Journal of Petrology Advance Access originally published online on September 6, 2005
Journal of Petrology 2006 47(1):145-189; doi:10.1093/petrology/egi071
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Sampling the Cape Verde Mantle Plume: Evolution of Melt Compositions on Santo Antão, Cape Verde Islands
1 GEOLOGICAL INSTITUTE, UNIVERSITY OF COPENHAGEN, GEOCENTER COPENHAGEN, DK-1350 COPENHAGEN, DENMARK
2 DEPARTMENT OF EARTH SCIENCES, AARHUS UNIVERSITY, DK-8000 ÅRHUS C, DENMARK
RECEIVED OCTOBER 10, 2003; ACCEPTED JULY 11, 2005
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONTHE GEOLOGICAL SETTING OF...PETROGRAPHY OF THE VOLCANIC...ANALYTICAL METHODSWHOLE-ROCK GEOCHEMISTRYDISCUSSIONSUMMARY OF CONCLUSIONSSUPPLEMENTARY DATAREFERENCES |
|---|
The volcanic history of Santo Antão, NW Cape Verde Islands, includes the eruption of basanite–phonolite series magmas between 7·5 and 0·3 Ma and (melilite) nephelinite–phonolite series magmas from 0·7 to 0·1 Ma. The most primitive volcanic rocks are olivine ± clinopyroxene-phyric, whereas the more evolved rocks have phenocrysts of clinopyroxene ±Fe–Ti oxide ± kaersutite ± haüyne ± titanite ± sanidine; plagioclase occurs in some intermediate rocks. The analysed samples span a range of 19–0·03% MgO; the most primitive have 37–46% SiO2, 2·5–7% TiO2 and are enriched 50–200 x primitive mantle in highly incompatible elements; the basanitic series is less enriched than the nephelinitic series. Geochemical trends in each series can be modelled by fractional crystallization of phenocryst assemblages from basanitic and nephelinitic parental magmas. There is little evidence for mineral–melt disequilibrium, and thus magma mixing is not of major importance in controlling bulk-rock compositions. Mantle melting processes are modelled using fractionation-corrected magma compositions; the models suggest 1–4% partial melting of a heterogeneous mantle peridotite source at depths of 90–125 km. Incompatible element enrichment among the most primitive magma types is typical of HIMU OIB. The Sr, Nd and Pb isotopic compositions of the Santo Antão volcanic sequence and geochemical character change systematically with time. The older volcanic rocks (7·5–2 Ma) vary between two main mantle source components, one of which is a young HIMU type with 206Pb/204Pb = 19·88,
7/4 = –5, 8/4 
0, 87Sr/86Sr = 0·7033 and 143Nd/144Nd = 0·51288, whereas the other has somewhat less radiogenic Sr and Pb and more radiogenic Nd. The intermediate age volcanic rocks (2–0·3 Ma) show a change of sources to two-component mixing between a carbonatite-related young HIMU-type source (206Pb/204Pb = 19·93, 7/4 = –5, 8/4 = –38, 87Sr/86Sr = 0·70304) and a DM-like source. A more incompatible element-enriched component with 7/4 > 0 (old HIMU type) is prominent in the young volcanic rocks (0·3–0·1 Ma). The EM1 component that is important in the southern Cape Verde Islands appears to have played no role in the petrogenesis of the Santo Antão magmas. The primary magmas are argued to be derived by partial melting in the Cape Verde mantle plume; temporal changes in composition are suggested to reflect layering in the plume conduit. KEY WORDS: radiogenic isotopes; geochemistry; mantle melting; Cape Verde
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONTHE GEOLOGICAL SETTING OF...PETROGRAPHY OF THE VOLCANIC...ANALYTICAL METHODSWHOLE-ROCK GEOCHEMISTRYDISCUSSIONSUMMARY OF CONCLUSIONSSUPPLEMENTARY DATAREFERENCES |
|---|
It is widely recognized that the mantle source of hotspot volcanism is complex, and that the volcanism is caused by decompression melting in rising mantle plumes. A number of workers have proposed that plumes may be zoned or change composition with time, e.g. Hawaii (Kurz et al., 1995
; Hauri et al., 1996; Lassiter et al., 1996; Abouchami et al., 2000; DePaolo et al., 2001; Mukhopadhyay et al., 2003), Iceland (Hards et al., 1995; Fitton et al., 1997; Chauvel & Hémond, 2000; Holm et al., 2000; Kempton et al., 2000) and the Galapagos (Hoernle et al., 2000). Understanding the geochemical heterogeneities of intraplate volcanism is fundamental to the understanding of mantle convection.
The extreme range of isotopic variation in the most primitive mafic igneous rocks within the Cape Verde islands (e.g. Gerlach et al., 1988) indicates that this hotspot may be useful further to constrain the mantle processes which lead to hotspot volcanism and the nature of mantle plumes. The Cape Verde hotspot (Fig. 1) is virtually stationary relative to the lithosphere, as it is situated close to the rotational pole of the slowly moving African plate (Pollitz, 1991). This setting allows the study of interactions between the plume and the overlying lithosphere without the complication of plate movement. Igneous activity related to the Cape Verde hotspot began at least 19 Myr ago (Duncan & Jackson, 1977; Mitchell et al., 1983) and, therefore, it should be possible to evaluate the changing nature of plume–lithosphere interactions and to test competing hypotheses of whether the parental magmas are derived from the convecting mantle (Gerlach et al., 1988; Davies et al., 1989; Hart et al., 1992; Hanan & Graham, 1996; Hofmann, 1997; Thirlwall, 1997) or the lithospheric mantle (Halliday et al., 1995; Chauvel et al., 1997; Class & Goldstein, 1997). Furthermore, the mantle source volumes for small degrees of melting under old, thick oceanic lithosphere may, potentially, be minor and adequate for high-resolution modelling of mantle composition. The complexity of the Cape Verde mantle plume may, therefore, be tested. Previous studies of the volcanic rocks of the Cape Verde archipelago have revealed a strong contrast between the southern islands, characterized by an EM1-type mantle source component, and the northern islands, which have greater affinity to HIMU (Gerlach et al., 1988; Davies et al., 1989; Christensen et al., 2001; Jørgensen & Holm, 2002). The presence of carbonatites on five of the 10 major islands is a unique feature of Cape Verde, suggesting that the mantle source is CO2-rich.
