Journal of Petrology Volume 42 Number 5 Pages 877-900 2001
© Oxford University Press 2001
Plume–Lithosphere Interactions in the Generation of the Basalts of the Kenya Rift, East Africa
R. MACDONALD1,*,
N. W. ROGERS2,
J. G. FITTON3,
S. BLACK4 and
M. SMITH5
1ENVIRONMENTAL SCIENCE DIVISION, IENS, LANCASTER UNIVERSITY, LANCASTER LA1 4YQ, UK
2DEPARTMENT OF EARTH SCIENCES, THE OPEN UNIVERSITY, MILTON KEYNES MK7 6AA, UK
3DEPARTMENT OF GEOLOGY AND GEOPHYSICS, UNIVERSITY OF EDINBURGH, WEST MAINS ROAD, EDINBURGH EH9 3JW, UK
4PRIS, UNIVERSITY OF READING, WHITEKNIGHTS, READING RG6 2AB, UK
5BRITISH GEOLOGICAL SURVEY, MURCHISON HOUSE, WEST MAINS ROAD, EDINBURGH EH9 3LA, UK
Received
January 1, 2000;
Revised typescript accepted
July 19, 2000
 |
ABSTRACT
|
|---|
Major and trace element and Sr–Nd–Pb isotopic data
for mafic volcanic rocks are used to assess the number of mantle
plumes contributing to the Tertiary–Holocene magmatism
of the Kenya Rift Valley, current estimates of which vary from
none to three. Rocks ranging in composition from nephelinite
to hypersthene-normative basalt have been sampled from three
lithospheric zones: the Tanzanian craton, the craton margin
reworked during the late Proterozoic, and the Mozambique mobile
belt. The magmas are interpreted as the products of variable
degrees of partial melting within the spinel–garnet peridotite
transition zone. Trace element and isotopic compositions from
all three zones are broadly similar to those of oceanic island
basalts, but there is considerable compositional variation,
which is related to a strong overprint from the lithosphere
on plume-derived melts. Sr and Nd isotopic ratios provide the
only clear distinction between magmatic rocks from the three
lithospheric domains. Within each setting, mafic magmatism has
tended to become less silica undersaturated with time, and at
any one locality magmatism has migrated towards the centre of
the rift. Magmas may have formed as a result of the infiltration
of plume-derived melts into the base of the lithosphere. The
extent of interaction of inferred plume melts with the lithosphere
has not varied systematically in time or space. The plume component
appears to be similar to the source of oceanic island basalts.
KEY WORDS: Kenya Rift Valley; mantle plumes; geochemistry; metasomatism
|
INTRODUCTION
|
|---|
Magmatism along the Kenyan and Ethiopian Rifts, the associated
flood basalts of northern Ethiopia and the volcanism of the
Afar Depression (collectively the East African Rift System)
represent one of the largest, currently active, continental
igneous provinces (Fig.
1). In the Kenya sector of the East
African Rift System alone, an estimated 924 000 km
3 of mafic
magma, including underplated material, has been generated in
the past 30 my (Latin
et al., 1993
). Such volumes are far too
large to have been produced solely within (anhydrous) lithosphere,
especially given the evidence for small melt fractions provided
by rare earth element (REE) modelling; the melt zone must actively
have been fed from upwelling asthenosphere (Latin
et al., 1993
).
Furthermore, estimated extension rates across the Kenya rift
are probably too small (ß-factors <1·7;
Hendrie
et al., 1994
; Mechie
et al., 1997
) to explain the magma
volumes and the protracted duration of magmatism (
35 my) without
the added effect of elevated mantle temperatures. The fact that,
in both the northern and southern parts of the rift, volcanism
pre-dated faulting by a few million years (Morley
et al., 1992
;
Smith, 1994
) also suggests that the initial magmatism was related
to the presence of hot mantle material beneath the site of the
present rift, such mantle having a potential temperature of
perhaps 1500°C (Mechie
et al., 1997
).
View larger version (47K):
[in this window]
[in a new window]
|
Fig. 1. Location map of Cenozoic volcanic rocks (black) of the East African Rift System and of the Ethiopian and Kenyan plateaux or domes (dashed lines). The arrow shows the apparent migration as a result of plate motions of a mantle plume, currently centred on Lake Victoria, over the past 50 my. Modified from George et al. (1998, fig. 1).
|
|
Geophysical and geochemical evidence strongly points to the involvement of the Afar mantle plume in the generation of the Ethiopian flood basalts and the magmatism in Afar (Marty et al., 1996; Hoffman et al., 1997). There is, however, considerable uncertainty about the number of plumes influencing magmatism and tectonics beneath the southern Ethiopian, Kenyan and Tanzanian sectors of the rift system. Gravity and topographic anomalies reveal two broadly circular, dynamically supported plateaux, implying two distinct mantle plumes (Ebinger et al., 1989). However, a more recent suggestion (Ebinger & Sleep, 1998) involved impact of the Afar plume beneath southern Ethiopia at 45 Ma, followed by rapid lateral flow of plume material guided by the topography at the base of the lithosphere to feed magmatism within lithospheric thin-spots generated by Mesozoic and early Tertiary extension. Smith (1994) related magmatism in the main, Gregory, part of the Kenya rift to the rise of a plume focused beneath a weak zone marking the boundary between the Tanzanian craton and the adjacent Proterozoic mobile belt. He suggested that the mantle plume may lie beneath the craton or much further north, beneath Ethiopia. Birt et al. (1997) and Mechie et al. (1997) envisaged a mantle plume confined beneath the Tanzanian craton, from which hot material rises to shallower depths under the outer rims formed by the western and eastern arms of the East African Rift System.
