The nature and causes of mantle heterogeneity

Don L. Anderson

Caltech, MS 252-21, Pasadena, CA 91125 USA,


Hotspots and hotspot tracks are on, or start on, preexisting lithospheric features such as fracture zones, transform faults,  continental sutures, ridges and former plate boundaries.  Volcanism is associated with regions of lithospheric extension or preexisting thin spots. The lithosphere controls the location of volcanism.  The nature of the volcanism and the occurence of "melting anomalies" or "hotspots", however, reflect the intrinsic chemical and lithologic heterogeneity of the upper mantle. The magnitude of magmatism reflects the fertility, not the absolute temperature, of the asthenosphere. Melting anomalies – volcanic chains, flood basalts, radial dike swarms – and continental breakup are frequently attributed to the impingement of deep mantle thermal plumes on the base of the lithosphere. The heat required for volcanism in this hypothesis is from the core. Alternatively, the volumes and rates of magmatism may be controlled by mantle fertility, melting point, ponding, focusing, and edge effects, i.e. plate tectonic and near-surface phenomena. I attribute the chemical heterogeneity of the upper mantle to subduction of young plates and seamount chains. As slabs warm up past the melting point of their crustal carapaces they become buoyant low-velocity diapirs that undergo further – adiabatic decompression – melting. The heat required for the melting of cold subducted material is extracted from the essentially infinite heat reservoir of the mantle, not the core. Melts from recycled oceanic crust, and seamounts – and possibly even plateaus – pond beneath  the lithosphere, particularly beneath basins, with locally thin or young lithosphere. The characteristic scale lengths – 150 km to 600 km – of variations in bathymetry and magma chemistry, and the variable productivity of volcanic chains, probably reflect compositional heterogeneity of the asthenosphere, not the scales of mantle convection or the spacings of hot plumes.

Mantle homogeneity; the old paradigm

Much of mantle geochemistry is based on the assumption of chemical and mineralogical homogeneity of the shallow mantle, with so-called  Normal Midocean Ridge Basalt (N-MORB) representative of the homogeneity and depletion of the entire upper mantle source ("the convecting upper mantle"). The entire upper mantle is perceived to be a homogeneous depleted olivine-rich lithology approximating pyrolite (pyroxene-olivine-rich rock) in composition. Venerable concepts such as reservoirs, plumes, temperature-crustal thickness correlations and others are products of these perceived constraints.

The perception that the mantle is lithologically homogeneous is based on two assumptions: 1) the bulk of the upper mantle is roughly isothermal (it has constant potential temperature) and 2) midocean ridge basalts are so uniform in composition ("the convecting mantle" is geochemical jargon for what is viewed as "the homogeneous well-stirred upper mantle") that departures from the basic composition of basalts-found along spreading ridges must come from somewhere else.   The only way thought of to do this is for narrow jets of hot, isotopically distinct, mantle to arrive from great depths and impinge on the plates. The fact that bathymetry follows the square root of age relation is an argument that the cooling plate is the only source of density variation in the upper mantle. The scatter of ocean depth and heat-flow as a fuction of age, however, strongly suggests that something else is going on. Plume influence is the usual, but non-unique, explanation for this scatter. Lithologic (major elements) and isotopic homogeneity, and constant temperature, of the upper mantle are linchpins of the plume hypothesis.  Another is that anomalous crustal thicknesses or eruption rates are proxies for mantle potential temperatures.

The isotopic homogeneity of MORB has strongly influenced thinking about the presumed homogeneity of the upper mantle. The homogeneity of MORB does not imply a homogeneous well-stirred upper mantle (e.g. Meibom & Anderson, 2003). The need to subdivide MORB [N-MORB, T-MORB, E-MORB, and P-MORB, for example] and the numerous "plume-influenced" sections of ridges, are indications that the basalts erupting along the global spreading ridge system are not completely uniform. It is common practice to avoid "anomalous" sections of the ridge when compiling MORB properties; this is another reason why MORB and the upper mantle are perceived as homogeneous.