|
To constrain magma generation processes in the Cape Verde archipelago, we have investigated the westernmost of the northern islands, Santo Antão; this island has not previously been studied in detail (Bebiano, 1932
; Gerlach et al., 1988; Christensen et al., 2001). We present volcanological, age and petrographical information as well as new major and trace element analyses and Sr–Nd–Pb isotope data (329 samples have been studied). These data are used to characterize the igneous evolution of the island and to provide a basis for modelling plume–lithosphere interactions. Systematic changes with time in the most primitive magma compositions on Santo Antão indicate that melt generation took place in a region of geochemically heterogeneous convecting mantle and that the lithospheric contribution to the melts was insignificant. We propose that changes in the composition of the most primitive magmas with time are related to layering in the plume and that a succession of layers yielded melts as they rose under the island. Lateral variations in the plume may explain inter-island compositional variations between the northern and southern chains of the Cape Verde archipelago, and perhaps even on the scale of individual neighbouring islands. |
THE GEOLOGICAL SETTING OF THE CAPE VERDE ISLANDS |
|---|
|
TOPABSTRACTINTRODUCTIONTHE GEOLOGICAL SETTING OF...PETROGRAPHY OF THE VOLCANIC...ANALYTICAL METHODSWHOLE-ROCK GEOCHEMISTRYDISCUSSIONSUMMARY OF CONCLUSIONSSUPPLEMENTARY DATAREFERENCES |
|---|
The Cape Verde Islands (4033 km2) consist of 10 major and several smaller islands located 500–800 km west of Africa (Fig. 1a). They are situated SW of the Cape Verde Plateau and 200 km from the centre of the Cape Verde Rise (McNutt, 1988
), which covers an area >3 x 105 km2 and is considered to represent a hotspot swell (Crough, 1978). Jurassic–Cretaceous sea-floor constitutes the basement for the islands (Hayes & Rabinowitz, 1975) and was intruded in early Miocene times (19 Ma) by tholeiitic sills, which are suggested to mark the initiation of Cape Verde magmatism (Duncan & Jackson, 1977; Natland, 1977). The N–S oriented Sal–Maio ridge is bounded to the east by large-displacement, N–S-trending, normal faults (Dash et al., 1976). The sea-floor has very steep gradients to depths of several kilometres north of Santiago and east of São Nicolau (Fig. 1). This may be related to lithospheric fracture zones, indicating that eruption sites for mantle melts are partly governed by deep faults. The Jurassic ocean floor is exposed on Maio and Santiago (De Paepe et al., 1974; Gerlach et al., 1988). Submarine Neogene lavas are found up to 400 m above the present coastline on several islands (Serralheiro, 1976; our unpublished data). These features demonstrate large local vertical movements in the lithosphere. Prior to initiation of volcanic activity, the sea-floor was covered by sediments (silt, clay and shale) which decreased in thickness towards the west from >1000 to <500 m. From early Miocene times, sediments, volcanic ash and mass wasting products from the islands were deposited around the islands in a flexural moat (Ali et al., 2003).
Age determinations on several islands (Mitchell et al., 1983; Plesner et al., 2002; Holm et al., unpublished data) suggest that most of the volcanic activity took place from 16 Ma until the present. Fogo may be entirely Quaternary and Maio relatively old (
12–7 Ma), whereas Santo Antão (7·5–0·1 Ma), São Vicente (7–0·3 Ma), Santiago (6–0·3 Ma) and Sal (16–1 Ma) developed over extended periods. The greatests amount of young (
0·5 Ma) volcanism is present on the western islands of Santo Antão and Fogo, as is evident from their immature morphology and age determinations. With recent (0·1–1 Ma) volcanism on several islands, and the oldest dated rock on Santo Antão only surpassed by rocks on Sal and Maio, there is not a simple age progression related to the geographical position of the islands.
The Neogene volcanic rocks consist of a large spectrum of eruption products with lavas being by far the most abundant. All the volcanic rocks are silica-undersaturated with basanites–tephrites and nephelinites representing the dominant lithologies. Phonolitic rocks constitute a few per cent of the total volcanic section. Small carbonatite intrusions are present on São Vicente, Maio, Santiago, Fogo and Brava.
Santo Antão
The geology of the 770 km2 island of Santo Antão (Fig. 2) was investigated by Bebiano (1932), who identified a wide variety of rock types. The island is almost entirely of subaerial volcanic origin, dominantly built up of large shields of gently seaward-dipping lavas, stratovolcanoes with numerous lateral vents, and hundreds of monogenetic scoria-cones. Strombolian to Hawaiian eruption products are typical and created the basic morphology of the island. Volcanoclastic fluviatile sediments and lahars occur locally, and are widespread in the Chã de Morte depression and SE towards the coast. Very minor amounts of aeolian sediments occur in the north. Most of the central part of the island is more than 1300 m asl. This highland is dissected, mainly to the N and NE, by deep gorges and steep valleys. The southern part of the island is covered by young volcanic products and is less deeply dissected.
|
The evolution of Santo Antão can be divided into three phases. We refer to these as the Older, Intermediate and Young Volcanic rocks (Fig. 2). In addition to field relations and age determinations, part of the grouping of the younger rocks has been based on geochemistry.
Older Volcanic rocks
The Older Volcanic rocks (OV) make up large parts of the island and constitute the bulk of its volume. Ages of 7·5 and 5·2 Ma have been obtained, but most of the succession is 3–2 Ma old (Plesner et al., 2002; Holm et al., unpublished data). A dyke swarm running SW–NE is exposed in the central Chã de Morte area and along the northern Ribeira Grande valley and represents part of the feeder system for the OV. Rocks older than 7·5 Ma probably exist below the present level of erosion and probably make up much of the submarine pedestal of the island. According to our calculations, the unexposed subaerial part and its submarine pedestal may represent around 50% of the total erupted mass for Santo Antão. The OV partly consist of a large volcanic complex centered near Chã de Morte and extending to the Escabecada, Tarrafal and Agua Nova valleys, where the lava successions have been sampled. The rest of the OV was sampled in the Chã de Igreja valley and at Fontainhas (Fig. 2).
Intermediate Volcanic rocks
This volcanic phase is divided into five groups. The volcanic complex in the Cova–Corda–Paul area (1·4–0·7 Ma) constitutes the COVA group. In the western part of the Chã de Morte depression, an edifice was built on the eroded Chã de Morte volcano; this comprises the Bordeira (BOR) group (2–1 Ma). A new fissure system opened parallel to, and SE of, the earlier (OV) dyke swarm, feeding the Cadela (CAD) volcanic rocks (0·7–0·3 Ma) that mainly erupted in the central highlands. The Sinagoa volcano also belongs to this group and was probably situated at the NE end of the fissure system at this time. In the west, the Agua Nova (AN) volcanic rocks were erupted between 0·5 and 0·3 Ma, at about the same time that the Monte Frado (FRA) volcano erupted further to the south (0·5–0·4 Ma).