George et al. (1998) favoured a two-plume model, proposing that the Ethiopian province was initially related to the thermal influence of the Kenyan, and subsequently the Afar, mantle plumes during northward drift of the African plate over the past 45 my. Rogers et al. (2000) used isotopic evidence to argue that a Kenya plume, whose origins are in the upper mantle, may be distinguished from the Afar plume, which has a deep mantle source. Burke (1996) favoured three plumes in the evolution of the East African Rift System. The first generated the basalts of southern Ethiopia between 45 and 35 Ma and perhaps also contributed to the Lokipiti volcanism at 35 Ma. The second, Afar, plume impinged on the base of the lithosphere and caused eruptions starting at 35 Ma. A third, Samburu, plume resulted in eruption of 30 000 km3 of basalt in north–central Kenya at 20 Ma.
In contrast, Bailey (1992) argued against the involvement of plumes at all, suggesting that Tertiary–Recent magmatism in the East African Rift System is related to lithospheric stresses from collisional tectonics between the African and Eurasian plates. Still others take the compromise view that rifting in Kenya is related to the rise of a plume (or plumes) under lithosphere already being extended in response to external forces. Zeyen et al. (1997) have argued that the Kenya rift area has been subjected to compressional far-field forces that arose partly from ridge push from the Indian Ocean and partly from the high topography of the Ethiopian Plateau above the Afar plume. Coblentz & Sandiford (1994), on the other hand, have suggested that a large part of the extensional forces in east Africa is a result of density variations within the lithosphere, rather than any far-field forces.
Attempts to use geochemical data for basalts to distinguish between the various plume (and non-plume) models are complicated by strong evidence of a lithospheric signature in the eruptive products (Norry et al., 1980; Davies & Macdonald, 1987; Rudnick et al., 1993; Class et al., 1994; Macdonald, 1994; Bell & Dawson, 1995; Paslick et al., 1995; Bell & Simonetti, 1996; Kalt et al., 1997; Rogers et al., 2000). Furthermore, the basement to the Kenya rift (implied here to be that part of the East African Rift System between Turkana and northern Tanzania) is complex, being divisible into three zones: the Archaean Tanzanian craton, the Panafrican Mozambique mobile belt and a zone of craton margin reworked during the Panafrican orogeny (Smith & Mosley, 1993). It might be expected that the nature of the lithosphere contribution to basalt composition would change at the boundaries between these zones.
In this paper, we use the first systematic survey of basalt geochemistry along the length of the Kenya rift to address three issues relating to the role of mantle plumes in rift evolution:
- is rift magmatism related to the ascent of one or more mantle plumes? Did the plumes contribute materially to the magmatism or did they provide only the thermal energy for partial melting?
- Has there been a significant lithospheric input to rift magmatism? Do the magmatic affinities of the eruptive rocks reflect variable inputs from the inferred plume and lithosphere components, and have the proportions changed either with time or with distance from any inferred plume axis?
- Do the magma compositions reflect changes in the nature of the lithosphere, particularly the change from mobile belt to craton?
|
GEOLOGICAL SUMMARY
|
|---|
The earliest volcanic rocks in the Kenya rift are those of the
Turkana region in the north, some 30–35 my old (Figs
1 and
2). Since then, the onset of magmatism has generally migrated
southwards through central and southern Kenya, finally reaching
northern Tanzania some 5–8 my ago. This represents a simplification
of a more complicated picture, however. As the rift opened out
into the rigid craton, it also developed and propagated north
of Baringo (Fig.
2) to form the inner rift and Quaternary volcanic
rocks. The overall southerly progression of the magmatism was
accompanied by a change in the nature of the lithosphere.The
Kenya rift straddles the margin of the Tanzanian craton (Fig.
2). This margin was reworked, overthrust and buried by rocks
of the Mozambique mobile belt during a late Proterozoic collisional
event. Integrating a wide range of geological and geophysical
results, Smith & Mosley (1993)
and Ebinger
et al. (1997)
showed that a major influence on the development of the rift
has been its location above a lithological heterogeneity, with
a contrast between the cold, thick, rigid Archaean lithosphere
and thinner, anisotropic, warmer, mobile belt lithosphere. During
the late Proterozoic, the mechanical anisotropy that existed
between the two types of lithosphere was modified by a series
of continental-scale NW–SE and north–south trending
ductile–brittle shear zones that affected the craton margin
and the mobile belt. Smith & Mosley (1993)
argued that reactivation
of these shear zones under varying stress field conditions has
controlled the location and geometry of graben structures and
the emplacement of magma since early Miocene times. Smith &
Mosley (1993)
recognized, therefore, three major lithospheric
blocks in the structural framework of the rift (Fig.
2): the
craton, reworked craton margin and the mobile belt.
