oceans. The buoyancy thus created is an essential feature ensuring the longevity of cratons. The possible causes, and other consequences, of the melting responsible for it are discussed below.
Changes to mantle composition and physical properties as a result of melting, from Priestley & McKenzie (2006). The dramatic change in density is caused by the depletion in garnet (see mineral mode fraction) and loss of iron (see magnesium number, mg#). The density variation shown here is the same as that produced by a temperature rise of about 500 °C. By contrast, the corresponding change in Vs is equivalent to a temperature change of only 96 °C, when the temperature is below 1100 °C.
Seismicity of the Himalayan collision zone
The Himalayan collision zone has always been a focus of argument in debates on continental rheology. It has been known for some time that earthquakes as deep as 80–90 km occur beneath parts of southern and NW Tibet (Fig. 8), where seismicity is otherwise restricted to the top 10–15 km of the crust (e.g. Chen & Molnar 1983). This vertical distribution of seismicity was seen to typify the bimodal ‘jelly-sandwich’ model, and is still used as an argument to support that model today (e.g. Chen & Yang 2004; Schulte-Pelkum et al. 2005; Monsalve et al. 2006). Attention has focused on the possible occurrence of earthquakes in the uppermost mantle of southern Tibet (e.g. Monsalve et al. 2006), and their evident invalidation of the earliest ideas of Maggi et al. (2000a, b). The situation has recently been reviewed by Priestley et al. (2008), and the key points are as follows (Fig. 9).
Fault-plane solutions of earthquakes larger than mb c. 5.5 beneath Tibet. The red focal spheres are for earthquakes with centroid depths in the range 70–90 km, discussed in the text and by Priestley et al. (2008). The grey focal spheres are all for earthquakes with centroids shallower than 15 km.
Schematic cross-section through the Himalaya to illustrate the relations between structure and earthquake distribution. The actual data have been discussed in detail by Priestley et al. (2008). The extra vertical exaggeration above sea level should be noted. Black dots are earthquakes within the underthrusting Indian shield, and occur down to the Moho. White dots are shallow normal faulting events, mostly in Tibet but also at the top of the flexing Indian shield beneath the Ganges foreland basin. Blue dots are shallow-dipping thrusts at c. 15 km depth on the main underthrusting interface(s) beneath the Himalaya. The box outlines the area where Schulte-Pelkum et al. (2005) and Monsalve et al. (2006) located microearthquakes in the uppermost mantle as well as in the lower crust. The rigid, strong, granulitic lower crust of India is underthrust beneath the Himalaya, and to an unknown distance further north, indicated by question marks. This Indian lower crust helps support the elevation of Tibet, because the only way Tibet can collapse is by flow of the warm, weak lower Tibetan crust over this rigid base, which is a slow, dissipative processes (Copley & McKenzie 2007). MCT, Main Central Thrust; MBT, Main Boundary Thrust. The precise configurations of geological boundaries indicated by question marks beneath the Indus–Tsangpo suture zone (ISZ) are unknown.
(1) South of the Himalayan range front earthquakes occur throughout the thickness of the Indian shield crust, the lower part of which is thought to consist of dry granulite, responsible both for its seismogenic behaviour and its relatively large elastic thickness.
(2) However, the crystalline crust of the northern India is unusually thin (c. 35 km beneath the sediments of the Ganges basin) for an Archaean shield, leading to a steady-state Moho temperature, based on mantle nodule geochemistry, that could be as low as c. 500 °C.
(3) When this shield is thrust beneath the Himalaya in Nepal, the relatively low mantle temperature, together with the high strain rates associated with it adopting a ‘ramp-and-flat’ geometry, may be responsible for the uppermost mantle microearthquakes, recorded by Monsalve et al. (2006) and others, that accompany other earthquakes in the lower crust.
(4) In southern Tibet, the upper crust of India has been shortened south of the Indus Suture Zone, the uppermost lower crust of India has become hotter, and seismicity is restricted to a few earthquakes very close to the Moho at 80–90 km, where errors in Moho and earthquake depth determinations make it unclear whether these events are in the crust or mantle. A similar situation exists in NW Tibet beneath the Kunlun, where earthquakes at 80–90 km occur very close to the Moho. Both places are about 400 km NW of the Himalayan front, and may represent the minimum distance India has underthrust Tibet, so that the lower crust and mantle of India underlie most of the SE and nearly all of the NW Tibetan plateau.