Young Volcanic rocks
Volcanism younger than about 0·4 Ma is divided into four groups: YTAR, PRC, COROA and LAGOA. The volcanic rocks in the Tarrafal area and to the NE constitute the YTAR (0·4–0·2 Ma) (young Tarrafal) group. In the western Tope de Coroa area, magmas were erupted through numerous aligned sets of SW-NE-trending fissures. The proto Coroa (PRC) group outcrops around Tope de Coroa (0·4–0·2 Ma), and is partly covered by lavas of the central volcanic COROA complex (0·20–0·17 Ma). The other main site for young volcanism is the LAGOA fissure system (0·4–0·10 Ma) in the central part of the island forming several aligned sets of scoria cones over the most recent fissures in the central highlands. Lavas from these areas, including the youngest lava on Santo Antão, descended the SE slopes and reached the coast W of Porto Novo (Fig. 2).
The LAGOA (and earlier CAD) magmas dominantly flowed towards the SE, giving rise to a youthful morphology on this side of the island. The NW and SE coastlines have migrated towards the SE. A major Plinian eruption at 0·2 Ma resulted in deposition of the phonolitic Cão Grande pumice (Mortensen et al., in preparation) from a vent which was centered just south of the Coroa complex but has since been buried by younger volcanic rocks. The pumice consists mainly of two layered sequences (CG1 and CG2). CG1 can be traced over most of the island and serves as a valuable marker below the most recent volcanic products.
The bulk of the island is built of lava flows. Pyroclastic rocks, mainly lapilli and bombs, are commonly prominent near the main eruption centres. A few ignimbrites have been observed locally but the Cão Grande phonolitic pumice (CG1) is by far the most voluminous deposit from a single eruption.
|
PETROGRAPHY OF THE VOLCANIC ROCKS |
|---|
|
TOPABSTRACTINTRODUCTIONTHE GEOLOGICAL SETTING OF...PETROGRAPHY OF THE VOLCANIC...ANALYTICAL METHODSWHOLE-ROCK GEOCHEMISTRYDISCUSSIONSUMMARY OF CONCLUSIONSSUPPLEMENTARY DATAREFERENCES |
|---|
Rock types by modal composition and mineralogy
The volcanic rocks have been divided into a nephelinite–phonolite series and a basanite–phonolite series on the basis of their geochemical characteristics. These are referred to as the nephelinite and basanite series, respectively. Figure 3 shows the average distribution of phenocryst phases in members of the two series based on visual estimation.
|
Most of the lavas contain clinopyroxene ± olivine ± Fe–Ti oxide phenocrysts in aphanitic to glassy matrices. The lavas show considerable variations in phenocryst content from almost aphyric to rocks with >50% phenocrysts. The phenocryst content generally decreases with increasing magmatic differentiation from around 20 vol. % (range: 5–55) in the most primitive samples to 5 vol. % (range <1–20) in the most evolved. Petrographic studies reveal complex crystallization histories with reverse zonation in some clinopyroxenes in many of the lavas in which Fe-rich green or Ti-rich purple cores have light-coloured diopsidic rims. Picritic and ankaramitic lava types constitute around 10 and 20% of the lavas, respectively; these commonly contain small chromite phenocrysts. Phonolitic rocks contain abundant aegirine–augite phenocrysts with minor amounts of one or more of magnetite, haüyne, titanite and sanidine. Compositionally intermediate rock types (8–3% MgO) have phenocrysts of Ti-rich clinopyroxene, titanomagnetite, amphibole, ± olivine, ± haüyne, ± plagioclase. Many of the lavas carry xenoliths of cumulate rocks derived from subvolcanic magma chambers; some of the primitive lavas contain nodules of mantle peridotite.
Clinopyroxene is present in all rock types, except a few picrites, and is generally the most abundant phase in the most primitive rock types, except when exceeded by olivine in the nephelinites. When present, reversely zoned clinopyroxenes usually constitute only a small proportion of the clinopyroxene population. The colour of the clinopyroxene in thin section varies with its Fe/Mg ratio from pale brown diopside through brownish-purple, Ti-rich intermediate salitic types to green, pleochroic acmitic salites. Zoning is commonly very complex with resorbed cores and is of oscillatory or hourglass type. Not uncommonly, clinopyroxene generations with both reversely zoned brownish-purple rims around green cores, and brownish-purple rims with pale brown cores are present in the same rock. Chromite is present in rocks with 12% MgO (and in one sample, with 8% MgO), whereas olivine occurs only in rocks with >6% MgO. Olivine is optically homogeneous and commonly has embayed rims; replacement with iddingsite is very rarely complete. Titanomagnetite occurs in the entire range of rock types but is most abundant in samples with 3–7% MgO. Amphibole is common in the intermediate members but rare in the most evolved (<3% MgO) members of the nephelinite series; it becomes increasingly common in the evolved rocks of the basanite series. Most amphiboles have decomposed to aggregates of Fe–Ti oxide crystallites, but are otherwise strongly pleochroic (brown to red–brown) and commonly show complex zonation. Plagioclase is only a common phenocryst phase in 20% of the tephrites (with 5–7% MgO) within the basanite series rocks, some phonolites, and a few primitive rocks. Oscillatory zonation is common in plagioclase that is sometimes overgrown by sanidine. Haüyne phenocrysts, forming clear blue to colourless equant crystals, typically with a black reaction rim, are common (up to 10 vol. %) in the nephelinite series in rocks with <5% MgO. Trace amounts of haüyne also occur in some phonolites of the COVA group of intermediate age which belong to the basanite series. Apatite is a common accessory in rocks of intermediate composition. Minor amounts of titanite occur in intermediate to evolved rocks with MgO < 7% and biotite is rare, occurring only in trace amounts in some phonolites (e.g. the COVA group).
|
ANALYTICAL METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONTHE GEOLOGICAL SETTING OF...PETROGRAPHY OF THE VOLCANIC...ANALYTICAL METHODSWHOLE-ROCK GEOCHEMISTRYDISCUSSIONSUMMARY OF CONCLUSIONSSUPPLEMENTARY DATAREFERENCES |
|---|
Sample preparation
Samples of 1–2 kg were prepared and analysed (unless otherwise indicated) at the Geological Institute, University of Copenhagen. To remove secondary alteration associated with the presence of caliche, specimens were initially leached with cold 0·1 M (or slightly stronger) HCl overnight and dried at 40°C. Subsequently, jaw-crushed material was used for Pb isotope analyses or crushed to fine powder (<10 µm) in an agate swing mill (for major and trace element and Nd–Sr isotopic analysis).