Deep rift structure
Crustal thickness varies along the axial part of the rift from 20 km beneath Turkana to 35 km beneath the Kenya dome in south–central Kenya to 38 km beneath northern Tanzania (Mechie et al., 1994, 1997; Last et al., 1996). The thinning has been accomplished by the northerly thinning of the lowermost crustal layer from 9 to 2 km. The layer has a P-wave velocity of 6·8 km/s, indicating that the lower crust may be a mix of high-grade metamorphic rocks, and underplating mafic and ultramafic material (Mooney & Christensen, 1994; Hay et al., 1995; Mechie et al., 1997).
Along the length of the rift, seismic refraction surveys have indicated Pn velocities of the uppermost mantle to be 7·5–7·7 km/s, which contrasts with values of 8·0–8·2 km/s beneath the rift shoulders and flanks. Teleseismic tomography shows a velocity reduction of 6–12% in an essentially vertical channel beneath the rift axis, extending to 160 km depth or more (Achauer et al., 1994; Slack et al., 1994). The velocity reduction has been explained by the presence of small amounts (3–6%) of partial melt (Green et al., 1991; Achauer et al., 1994) and/or an rise in potential temperature by 300 K beneath the rift (Sobolev et al., 1996). The presence of steep-sided low-velocity zones extending deep into the mantle is consistent with gravity models of low-density material reaching deep into the mantle (Ebinger et al., 1989; Hay et al., 1995; Birt et al., 1997; Simiyu & Keller, 1997).
Temperature distribution with depth beneath the rift is poorly constrained. As a guide, Mechie et al. (1997) used the following values for modelling velocity distribution. Under the northern and southern sections of the Kenya rift, they estimated Moho temperatures of 630°C (21 km depth) and 1000°C (35 km), respectively. These values correspond to surface heat flows of 105–120 mW/m2, close to the average of 105 mW/m2 measured from the rift by Morgan (1983). For greater depths, Mechie et al. (1997) estimated, for the northern rift, temperatures rising from 940°C at 45 km to 1100°C at 63 km; for the southern rift, the temperatures for the same depths are 1200°C and 1350°C, respectively.
Surface heat flow from the Tanzania craton is much lower (<50 mW/m2), approaching normal values for stable cratonic lithosphere (Nyblade et al., 1990, 1996). Lower lithospheric temperatures are supported by microseismic studies; focal depths are greater in northern Tanzania than in Kenya. Thus heat flow and seismic evidence are consistent with the idea of a southward propagating rift system, such that at least the upper parts of the craton have not been thermally modified by the action of any inferred plume.
|
ANALYTICAL TECHNIQUES
|
|---|
A total of 128 samples were analysed for major and trace elements
by X-ray fluorescence (XRF) at the University of Edinburgh,
using techniques outlined by Fitton
et al. (1998)
. The samples
came from several sources, including the collections of the
British Geological Survey, the East African Geological Research
Unit (University of Leicester), Lancaster University, the University
of Oregon and Professor J. B. Dawson. Specimen details are given
in the Appendix. Localities referred to in the text are shown
in the Appendix (Fig.
A1). We have also used the XRF datasets
for Tanzanian eruptive rocks of Paslick
et al. (1995; 27 samples)
and Kenyan eruptive rocks of Rogers
et al. (2000; 30 samples)
,
because they were analysed in the Edinburgh laboratory. The
total number of analyses is thus 185. A subset of 59 samples
was also analysed by instrumental neutron activation analysis
(INAA; trace elements) at the Open University (Potts
et al.,
1985
). Representative analyses are given in Table
1; the complete
dataset is available for downloading from the
Journal of Petrology Web site, at
http://www.petrology.oup.journals.org.
View larger version (19K):
[in this window]
[in a new window]
|
Fig. A1. Map, simplified from Fig. 2a, showing localities referred to in the text. Lor, Lorikipi or Loriu; Na, Namarunu; K, Kerio Valley; Ka, Kaparaina; Ng, Ngelesha Escarpment; LB, Lake Bogoria; Lois, Loisiomurto; Olg, Olokisalie (Olorgesailie).
|
|
Sr, Nd and Pb isotopic data for a sub-set of these samples are taken from the literature (Davies & Macdonald, 1987; Macdonald et al., 1995; Paslick et al., 1995; Black et al., 1998; Rogers et al., 2000).
|
NOMENCLATURE AND DISTRIBUTION
|
|---|
For this study we decided to work on mafic igneous rocks from
the axial regions and shoulders of the rift, thus ensuring,
first, that we had the most complete geophysical information
on deep structure, and, second, that we could confidently relate
our samples to the basement zones established by Smith &
Mosley (1993)
. This approach also allowed us to have ‘time-slices’
through geographically contiguous successions, in that earlier
work has shown that there has been a tendency for volcanism
in any given region of the rift to migrate with time towards
the axial zone (Lippard & Truckle, 1978
; Truckle, 1980
).
The main data gap is for the older (>20 Ma) successions of
the Turkana rift (Bellieni
et al., 1987
; Morley
et al., 1992
;
Ebinger
et al., 1993
). Most of the rocks from northern Kenya
were selected to lie on or close to three cross-rift traverses,
all within the mobile belt (Fig.
