| Journal of Petrology | Pages |
© 1998 Oxford University Press |
|
|
Mafic potassic magmas are generally restricted to continental regions and occur in a variety of tectonic environments. Their source regions are now widely believed to lie within the continental mantle lithosphere (e.g. Fraser et al., 1986; Nelson et al., 1986; Dudas, 1991; Rogers et al., 1992; Gibson et al., 1995) of which they are a complementary sample to that provided by peridotite xenoliths (e.g. Menzies et al., 1987; Jochum et al., 1989). In addition, their trace element and isotopic characteristics can be used to infer the temporal evolution of the mantle lithosphere and the processes involved in lithosphere stabilization. Potassic magmas are also among the most compositionally extreme products of processes that scavenge and fractionate incompatible elements in the upper mantle. Hence their geochemical variations can also be used to investigate the time-integrated effects of these processes on the radiogenic isotope evolution of the mantle lithosphere and to explore links between their lithospheric source regions and those of ocean-island basalt (OIB) (McKenzie & O'Nions, 1983, , 1995; Turner et al., 1996). The Virunga province of the western branch of the African Rift is a classic example of intra-plate potassic magmatism, and the Ugandan part of the province was first described in remarkable detail by
Holmes & Harwood, (1937). They presented both excellent field and petrographic descriptions of mineralogically and texturally diverse lavas and an unusually large number of classical chemical analyses of whole-rock samples. These were used to establish a taxonomic system for potassic lavas, some terms from which are still used today. Subsequent geochemical studies of this province (e.g. Mitchell & Bell, 1976; Vollmer & Norry, 1983a, 1983b) were based largely on the Holmes collection, and it is only recently that additional collections have been made from the Rwandan and Zairean sectors (Aoki et al., 1985; De Mulder, 1985; De Mulder et al., 1986; Demant et al., 1994). These later geochemical studies confirm the strongly incompatible element enriched, but generally mafic, nature of the lavas (e.g. De Mulder, 1985; De Mulder et al., 1986; Rogers et al., 1992) and emphasize their anomalous and sometimes extreme isotopic characteristics (Vollmer & Norry, 1983a, 1983b; Rogers et al., 1992), both of which point to magma sources located in the mantle lithosphere. In their study of the volcano Karisimbi, De Mulder et al., (1986) and
Rogers et al., (1992) presented combined major and trace element andradiogenic isotope analyses of a range of potassic lavas. The present investigation was designed as an eastward continuation of that work to the volcanoes of Muhavura, Gahinga and Sabinyo. These centres lie on the border between Rwanda and Uganda and hence overlap with the area covered by Holmes & Harwood. In this paper we present representative major and trace element and radiogenic isotope analyses of a suite of 60 lavas from these three volcanoes along with some preliminary 40Ar/39Ar dates, and discuss the petrogenesis of the Virunga lavas. We then explore implications for the structure and age of the lithospheric source regions beneath the Virunga and investigate the rates of lava eruption and causes of melt generation. The African Rift (Fig. 1) is arguably the best known example of a magmatic continental rift. It stretches from the Red Sea in the north to its southern extremities in Malawi and can be broadly divided into three major sections: the Ethiopian Rift, the Kenyan or Gregory Rift and the Western Rift. It has long been appreciated that these three rifts are located on plateaux (e.g. Holmes, 1965), the Ethiopian Rift cutting across the Ethiopian Highlands and the Kenyan and Western Rifts across the East African Plateau. The dimensions and circular planform of these plateaux are similar to those of uplifted areas of the oceans associated with ocean islands (e.g. Courtney & White, 1986). The topography broadly correlates with a regional negative Bouguer gravity anomaly, suggesting that the plateaux are dynamically supported by convective activity in the underlying asthenosphere (Ebinger et al., 1989). This conclusion has encouraged a consensus that the plateaux, and hence the African Rift system, are underlain by possibly two mantle plumes (e.g. Hart et al., 1989; Latin et al., 1993; Macdonald, 1994). Figure
The location and regional trend of the rift basins are strongly influenced by basement structure. This is evident on both a regional scale, in that the rifts are located in Proterozoic terrain surrounding the Archaean Tanzanian craton (Fig. 1), and on a local scale, as for example in Kenya where the strike of the rift closely follows the structural grain of the Pan-African basement (Smith & Mosely, 1993). Estimates of the amount of extension associated with the rift basins are generally small, with maximum [beta] values of <1·1 in most of the major rift sections (Ebinger et al., 1989). This value is generally supported by seismic profiling along the Kenyan Rift, which reveals a crustal thickness of ~35 km (KRISP Working Group, 1987; KRISP Working Party, 1991). Many previous studies of the Kenyan rift system have emphasized the control exerted by extensional tectonics on the location of volcanic centres (e.g. Rosendahl, 1987). In particular, they show that volcanoes are frequently associated with high-relief accommodation zones between extensional basins, not within them nor along the main border faults. However, it is equally clear that volcanism can occur at almost any time during rift development; this implies that extension is not the only cause of magmatism. The Western Rift contrasts with the eastern branch in that volcanic rocks are much less abundant, but frequently potassic, and the rift grabens themselves are narrower; hence it may be considered to be at a less mature stage than the eastern branch. Consequently, the Virunga volcanic rocks are of additional interest as an example of the initial products of the magmatic evolution of a continental rift over a mantle plume. The Virunga province is located on the international boundaries of Zaire, Rwanda and Uganda, and sits astride the western branch of the East African Rift system (Fig. 1). The basement geology of northern Rwanda and neighbouring parts of Zaire and Uganda is dominated by the Kibaran mid-Proterozoic mobile belt (1·3-1·4 Ga) with local intrusions of anorogenic granites and syenites, dated at between 0·7 and 1 Ga (Cahen & Snelling, 1984). There are also remnants of late Archaean-early Proterozoic (2·0-2·6 Ga) Ubendian terrain within this complex section. Within the Rift, basins not filled with volcanic rocks contain Neogene to Recent lacustrine deposits