Mantle heterogeneity: toward a new paradigm

It is increasingly clear that the upper mantle is heterogeneous in all parameters at all scales. The parameters include mineralogy, major and trace element chemistry, isotopes, melting point, and temperature. An isothermal homogeneous upper mantle, however, has been the underlying assumption in much of mantle geochemistry for the past 35 years (e.g. Zindler et al., 1984; Meibom & Anderson, 2003).  Derived parameters such as degree and depth of melting and the age and history   of mantle "reservoirs" are based on these assumptions. There is now evidence for major element  (Butler et al., 1993, Natland, 1989, Korenaga & Kelemen, 2000), mineralogical  (Bonatti & Michael, 1984; Dick et al., 1984; Dick, 1989; Dick & Natland, 1996, Niu et al., 2002, Salters & Dick, 2002), trace element (Cousens, 1996, Weaver, 1991, Hofmann & Jochum, 1996) and isotopic heterogeneity (e.g. Anderson, 1989, Gerlack, 1990), on various scales (grain size to hemispheric) and for lateral variations in temperature and melting point.

One must distinguish "fertility" from (trace element) "enrichment", although these properties may be related  (e.g. Anderson, 1989). Fertility implies a high basalt-eclogite or plagioclase-garnet content. Enrichment implies high contents of incompatible elements and long term high Rb/Sr, U/Pb, Nd/Sm etc. ratios. Because of buoyancy considerations, the most refractory products of mantle differentiation-harzburgite, lherzolite, and pyrolite-may collect at the top of the mantle and bias our estimates of mantle composition.  The volume fractions and the dimensions of the "fertile" components – basalt, eclogite, piclogite – of the mantle are unknown. There is also no reason to suppose that the upper mantle is equally fertile everywhere or that the fertile patches in hand specimens and outcrops are representative of the scale of heterogeneity in the mantle.

There are two kinds of heterogeneity of interest to petrologists and seismologists, radial and lateral. Melting and gravitational differentiation stratifies the mantle. Plate tectonic processes introduce lateral heterogeneities, some of which can be mapped by geophysical techniques. Convection is thought to homogenize the mantle although this is far from proved.  Free convection is not the same as stirring by an outside agent. Melting of large volumes of the mantle, as at ridges, tends to homogenize the basalts that are erupted, even if they come from a heterogeneous mantle.

Source of mantle Heterogeneity

Oceanic plates including basalts (often hydrothermally altered), mafic and ultramafic cumulates and depleted harzburgitic rock, are constantly formed along the 60,000 km long mid-ocean spreading ridge system. The mantle underlying diverging and converging plate boundaries undergoes partial melting down to depths of order 50-200 km in a region up to several hundred km wide, the processing zone for the formation of MORB, backarc basin basalts, and island arc basalts. Midplate volcanoes and off-axis seamounts process a much smaller volume of mantle, and the resulting basalts can therefore be much more heterogeneous. Before the oceanic plate is returned to the upper mantle in a subduction zone, it accumulates sediments and becomes hydrated.  Plateaus, aseismic ridges and seamount chains also enter subduction zones but their ultimate fate is uncertain.

Young plates, or slabs with thick oceanic crust, will not sink far into the mantle and are likely to reside in the shallow mantle after subduction, if they subduct. The distribution of ages of subducting plates is highly variable . There is a large amount of material of age  0-20 Myr and 40-60 Myr at subduction zones. About 15% of the surface area of plates currently approaching trenches is young (< 20 My) lithosphere (Rowley, 2002); more is in young back-arc basins. More than 10 % of the seafloor area is composed of seamounts and plateaus. Seamounts constitute up to 25% by volume of the oceanic crust (Gerlack,1990). This material, if subducted at all (Oxburgh & Parmentier, 1977) will warm up on short times scales and become neutrally buoyant. The basaltic parts will melt, even if the ambient mantle temperature is well below the normal mantle solidus.  Thick oceanic plateaus may accrete to continental margins and some may get trapped in suture zones between converging cratons.

The rate at which young crust-and potentially buoyant litosphere-enters the mantle is about 2 to 4  km3/yr (Rowley, 2002). The global rate of "hotspot" volcanism is  ~2 km3/yr  (Phipps Morgan, 1997). This  encourages us to think that "melting anomalies" may be due to  fertile patches of subducted oceanic crust that was young or thick at the time of subduction.  The fate of older plates and deeper slabs need not concern us, at the moment. Evidence for deep subduction (Grand et al.,1997) does not imply that all subducted material sinks into the lower mantle (Anderson, 1989, 2002).