Whole-rock major and trace element analysis
Major elements were analysed at the Geological Survey of Denmark and Greenland (GEUS) by X-ray fluorescence (XRF) using a Philips PW1606 instrument; volatile contents were estimated based on loss on ignition (weight loss 110–1000°C) corrected for the oxidation of Fe. Na was analysed by flame photometry and Fe2+ by titration (Kystol & Larsen, 1999). The analytical precision of the data is (1
, abs. wt %): SiO2: 0·15; TiO2: 0·015; Al2O3: 0·05; Fe2O3total: 0·1; FeO: 0·1; MnO: 0·003; MgO: 0·05; CaO: 0·03; Na2O: 0·05; K2O: 0·005; P2O5: 0·005; volatiles: 0·10%.
Trace elements were analysed by XRF on pressed powder pellets using a Philips PW 1400 instrument. The international standards G-2, GSP-1, AGV-1, W-1, BCR-1, PCC-1 and DTS-1 were used for calibration and AGV-1 as monitor (Table 1).
|
A subset of 60 samples was analysed by instrumental neutron activation analysis (INAA) using a Ge detector on a multi-channel analyser at the Geological Institute (R. Gwozdz, Tracechem) on 100 mg powder irradiated at the Danish Nuclear Research Reactor at Risø. Iron was used for calibration and the analytical quality monitored by BHVO-1 (Appendix A).
|
Another subset of 40 samples was analysed by ICP-MS at GEUS. Calibrations were based on synthetic standards and BHVO-1, BIR-1 and GH. The international standards BE-N and BCR-1 were used to monitor the analytical quality, and were dissolved and analysed along with the samples (Appendix A).
Precision and detection limits for the analysed elements and analyses of standard materials are given in Table 1. Incompatible trace element levels are, in general, relatively high and analytical uncertainties low: e.g. La, Zr and Nb 2 
4% for typical concentrations, and 2(La/Nb) 0·03 and 2(Zr/Nb) 0·3.
In the subsequent discussion, XRF results are plotted for the elements Sc, V, Cr, Co, Ni, Cu, Zn, Ga, Rb, Sr, Y, Zr, Nb, Y, Ba, La, Ce, Nd, Pb and Th for consistency. For other trace elements, ICP-MS analyses are preferred to INAA when both are available (for Cs, REE, Hf and Ta).
Mineral compositions
Mineral analyses were performed on polished thin sections using a fully automated JEOL 733 electron microprobe at the Geological Institute, University of Copenhagen, and the Department of Earth Sciences, Aarhus University. An acceleration voltage of 15 kV and a beam current of 15 nA were used. Except for alkali-rich minerals, the electron beam was focused to 2–4 µm diameter. Energy dispersive analysis was used for most elements; Na, Cr and Ni were analysed by wavelength dispersive methods. ZAF corrections were performed using the Tracor Northern software. Standardization was performed using international natural and industrial mineral standards. Typical uncertainties are 2–5% for the major oxides and <1% of the total. Detection limits are about 0·1% oxide.
Sr–Nd–Pb isotope analysis
Chips of rock were leached for 1 h in hot 6N HCl and dissolved in concentrated HF + HNO3 (in the ratio 10:1). Pb was extracted in two passes in 6N HCl from ion exchange (AG1 x 8, 100–200#) columns of Pb complexed in 1M HBr. Pb was loaded on Re filaments in silica gel and phosphoric acid and run on a VG 54-30 TIMS instrument in the static mode with filament temperatures of 1180–1270°C. Many of the Pb isotope analyses were duplicated by total rock analyses from dissolution of chips. Only results with within-run statistics better than 1 SEM (standard error of the mean) = 0·2% were accepted. If duplication resulted in large differences, a third or fourth analysis was performed and an eventual outlier discarded. The resulting reproducibility for 23 different samples was: 1(206Pb/204Pb) = 0·006 (excluding two analyses), 1(207Pb/204Pb) = 0·003 (excluding three analyses) and 1(208Pb/204Pb) = 0·011 (excluding five analyses). These uncertainties are actually smaller than the reproducibility of the standard NBS981 used for mass fractionation corrections, which amounted to, on average, 0·12 ± 0·01(1)%/AMU (for all three ratios, N = 20) relative to the values of Todt et al. (1993). Two or three NBS981 standards were run with each sample batch and used for corrections for that batch. We estimate that the reproducibility of NBS981 is our analytical uncertainty. Total chemical procedure blanks were 50–200 pg and insignificant compared with >6 ng dissolved sample Pb.
Sr was extracted using a cut from 2M HCl passing through the anion exchange columns and cleaned using Sr-spec©. REE were collected from the columns after Sr. Nd was then extracted from the REE fraction by ion exchange on Teflon-coated resin using 0·25N HCl. Sr was loaded in 1M phosphoric acid on Ta filaments and Nd loaded in 0·1M HCl on Ta filaments in a triple-filament arrangement with a centre Re filament for high T (1850–1900°C) ionization. Both Sr and Nd were run dynamically with a three-peak jump sequence. Sr results were corrected for mass fractionation using the exponential law and 86Sr/88Sr = 0·1194. Nd was linearly corrected to 146Nd/144Nd = 0·7219. 87Rb interference on Sr was corrected using 85Rb, and 144Sm on Nd using 147Sm. Age corrections of isotope ratios would be insignificant and were not applied.
|
WHOLE-ROCK GEOCHEMISTRY |
|---|
|
TOPABSTRACTINTRODUCTIONTHE GEOLOGICAL SETTING OF...PETROGRAPHY OF THE VOLCANIC...ANALYTICAL METHODSWHOLE-ROCK GEOCHEMISTRYDISCUSSIONSUMMARY OF CONCLUSIONSSUPPLEMENTARY DATAREFERENCES |
|---|
Rock classification
According to the total alkalis vs silica diagram (Fig. 4) (Le Maitre, 1989
), most samples are nephelinites or basanites–tephrites and need to be classified using CIPW norms according to the revised IUGS nomenclature (Le Bas, 2000). One problem with this approach is that the unknown degree of secondary oxidation of Fe may lead to significant deviation of the actual norm from the pristine norm. We have calculated CIPW norms after correcting Fe to the lower limit of oxidation using the minimum of measured Fe3+/Fetotal, which increases systematically from 0·27 in samples with >5% MgO to 0·50 in the phonolites. There is a clear trend towards decreasing silica and increasing alkalis from the OV picrobasalts towards the young mafic volcanic rocks, including basanites, nephelinites and melilite nephelinites. Evolved rocks trend towards phonolite with very few trachytic compositions. The most evolved rocks belong to the intermediate (COVA) and young volcanic phases.