of variable thickness (Ebinger, 1989). The Virunga province comprises two active and six extinct volcanoes and numerous smaller volcanic cones. Current volcanic activity is concentrated in the west of the province that overlies the central region of the Western Rift. However, even the most easterly volcano, Muhavura, has a very youthful morphology and appears to be intermediate in age between the currently active volcanoes and the more deeply eroded Sabinyo. This chronology is verified with new 40Ar/39Ar ages on potassic phases from Sabinyo and Muhavura (see below). The petrographic diversity of the Virunga lavas is emphasized by the contrasting volcanic products of the two currently active volcanoes, Nyiragongo and Nyamuragira; a contrast made all the more dramatic by the fact that these volcanoes are separated by <15 km. Nyamuragira is a more typical Virunga volcano erupting K-basanites and evolved derivatives such as K-hawaiites (Aoki et al., 1985). These are feldspar, olivine and clinopyroxene phyric, but frequently with leucite phenocrysts or leucite in the groundmass. Apart from the presence of leucite, they have the appearance of typical alkaline basalts, as found in other continental rift and ocean-island localities. By contrast, the leucite-bearing nephelinites and melilitites of Nyiragongo are much more distinctive in that they are virtually feldspar free, comprising nepheline, leucite and melilite with olivine and clinopyroxene and an abundance of silica-poor accessory phases (Demant et al., 1994). This study focuses on the three most easterly volcanoes, Muhavura, Gahinga and Sabinyo (Fig. 2). They lie along an approximately east-west line, suggesting their positions are fault controlled. Muhavura and Gahinga erupt lavas of the K-basanite series and thus resemble Nyamuragira, Mikeno and Karisimbi to the west. Sabinyo, however, is unique in the Virunga in that its lavas are silica-saturated, orthopyroxene-bearing K-trachytes or latites. Relatively few new analyses are presented in this study because of poor exposure on the flanks of Sabinyo, which are covered by the so-called `boulder beds' (Holmes & Harwood, 1937). However, the samples described were taken from in situ lava flows, and not random blocks from within the boulder beds, and are of significance because they illustrate an important aspect of the interaction between mantle-derived magmas and the continental crust. Figure
The nomenclature of potassic rocks is complex and often confusing. In an attempt to reduce that complexity we have adopted a compositional classification in this study, based on the total alkali-silica diagram (Fig. 3a, after
Cox et al., 1979). Because all Virunga lavas have K2O > Na2O, rock names are prefixed K-basanites, K-hawaiites, etc., similar to the scheme adopted by
De Mulder, (1985) in his original study of Karisimbi. This system has the double benefit of denoting the potassic composition of the lavas and emphasizing their overall similarity to non-potassic alkaline lavas from other localities. The exception to this scheme is in the term latite used to describe the potassic silica-saturated lavas from Sabinyo. Figure
The majority of the lavas from Gahinga and Muhavura are phenocryst rich, containing plagioclase, clinopyroxene, olivine, iron oxide and leucite, and are classified as K-basanites and K-hawaiites. The frequently large phenocrysts (up to 5 mm) are generally subhedral to euhedral, but often fragmented, whereas the groundmass varies from glassy to finely crystalline. Olivine is forsteritic (Fo70+) and clinopyroxenes are titaniferous with up to 3 wt % TiO2. Variations in the modal abundances of plagioclase and olivine phenocrysts give rise to basanitic varieties richer in olivine and mugearitic rocks that are essentially olivine free and plagioclase rich. Leucite varies in abundance and grain size from being virtually absent in the more basanitic rocks to a modal abundance of 10% and a grain size of 1-2 mm in the more fractionated K-hawaiites. Iron oxide is a ubiquitous phase as a micro-phenocryst and apatite is also present in many samples. Xenocrysts of amphibole and mica are also found in some of the K-basanites and K-hawaiites. The latites from Sabinyo are finer grained and less porphyritic than the K-basanite series, with phenocrysts rarely exceeding 2 mm in length. Olivine is absent, its place taken by orthopyroxene. Clinopyroxene is present but it is less titaniferous than that in the K-basanite series. Phenocrysts or xenocrysts of amphibole and mica are also present, but the dominant potassic phase is an alkali feldspar, clearly reflecting the more silica-saturated composition of these lavas. All of the lavas analysed in this study are petrographically fresh, any minor alteration being limited to minor serpentinization of olivine. Xenoliths are rare and no examples of the olivine-biotite-pyroxene (OBP) assemblage, originally recognized by
Holmes & Harwood, (1937), and so central to their petrogenetic arguments, were discovered. The paucity of xenoliths is consistent with the evolved nature of many of the lavas and the presence of high-level magma chambers indicated by the occurrence of leucite. Major and trace elements were determined by a combination of X-ray fluorescence at Edinburgh (Fitton & Dunlop, 1985) and instrumental neutron activation analytical techniques at the Open University (Potts et al., 1985). Sr, Nd and Pb were separated by standard ion-exchange techniques and analysed on Finnigan MAT261 (Sr and Pb) and MAT 262 (Nd) multi-collector mass spectrometers. Repeat analyses of NBS987 gave a mean value of 0·71028 ± 0·00003 (2[sgr]) and our in-house J&M Nd standard a mean value of 0·511865 ± 0·000016 (2[sgr]), both on 10 analyses. Blank levels were <1·5 ng for Sr and <0·4 ng for Nd. Pb was analysed in temperature-controlled runs at 1100°C and ratios were corrected for 1%° per a.m.u. relative to the recommended values for NBS981. Replicate analyses indicate an external precision of ±0·2%° for all three Pb isotope ratios. Pb blanks were <1 ng. Major and trace element analyses of representative samples from the volcanoes Muhavura, Gahinga and Sabinyo are listed in Table 1, and the data are illustrated in a total alkali-silica diagram in Fig. 3a. All of the samples are alkaline and range in composition from K-basalt and K-basanite through K-hawaiite to K-mugearite. More evolved samples from Muhavura and Gahinga include two K-benmoreites, whereas samples from Sabinyo are latites. All analyses are rich in potassium, with K2O/Na2O ratios between one and two (Fig. 3b), and may therefore be described as shoshonitic. The analyses accord generally with the major element analyses presented by
Holmes & Harwood, (1937), but Fig. 3 places these rocks in a more contemporary taxonomic framework.
