Journal of Petrology Volume 42 Number 2 Pages 321-354 2001
© Oxford University Press 2001
Melt Generation and Movement beneath Theistareykir, NE Iceland*
1INSTITUTE OF THEORETICAL GEOPHYSICS, BULLARD LABORATORIES OF THE DEPARTMENT OF EARTH SCIENCES, CAMBRIDGE UNIVERSITY, MADINGLEY ROAD, CAMBRIDGE CB3 OEZ, UK
2NORDIC INSTITUTE OF VOLCANOLOGY, REYKJAVIK, ICELAND
3WOODS HOLE OCEANOGRAPHIC INSTITUTION, WOODS HOLE, MA 02543, USA
Received September 17, 1999; Revised typescript accepted June 5, 2000
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
A detailed study of the volume and composition of all the lavas from the Theistareykir segment of the Northern Volcanic Zone of Iceland was designed to study basaltic melt generation and movement beneath a spreading ridge. The trace element compositions of the lavas are variable, and those of melt inclusions in olivine, clinopyroxene and plagioclase phenocrysts even more so. We show that this variability can be produced by mixing instantaneous melts produced by isentropic decompression of mantle whose initial potential temperature is 1480°C, and that the calculated volume and composition of the average melt is consistent with geophysical and petrological observations. Pressure and temperature estimates suggest that the phenocrysts form in the upper mantle, at depths of 30–40 km, and trap melts formed at greater depths. Some mixing of the instantaneous melts occurs before the melt is trapped, and more mixing occurs before the lavas are erupted. A similar model can account for the composition of melt inclusions from the FAMOUS area of the Mid-Atlantic Ridge, and from the Gorda and Juan de Fuca Ridges.
KEY WORDS: basalt; Iceland; melt inclusions; melting; ridges
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
We still know little about how the mantle melts, and how melt separates from its residue and moves to the surface, despite the central importance of this process to igneous petrology and geochemistry. The equations that govern the movement of two interconnected fluids of widely different viscosity are known, and have been used to study the fluid dynamical processes involved in melt separation and movement. It is, however, uncertain how well these equations describe the processes that occur within the Earth. The equations have been shown to support solitary waves in one, two and three dimensions. Only three-dimensional solitary waves are stable, and they possess spherically symmetric contours of porosity (Barcilon & Richter, 1986
; Barcilon & Lovera, 1989; Wiggins & Spiegelman, 1995). As yet there is no evidence that such waves exist in the source regions of magma. There is increasing evidence (see McKenzie, 2000) that the time scale for melt transport from the source regions to the surface is no longer than 
103 a, and may be as short as 20 a. Even a time scale of 103 a is not compatible with solitary wave transport, and requires the melt to move in channels whose minimum dimension must exceed a few millimetres. Although Aharonov et al. (1995) and Kelemen et al. (1997) have argued that such channels are formed by chemical processes, they can also be produced by mechanical processes that move grains apart. At present there is almost no direct information about the physical processes involved in the formation and movement of melt. The theoretical work that has been carried out shows that the behaviour of the idealized two-phase system is rather complicated. It is therefore desirable to obtain some observational constraints on the processes involved. The principal difficulty in doing so is that geophysical methods can only provide information about the average seismic velocities and electrical conductivities of the region where melts are generated, because the length scales involved in melt separation and movement are much smaller than can be imaged geophysically. The resolution of other remote sensing methods is even poorer. Hence geochemical and petrological studies appear at present to be the most promising method of studying melting processes.
Melting by isentropic upwelling beneath ridges has been widely studied (see Langmuir et al., 1992), and is better understood than is the generation of basalts beneath intra-plate volcanoes and of calc-alkaline magmas at subduction zones. Most ridges are submarine, and therefore hard to study. In Iceland, however, the Mid-Atlantic Ridge is above water and can be mapped in detail. Although the presence of a plume beneath the island is in some ways a disadvantage, it is also the reason why this part of the ridge is subaerial. We chose to work on the Northern Volcanic Zone (NVZ, Fig. 1a) where the ridge consists of a single plate boundary that is approximately normal to the east–west spreading direction with a separation rate of 20 mm/a. The study described here is restricted to the Theistareykir segment (Fig. 1b), which was chosen because it previously had been mapped in detail by Grönvold (O’Nions et al., 1976), and was known to contain MgO-rich lavas that include small volumes of picrites (Fig. 1c). The field work was designed to map all flows, both on the ground and using aerial photographs, and to estimate their volumes and compositions. The discussion below is restricted to the melt that has been produced from beneath Theistareykir since the end of the most recent glaciation. Because subglacial eruptions form steep-sided ridges and table mountains of hyaloclastite and pillow lavas, in contrast to the shallow dips of the surfaces of postglacial lava flows, it is easy to use the geomorphology of the flows to determine when they were erupted. Flows erupted in the present interglacial can be distinguished from those produced during previous interglacials by the absence of glacial striations and overlying glacial drift. The absolute ages of the postglacial lavas are constrained by using ash layers, whose ages are known from radiocarbon measurements. Sixty-six new samples of postglacial lavas have been analysed for major, minor and trace element abundances, and maps such as that shown in Fig. 1b were produced for all flows (Slater, 1996). Two specimens were collected at each location; one of which is stored at the Nordic Institute of Volcanology in Reykjavik and one at Cambridge. Together with existing analyses, these provided information on 99 samples. Most basic flows in Theistareykir contain abundant phenocrysts of olivine, and some contain those of clinopyroxene and plagioclase. All phenocrysts contain abundant melt inclusions.