2). The geology of much of
the area has been described by Dunkley
et al. (1993)
. Traverse
1 runs from the Kerio valley in the west through Kaparaina to
Baringo in the central rift to the Ngelesha escarpment. It incorporates
successions ranging from Miocene (20–15 Ma) to Recent
(<120 ka). Rocks of Traverse 2 (Ribkwo–Silali volcano–Laikipia)
cover a similar age range (16 Ma–a few hundreds of years).
Traverse 3 runs from Lorikipi–Loriu in the west through
Namurunu on the rift floor to the Tirr Tirr plateau in the east,
samples ranging from 16·8 ± 2 Ma to <10 ka.
The range in ages of the rocks within each traverse allows us
to examine compositional variations with time at the local,
as well as regional, levels. All rocks in the dataset were chosen
to be essentially unaltered, with an absence or scarcity of
secondary carbonates or zeolitization and, at worst, minor serpentinization
or oxidation of olivine.
All samples have mg-number 
40 (calculated with Fe2O3 = 0·15 Fe2O3*) and SiO2 < 51 wt %. According to the LeBas et al. (1986) total alkali–silica (TAS) classification scheme, the rocks are foidites, picrobasalts, basanites, basalts and hawaiites. Here we use a simplified classification, recognizing four groups on the basis of CIPW normative characteristics: (1) nephelinites (ne + leu > 10%); (2) basanites (5 < ne <10%); (3) alkali olivine basalts (0% < ne < 5%); (4) hypersthene (hy)-normative basalts. The divisions are somewhat arbitrary, in that there is a complete compositional spectrum between the most and least silica-undersaturated rocks, and rocks of more than one type are normally intimately related in the field.
The number of samples in each lithospheric domain is shown in Table 2. An important difference is in the proportions of more and less silica-undersaturated rock types. For example, in the craton, nephelinites and basanites form 80% of our samples and alkali olivine basalts and hy-normative basalts 20%. The corresponding figures for the reworked craton margin are 33% and 67%, and for the mobile belt 21% and 74%. These proportions would strictly be applicable only if our sampling had been truly random. Nevertheless, they are consistent with previous observations of the distribution of magma types based on detailed field mapping (e.g. LeBas, 1987; Smith & Mosley, 1993; Smith, 1994). In the Kenya rift, olivine-poor nephelinitic and carbonatitic volcanism is largely restricted to the craton and reworked craton margins. In the mobile belt, olivine-rich nephelinitic volcanism is associated with large shield volcanoes and high-volume eruptions of alkali and transitional basalts. Smith (1994) has related the differences to thinner lithosphere beneath the mobile belt, which is more easily eroded and penetrated by rising magmas, partly because it had previously undergone significant stretching and warming during Cretaceous and Palaeogene rifting events.
Mg-numbers for the mafic rocks vary from 40 to 80 but only 19 samples possibly represent primary magmas, i.e. mg-number >65. We appreciate that although the use of mafic rocks minimizes the effects of polybaric fractionation on incompatible trace element ratios, it does not totally preclude them. Nevertheless, the spread of compositions that we shall report here, e.g. in incompatible trace element ratios, is independent of mg-number, such that the trace element characteristics of the eruptive rocks dominantly reflect mantle source composition and mantle processes.
Compositional relationships between magma types
Within each setting, the transition from average nephelinite through basanite and alkali olivine basalt to hy-normative basalt is marked overall by decreases in Na2O, K2O and the incompatible trace elements (ITE), and increases in SiO2 (Table 2). This relationship is normally considered to reflect different degrees of partial melting of compositionally similar mantle sources. Figure 3 is a Ce/Y vs Zr/Nb plot, where the continuous lines represent non-modal fractional melting curves for partial melting of fertile and depleted spinel peridotite and garnet peridotite (Hardarson & Fitton, 1991). Two features may be noted. First, within each setting, as exemplified by the rocks of the reworked craton margin in the inset in Fig. 3, the sequence of average compositions nephelinite 
hy-normative basalt appears to represent progressively shallower (increased component of spinel-facies peridotite) melting, rather than larger degrees of partial melting, although the spread within the dataset for each setting indicates variable melt fractions. Second, regardless of magma type, the craton rocks appear to have been generated at the highest pressures and the mobile belt rocks at the lowest, although there is substantial overlap between the magma types. The lower heat flow through the craton might be expected to result in smaller volumes of melt and smaller melt fractions. Whereas the former is clearly true, the latter does not seem to be so because the trace element ratios suggest similar melt fractions, albeit from greater depth. Therefore controls on melt fraction are not directly related to lithosphere thickness, although depth of melting is. However, this assumes that the characteristics of the melts are totally controlled by the melt regime, whereas they must also reflect the nature of the source region, e.g. whether it is enriched or not and the style of that enrichment.