In more detail, three groups can be distinguished from the alkali-silica diagram. The majority of the samples from Muhavura and all from Gahinga have SiO2 between 45 and 55 wt % and total alkalis in excess of 6 wt %, the abundances of which increase with that of silica. Samples in this group tend to be multiply saturated with olivine, clinopyroxene, plagioclase, magnetite and, in many cases, leucite. Nine samples from Muhavura, by contrast, lie on a negative trend at low silica and are grouped as the Muhavura low-silica lavas. Total alkalis in these samples range from ~3 wt % at 46 wt % SiO2 to >6 wt % at ~43 wt % SiO2. They are also the most magnesian samples in this suite (mg-number ranging from 0·56 to 0·74) and have high Fe2O3 which ranges from 12 to 14 wt %. The third group of samples comprises the five latites from Sabinyo. They represent the most evolved samples in the suite but lie off the main trends defined by the Gahinga and Muhavura lavas, although they retain shoshonitic characteristics. These three groups are discernible on other major element plots. On CaO vs Al2O3 (Fig. 4a), for example, the whole suite broadly defines a negative trend which indicates that fractionation is not controlled by feldspars and/or feldspathoids. However, the Sabinyo latites have lower Al2O3 than rocks with similar CaO contents from Gahinga and Muhavura, suggesting that feldspar was probably involved in their evolution. Fe2O3 and TiO2 (Fig. 4b) are also strongly correlated, reflecting the importance of iron-titanium oxides in controlling the compositions of all rock types. Qualitatively, major element variations indicate that differentiation was dominated by the separation of the dense phases, olivine, clinopyroxene and iron oxide, and that feldspar control is only important in the evolution of the Sabinyo latites. Figure
The compatible trace elements, Ni, Cr, Sc, V, and Co correlate positively with mafic major element indices of fractionation such as MgO, Fe2O3, and CaO (e.g. Fig. 4c) and negatively with SiO2, although the trends against silica are less well defined. These variations are again indicative of fractionation dominated by olivine and pyroxenes. Even in the low-silica group from Muhavura, Ni exceeds 150 ppm only in the olivine-rich sample M78, and all of the samples plot well below the values of primary mantle melts in Fig. 4c, consistent with their moderate to low mg-number. Thus primary mantle-derived magmas are not present, although the strong tendency for most of the analyses to plot below the example fractionation curve requires the parental magmas to have MgO contents >11 wt % MgO. All of the lavas are strongly enriched in the light rare earth elements (LREE), (La/Yb)n varying from 18 to 33 and La abundances from 65 to 132 ppm. The lavas are similarly enriched in other incompatible elements and the convex-upward mantle-normalized abundance curves are broadly similar to those of other intra-plate mafic alkaline rocks (Fig. 5). However, the most incompatible elements, including potassium, are more enriched in the Virunga K-basanites than in most other within-plate basalts, and this enrichment is the source of their potassic characteristics. Small negative Ti anomalies are developed in the fractionated Muhavura and Gahinga lavas, with the largest anomaly occurring in the K-benmoreite sample G40/1. Again, the Sabinyo latites are distinct among Virunga volcanics, with high Th, Pb and Rb relative to Ba and slight depletions of Ta and Nb relative to La. They also have negative Eu anomalies ranging in value from 0·89 in S46 to 0·58 in the most silicic latite, S53, compared with values generally >0·9 in the Muhavura low-silica lavas. Negative Ti anomalies are also characteristic of the latites. Figure
Incompatible element abundances generally increase with fractionation, as revealed by the negative correlation between CaO and Zr (Fig. 4d), although the trend is scattered and the Sabinyo latites have low Zr for their SiO2 contents. However, the overall increase in incompatible element abundances is not a simple function of fractional crystallization. Within the whole suite, Zr/Nb ratios, for example, vary from 2·5 in lavas from Muhavura to >4 in the most silica-rich latites from Sabinyo, with the Gahinga lavas lying between (Fig. 6). Moreover, there is considerable variation even within one volcano, although some groups of samples lie along lines of constant Zr/Nb ratio. Figure
Similarly, La/Yb ratios vary from values as low as 30 to >55, although these variations are systematic within the groups defined on major element and petrographic criteria. Within the low-silica lavas from Muhavura REE fractionation varies from La/Yb = 28 at ~47 wt % SiO2 to La/Yb = 46 in samples with ~43 wt % SiO2 (Fig. 7). Other indices of incompatible element enrichment, such as Nb/Y and Zr/Nb, also vary systematically with silica within this group. This behaviour contrasts with the opposite trend shown by the Sabinyo latites, within which La/Yb increases from 36 to 58 as SiO2 increases from 48 wt % to 62 wt %. The remaining samples do not show any particular trend but have generally lower La/Yb ratios than the Sabinyo latites and intermediate silica contents. Thus the three groups identified on major element grounds are emphasized by incompatible element ratios. Figure
The isotope ratios of Sr, Nd and Pb of selected eastern Virunga lavas are listed in Table 1 and illustrated in conventional isotope diagrams in Figs 8 and 9. All of the lavas analysed plot in the enriched quadrant of a conventional Sr-Nd isotope diagram. The Sabinyo lavas have the most extreme Nd and Sr isotope ratios, with 143Nd/144Nd as low as 0·51204 and 87Sr/86Sr up to 0·7099. The Sabinyo data presented here and those analyses published by Vollmer & Norry, (1983b) define a linear trend back towards the cluster of analyses of Muhavura and Gahinga lavas at values slightly more enriched than Bulk Earth. These latter analyses are in general agreement with previously published values from the Virunga (Vollmer & Norry, 1983b; De Mulder et al., 1986), whereas the isotope ratios of the Muhavura low-silica lavas overlap the field defined by so-called primitive K-basanites from Karisimbi (Rogers et al., 1992). Figure
Figure