|
The Theistareykir segment is dominated by Storaviti, a lava shield with a volume of
30 km3, 40 times larger than any of the other post-glacial flows in Fig. 1b. This large volume of melt was erupted shortly after deglaciation, as was that of a number of other lava shields in Iceland. Jull & McKenzie (1996) have argued that the large volume of melt involved was generated by glacial unloading. Whatever the process that generated the magma, the mean composition of the melt produced from the Theistareykir segment is essentially the composition of the Storaviti lavas given in Table 1. This average was estimated from samples collected from the surface flows. Whether it is representative of the much larger volume of buried flows is unknown, as no drill core is available and the volcano has not been eroded.
|
The composition of some of the smaller flows differs widely from that of the average Storaviti lava, and that of melt inclusions in the phenocrysts is even more variable. The principal purpose of this paper is to find out whether this variability can be produced by fractional melting of an initially homogeneous source of uniform composition, followed by mixing of melts produced at different depths. These processes are a feature of all models of melt generation beneath ridges. The aim of this paper is to discover whether they can account for the elemental compositions of the lavas, melt inclusions and residue, or whether more complicated models, such as infiltration–reaction processes discussed by Kelemen et al. (1997) and Shimizu (1998), or variations in source composition (Sours-Page et al., 1999) are required by the elemental observations (rather than being simply compatible with them). Such a discussion requires estimates of the fractional melt compositions generated in the melting zone beneath the Theistareykir segment. The variation of melt fraction with depth must therefore be estimated, by inverting the observed mean composition of the Storaviti lavas (McKenzie & O’Nions, 1991, 1995), using the source composition constrained by using the Nd isotopic compositions (Elliott et al., 1991).
The compositions of instantaneous fractional melts generated by this melting model are then compared with those of the melt inclusions and lavas using principal component analysis. This comparison shows that all the observed compositions can be generated by mixing fractional melts from different depths. Four other compositional datasets, from diopsides in abyssal peridotites (Johnson et al., 1990), from melt inclusions in high-An plagioclase phenocrysts from the Gorda (Nielsen et al., 1995) and Juan de Fuca (Sours-Page et al., 1999) Ridges, and from melt inclusions in olivine from the FAMOUS area (Shimizu, 1998), are discussed in the same way, and are all consistent with the same model. Finally, the depth at which melt inclusions in diopside form is estimated from the compositions of the host lavas and phenocrysts, and is compared with geophysical information from NE Iceland. The depth estimates give values that exceed that of the Moho.
There is no doubt that the isotopic observations reported by Elliott et al. (1991) from Theistareykir require variations in source composition. Clearly, understanding the implications of such isotopic variability is much easier if it is possible to show that it is associated with trace element variations which require compositional variations in the source. Sadly, the discussion below shows that this is not the case for any of the regions discussed.
|
USEFUL THEORETICAL IDEAS |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
Passive upwelling
Melt generation by passive upwelling beneath ridges produces
20 km3 of melt each year, and is the dominant process of melt production on Earth. Richardson & McKenzie (1994) used a stream function to show that the mean composition of the melt produced by fractional melting depends only on X(z), the variation of melt fraction with depth z, when the upwelling is passive. This important result greatly simplifies geochemical calculations, as the mean concentration is therefore independent of the shape of the melting region and of the matrix velocity within it. Because of this result, the simple (and unrealistic) model of melting in Fig. 2 can be used to calculate the mean concentration. McKenzie & O’Nions (1991) and White et al. (1992) exploited this result to obtain X(z) by inversion from the mean concentrations Ci of a number of elements i. If the vertical velocity of the matrix is W(z), then the melting rate 
(z) is
 |
is a function of z alone because all elements of matrix follow the same isentropic path through P,T space if their initial temperatures are the same and the source is homogeneous. Ci is then given by
 |
i is the first moment of the instantaneous melt composition. Although this integral for i is in general difficult to evaluate, it is simple to do so for the model in Fig. 2 because the melting zone is triangular and the vertical velocity of the matrix W is constant. Hence
 |
i to obtain X(z) possible.