Convection and diffusive equilibration  are  extremely sluggish.  Once in the mantle crustal materials and depleted residues of different ages are mechanically juxtaposed, but not chemically mixed or vigorously stirred. The resulting state of the upper mantle is a highly heterogeneous assemblage of enriched and depleted lithologies representing a wide range in chemical composition, melting point and fertility and, as a result of different ages of these lithologies, widely different isotopic compositions. Large-scale chemical heterogeneity of basalts sampled along midocean ridge systems occur on length scales of 150 to 1400 km. This heterogeneity presumably exists in the mantle whether a migrating ridge is sampling it or not. Fertile patches, however, are most easily sampled at ridges.

Mantle heterogeneity is not due to random or unknown effects. It is due to recycling of materials of known chemistry, dimensions and ages-in most cases. These materials were all at or near the surface of the Earth. They mostly remain and evolve at shallow depths. They are sampled as ridges move about and as fissures open up (e.g. Natland & Winterer, 2004). The variations in volume and chemistry observed at so-called hotspots may reflect the distribution and ages of slabs, particularly those that were young or had thick crust upon subduction.

Subducted material in the shallow mantle is resampled on charac-teristic time scales of  300 Ma to 2 Gy at leaky transform faults, extensional regions of the litho-sphere, by migrating ridges and upon continental breakup. In contrast, very old and cold slabs are more likely to sink deeper into the mantle, where they will reside for longer periods of time.  However, even the thicker slabs contribute their sediments and fluids, and possibly their crusts, to the shallow mantle. And even if they sink into the transition region they will may back into the shallow mantle as they heat up.

The fate of the crustal part of the slab depends on the heating rate vs. the sinking rate; the oldest slabs are expected to sink the deepest, and the fastest.  The time between subduction and island arc formation is too small for recycled crust to warm up and melt and contribute substantially to arc volcanism, except where young slabs, or ridges, subduct. This does not rule out subsequent melting of recycled basalt and eclogite as a contributor to the heterogeneity  of the asthenosphere and to ocean island basalts, seamounts and melting anomalies along midocean ridges.

Middle-aged plates reside mainly in the bottom part of the transition region, near and just below 650 km (Anderson, 1989a,b, 2002). Plates which were young (< 30 Ma ) at the time of subduction (e.g.  Farallon slab under western North America) and slabs subducted in the past 30 Myr are still in the upper mantle (Wen & Anderson, 1995, 1997a). Old, thick slabs appear to collect at 750-900 km (Wen & Anderson, 1997, Becker & Boschi, 2002).

The source of heat for large-scale eclogite melting is the huge volume of warm mantle enveloping a subducting slab (see  Oceanic plates in narrow closing ocean basins and backarc basins are much thinner than those at the subduction margins of old, huge plates, and do not require much reheating to become neutrally buoyant and even partially molten in the shallow mantle. They are unlikely to sink into the lower mantle.

Scale of mantle heterogeneity

In the plume model isotopic differences are attributed to different reservoirs at different depths. In the marble cake and plume pudding models the characteristic dimensions of isotopic heterogeneities are centimeters to meters. Meibom & Anderson (2003) attribute chemical differences between ridge and island basalts to the nature of the sampling of a common heterogeneous region of the upper mantle.  In order for this to work there must be substantial chemical differences over dimensions comparable to the volume of mantle processed in order to fuel the volcano in question, e.g. tens to hundreds of km. Chemical differences along ridges have characteristic scales of 200 to 400 km (Graham et al., 2001, Butler et al.,1993). Inter-island differences in volcanic chains, and seamount chemical differences, occur over tens of km e.g. the Loa and Kea trends in Hawaii. If heterogeneities were entirely grain-sized or km-sized, then both OIB and MORB would average out the heterogeneity  in the sampling process. If heterogeneities were thousands of km in extent and separation, then again OIB and MORB chemistry would not care about the sampling process. Therefore, there must be an important component of chemical heterogeneity  at the tens of km scale, the scale of crustal and lithospheric thicknesses. The hundreds of km scales are comparable to the segmentation of ridges, trenches and fracture zones. Chunks of slabs having dimensions of tens by hundreds of km are inserted into the mantle at trenches. They are of variable age, and equilibate and are sampled over various time scales. Some of them are seamount chains.