(5) When taken out of geological context, the earthquake distribution beneath southern Tibet, with depths of <15 km or 80–90 km, resembles the bimodal depth distribution expected from the ‘jelly-sandwich’ model of Chen & Molnar (1983). However, that model is meant to be steady state and generic, whereas the geological context of southern Tibet clearly shows cold Indian material being thrust beneath the crust of Tibet.
(6) Priestley et al. (2008) concluded that the distribution of earthquake depths throughout the region is consistent with a generic global view of seismicity in which earthquakes occur in (a) ‘wet’ upper crustal material to a temperature of c. 350 °C, or (b) at higher temperatures in dry granulite-facies lower crust, or (c) in mantle that is colder than c. 600 °C.
Most importantly, it is impossible to understand the seismicity of the Himalayan collision zone without some realization that the situation is not in steady state. Material beneath southern Tibet was once south of the Himalaya, and has changed its temperature structure as it has underthrust northwards. The deep earthquakes close to the Moho at 80–90 km beneath SE and NW Tibet, and the mantle microearthquakes beneath Nepal, are interesting curiosities; but their contexts make clear that, contrary to the views of Chen & Yang (2004), Schulte-Pelkum et al. (2005) and Monsalve et al. (2006), they do not represent a bimodal depth distribution that vindicates the ‘jelly-sandwich’ model of Chen & Molnar (1983). Nor do they represent a sensible peg on which to hang a generic understanding of continental rheology.
Metastability, metamorphism and the nature of cratons
In addition to its unusual seismicity, the Himalayan collision zone is important for the insights it provides on the support of mountain belts and, recent work shows, on the origin of cratons and post-collisional regional metamorphism.
An important series of studies in the Caledonian root zone in western Norway by Austrheim and co-workers (e.g. Austrheim & Boundy 1994; Bjornerud et al. 2002; Lund et al. 2004) emphasized the importance of granulite metastability and the mechanical changes that accompanied the granulite–eclogite transformation. Granulite, if completely dry, can exist metastably at pressures and temperatures well beyond its nominal stability limit and, while doing so, remains ‘strong’ in the sense that it retains original fabrics for tens or hundreds of millions of years, does not flow appreciably on those time scales, and deforms, if at all, by brittle fracture in earthquakes. Water, if present, has a dramatic effect, acting as a catalyst that allows the transformation to eclogite, accompanied by a ductile deformation that completely removes all original fabrics. The importance of these processes for the Himalayan collision zone was summarized by Jackson et al. (2004): the foreland of India is strong (large Te) and seismogenic throughout the crust because it is likely to consist of dry granulite. It then remains as metastable granulite beneath the 80–90 km thick root zone of the High Himalaya and southern Tibet, and is responsible for the support of that elevation.
A popular misconception is that the mantle of peninsular India must be strong to support the Himalaya. If the rheology from the Indian plains to Tibet is treated as homogeneous, then the Indian mantle does indeed need to be relatively ‘strong’ (i.e. have high viscosity) to support the outward buoyancy force caused by the thick crust of Tibet adjacent to India, as models formulated in this way show (e.g. Hetenyi et al. 2006). However, if the lower crust of India, underthrust beneath southern Tibet, is essentially rigid, then the only way the Himalaya and Tibet can collapse is by flow of the Tibetan middle crust over this rigid base (Fig. 9). Such flows, involving strong vertical gradients of horizontal velocity within a channel that deforms by simple shear, are strongly dissipative, and have very long relaxation times associated with the removal of long-wavelength thickness variations (McKenzie et al. 2000). As there is now abundant evidence for the strong lower crust of peninsular India underthrusting the Himalaya (e.g. DeCelles et al. 2002), these flows are applicable to the collision zone (Copley & McKenzie 2007); in this sense, it is thus the crust of India that is responsible for supporting the Himalaya, and the rheology of its underlying mantle is unimportant.
The surface-wave tomography of Priestley & McKenzie (2006) shows a lithosphere beneath south Tibet approaching 250 km in thickness. A cross-section is shown in Figure 10, showing the high-velocity lithosphere lid in southern Tibet, which becomes less pronounced to the north, where it underlies a relatively low-velocity mantle immediately beneath the Moho. The presence of the mantle of the Indian shield beneath southern Tibet has been suggested before by others, based on SKS anisotropy observations (e.g. Chen & Ozalaybey 1998; Huang et al. 2000; DeCelles et al. 2002). Its presence confounds models suggesting that its delamination and resorption into the mantle is responsible for the normal faulting in southern Tibet, which instead can be explained by the lower-crustal flow of Tibet over the Indian lower crust (Copley & McKenzie 2007). McKenzie & Priestley (2007) went further still, and suggested that the low sub-Moho velocities beneath central–northern Tibet are related to a downward heating of the mantle lithosphere by the very great crustal thickness (60–90 km) above. This suggestion throws light on the possible origins of both post-collisional regional metamorphism and the generation of cratons.