|
Major elements
Representative major element compositions are given in Table 1. The complete geochemical dataset is available in Electronic Appendix A, which can be downloaded from http://www.oupjournals.org/. MgO contents in 329 analysed samples range from 19 to 0% (analyses reported as wt % on a volatile-free basis unless otherwise indicated) and are used as reference variable in major element variation diagrams (Table 1; Fig. 5). For samples with >5% MgO, there is a broad range of SiO2 contents between 37 and 45%, whereas for <5% MgO, there is a well-defined trend towards phonolitic compositions with around 57% SiO2. CaO shows a major inflection around 8% MgO. With decreasing MgO, FeO and TiO2 decrease from 8% MgO among the rocks most rich in FeO and TiO2, and from around 6% MgO among rocks with the lowest levels of FeO and TiO2, which are the OV. The same type of variation is seen for P2O5, where the high-P2O5 AN group shows a decrease with falling MgO from around 6% MgO, whereas the OV group starts to decrease in P2O5 only when MgO falls below 4–5%. There is a remarkably broad range of compositions at high MgO where the temporal rock groups can be identified, although there are considerable overlaps and several samples do not conform to the main group characters. An additional group is geochemically identified among the OV and is called the ‘enriched old volcanic rocks’ (EOV).
|
Alteration of samples
Volcanic rocks are typically highly permeable and subject to reaction with infiltrating hydrothermal fluids or meteoric water. Because volcanic glass is hygroscopic, and there is a large variation in the proportion of glass in the samples, high volatile contents (LOI) may indicate hydration of the glass but not, necessarily, significant post-magmatic mobilization of major or trace elements. Hydrated samples with >2·5% volatiles do not systematically show abnormal contents of fluid-mobile elements (Na, K, Rb, Ba) compared with the trends in variation diagrams (Figs 4–6), in plots vs an immobile element such as Al2O3 (not shown), and do not show other signs of alteration, such as elevated Fe3+/Fe2+ ratios. Samples (
5% of the dataset) with very high volatile contents, or with Fe3+/Fe2+, K/Rb or Na2O/Al2O3 ratios deviating from the main range have not been used in the discussion below.
Trace elements
The volcanic products from Santo Antão show relatively high abundances of mantle incompatible elements that increase with decreasing MgO (Table 1; Fig. 6). Only Zr, Hf, Rb, Cs and Th increase to the phonolite stage; REE, Y, Nb, Ta, Sr, Ba, Pb and U decrease at low MgO values. Inflections in the trends occur at varying MgO contents for different elements reflecting the onset of crystallization of various minerals. Sr decreases with falling MgO from 2% MgO, whereas the Nb, Ba and Ga trends inflect at 1% MgO. Levels of incompatible trace elements in the high-MgO rocks show considerable variations between the groups identified, e.g. La 30–130, Zr 150–600, Nb 50–150, Rb 10–70, Ba 250–1000 and Sr 500–2000 ppm (Table 2; Fig. 6).
|
|
The variations of selected compatible trace elements as a function of % MgO are presented in Fig. 7. Ni and Cr contents decrease to a few ppm at 5% MgO. The CAD, LAGOA and COROA groups have 100–400 ppm Cr at 12% MgO, whereas OV, COVA and FRA have 500–800 ppm Cr. There is the same qualitative difference for Ni. OV and FRA have higher Sc at high MgO than other groups. Sc decreases for all groups for MgO <9–10%. V decreases below
7% MgO.
|
Primitive mantle-normalized incompatible element patterns for the most primitive Santo Antão volcanic rocks show an overall picture for averages of old, intermediate and young rocks with MgO = 10–14 wt % (Fig. 8) of strongly fractionated moderately incompatible elements including the HREE, a maximum at Nb and Ta and variable negative K anomalies and an increase in abundance of incompatible elements with time. The enrichment of Nb varies from 60–210 x PM (Primordial Mantle; Sun & McDonough, 1989
) for all samples, and, on average, the enrichment in the OV samples is about half of that in the samples of the young groups for all very incompatible elements. The negative anomaly for K (K/K*; K* is interpolated from adjacent elements in Fig. 8) is, on average, 0·5 for the young volcanic rocks and 0·3 for the intermediate and OV. The abundances of HREEs are low, e.g. LuPM 4.
|
Distinction of the temporal groups
Of the most primitive samples (MgO >8%), the OV rocks have the highest SiO2 contents (43–45·5%; Fig. 5a) and lowest incompatible element contents (e.g. Figs 5 and 6: lowest TiO2, P2O5, La and Nb) and constitute a well-defined group. They are further characterized by their relatively low La/Nb (0·6–0·7), Sr/Nb (9–12), Rb/Nb (0·2–0·4), Ba/Nb (5–8) and high Ce/Pb (35–40) ratios relative to other Santo Antão high-MgO rocks (Table 1).
The intermediate age rocks (SiO2 40–43%) of the COVA group are more alkaline and incompatible element-enriched (Fig. 8) than the OV and have higher Sr/Nd (16–20), Sr/Nb (12–15), La/Nb (0·7–0·8) and Rb/Nb (0·3–0·8) ratios (Table 1). The BOR volcanic group includes the first nephelinites to be erupted on Santo Antão and these show even greater enrichment in very incompatible elements than the COVA samples (Figs 5 and 6). Geochemically, the FRA group is intermediate between the COVA group and the more enriched AN group.
The CAD group nephelinites and melilite nephelinites display a large range of compositions with SiO2 contents as low as 38%. Their LREE enrichment reaches the extreme for Santo Antão (Fig. 6d and e), and they have high La/Nb ratios (0·8–1·1). Furthermore, Ba/Nb in CAD is not higher than in OV and COVA. LAGOA has Sr/Nd = 13–15, whereas COROA and PRC have 10–14. All the latter three have low Sr/Nb (10–12) and relatively high La/Nb (0·7–0·9; one sample has 1·1).
Mineral chemistry
Phenocryst phases were analysed in representative rocks, and selected mineral analyses are presented in Table 2.