Pb isotope variations in the Virunga lavas are less systematic and therefore difficult to interpret. 206Pb/204Pb ratios are restricted, lying in the range 19·3-19·5, but all analyses have unusually high 208Pb/204Pb and 207Pb/204Pb ratios and plot well above the northern hemisphere reference line (NHRL) in conventional Pb isotope diagrams (Hart, 1984). Isotope ratios of the Muhavura low-silica group overlap with the field defined by lavas from Sabinyo whereas Gahinga lavas extend to some of the lowest values of the Eastern Virunga analyses presented here. What is clear, however, is that the Pb isotope ratios of the Sabinyo lavas are distinct from the trend defined by other evolved lavas from the Virunga province, most notably the K-trachytes from Karisimbi. The age and duration of Virunga volcanism have been the subject of debate, with ages ranging from <100 ka to 12 Ma [reviewed by
Cahen & Snelling, (1984)]. It is clear that the morphology of many of the volcanoes is youthful, although the steeply dissected structures of Mikeno and Sabinyo imply older ages. This is certainly the case for Mikeno, where
Guibert et al., (1975) reported 10 K-Ar ages on whole rocks and mineral separates ranging from 0 to 4·7 Ma. Two other minor volcanic occurrences in Zaire, Tongo and Mushebele, have ages of 8·9 and 12·6 Ma, respectively. By contrast, whole-rock K-Ar ages from Karisimbi range from 30 to 120 ka (De Mulder, 1985), consistent with the much younger morphology of that volcano. However, the older ages were derived from the stratigraphically youngest and most contaminated trachytes, and may have been influenced by the contamination process. Five mineral separates, three phlogopites from Sabinyo and two leucites from Muhavura, were analysed by laser ablation 40Ar/39Ar mass spectrometry using techniques described by
Kelley, (1995). The results of the analyses are listed in Table 2. The two leucites from Muhavura give weighted mean ages of 33 ± 9 ka and 52 ± 19 ka and are within error of each other. The two samples come from different altitudes, M87 being from near the volcano summit and M32 from near the base of the cone within the lava plain. Although these are only preliminary results, they confirm the very young age of this volcano and further imply that the cone may have been constructed over a relatively short period of time. Table 2. Argon isotope analyses and ages of mineral samples from the eastern Virunga province
The three ages from the Sabinyo mineral separates are all older, with ages of 64 ± 25 ka, 113 ± 35 ka and 176 ± 30 ka. These are consistent with the more dissected morphology of this volcano and also with the K-Ar age of 137 ka (no errors given) from
Bagdasarayan et al., (1973). The errors on all of these ages are not insignificant and so they should be considered as preliminary results. However, they confirm the relatively short-lived duration of volcanism in the eastern Virunga province and show that, of the major volcanic structures, only Mikeno is older than 200 ka. Therefore, much of the volume of the present volcanoes must have been erupted since that time and, as only Mikeno and the latites of Sabinyo show evidence of reliable ages >100 ka, it seems likely that much of the K-basanite magmatism is <100 ky old. Arguments concerning the petrogenesis of continental volcanic rocks focus on discriminating between crustal contamination and the influence of enriched source regions in the continental mantle lithosphere. Enriched isotope signatures can arguably be derived from both sources but criteria for deciding on their origin must depend on the covariation of radiogenic isotopes with major and trace elements. Previous discussions on the origins of the Virunga lavas have emphasized both crustal and mantle origins, even to the extent of adding crustal material at mantle depths (e.g. Vollmer & Norry, 1983b). These debates are not new and find their origins in the original
Holmes & Harwood, (1937) memoir. More recently, arguments have been proposed that place the source of the K-basanite series within the mantle lithosphere and relate the K enrichment to melt migration into the source region ~1 Gy ago (De Mulder et al., 1986; Davies & Lloyd, 1989; Rogers et al., 1992). However, the origin of the Sabinyo latites remains at issue. It has been shown above that the K-basanite series is geochemically distinct from the latites and hence the latter are inferred to be the products of a different petrogenetic process. Thus, the following discussion focuses on the clearly defined trends shown by the low-silica K-basanites and the Sabinyo latites. Although none of the lavas analysed can be regarded as primary, the covariation of Ni with MgO requires any primary Virunga magma to have had >11 wt % MgO (see Fig. 4 and caption for explanation). Major element trends, particularly the overall decrease in MgO and CaO as fractionation proceeds, are broadly consistent with the fractional crystallization of olivine and clinopyroxene. Similarly, the positive correlation between Fe2O3 and TiO2 indicates a strong control by Fe-Ti oxides, even in the most mafic of the Virunga lavas. This may in turn reflect high oxygen fugacities in the parent magma, a common aspect of potassic lavas from other localities (e.g. Kress & Carmichael, 1991). However, the steep increase in Al2O3 as CaO decreases (Fig. 4) militates against plagioclase fractionation in all but the most evolved lavas. Extrapolation of these trends implies a parental melt composition with MgO >11 wt %, Fe2O3 13-14 wt %, TiO2 3-4 wt %, Al2O3 and CaO ~12 wt %, and with SiO2 ~45 wt %. This is comparable with many basanites, and it is only the excess of K2O over Na2O that distinguishes the Virunga lavas from basanites from elsewhere. The high MgO and low SiO2 require a mantle origin for the K-basanite parent magma, and this conclusion is supported by the experimental results of
Sack et al., (1987). Moreover, the high K2O/Na2O ratio of M78, the most primitive lava in this suite, confirms that potassium enrichment is mantle derived. The subsequent evolution of the K-basanites is not simple. The lack of linear correlations between incompatible elements (e.g. Zr and Nb, Fig. 6) and the range of La/Yb ratios at a given SiO2 content (Fig. 7), both militate against simple fractional crystallization as the only process in the evolution of these lavas. Nevertheless, the volcanological evolution of the Virunga volcanoes involves the intermittent development of lava lakes (Burt et al., 1994) and magma storage in high-level, volcanic or sub-volcanic reservoirs. The presence of leucite testifies to the low-pressure origin of many of the lavas, as leucite is only stable at pressures of <2 kbar (~6 km depth) (Barton & Hamilton, 1978), and it is in these shallow reservoirs that fractional crystallization occurred. It may be that those small groups of analyses that plot along lines of constant Zr/Nb ratios in Fig. 6 are the products of these shallow reservoirs. However, these effects of fractional crystallization were superimposed on