View larger version (26K):
[in this window]
[in a new window]
|
Fig. 3. Ce/Y–Zr/Nb plot of mafic volcanic rocks. The continuous lines are non-modal fractional melting curves calculated by Hardarson & Fitton (1991) for four mantle compositions: GD, depleted garnet lherzolite; GP, primitive garnet lherzolite; SD, depleted spinel lherzolite; SP, primitive spinel lherzolite. Numbers on lines refer to percentages of melt. In the main diagram, all data are plotted, subdivided by structural setting. In the inset, the average compositions of each lithology in the reworked craton margin are shown. RCM, reworked craton margin; MB, mobile belt; BSN, basanite; AOB, alkali olivine basalt; HNB, hypersthene-normative basalt.
|
|
The major and trace element data indicate, therefore, that the mafic rocks were formed, within each setting, over a range of pressures, roughly spanning the garnet–spinel transition, and by variable, but apparently always small, degrees of melting.
|
COMPOSITIONAL DIFFERENCES BETWEEN SETTINGS
|
|---|
Major and trace element data
An aim of this project was to determine whether the composition
of the mafic eruptive rocks of the rift is related to lithosphere
type, in particular the primary distinction between mobile belt,
reworked craton margin and craton. At this stage, we shall examine
this question using average major and trace element compositions
for each magma type (Table
2), selected major element plots
(Fig.
4) and mantle-normalized plots for average lithologies
in each setting (Fig.
5). Comparisons are slightly complicated
by the different average
mg-number of rocks from each setting.
For example, highly magnesian, strongly undersaturated (leucite-normative)
rocks are virtually restricted to the craton and reworked craton
margin, such that the average nephelinites from these settings
are higher in CaO, K
2O, TiO
2 and P
2O
5, and lower in SiO
2 and
Na
2O than average nephelinite from the mobile belt. Allowing
for such differences in
mg-number, the major element compositions
for each magma type are seen to be very similar (Table
2). Minor
differences are that nephelinites and basanites from the craton
tend to be more potassic than those from the other settings
(Fig. 4a), whereas rocks from the mobile belt and reworked craton
margin tend, at given MgO, to be more aluminous than those from
the craton (Fig. 4b), possibly because they have equilibrated
at slightly lower pressures or were derived from a more enriched
or less depleted source.
View larger version (19K):
[in this window]
[in a new window]
|
Fig. 5. Mantle-normalized element plots showing the average composition of each magma type in the craton, reworked craton margin and mobile belt. Normalizing values from Sun & McDonough (1989).
|
|
Incompatible trace element variations in each magma type are summarized in mantle-normalized diagrams in Fig. 5. Patterns for the nephelinites are remarkably similar in each setting, average abundances increasing in the order mobile belt through reworked craton margin to craton. The basanite and alkali olivine basalt patterns are overall similar to those in the nephelinites, although the order of enrichment of individual incompatible trace elements is more variable; for example, in the basanites, the order of enrichment in Th is craton > reworked craton margin > mobile belt; for P the order is reversed. The patterns for the hy-normative basalts are slightly different from those of the other types in that the K trough is much less marked, especially in the craton margin and mobile belt rocks.
Mantle-normalized REE plots are given in Fig. 6. Light REE (LREE) enrichment relative to heavy REE (HREE) is generally strongest in more undersaturated rocks, i.e. nephelinites and basanites > alkali olivine basalts > hy-normative basalts. This is consistent with a greater garnet control in undersaturated rocks and greater spinel control in hy-normative basalts. A La/Yb–Tb/Yb plot can be used to summarize differences in REE behaviour between settings (Fig. 7). There is very significant overlap between rocks from each setting but basalts from the mobile belt show a slight tendency to have less fractionated REE patterns, with lower values of both ratios than the craton and craton margin samples. The data define a broad linear trend between melting curves calculated for fractional melting of fertile garnet-bearing and garnet-free peridotite. The steep trend indicates an important role for garnet in the craton and craton margin basalts, i.e. they were generated at higher pressures, which is consistent with Ce/Yb–Zr/Nb relationships (Fig. 3). As with the other incompatible trace elements, there are no major differences in REE patterns for the same lithologies in the different settings. There would appear, therefore, to be no compelling evidence from the major and trace element data that lithospheric type has strongly influenced magma composition.
View larger version (29K):
[in this window]
[in a new window]
|
Fig. 6. Mantle-normalized REE plots of the average composition of each magma type in craton, reworked craton margin and mobile belt. Normalizing values from Sun & McDonough (1989).
|
|
View larger version (35K):
[in this window]
[in a new window]
|
Fig. 7. La/Yb–Tb/Yb plot of rocks from the craton, reworked craton margin and mobile belt, with the field of oceanic island basalts (OIB) for comparison. Continuous lines are for melting of fertile lherzolitic mantle, the contours representing the amount of modal garnet. The small, 0%, field represents melting in the spinel peridotite facies. Partition coefficients from Halliday et al. (1995). Data for the OIB field in this and subsequent diagrams are from a compilation by N. W. Rogers of published analyses.
|
|
Sr–Nd–Pb isotopic evidence
87Sr/86Sr in the whole dataset ranges from 0·7030 to 0·7057, 143Nd/144Nd from 0·5130 to 0·5124 and 206Pb/204Pb from <18 to >21. These ratios generally lie within the ranges defined by ocean island basalts (OIB) (Hofmann, 1997). In contrast to major and trace element compositions, the boundary between the craton (including its reworked margin) and the mobile belt is reflected in the Sr and Nd isotope ratios of the basalts (Fig. 8). Those basalts erupted through the craton have 143Nd/144Nd < 0·51275 and 87Sr/86Sr > 0·7035, whereas those from the mobile belt have 143Nd/144Nd > 0·51275 and 87Sr/86Sr ( 0·7035. This is the only clear evidence of a control exerted by the underlying lithosphere on magma compositions. The data for rocks from the craton and craton margin define two sub-parallel trends in which 87Sr/86Sr increases with little variation in 143Nd/144Nd. Such trends have been shown locally to be related to crustal contamination, e.g. at the Naivasha (Davies & Macdonald, 1987) and Silali (Macdonald et al., 1995) complexes.