Clinopyroxenes show a continuous range and variation from diopside-rich with 1·2% Cr2O3 (3 vol. % Cr-diopside component) and Mg-number (based on Fe2+ + Fe3+) = 89 to ferro-salite (Mg-number = 42) with 10 vol. % acmite and 10 vol. % other ferri-components (Fe3+viFe3+iv and Fe3+viAliv). Clinopyroxenes in the intermediate composition lavas are Ti-salites with up to 10 vol. % of the CaTiAl2O6 molecule and rich in Fe3+ (calculated Fe3+/Fetotalmean 0·5). TiO2 reaches a maximum of 7·1%. Low-pressure crystallization is indicated by very low Al3+vi values (<0·1 and commonly 0). Abrupt zonation with core–mantle differences in a small fraction of the phenocrysts may represent a break in composition of up to half the entire range for Santo Antão clinopyroxenes.
Olivine phenocrysts are in the range Fo91–74 with most >Fo80. Almost all of the most forsteritic olivines (>Fo88) have 0·2–0·4% CaO. Ni in olivine phenocrysts correlates with Mg-number and reaches >3000 ppm. There is only a general and very broad correlation between the MgO in host rocks and mol. % Fo in olivine. Olivines display a compositional variation in individual samples of Fo = 5–15. In 111832, olivine is in the range Fo90–81.
Amphiboles are mostly kaersutites but pargasite–ferropargasite (lower Ti) occurs in the phonolites and tephriphonolites. They cover a wide compositional range (Mg-number 75–42). TiO2 contents (6·6–3·7%) decrease with decreasing Mg-number. K2O/Na2O ratios are mostly in the range 0·5–0·7.
The most common oxide in the intermediate rocks is titanomagnetite, although chromite is present in many of the more primitive samples. Both magnetite and chromite, however, occur in rocks of various stages of evolution, suggesting magma mixing. The titanomagnetite variation follows the Fe–Ti trend of Barnes & Roeder (2001). There is a clear trend of decreasing ulvöspinel component from up to Usp80–Mag20 in rocks with 12% MgO to Usp5–Mag95 in rocks with 1% MgO. Most analyses have a (titano-)maghemite component of 10 vol. %, probably resulting from oxidation during cooling. Chromite has up to 42 % Cr2O3, which correlates with Mg-number in the interval 20–65. MnO in magnetite increases to >6% in some phonolites.
Haüynes are relatively rich in Ca and S for this sodalite group mineral. The Na/(Na + K + Ca) ratio in Santo Antão haüyne increases with host rock evolution from nephelinite to phonolite; this is accompanied by an increase in the Cl/SO4 ratio (Table 2).
Some basanite series rocks have plagioclase in the compositional range An87–26. Plagioclase in phonolites is An39–23 and sodic varieties grade into anorthoclase. Zoning is both normal and reverse, but mostly optically undetectable with An < 4%, but ranging to 11%. Sanidine is typically Or50–60 but wide ranges (Or40–82) occur in the evolved nephelinites, tephriphonolites and phonolites of the nephelinite series. Rare nepheline is around Ne84Ks16·
Sr–Nd–Pb isotopic compositions
The Sr, Nd and Pb isotope compositions of selected samples from the Santo Antão volcanic sequence are presented in Table 3 and cover the ranges: 87Sr/86Sr 0·70299–0·70333; 143Nd/144Nd 0·51297–0·51282; 206Pb/204Pb 19·17–19·89; 207Pb/204Pb 15·58–15·65 and 208Pb/204Pb 38·79–39·66 (Fig. 9). The new data for Santo Antão considerably extend the range of Pb isotopic compositions for the Cape Verdes reported by Gerlach et al. (1988), whereas two of their samples—NA15 and NA80—have slightly lower 87Sr/86Sr and higher 143Nd/144Nd (Fig. 9a). Because the data of Gerlach et al. (1988) lack stratigraphical constraints, they will not be considered further in this work. There is a negative correlation between 87Sr/86Sr and 143Nd/144Nd and a positive correlation between 208Pb/204Pb and 206Pb/204Pb (Fig. 9c and d). However, neither 207Pb/204Pb, 143Nd/144Nd nor 87Sr/86Sr correlates with 206Pb/204Pb (Fig. 9b, c and e). Overall, the Santo Antão volcanic rocks have Sr and Nd isotopic compositions between those of Central Atlantic N-MORB and the southern Cape Verde Islands (Fogo and Santiago in Fig. 9), whereas Pb isotopes are more radiogenic than either.
|
|
The isotopic compositions of the Santo Antão magmas have changed over time and the various temporal groups can, to some extent, be identified by their isotopic compositions, although there is considerable overlap. The OV extend to the highest 87Sr/86Sr, 206Pb/204Pb and 208Pb/204Pb and most unradiogenic Nd, whereas the intermediate AN and COVA groups have the most MORB-like Sr and Nd. The most unradiogenic Pb is found in the intermediate CAD group. The COROA, PRC and LAGOA groups of the young volcanic rocks have intermediate Sr and Nd isotopic compositions, but trend towards higher 207Pb/204Pb than the older rocks (Fig. 9c and f). Intermediate BOR and FRA group rocks have the most uranogenic Pb with the lowest time-integrated Th/U (Fig. 9d; Table 3).
All measured µ (238U/204Pb ratios) are high, mostly 26–43, with a total range of 18–71 and median of 35, but show no correlation with Pb isotope ratios or sample groups. This range of µ is high relative to MORB (average µ = 11 ± 5; White, 1993) and comparable with many OIB (µ = 20–43; Thirlwall, 1997). The decoupling of µ and the isotopic composition of Pb are common features in both MORB and OIB (White, 1993; Thirlwall, 1997).