variations developed at an earlier stage in the evolution of the magmas that resulted in the isotopic and trace element heterogeneties in the erupted lavas. The contrasting behaviour of trace elements in the Sabinyo latites and the low-silica lavas from Muhavura is clearly illustrated on a plot of Zr/Nb against La/Yb (Fig. 10). The low-silica group defines a negative trend that overlaps and continues the trend defined by the most primitive lavas from Karisimbi, whereas the Sabinyo latites define a positive trend to high values of both La/Yb and Zr/Nb. The increase in La/Yb and Zr/Nb in the latites is accompanied by systematic changes in 87Sr/86Sr (Fig. 11a) and an increase in SiO2. In addition, the latites have low Ce/Pb ratios compared with the K-basanites (Fig. 11b) and the trend to higher Zr/Nb and La/Yb ratios is also accompanied by an increase in the La/Nb and Th/Nb ratios. For example, La/Nb increases from 1·1 to 1·5 in the latites, compared with values between 0·8 and 1·0 in the K-basanite series. Figure
Figure
In Fig. 12 the Sabinyo lavas from this and other studies define an excellent near-linear trend that is consistent with a simple mixing model. Samples with high 87Sr/86Sr ratios have the lowest Sr concentrations (500 ppm) and high silica contents (64 wt %), and the overall trend indicates an end-member with Sr <500 ppm, a 87Sr/86Sr ratio >0·710, and a silica content >64 wt %. Taken with the trace element evidence summarized above, these observations strongly suggest a crustal source for this end-member. Yet, despite its crustal characteristics it must also have a La/Yb ratio >60 to generate the trend in Fig. 11b, in which it is clear that the Sabinyo latites with the highest SiO2 also have the highest La/Yb ratios of the whole Virunga province. By contrast, most crustal melts have much lower La/Yb ratios, values between 20 and 30 being more typical (e.g. Harris & Inger, 1992; McDermott et al., 1996). Moreover, evolved magmas from rift environments (e.g. trachytes, phonolites and comendites) also have La/Yb ratios in the range 20-30 (e.g. Baker, 1987) and so could not produce the trend to high La/Yb seen in the Sabinyo latites. Figure
Melts generated in the deep crust in the presence of residual garnet, by contrast, do have very high and variable La/Yb ratios (40 < La/Yb < 120, e.g. Gilbert & Rogers, 1989; Hawkesworth & Clarke, 1994) while retaining typical crustal Zr/Nb ratios of ~10, high 87Sr/86Sr ratios, high SiO2 and low Sr abundances. Analyses of such deep crustal melts are illustrated in the inset to Fig. 10 and a hypothetical mixing line between a K-basanite and a melt with similar characteristics to the garnet-bearing dacites from the Pyrenees. This example simply demonstrates that the trends defined by the Sabinyo latites are best explained by mixing between a mafic end-member and a silicic melt with the characteristics of a melt from a garnet-bearing crustal source region. Although such melts have not been recorded from any part of the African rift system, the trend to high La/Yb and Zr/Yb is inconsistent with a crustal contaminant with low La/Yb ratios. Projecting the linear trend in Fig. 12 to an Sr content of 200 ppm (1000/Sr = 5), representative of many crustal melts, gives a 87Sr/86Sr ratio for the crustal end-member of 0·720. Assuming this value and an Nd content of 25 ppm (Sr/Nd = 8), the Sabinyo latite trend in Fig. 13 can then be used to estimate the 143Nd/144Nd ratio of the crustal end-member to a value close to 0·5114. Assuming a crustal 147Sm/144Nd ratio of 0·132, this ratio corresponds to an NdCHUR model age of ~2 Ga, broadly consistent with the age of the local Ubendian crustal remnants. Figure
Inferring the composition of the second, mafic end-member is more problematic. The trend in Fig. 12 does not project back towards the composition of the Muhavura low-silica K-basanites or the K-basanites from Karisimbi. These latter samples have similar Sr concentrations to the latites but lower 87Sr/86Sr ratios. Instead, the Sr isotope trend points to a component that is more Sr rich, and it was originally suggested that the low 87Sr/86Sr end-member was represented by the Nyiragongo nephelinites (Rogers et al., 1992). However, in other diagrams, especially Figs 10 and 11b, the Sabinyo trend projects towards the bulk of the K-basanites, and in Fig. 12 it is anchored in the cluster of points defined by the more evolved K-basanites and K-hawaiites from Muhavura and Gahinga. The combination of these observations precludes the involvement of a nephelinitic melt. Rather, it strongly suggests that the mafic end-member in the Sabinyo magmatic system was a fractionated K-basanite or K-hawaiite with a slightly elevated 87Sr/86Sr ratio. Thus, we conclude that the Sabinyo latites are predominantly silicic melts from the deep crust that have interacted with a more mafic potassic basanite melt at depth. The Sabinyo latites are not the only representatives of contaminated rocks in the Virunga province; De Mulder et al., (1986) recognized a trend to high 87Sr/86Sr and low 143Nd/144Nd in mugearitic and trachytic rocks from Karisimbi, which they modelled as being the product of combined assimilation and fractional crystallization (AFC) in a sub-volcanic magma chamber. The contaminant was regarded as being derived from the local upper crust, and the close association of the trachytes with a caldera collapse phase, together with the trend to high 206Pb/204Pb ratios in the same rocks noted by
Rogers et al., (1992), supported this conclusion. Analyses of Pb isotopes from this study and previous work (Vollmer & Norry, 1983b) show that the trend defined by the Sabinyo latites is isotopically distinct from that of the K-trachytes in Karisimbi (Fig. 9). The higher 206Pb/204Pb ratios in the K-trachytes are not seen in the Sabinyo latites, which plot on much steeper trends in both Pb isotope diagrams, suggesting that the two contaminants are isotopically distinct. This, however, is not an unexpected result if, as indicated by the trace elements, the Sabinyo contaminant is derived from the lower crust. Although the striking linear trend defined by the Sabinyo latites in Fig. 12 is consistent with simple mixing between two melts, it could also be the result of assimilation combined with fractional crystallization. AFC trends define curved trajectories on this diagram for low values of r, where r is the ratio of assimilation to fractionation as defined by