View larger version (23K):
[in this window]
[in a new window]
|
Fig. 8. 87Sr/86Sr–143Nd/144Nd plot for all data, subdivided by structural setting. (Note the approximately horizontal data arrays at 143Nd/144Nd values of 0·5127 and 0·5126.)
|
|
Variations in Pb isotopic ratios in the whole dataset are parallel to, but generally displaced above, the Northern Hemisphere Reference Line (NHRL) (Fig. 9). Mobile belt rocks are restricted to the lower and middle parts of the range (206Pb/204Pb 18·5–19·5). Rocks from the reworked craton margin also show a rather restricted range and are displaced to higher 207Pb/204Pb than those from the mobile belt. Pb isotopic ratios in the craton rocks define a broadly linear array that crosses the NHRL. If the array has age significance, an age of 2–2·5 Ga is inferred, which is comparable with the 2·5–2·7 Ga age of the oldest rocks in the Tanzania craton (Bell & Dobson, 1980) and the age of 2·5–2·9 Ga for lithosphere stabilization, deduced from Re/Os studies of Tanzanian mantle xenoliths by Chesley et al. (1998).
View larger version (20K):
[in this window]
[in a new window]
|
Fig. 9. 207Pb/204Pb–206Pb/204Pb and 208Pb/204Pb–206Pb/204Pb plots, using all data, subdivided by structural setting. Northern Hemisphere Reference Line (NHRL) from Hart (1984).
|
|
The fact that Pb isotopes do not distinguish between mafic rocks erupted through the craton and mobile belt as effectively as Sr and Nd isotopes may be in part because Pb isotopes are more susceptible to crustal contamination than are Sr and Nd. However, there is no correlation between such indices of crustal contamination as Ce/Pb and 207Pb/204Pb, suggesting that the major control on Pb isotopes has been source composition. The complexity of isotopic relationships in the craton was noted by Paslick et al. (1995), who showed, for example, that the combination of HIMU Pb and EMI Sr and Nd means that the lavas did not acquire their isotopic compositions simply from an asthenospheric OIB source. Attempts to identify mantle components are even more complicated when the full dataset is considered. Excluding any crustal component, rocks from all three lithospheric domains would require three- or four-component mixes of the hypothetical end-members recognized in oceanic basalts—HIMU, EMI, EMII and DM (Fig. 10).
View larger version (18K):
[in this window]
[in a new window]
|
Fig. 10. 87Sr/86Sr–206Pb/204Pb plot, using all data, subdivided by structural setting. Values for HIMU, EMI, EMII and DM end-member compositions from Hart et al. (1992).
|
|
|
BASALT COMPOSITION VS TIME: SOME GENERALIZATIONS
|
|---|
On the basis of spatial and temporal variations in basalt composition
along the northern sector of the rift, Lippard & Truckle
(1978)
and Truckle (1980)
suggested that (1) mafic lavas west
and east of the rift are generally more silica undersaturated
than are lavas within the rift; (2) most sectors of the within-rift
sequences show a decrease of silica undersaturation of mafic
lavas with time; (3) during the Miocene and Pliocene, mafic
lavas became increasingly silica undersaturated along the rift
axis towards the south; (4) during the Quaternary, mafic lavas
have shown a trend of increasing silica undersaturation northwards
and southwards away from the Naivasha area, the culmination
of the so-called Kenya Dome.
These relationships do not always hold at the regional and local levels. Crossley & Knight (1981) found that there is no simple relationship between magma alkalinity and distribution in space and time in the western part of the southern rift in Kenya, whereas lavas ranging from Tertiary to Recent age in the Turkana rift seem to have been produced from similar sources by similar processes (Bloomer et al., 1989). Individual central volcanoes, such as Olokisalie (Olorgesaillie) in the southern rift, can contain lavas ranging from nephelinites to hy-normative basalts (Baker, 1987). In the part of the Kenya rift between Lake Bogoria and Turkana, Dunkley et al. (1993) reported that Quaternary basalts become increasingly undersaturated northwards and that, in centres that have erupted both types of basalts, the hy-normative varieties pre-date the ne-normative types.
Our new data largely corroborate the generalizations about the distribution of magma types in time, at least for the 20 my period represented by our dataset. At scales varying from that of the rift through regional (e.g. Traverse 2; Samburu volcanic rocks; southern rift) to that of individual centres (Olorgesaillie), magmatism tends on average to become less silica undersaturated with time (Fig. 11). This is caused by two effects; nephelinites disappear from the sequences and hy-normative basalts become more important volumetrically. For example, there is only one hy-normative basalt older than 10 Ma in our samples; in the mobile belt and reworked craton margin, all nephelinites are older than 11 Ma and 2·5 Ma, respectively. We have no samples of very recent nephelinitic magmatism from the mobile belt and reworked craton margin. Because at any one locality magmatism has tended to migrate towards the centre of the rift, it follows that the more silica-saturated rocks are concentrated towards the axis, although not exclusively.
