398 | Nature | Vol 586 | 15 October 2020
Article
they feature a more fertile upper part (composed predominantly of
fertile lherzolites), a middle to lower portion that is more depleted
(predominantly harzburgites and depleted lherzolites), and a lower-
most part strongly affected by melt-related metasomatism. Depleted
volumes of the keels are located at two levels (Fig. 3a, shown in blue
and yellow). The lower level (120–200 km depth) contains most of
the keels and agrees well with the low-density anomaly related to the
viscous underplate in the numerical model (Fig. 1d and Fig. 3c). Thus,
the observation of a shallow fertile layer underlain by a depleted layer,
which is in turn underlain by a metasomatized layer, provides strong
support of the proposed model.
It should be noted that many lithospheric sections, especially
beneath Proterozoic areas, have been so affected by metasomatism
that they contain little clear evidence for primary layering in terms
of mantle depletion. The extent of the stratification in the Archaean
sections, whether primary or metasomatic, can be evaluated using
subsidiary data such as the distribution and composition of chromites,
and profiles of Al 2 O 3 in the whole rock and the fraction of magnesium
XMg in olivine (see Extended Data Figs. 4–7, and their legends). Some
types of metasomatism raise the Ca content of the protolith, introduc-
ing clinopyroxene without affecting the composition of pre-existing
chromite; the presence of high-Cr chromites in a lherzolite may be a
signature of such processes. Metasomatism by mafic melts, on the other
hand, introduces Al as well as Ca, leading to the growth of garnet, and
any residual chromite tends to have high TiO 2 content.
For example, in the southern part of the Kaapvaal craton^5 ,^29 ,^30 kimber-
lites older than about 110 million years (Myr) sampled a 190-km-thick,
highly depleted lithospheric root with harzburgites concentrated
below depths of 140 km (compare with the Limpopo Belt; Extended
Data Fig. 6a). However, sampling by younger kimberlites shows that
the sub-cratonic lithospheric mantle beneath this area underwent
large-scale metasomatism around 100 million years ago, character-
ized by melt-related refertilization (enrichment in Fe and Al) in the
deeper parts, a thinning of the lithospheric mantle root (keel) to about
160 km in thickness, and more carbonatite-related, phlogopite-rich
metasomatism in the upper part, producing a ‘fertile’ layer. In this
case the observed compositional layering can be clearly recognized
as largely secondary, obscuring a layering that is more pronounced
in older sections.
In summary, it is possible to identify many sections of cratonic
lithospheric mantle that show apparent boundaries between a more
depleted deeper layer and a more fertile shallow layer, or the inverse
in some cases (Slave craton; Extended Data Fig. 5). These boundaries
typically occur at around 120–140 km depth, and the deeper, more
depleted layers (blue in Fig. 3a) may be from 20 km to 70 km thick,
depending on the obscuring effects of melt-related metasomatism
(red in Fig. 3b).
Further evolution of the protokeels might be suggested, based on
numerical experiments in which the viscous underplate detached from
its parent subducting plate and because of that formed an independ-
ent depleted low-density layer beneath the arrested mantle (Fig. 1d,
Extended Data Fig. 3b). One can also speculate that several layers origi-
nating in this way may be superposed and accreted by subsequent sub-
duction episodes, contributing to the gradual growth of cratonic roots.
The complex layering of cratonic keels predicted by our model is also
consistent with seismic studies suggesting the presence of one or sev-
eral mid-lithosphere discontinuities at depths of 80–160 km (refs.^31 ,^32 ).
Although there is probably no universal cause for these discontinui-
ties^32 , they are likely to be frequently associated with the presence of
hydrated amphibole-bearing lithospheric mantle^32 , similar to what
is produced by hydrous diapiric upwellings in our models. Another
potential source of a discontinuity is the compositional boundary
between the depleted viscous underplate and the more fertile overly-
ing lithospheric mantle (Fig. 1d). The presence of the mid-lithosphere
discontinuity at relatively shallow depths (80–120 km; ref.^31 ) in some
cratons may again imply that their cratonic keels were built up by super-
position of several episodes of subduction-related viscous underplating
and hydrous diapirism.
Subcratonic mantle xenoliths in kimberlite pipes comprise mainly
peridotites and minor eclogites; this is also consistent with our model.
Small bodies of eclogite might be formed in the subcratonic mantle as
basaltic melts are extracted from the ascending mantle diapirs/plumes2,200 km
2,450 km020406080 100
Metasomatized peridotites (%)70
90110
130150170190210230250270
020406080 100
Depleted peridotites (%)Depth (km)
BotswanaSiberianLesothoGawlerLimpopoabRelative density in the model (kg m–3)10 0 –10 –20 –30 –40 –5 0Viscous
underplateArrested
mantlec
KaapvaalSlaveBotswanaLimpopoSiberianLesothoKaapvaalSlaveFig. 3 | Compositional stratification of lithospheric keels beneath well
known cratons compared to the predictions of the model. a, Profiles of
depleted mantle rocks (harzburgites and depleted lherzolites) in the keels
beneath the 7 cratons named on the diagram. b, Distributions of
metasomatized mantle rocks in the keels. c, Relative mantle density profiles
from the hard surface of the Earth across the protokeel areas (at distances of
2,200 km and 2,450 km, orange and magenta lines, respectively) computed
with respect to the protokeel-free cross-section (at 2,000 km) for the
numerical experiment presented in Fig. 1d. The layered structure of the keels
includes a middle to lower portion that is more depleted (probably related to
the viscous underplates), and a lowermost part strongly affected by
melt-related metasomatism^5 (probably related to plumes). Volumes related to
the mantle keels beneath two cratons (Slave and Kaapvaal) are shown in yellow
to emphasize the different depths of their melt-depleted layers. The structure
of the lithospheric keels is evaluated using garnet and chromite xenocrysts
from kimberlites and related rocks^14 (see Methods). More detailed descriptions
are presented in Extended Data Figs. 2–7.