DePaolo, (1980). The strong linear trend defined by the data, therefore, constrains any AFC process to high values of r and so indistinguishable from mixing. By contrast, the calculated value of r in the contamination of the Karisimbi trachytes is low (De Mulder et al., 1986). Low values of r are consistent with the large amount of heat required to drive contamination in a shallow crustal environment, whereas high values of r are predicted for assimilation of hotter, deeper crustal rocks (DePaolo, 1980). Thus, the trace element variations and the implied ratios of assimilation to fractionation are both consistent with a deeper (mid to lower) crustal origin for the Sabinyo latites. Within the Muhavura low-silica lavas, incompatible element ratios vary significantly. For example, as Zr/Nb increases from 2·6 to 3·5 so La/Yb decreases from 49 to 28. This negative trend is consistent with varying melt fractions of a garnet-bearing mantle or mixing between different mantle melt fractions. However, isotope ratios also change systematically with trace elements within this group and so trace element variations are more likely to reflect mixing than melting. On the Sr mixing diagram (Fig. 12) the low-silica lavas define a positive array within the field defined by the Karisimbi K-basanites, but at lower 87Sr/86Sr ratios than the Sabinyo latites, projecting back towards the Nyiragongo nephelinites. Significantly, the low-silica K-basanites plot below the Sabinyo trend on this diagram, from which it is concluded that the K-basanites cannot be the product of mixing between nephelinitic magma and crustally derived silicic melts, as originally proposed by
Holmes & Harwood, (1937). Such a model is also precluded by the mafic composition of the K-basanites. Hence, the low silica K-basanites are interpreted to be mixtures of a silica-undersaturated nephelinite and a primitive potassic basanite or potassic basalt (see
Rogers et al., 1992). All of the samples in this group have low La/Nb ratios (<0·9) and high Nb/Th and Ce/Pb ratios (>20) (Fig. 11) and are magnesian (MgO >7 wt %), indicating a strong mantle influence. Finally, these characteristics are accompanied by slightly elevated 87Sr/86Sr and low 143Nd/144Nd ratios relative to Bulk Earth, supporting the conclusion from earlier work (Rogers et al., 1992) that enriched isotope ratios are a feature of the Virunga source region. Mixing between the most isotopically extreme low-silica K-basanite and Nyiragongo nephelinite (Table 3) for both Sr and Nd isotope ratios is illustrated in Fig. 13. The curve shown matches the relatively flat-lying trend of the Muhavura low-silica lavas (Fig. 13) although it does not readily explain some of the lower 87Sr/86Sr ratios in the Karisimbi K-basanites. Nephelinite involvement is limited to 20-30%, reflecting the highly incompatible element-enriched composition of that end-member. Table 3. Mixing parameters used in calculations illustrated on Figs 10 and 13
The source of the dominant component in the K-basanites has 87Sr/86Sr ~0·707 and 143Nd/144Nd ~0·51243, similar to values from previous studies (Davies & Lloyd, 1989; Rogers et al., 1992), and is inferred to be located within the mantle lithosphere. Model age calculations, based on a depleted mantle composition, give an age of 900 Ma if the Sm/Nd ratio of the melt is used or 1100 Ma if a higher Sm/Nd ratio for the source is assumed. These ages are similar to those of the most recent episodes of post-orogenic granite magmatism in the basement (Cahen & Snelling, 1984). The nephelinitic component is isotopically less extreme, but the source of this magma may also be located within the mantle lithosphere. The Nyiragongo nephelinites are highly enriched in incompatible elements and have strongly fractionated REE profiles. Vollmer & Norry (1983a) and
Vollmer et al., (1985) also cited extreme Pb isotope ratios from Nyiragongo that define an ~500 Ma age, which they interpreted as the age of source enrichment. Although this age is younger than that of the K-basanite source, it is consistent with the higher 143Nd/144Nd and lower 87Sr/86Sr ratios that characterize the Nyiragongo lavas. Hence, we infer that the Nyiragongo source is distinct from the K-basanite source but is also located within the mantle lithosphere. The depths from which all magmas in the Western Rift have been derived have been discussed at length by
Furman, (1994). In general, the Virunga K-basanites were inferred to have been derived from similar depths to the Rungwe basanites and the Kivu basalts, whereas the potassic lavas from Katwe, and the nephelinites from Rungwe and Nyiragongo showed a greater involvement of garnet in their source regions. Evidence for this can be derived from the major elements (CaO/Al2O3 ratios), REE fractionation [heavy REE (HREE) depletion in the deeper melts] and from experimental studies (Edgar et al., 1976; Sack et al., 1987). These relative depth constraints can be further refined by considering the minor phases involved in the generation of the different magmas. For example, the development of such silica-undersaturated lavas as the Nyiragongo nephelinites requires the presence of carbonate in the magma source region. Carbonate is only stable within the mantle at depths greater than ~100 km (e.g. Olafsson & Eggler, 1983), consistent with the requirement for residual garnet in the magma source region. By contrast, the K-basanite-basalt end-member of the Muhavura lavas has a silica content of 45-47% and lavas contain xenocrysts of amphibole. Such levels of silica do not require a carbonated source region and if the amphibole xenocrysts are source related this would imply an origin from depths shallower than ~90 km (~30 kbar). These two source regions have distinct isotopic characteristics which can be related to their ages. The ~1 Ga model age calculated for the source of the Muhavura potassic basanites and the primitive K-basanites from Karisimbi (Rogers et al., 1992) is similar to that determined by