At first sight, the disappearance of nephelinites from the eruptive sequences might appear consistent with the idea of Thompson & Gibson (1994) that prolonged magmatism at a rift site removes most or all of the (usually potassic) lithospheric-source melt, allowing subsequent melts from convecting mantle to penetrate to the surface without significant interaction with lithosphere-derived melts. We shall show below, however, that at any site in the Kenya rift, the lithospheric signature in the mafic rocks does not decrease systematically with time. Instead, the time trends (Fig. 11) seem to indicate that soon after initiation of magmatism in each area, the melt column extended from the generation depths of nephelinites (100 km; Macdonald et al., 1994) to those of the hy-normative basalts (<50 km). With time, the column tended to move upwards, such that nephelinitic magmatism ceased and alkali olivine basalts and hy-normative basalts became increasingly important volumetrically.
|
COMPARISONS WITH OCEAN ISLAND BASALTS
|
|---|
Deviations of the Kenya data from OIB are clearly seen in plots
of elements of different compatibilities, normalized to Y to
minimize the effects of fractional crystallization. Thus in
the Nb/Y–Zr/Y plot (Fig.
12), several rocks from all three
tectonic settings show elevated Nb/Y at a given Zr/Y relative
to OIB. Nevertheless, the large degree of overlap between the
Kenya and OIB fields suggests some common source and/or process.
None of the chemical features is ubiquitous. Zr depletion relative
to Ti, noted previously in Kenyan basalts by Davies & Macdonald
(1987)
and Macdonald (1994)
, is mainly developed in rocks of
the mobile belt. Phosphorus enrichment relative to Nd is essentially
restricted to the mobile belt. Rb depletion relative to Ba and
K, and Nb enrichment relative to K and La are common, but not
universal, characteristics. Most, but not all, of the rocks
show Ba enrichment relative to mean OIB. A measure of the complexity
is given by Ce/Pb and P/Nd ratios, which are relatively constant
in OIB (25 ± 5 and 74 ± 13, respectively; Sun
& McDonough, 1989
) but show large ranges in the rift rocks
(5–100 and 20–155). The great diversity in the rift
eruptive rocks indicates that they did not acquire their trace
element features simply from an asthenospheric OIB-like source
(see Macdonald, 1994
). We noted above that, with the exception
of rocks of the mobile belt, isotopic evidence also presents
a more complicated picture than a simple OIB-like source for
the rift magmas, a point made earlier for Tanzanian rocks by
Paslick
et al. (1995)
and for Kenya rift carbonatites by Kalt
et al. (1997)
. It would appear that the mafic rocks contain
a very heterogeneous, lithospheric component. We now assess
the contribution of crustal contamination to the lithospheric
component.
View larger version (38K):
[in this window]
[in a new window]
|
Fig. 12. Zr/Y–Nb/Y plot to show the Zr depletion relative to Nb of many Kenyan eruptive rocks compared with OIB.
|
|
|
CRUSTAL CONTAMINATION
|
|---|
Various lines of major and trace element evidence suggest that
crustal contamination has not contributed significantly to the
compositional spread in the mafic rocks of the Kenya rift: (1)
the majority of specimens (81%) are nepheline-normative, precluding
an extensive crustal fractionation history, a point already
made for Tanzanian rocks by Paslick
et al. (1995)
; (2) 58% of
our samples contain higher abundances of Ba than the average
upper continental crust (500 ppm; Taylor & McLennan, 1985
);
Ba concentrations in the Kenya rocks have been buffered by high
concentrations in the melt derived from the mantle sources and
any evidence of crustal contamination is obscured (see Paslick
et al., 1995
); (3) the spread of data in a La/Yb–Sm/Eu
plot (Fig.
13) precludes significant contamination of OIB-type
magmas with rocks comparable with either average upper or lower
continental crust.
View larger version (28K):
[in this window]
[in a new window]
|
Fig. 13. Sm/Eu–La/Yb plot for rocks of all three structural settings. The distribution is not consistent with significant amounts of mixing of OIB with either average upper (UCC) or lower (LCC) continental crust (Taylor & McLennan, 1985).
|
|
Within the rocks of the craton and craton margin, Sr and Nd isotopic data define two flat-lying trends in which 87Sr/86Sr increases with little variation in 143Nd/144Nd (Fig. 8). In basalts from the Naivasha area in south–central Kenya, this increase in 87Sr/86Sr is associated with an increase in 207Pb/206Pb, a decrease in 206Pb/204Pb and little change in La/Y or Nb/Y, and has been related by Davies & Macdonald (1987) to crustal contamination. Paslick et al. (1995) have also used Pb isotopic data to show that minor crustal contamination may have affected some rocks from Tanzania, such as the hy-normative basalt MD93-7, which has the lowest 206Pb/204Pb ratio (17·632) in our dataset. Samples with high 87Sr/86Sr are not generally associated with strong indicators of crustal interaction. On the contrary, they appear to have the characteristics of small melt fractions from the mantle, in particular low SiO2 contents and Rb/Sr ratios and high Sr contents, and, within the craton samples, high Nb/Y and La/Y. The samples from the reworked craton margin show elevated 87Sr/86Sr with little change in La/Y or Nb/Y and these may well have been influenced by crustal interaction.