Compared with other Atlantic hotspots, the Cape Verde Islands trend towards Tristan da Cunha [an EM1-type OIB (Zindler & Hart, 1986)] in Sr–Nd–Pb isotopic space at higher 143Nd/144Nd than the Canaries and St Helena hotspots (Fig. 9a), and lower 143Nd/144Nd than the Iceland trend (not shown). 208Pb/204Pb in samples from the southern Cape Verde Islands trend towards unradiogenic Pb compositions above the NHRL (Northern Hemisphere Reference Line, Hart et al., 1992) (Fig. 9d). Santo Antão OV samples trend towards radiogenic Pb below the NHRL for 207Pb/204Pb and lie well correlated on the NHRL for 208Pb/204Pb, whereas the intermediate and young samples lie on the NHRL for 207Pb/204Pb and below for 208Pb/204Pb (Fig. 9c and d). The Canaries and Santo Antão overlap in Pb isotopic composition. Whereas the Canaries are situated between St Helena [HIMU-type OIB (Zindler & Hart, 1986)] and the southern Cape Verde Islands in all diagrams in Fig. 9, the Santo Antão rocks have a component of variation perpendicular to the Canaries trend in Fig. 9e (208Pb/204Pb vs 87Sr/86Sr).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONTHE GEOLOGICAL SETTING OF...PETROGRAPHY OF THE VOLCANIC...ANALYTICAL METHODSWHOLE-ROCK GEOCHEMISTRYDISCUSSIONSUMMARY OF CONCLUSIONSSUPPLEMENTARY DATAREFERENCES |
|---|
We first discuss magmatic evolution and the relative roles of fractional crystallization, magma mixing and crustal contamination. Mineral compositions are used for quantification of the liquid line of descent of the magmas and the primary magmatic compositions are derived. Mantle melting is subsequently discussed and the primary magmas are characterized. We then turn to the systematic changes in mantle melt compositions during the evolution of Santo Antão and identify the local end-members that explain the variation and relate these to global mantle components. Finally, we address plume–lithosphere processes and speculate on the origin of the Cape Verde mantle components.
Evidence for mineral–melt disequilibria
There is a lack of equilibrium between some of the phenocrysts of olivine and clinopyroxene and their host rocks as illustrated by an up to threefold variation of Fe/Mgol and up to tenfold variation of Fe/Mgcpx in individual rocks. This cannot be a result of changes in melt composition during closed-system fractional crystallization because clinopyroxene zonation is abrupt and commonly reverse. Neither could the olivines conceivably have formed from melts at elevated pressures >1 GPa, where Kol/melt is higher (Ulmer, 1989), because low-pressure crystallization is invariably indicated by the low Al contents of all clinopyroxenes. Although some samples have accumulated crystals and this can explain apparent host rock–mineral disequilibrium, others with too Fe-rich or too Mg-rich olivines in rocks with only modest amounts of phenocrysts cannot be explained in this way. This, together with the reverse zonation of some clinopyroxenes in many rocks, the occurrence of Fe–Ti oxides in chromite-bearing MgO-rich rocks, and chromite in relatively evolved magmas with 7–8% MgO, indicates that many crystals are xenocrysts. Green clinopyroxene cores in relatively primitive lavas are commonly identical to equilibrium phenocrysts in phonolites. The rock trends, such as FeO–MgO and CaO–MgO (Fig. 5), however, allow only a small fraction of phonolite in any mixture with primitive magma and mixing seems, in general, not to be volumetrically important for Santo Antão magmas. It seems likely that the xenocrysts were generally mobilized from semi-solid cumulates by rising magma, or sank into relatively primitive dense magma in a stratified magma chamber from overlying more evolved, buoyant magma.
Magmatic differentiation
Olivine
The wide spectrum of volcanic rocks, from picrites to phonolites, on Santo Antão demonstrates the operation of fractional crystallization. The most magnesian olivine phenocrysts (Fo88-91 with Ni = 1800–3400 ppm) have higher CaO contents than typical mantle olivine (Larsen & Pedersen, 2000). This suggests that essentially unmodified mantle melts were emplaced into the shallow crust. The wide compositional range of olivines in many samples indicates that truly comagmatic suites may not be found, as implied by the occurrence of mineral disequilibrium discussed above. Constant whole-rock SiO2 and FeOtotal within the various volcanic groups for MgO > 10%, accompanied by rising TiO2 and CaO and decreasing Ni with falling MgO, attest to forsteritic olivine control. Olivine is accompanied by traces of chromite and Cr decreases strongly with MgO. Samples with >11% MgO have 1–40 vol. % phenocrysts and many of these are probably xenocrystal. Simple calculations of the MgO content of the melt by subtraction of olivine phenocryst compositions from the host whole-rock compositions yield a maximum of 12·4% MgO for the matrix. This implies that primary melts may have had at least 12% MgO.
Clinopyroxene
The decrease in whole-rock Sc and CaO starting at 8–10% MgO shows that clinopyroxene began to fractionate at this stage. It is, therefore, clear that mainly olivine controlled the major element variation for MgO >8%. Ti substitution in clinopyroxene controls the Al variation and reflects the magmatic evolution of Ti, whereas Al in clinopyroxene is uncorrelated with Al in the magmas. Moreover, the variation in Al/Ti ratio from 2 to 4 in the most primitive diopsides is similar to the variation of this ratio in the primitive melts. Kd(Ti)cpx/rock values were calculated for clinopyroxenes found to be in approximate equilibrium with their host rocks (assuming Kd(Fe/Mg)cpx/rock = 0·20–0·35). There is an increase in Kd(Ti)cpx/rock from around 0·5 to 0·6 [higher than reported values of 0·2–0·4 (e.g. Hart & Dunn, 1993; Halliday et al., 1995)] in the interval 8–6% MgO. Ti becomes compatible in the more evolved magmas with <6% MgO, reaching K(Ti)cpx/melt 2 in the phonolites. The maximum TiO2 content in these clinopyroxenes occurs when Mg-number is around 70 (67–73). The maximum whole-rock TiO2 occurs in rocks with Mg-number (Fetotal) = 44 (42–45). This allows the distribution of Fe/Mg to be calculated; K(Fetotal/Mg)cpx/melt 0·33 ± 0·03, which is somewhat higher than the value of 0·23 ± 0·05 reported by Grove & Bryan (1983) and Sack & Carmichael (1984). The difference in partitioning of Zr and Hf between clinopyroxene and melt is also noteworthy. Below 8% MgO, the Zr/Hf ratio steadily increases from around 41 to 60–70 in phonolites, indicating that Hf is less incompatible than Zr, as also reported by Hart & Dunn (1993).
The wide range of element concentrations at MgO >9%, reaching fourfold variation for some incompatible elements, cannot be explained by fractionation of the mineral phases present. This is evidence that parental melt compositions were far from constant.