Davies & Lloyd, (1989) for the lavas from the Katwe-Kikorongo field to the north of the Virunga. The Nyiragongo source region, by contrast, has a younger age, estimated to be ~500 Ma from extreme Pb isotope variations in one unusual sample. However, all Nyiragongo lavas have lower Sr and higher Nd isotope ratios than any of the other Virunga lavas, consistent with a younger source age. Additional age constraints were derived from xenocrystic clinopyroxenes in the Katwe-Kikorongo lavas and the Pb isotope systematics of the Karisimbi primitive K-basanites. The clinopyroxene xenocrysts define a 1·8 Ga secondary Pb isochron (Davies & Lloyd, 1989) and, whereas the linear trend was interpreted to reflect mixing between the host magma and lithospheric material, it is clear that the xenocrysts incorporate a significantly older component. Moreover, the xenocrystic nature of the pyroxenes implies that they were derived from depths similar to or shallower than the magma source region. In the Karisimbi primitive K-basanites, Rogers et al., (1992) also identified evidence for an early Proterozoic or even Archaean age in the high 207Pb/204Pb ratios of the Virunga lavas. These high 207Pb/204Pb ratios are not the result of crustal contamination, as lavas with these characteristics still retain Ce/Pb ratios >20. These depth and age constraints can be tentatively assembled to give a picture of the vertical age structure of the lithosphere beneath the Virunga (Fig. 14). If this structure is correct, it implies a downward growth of the lithosphere through time, in part resulting from conductive cooling of a thermal boundary layer. As layers of mantle thermally accrete to the base of the lithosphere, they are able to trap melts percolating up from the asthenosphere and stabilize them in the form of veins and metasomatic mineral phases, as previously described by numerous studies of mantle xenoliths (e.g. Kempton, 1987). The trace element abundances of the K-basanites show little fractionation between the large ion lithophile elements (LILE) and the high field strength elements (HFSE), characteristics that are typical of mantle-derived silicate melts. A process of melt enrichment in the late Proterozoic is therefore preferred over fluid addition from, for example, a subducting slab. The ages recorded by the magmatic rocks and xenocrysts are therefore ages of enrichment and must be regarded as minimum ages of lithosphere stabilization. Nevertheless, the data still imply that, as the deeper layers record the youngest enrichment ages, they stabilized most recently and may have served to shield shallower levels from the effects of subsequent melt infiltration. Figure
The rate of eruption, and possibly melt generation, in the Virunga province can be assessed from the total volume of the province and the length of time over which it has been active. The 40Ar/39Ar dates presented in this study show that the oldest Sabinyo lavas are between60 and 180 ka and that most of the K-basanite andnephelinitic magmatism appears to be <= 100 ka in age, notwithstanding the evidence for older ages from Mikeno and other minor localities back to possibly 12 Ma. Volume estimates have been made of two of the volcanoes. Pouclet, (1976) estimated the volume of Nyiragongo at 500 km3, and
De Mulder, (1985) came to a similar figure for the volume of Karisimbi. Of the remaining volcanoes, Nyamuragira and Muhavura-Gahinga combined are similar in their dimensions to Nyiragongo (500 km3 each), whereas smaller Sabinyo and Visoke, little more than a satellite to Karisimbi, probably amount to <500 km3 in total. Mikeno is the least known and most eroded structure but may be assumed to have originally had a similar volume. However, the age information from Mikeno suggests that it is considerably older than the rest and may be excluded from an analysis of eruption rates over the most recent history of the Virunga. Thus the total volume is ~3000 km3, although the amount erupted in the last 100 ky is of the order of 2000 km3. Assuming a volume of 2000 km3, this leads to an eruption rate of ~0·02 km3/yr integrated over the last 100 ky. Eruption rates have been independently determined for Nyamuragira from the historic eruptive record (Burt et al., 1994) and these reveal a rapid increase in productivity from 0·55 m3/s to 1·37 m3/s in (1980). These values are equivalent to 0·017 km3/yr and 0·043 km3/yr, respectively, and the first of these agrees remarkably well with the estimate of the eruption rate for the whole province over the past 100 ky. Although this result may suggest that the most recent phase of activity at Nyamuragira is unusually productive, the overall similarity of the long and short time-scale productivities implies that most of the K-basanites and nephelinite-related volcanism in the Virunga have been erupted in the most recent 100 ky. Estimating the melt generation rate in the mantle source region of the Virunga is more complex in that it requires an estimate of the amounts of fractionation and contamination that have taken place. If, however, we assume 50% of the melt is trapped at depth as cumulates and intrusives (Cox, 1993), then the melt production rate may be as high as 0·04 km3/yr. This figure is probably subject to considerable error, as it is based on some poorly constrained volume estimates and incomplete geochronological information, and ignores any possible effects from crustal contamination. Despite these reservations, the value of 0.04 km3/yr is comparable with melt generation rates derived from smaller ocean islands that are located on thick oceanic lithosphere, such as the Cape Verdes (0·03 km3/yr) (White, 1993), and it is similar to estimated magma production rates in the Gregory Rift over the past 30 My (0·03 km3/yr, Latin et al., 1993). However, it is considerably less than that of Hawaii (0·16 km3/yr) and other large ocean islands. Although the Virunga is clearly associated with the Western Rift, extension across the latter is relatively minor with estimated maximum [beta] factors of 1·1 (Ebinger, 1989). Moreover, extension began at ~12 Ma, and although there is evidence of magmatism in the Virunga about this time, the most productive phase has been restricted to the most recent 100 ky. Thus, it is difficult to demonstrate a causative link between extensional tectonics and magmatism. The association of magmatism with the rift may be more probably the result of the coincidence of easily fusible source regions within the lithosphere and extensional tectonics providing easy routes to the surface. In this respect, the location of magmatism is controlled by tectonics, but its occurrence is controlled by the presence of suitable source material at depth (see
Furman, 1994). Furthermore, given the volcanic products of both the Virunga and the other Western rift provinces (Rungwe, Furman, 1994; Katwe-Kikorongo, Davies & Lloyd, 1989) are dominated by lithospheric source regions, the most probable cause of melting involves heating of the lithosphere by the subjacent East African mantle plume. Conductive heating and melting of the lithosphere at the volatile-enriched mantle solidus has been modeled by
Turner et al., (1996). The amount of melt produced, represented by the melt thickness, depends on the duration of heating, the potential temperature of the underlying mantle plume (Tp) and the thickness of the lithosphere. A melt thickness of 1-2 km can be calculated for the Virunga province from the above total volume estimates integrated over the outcrop area of the whole province (~3500 km2). If it is assumed that East Africa arrived over the plume during the early Miocene, then the lithosphere has been heated by the plume for between 10 and 15 My. Applying these values to the model of conductive heating described by