In contrast, basalts erupted through the mobile belt show no systematic correlation between isotopic variation and indices of crustal contamination, such as Rb/Sr or Ce/Pb. This apparent lack of crustal interaction may be related to the relative thinness of the crust. Seismic profiles along the rift have revealed that the crust thins from 35 km close to the craton margin to 20 km beneath the Turkana Depression. The attenuated crust is probably related to the numerous extensional events that have affected northern Kenya during the Mesozoic and early Tertiary (Mechie et al., 1997).
We conclude that although some compositional spread in the dataset is undoubtedly related to minor crustal contamination, the effects tend to be swamped by mantle-derived variability.
|
MANTLE HETEROGENEITY
|
|---|
Thus far, we have discussed major and trace element data partly
in terms of average compositions. This approach disguises considerable
compositional variation within groups, particularly in the incompatible
trace elements, taken to include K, Ti and P. Thus, incompatible
trace element abundances commonly have a range, within each
lithological type, of a factor of four at given
mg-number. Class
& Goldstein (1997)
also reported fractionation-corrected,
incompatible trace element concentrations with a range of a
factor of four in alkali basalts from La Grille, Grande Comore.
In the Kenya case, the large ranges result in large overlaps
between magma types. We now point out some important features
of the variability in incompatible trace element abundances,
first with reference to P. Whereas the great majority (>160)
of rocks show a fair TiO
2–P
2O
5 correlation (and TiO
2/P
2O
5 ratio >3), around 20 have lower ratios. On a P/Y–Zr/Y
plot (Fig.
14), these show a trend of steeply increasing P/Y
with little change in Zr/Y. The high-P rocks are restricted
to the reworked craton margin and mobile belt. Although they
are particularly common along Traverse 2 (in successions ranging
in age from Miocene to Recent), they also occur as Miocene lavas
at Loisiumurto in SW Kenya, in Pleistocene basalts at Thiba
(central Kenya) and in young basalts at the Barrier. Thus the
anomalously high-P rocks are found throughout Kenya and in rocks
covering the age span of our dataset.
View larger version (44K):
[in this window]
[in a new window]
|
Fig. 14. P/Y–Zr/Y plot to show the sub-group of Kenyan mafic rocks with relative P enrichment. Field of OIB also shown.
|
|
Relationships between many pairs of incompatible trace elements show similar features to P, namely a majority of specimens showing a strong positive correlation and a smaller number of samples adding wide scatter. The anomalous features do not characterize all rocks in a given volcanic succession; for example, the volcano Silali contains high-P and normal-P basalts of essentially the same age. The ‘outriders’ for one element are not usually those for another, i.e. there is substantial decoupling of the incompatible trace elements, reflected, inter alia, in the ranges of ratios and in different mantle-normalized patterns, even for rocks of the same magma type from the same successions (Fig. 15).
The variability in incompatible trace element ratios is present in all four compositional groups and in rocks with mg-number ranging from 40 to 80. Some of the relative enrichments and depletions must, therefore, be mantle derived and we can envisage one or more of the mantle sources as being strongly heterogeneous on a local scale. As it is unlikely that convecting mantle can show heterogeneity on such a scale, we infer that this highly variable component must be lithospheric.
|
MANTLE END-MEMBERS
|
|---|
On the basis of the trace element and isotopic characteristics
and the strong heterogeneity implied by incompatible trace element
variability, it would appear, therefore, that we must consider
at least a two source component model for the rift magmas, one
of sublithospheric (plume?) origin, as is also required by magma
volumes (Latin
et al., 1993
), and the other originating within
the lithosphere.
The lithospheric component
We consider first the possibility that the variable composition of the lithospheric mantle is a result of metasomatism by either subduction-related fluids or carbonatitic melts.
On the basis of xenolith studies in the Tanzanian craton, Rehkamper et al. (1997) identified a metasomatic component derived from fluids released from a subducting slab. Such a component would have the advantage, in the broader rift context, of explaining the Ba enrichment in many rocks. We test it here using Ce/Pb–La/Nb relationships. Subduction-related fluids carry Pb and La in preference to Ce and Nb, respectively (Keppler, 1996; You et al., 1996). Thus arc rocks tend to have higher La/Nb and lower Ce/Pb than OIB. Using these criteria, Fitton (1995) has interpreted older (>5 Ma) lavas of the Basin and Range Province in the western USA as having a s
TOP
ABSTRACT
INTRODUCTION
40 wt %) and low Na2O/K2O (
, 
, •, rocks from the Nasorut Basalt Formation (specimen numbers 12/1012, 12/384 and 12/980); 
, from the Lokweilabit Basalt (3/190); 
, from the Ribkwo Basalt (5/104); 
, from the Kaparaina Basalt (3/490).