Other phases
The spinel crystallization gap expected between the crystallization of chromite and titanomagnetite (Irvine, 1967; Sack & Ghiorso, 1991) is missing and probably reflects the highly oxidized nature of the magmas; this is evident from the high Fe3+/Fe2+ ratio in the bulk rocks (Fe3+/Fetotalmedian = 0·44, Fe3+/Fetotalmin = 0·26) and clinopyroxenes (Fe3+/Fetotalmedian = 0·59, Fe3+/Fetotalmin = 0·20). Chromite ceases to crystallize when clinopyroxene starts and subsequently clinopyroxene accommodates Cr. The continuous spinel crystallization could be explained if titanomagnetite was already a (possibly minor) liquidus phase at this early stage of evolution (8% MgO), which is clearly indicated for the most silica-undersaturated magmas by inflections in variation diagrams for SiO2, FeO, TiO2 and V (Figs 5 and 7). This indicates that the majority of the magmas were more oxidized than the OV magmas that fractionated Fe–Ti oxides from melts with 5–6% MgO. Mg-number (Fe corrected) decreases over the entire range of MgO contents but only slightly in the interval 6–3% MgO. This would indicate that, at this stage, Fe–Ti oxides make up a large proportion of the crystal fractionate. Major element mass balance calculations show that 7% Fe–Ti oxide crystals and 33% ferromagnesian minerals, mainly clinopyroxene, could have been extracted from the magma in this MgO interval. With compositions close to the host melts, large amounts of amphibole would need to fractionate to significantly change melt compositions, but only small amounts can be accommodated in incompatible element modelling and amphibole rarely constitutes a large proportion of the phenocryst assemblage. This is in accord with the lack of evidence for noteworthy amphibole fractionation in the variation diagrams (Figs 5 and 7). The initiation of apatite fractionation is clearly marked by the inflection in the P2O5 trend at 5–6% MgO. Titanite fractionation is probably responsible for the trend in the young volcanic rocks of declining Nd, Y and HREE and stable levels of La and Ce from 4–5% MgO (Fig. 6). Whereas plagioclase is rarely present in the evolved rocks, sanidine does occur and clearly fractionated in magmas with <1% MgO, as demonstrated by the decreasing Sr and Ba contents. Nb is relatively low in some phonolites and is decoupled from HREE variation. This probably reflects the fractionation of pyrochlore, although this phase has not been observed petrographically.
The evolved magmas
Rb and Th remain incompatible throughout the fractionation range. Rb is used as an index of differentiation because the XRF analytical precision for Rb is far better than that for Th in the primitive rocks (Th is analysed by INAA in only a third of the samples), and the trend for Rb (Fig. 6a) is as good as for other incompatible elements. We have assumed that any evolved compositions that occur within a sequence of volcanic rocks with closely related geochemical features belong to the same magmatic suite. The locations of inflections in variation diagrams reflect the onset of crystallization of various phases in the evolving magmas. The wide range of magma compositions observed can be divided into two broad trends: one for magmas derived from nephelinitic parent magmas (the nephelinite trend), and one for the basanitic parent magmas (basanite trend). These magma differentiation trends become much more distinct at the most evolved end of the compositional spectrum (Fig. 10a). At 65 ppm Rb (5% MgO, Fig. 6), Y (Fig. 10d), Ce, Nd, HREE and Zn all become compatible in many nephelinite-trend magmas, as discussed above, and at 75 ppm Rb for some basanite-trend magmas. At 115 ppm Rb, decreasing Ba and Sr mark the start of sanidine crystallization; La, Nb and Ta become compatible and Nb/Ta increases, probably marking the start of titanite fractionation. Plagioclase did not fractionate in significant amounts for the majority of the Santo Antão magmas and sanidine fractionated only during extremely advanced stages of crystallization (with <1% MgO). The proportion of felsic fractionating phases increased at around 5% MgO (Fig. 3) and 70 ppm Rb, but this is not evident in Fig. 6. However, a decrease in SiO2/Al2O3 (not shown) from 4–5 at high MgO to 2·8 for MgO < 5% is caused by the co-fractionation of olivine and clinopyroxene which ceases at this stage of evolution, after which SiO2/Al2O3 remains fairly constant at 2·4–2·8.
|
Modelling magmatic differentiation processes
We have adopted a broad approach for discussion of the liquid line of descent of the Santo Antão magmas. In general terms, one may consider that there are two dominant magma types at Santo Antão: a younger nephelinitic low-silica type (LAGOA, COROA, PRC, CAD & BOR) and an older basanitic type (OV, OAN, COVA, AN & FRA) with distinctly higher SiO2 and less enrichment in incompatible elements (Figs 5 and 6). Both magma series evolve towards phonolite, but the nephelinite trend is characterized by high Zr and Th, with up to twice the amounts in the basanite trend at all stages of magmatic evolution (Fig. 6). It is evident that the CG1 (Cão Grande pumice no. 1) sequence of phonolitic pumice belongs to the nephelinite trend, whereas the subsequent CG2 (Cão Grande pumice no. 2) represents a late example of a basanite-trend evolved magma-type which is otherwise typical for the COVA and FRA groups (Fig. 10a).
Two models for the liquid line of descent demonstrate the evolution of the two main magma series. The melilite nephelinites–phonolites of the nephelinite trend (COROA, PRC) and the basanites–phonolites of the basanite trend (OV, COVA) span the entire range of geochemical variation for most elements, and we have modelled the evolution of two primitive melt compositions. Fractional crystallization of the magmas was modelled in steps of 1% crystallization for chosen parental magma compositions, melilite nephelinite 110052 and basanite 111750. The equilibrium mineral compositions were calculated for each step based on mineral/melt distribution coefficients for Fe/Mg, and the mineral compositions parameterized as functions of Fe/Mg based on the measured mineral compositions. This allows the large and non-monotonous variation in, for example, TiO2 and Al2O3 in clinopyroxene and spinel minerals to be appropriately incorporated into the fractionation modelling. The fractionating mineral assemblage was assessed from the observed phenocryst assemblages (Fig. 3) and whole-rock geochemical features, such as inflections on variation diagrams (Fig. 4). The results for fractionation in the interval 12–5% MgO are reported in Table 4. For the most evolved rocks, magmatic evolution may be monitored by the incompatible elements Rb (Fig. 10) and Th; these results are reported in Table 4. For the nephelinite parental melt, 12% olivine fractionation is modelled before 8% fractionation of the assemblage olivine + clinopyroxene + magnetite with an increasing relative amount of magnetite. Amphibole, haüyne and apatite fractionation set in after 20, 25 and 35% crystallization, respectively (Table 4). After 40% fractionation, the remaining melt has 4·8% MgO. For the basanite parental melt, Fe–Ti oxide fractionation starts relatively late: 10% olivine is followed by 10% olivine + clinopyroxene, and then 25% clinopyroxene + Fe–Ti oxide + amphibole. After another 15% fractionation of the same assemblage plus apati