Turner et al., (1996) and assuming a lithosphere thickness of 150 km (Fig. 15) gives a maximum Tp for the subjacent plume of 1430°C. The calculated melt thickness, which assumes 50% melt trapped at depth, is almost certainly an overestimate, and so 1430°C must be regarded as a maximum temperature for the underlying plume. The temperature derived from Fig. 15 is considerably lower than that often assumed for mantle plumes (Tp > 1500°C), but it is consistent with the location of the Virunga province at the periphery of the East African plateau, the present-day topographic and geophysical expression of the East African plume (Fig. 1). Although the presence of small volumes of magmatism in and around the Virunga province back to 12 Ma suggests that contact with the plume occurred between 10 and 15 My ago, if the duration of conductive heating has been <10 My, then a melt thickness of 2 km implies higher plume temperatures. However, the duration of heating would have to be less than ~7 My to necessitate plume temperatures >1500°C. Figure
In comparison with models of melt generation for the Kenyan Rift, which
Latin et al., (1993) have related to the decompression of the East African plume as a consequence of extension across the Kenyan Rift, it must be emphasized that the duration of magmatism in the Virunga province is shorter, its volume much smaller, and the time of onset is generally later than that of the Kenyan Rift. However, as the relative contributions from lithospheric and sub-lithosphere sources to Kenyan basalts have yet to be determined, the similarity between melt generation rates between the Virunga province and the Gregory Rift must be regarded as being largely fortuitous. Finally, the rates of melt generation in both the Virunga province and the Kenyan Rift are at least an order of magnitude less than those of flood basalt provinces such as the Paraná (~0·4 km3/yr, Stewart et al., 1996) and almost two orders of magnitude less than for the Deccan (Courtillot et al., 1988; Duncan & Pyle, 1988). Although this observation is not unexpected for the Virunga, it is surprising that the Kenyan Rift has such a low production rate, considering the [beta] factor of 2·5 proposed by
Latin et al., (1993) against that inferred for the Paraná (1·5). It may be that amounts of extension across the Kenyan Rift are much less than estimated and/or the underlying plume is significantly cooler than that involved in the generation of continental flood basalt provinces. The potassic lavas from the Eastern Virunga province represent a suite of K-basanites and more evolved derivatives that reflect processes operating in the mantle and in shallower crustal magma chambers. Systematic variations in trace element ratios with radiogenic isotopes and major elements reveal how a strongly silica-undersaturated magma interacts with the mantle lithosphere to produce the range of K-basanite magmas, which in turn interact with melts derived from the middle or lower crust to produce silicic latites. Qualitative estimates of the relative depths of the source regions of the different magma types, together with their isotopic characteristics, provide further insights into the depth structure of the lithosphere beneath the Virunga province, and suggest that the mantle lithosphere has grown in thickness through time. The Nyiragongo nephelinites represent melts derived from the deepest parts of the lithosphere, which must be >100 km thick. The older source of the K-basanites lies at a shallower depth and is isotopically much more complex, with indications of early Proterozoic or even Archaean ages. The deepest sections of the lithosphere are therefore the youngest and this structure is consistent with the downward growth of the deeper parts of the lithosphere as a thermal boundary layer. Only the carbonated source of the melilitites has any similarity to OIB, and so models that invoke the convective removal of the lithosphere to account for the unusual isotopic characteristics of OIB may be restricted to material derived from this lowermost layer. The Sabinyo latites represent mixtures between a K-basanitic magma and a silicic melt derived from a garnet-bearing source in the deep crust. This melt is distinct from the contaminant seen in the trachytic lavas from Karisimbi. In the latter case, contamination from upper-crustal material was associated with fractional crystallization, whereas the Sabinyo latites are very close to binary mixtures between two end-members. Melt production rates, derived from an estimate of the volume of the whole province and preliminary 40Ar/39Ar ages, are low and comparable with those for small ocean islands. This low production rate, and the lack of a significant plume contribution to magmatism, is consistent with both the thick lithosphere inferred from the depth of magma sources and the remoteness of the Virunga province from the axis of the putative East African mantle plume. The control exerted by the lithosphere on the location of both extension and magmatism and on the composition of magmatism in the early stages of continental extension is dominant, and the timing of magmatism is consistent with a model of conductive heating of the lithosphere by the cooler peripheral parts of the East African mantle plume. We would like to thank Sandro Conticelli, Rick Wendlandt and Ray Kent for constructive reviews, and Simon Turner for comments on an earlier version of this paper. As always, we have also benefited from discussions with many colleagues, including David Graham, Chris Hawkesworth, Dave Peate and Cindy Ebinger. Mabs Gilmour and Peter van Calsteren are also thanked for maintaining the radiogenic isotope laboratory at the Open University. This research was partly funded by the NERC.INTRODUCTION
REGIONAL OVERVIEW AND BACKGROUND GEOLOGY
NOMENCLATURE
PETROGRAPHY
ANALYTICAL TECHNIQUES
RESULTS
Major and compatible trace elements
Table 1. Major and trace element and radiogenic isotope ratio analyses of selected rocks from the eastern Virunga
Incompatible trace elements
Radiogenic isotopes
The age of Virunga volcanism and preliminary 40Ar/39Ar geochronology
Discussion
Magmatic differentiation
Sabinyo latites: examples of lower-crustal contamination
Low-silica K-basanites: magma mixing and mantle lithosphere source regions
Melt generation rates and mechanisms
CONCLUSIONS
ACKNOWLEDGEMENTS
REFERENCES