SW–NE tomographic profile showing Sv deviations, obtained from surface waves, relative to a regional reference model (adapted from Priestley et al. 2006, 2008). The positions of the Main Central Thrust (MCT), Bangong suture (BS) and Kunlun fault (KF) are also shown. Mantle velocity deviations are shown only below 100 km, because of the extreme thickness of the Tibetan crust. The almost continuous high-velocity lid at depths of 150–250 km is clear, although it is more pronounced in the south. The low velocity of the uppermost mantle shallower than 150 km in northern Tibet is also clear in this image.
The internal generation of heat by very large crustal thicknesses (England & Thompson 1984; Le Pichon et al. 1997) removes the need for the advection of heat to the base of the crust by mantle delamination, which is a popular explanation for the occurrence of regional metamorphism and post-collisional granites on a time scale too short to allow for conductive heat transport through the lithosphere (e.g. Houseman et al. 1981). Figure 11a shows that significant internal heating of the crust can occur within a few tens of millions of years. This heating can cause partial melting of the lower crust, which will result in its dehydration, as any water will move into granites that migrate to higher levels in the crust. The remaining lower-crustal residue will be broadly granulitic in mineralogy (Burton & O'Nions 1990). McKenzie & Priestley (2007) discussed geochemical evidence for such processes beneath Tibet today. These processes act to generate a dry granulitic lower crust, thus producing an essential characteristic of the cratons (and, probably, the ‘cores’).
(a) Evolution of temperature within the lithosphere after sudden thickening by shortening. Before shortening by a factor of three at time t = 0, the thicknesses of the lithosphere and crust were 80 km and 30 km. The dashed line is the location of the Moho after thickening. Adapted from McKenzie & Priestley (2007). It should be noted how the temperatures both above and below the Moho increase substantially after a few tens of million years. (b) Thick crust above thick lithosphere can be generated by uniform thickening, as long as the mantle lithosphere is already buoyant through earlier depletion during melting, as it then does not delaminate when thickened. The thickened crust above it becomes internally heated and dehydrates at depth to form granulite. The hydrated upper crust can then be removed by erosion to leave a craton. Variations on this theme, with different initial structures and thickening modes, have been discussed by McKenzie & Priestley (2007).
The downward heating of the uppermost mantle is sufficient to reduce Vs to the values seen in the surface-wave tomography of Figure 11a (McKenzie & Priestley 2007), but it is not enough to generate the melting responsible for mantle depletion and density stabilization, which requires extraction of about 30% melt. McKenzie & Priestley (2007) discussed geochemical and petrological reasons for believing that depletion and formation of the harzburgites seen in kimberlite nodules occurred while the mantle lithosphere was still relatively thin (c. 120 km), perhaps during generation of komatiites in the Precambrian. What the modern example of Tibet shows is that such mantle can be subsequently thickened in collision zones, which can also generate dry granulite crust on top (Fig. 11b).
Thus cratonization may be occurring today in Tibet, and possibly in Iran too (McKenzie & Priestley 2007). The requirements for such cratonization are those of major continental collision zones involving depleted mantle lithosphere. Once formed, such cratons are very difficult to destroy.
Conclusions
A series of developments over the last decade are leading towards a new, coherent view of lithosphere structure, rheology and evolution. These developments began with simple questions related to regional variations in earthquake focal depths and effective elastic thickness estimates, but progressed rapidly into a much wider range of issues, all of which (perhaps surprisingly) are related to those original questions. The lithosphere characteristic that has probably had the greatest influence on continental geology is its thickness, which is related to its strength and buoyancy. Areas of thick continental lithosphere, which can now be identified as ‘cores’ and some of which are associated with shields, are stabilized by their buoyancy, derived from the depletion of their mantle through melting, and by their great strength, derived mostly from their dry granulitic crust. The thick lithosphere reduces the temperature of the crust, and so also plays a subsidiary role in preserving continental lithosphere strength over geological time. Such thick, strong lithosphere may be forming today in Tibet and Iran, where continental collision and the consequent internal heat generation from highly thickened crust is probably the essential mechanism.
Acknowledgements
This paper is Cambridge Earth Sciences contribution number 8962.
- © 2008 The Geological Society of London