The Central Limit Theorem (CLT) is essential in trying to understand  the range and variability  of mantle products extracted from a heterogeneous mantle. In the standard geochemical model, differences are ascribed to separate reservoirs and convective homogenization of some (Hofmann,1997). The lower mantle is taken as the main isolated reservoir because of its remoteness and high viscosity. The crust, lithosphere, and perisphere are also isolated in the sense that isotopic anomalies can develop outside "the convecting mantle". Depending on circumstances , small domains-tens to hundreds of km in extent-can also be isolated for long periods of time until  brought to a ridge or  across the melting  zone. Mineralogy, diffusivity, and solubility are issues in determining the size of isolatable domains. When a multicomponent mantle is brought across the solidus and erupted, the magmas can be variable or homogeneous; this is controlled by sampling theory, the statistics of large numbers and the CLT.  Blending of magmas is an alternate to the point of view that convection is the main homogenizing agent of mantle basalts. There are also differences from place to place and with depth, i.e. large scale heterogeneities; Samoa doesn't necessarily represent just a different way of sampling the same mantle that the EPR does. For example, the perisphere concept (Anderson, 1989) places an enriched layer at the top of the mantle but this is attenuated or absent beneath ridges. The base of the plate collects melts from the asthenosphere and may become such an enriched layer.

The isolation time of the upper mantle is related to the time between visits of a trench or a ridge. With current migration rates a domain of the upper  mantle can be isolated for as long as 1 to 2 Gyr (Anderson, 1993). These are typical mantle isotopic ages and are usually ascribed to a convective overturn time.  Either interpretation is circumstantial.

Spectral analysis results

Geoid anomalies over the Pacific plate show linear undulations approximately oriented in the direction of absolute plate motion. Spectral analyses have revealed a broad range of dominant wavelengths, in the geoid and bathymetry, centered on wavelengths of 160 km, 225 km, 287 km, 400 km, 560 km, 660 km, 850 km, 1000 km, and 1400 km (Wessel et al., 2001, Cazenave et al., 1992).  Although these have been interpreted as the scales of convection and thermal variations they could also be caused by density variations due to chemistry and, perhaps, partial melt content. Several of these spectral peaks are similar in wavelength to chemical variations along the ridges, i.e. perpendicular to the spreading direction. The shorter wavelengths may be related to thermal contraction and bending of the lithosphere. The longer wavelengths probably correspond to lithologic (major element)  variations in the asthenosphere and, possibly, fertility and melting point variations.

Intermediate-wavelength (400-600 km) geoid undulations have been detected after along-track filtering of the Seasat altimeter data (Baudry & Kroenke, 1991; Maia & Diament, 1991). These lineations-roughly parallel to the absolute motion of the Pacific plate-are continuous across fracture zones and some have linear volcanic seamount chains at their crests. Systematic filtering has also revealed geoid anomalies  elongated in the east-west direction with dominant wavelengths of 750 km and 1100 km (Cazenave et al., 1992).

The basalts along midocean ridges are fairly uniform in composition but nevertheless show variations in major oxide and isotopic compositions. Long-wavelength periodicities in chemistry have been determined along an approximately 1100 km section of the southern East Pacific Rise (Butler et al., 1993).  Major and minor element chemistry shows spectral peaks with wavelengths of 225 km and 575. The length scales of the mantle compositions being melted are uncorrelated with those of magmatic temperature variations.

Indicators of the degree and depth of partial melting show a strong spectral peak near a wavelength of 430 km. There is significant power in the concentration spectrum of Na-an index of the amount of melting- near 260 km and of Fe-an index of depth of melting- near 200 km, bounding the average spectral peak for the oxides at 225 km. There appears to be strong coupling between the degree and depth of melting, and magmatic temperature or composition at length scales around 225 km and 400-600 km, about the wavelengths of geoid undulations observed in the vicinity of the East Pacific Rise.

Helium isotope data for MORB glasses recovered along 5,800 km of the southeast Indian ridge reveals structure at length scales of 150 and 400 km (Graham et al., 2001) that may be related to intrinsic heterogeneity of the mantle. Isotope variations in igneous rocks are generally interpreted in terms of convective mixing in the upper mantle, on the one hand, and unassimilated deep mantle material on the other. High 3He/4He ratios at some ocean islands, along with lower and relatively uniform values in mid-ocean-ridge basalts (MORBs), are assumed to result from a well mixed upper-mantle source for MORB and a distinct deeper-mantle source for ocean island basalts. Alternatively, this could be a result of sampling and magma mixing under the volcano (Meibom & Anderson, 2003). Large variations in magma output along volcanic chains occur over distances of hundreds to thousands of km and most chains are less than a thousand km long. I interpret these dimensions as the characteristic scales of mantle chemical and fertility variations. This provides a straightforward explanation of the order of magnitude variations in volcanic output along long volcanic chains.