|
Exactly the same argument shows that the second moment Cij of the distribution
 |
 |
Principal components
Plots of concentration ratios are widely used in petrology, although they have a number of important disadvantages. As they are two dimensional, at most four elemental concentrations can be used, whereas for many of the inclusions 14 elemental concentrations are measured. Which four elements are selected depends on chemical arguments and on the judgement and taste of the researcher. Whatever choice is made, most of the data are not used. Another disadvantage of such plots is that mixing lines between end members are in general curves, whose shape depends on the ratios of the concentrations of the elements in the end members. Mixing curves between end members whose composition is unknown must therefore be calculated. But perhaps the most important problem with such plots is that they may produce unpredictable and nonintuitive effects when they reduce the dimensionality of the dataset from four to two. Nielsen et al. (1995) and Shimizu (1998) plotted Ti/Zr against La/Sm. As these four elements behave in different ways during melting, the system Ti, Zr, La, Sm contains four degrees of freedom. The plots that these workers used can represent only two of these degrees of freedom, neither of which is related to the chemical processes that occur in the source region in a simple manner. For instance, La/Sm may increase because the La concentration increases, or the Sm concentration decreases.
A simple method of presenting the data that avoids these problems is to project the data onto planes defined by the principal components of the datasets themselves. This approach is widely used in many fields, although it has as yet been exploited to only a limited extent (Le Maitre, 1968; Allègre et al., 1987, 1995; Albarède, 1995, 1996) in geochemistry. The principle of the method is easily understood in two dimensions. If the elemental concentrations of two elements that behave in a similar, but not identical, manner, such as La and Ce, are plotted for a number of samples, the plot shows an elliptical cloud of points whose long axis is not in general along either the Ce or the La axis. This long axis can be made parallel to the x axis by rotating the coordinate system through some angle 
, where
 |
 |
is the angle between the x axis and the La axis of the original plot. The angle is determined from the distribution of the data themselves. As equation (5) shows, the new variables x and y depend linearly on [La] and [Ce]. The extension of these ideas to three or more dimensions is straightforward, and standard mathematical subroutines are available that can find the components of the orthogonal matrix O that rotates the principal axes of the multidimensional ellipse to be parallel to the coordinate axes (Press et al., 1992). A considerable number of datasets of trace elements have now been processed from a wide variety of lava types, geographic regions and tectonic settings. All have shown that the two largest principal components are considerably larger than the remainder, and together describe most of the observed variability. It is therefore possible to display most of the variability using one plot whose x axis is taken to be parallel to the largest, and y axis to the second largest, principal component. These components are labelled 1 and 2, respectively, throughout the discussion below. A major advantage of using principal components is that mixing lines in this space are straight, and that mixing obeys the lever rule, with the proportions of the end members required to generate a particular composition being inversely proportional to the distance from the points representing the end-member compositions. This ability to describe mixing is important, because the variation in the compositions of melt inclusions shows that mixing of melts with different compositions is a dominant process in magma production. One further advantage of using principal components is that random noise is reduced, because the concentrations of elements that behave coherently are combined. In contrast, the calculation of elemental ratios decreases the signal-to-noise ratio.
In practice, the first step in obtaining the principal components is to calculate the auto- and cross-correlations, Cij, between the concentrations ci and cj of two elements i and j in N samples
 |
. The standard deviation of any component is the square root of its eigenvalue. If a vector e is an eigenvector, so is -e. The sign of any eigenvector can therefore be reversed. This relationship is especially useful for the eigenvector with the largest eigenvalue, as it allows the sign of the eigenvector to be chosen to make this component of the projected compositions positive. The only subjective operation in the use of principal components concerns the normalization of the observed concentrations. If the measured concentrations are used with no normalization, the principal components are dominated by the element that shows the largest variations in concentration in parts per million. The variability of incompatible elements that are present in low concentrations, such as La, then has little effect on the eigenvalues and eigenvectors. The influence of the trace element concentrations on the principal components can be increased by normalizing the concentrations with respect to the source. We use the concentrations of the MORB source, given in Fig. 3, rather than chondritic values, largely to allow the principal components to reflect the importance of the large variations in La and Ce in the inclusions.
|
S of 330 J/kg K.
Unlike the auto- and cross-correlations, the auto- and cross-correlation coefficients rij, defined by
 |
We can now combine the ideas discussed above to provide a convenient method of comparing the variability of the observed and calculated concentrations by defining the correlation between two principal components Xij
 |
 |
 |
|
ANALYTICAL METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
The samples were crushed with a steel pestle and mortar and ground in an agate swing mill. The rare earth element (REE) concentrations were measured using an inductively coupled plasma source mass spectrometer (ICP-MS), a VG PlasmaQuad PQ2 Turbo, at the NERC ICP-MS facility at Silwood Park (see Slater, 1996
). Major, minor and trace element concentrations were measured using X-ray fluorescence spectrometry (XRF) at Edinburgh University. Major elements were determined after fusion with a lithium borate flux containing La2O3 as a heavy absorber. Trace element concentrations were measured using pressed powder samples [see Slater (1996) and Fitton et al. (1998) for details]. To prepare samples for melt inclusion studies, the rocks were first coarsely crushed and the phenocrysts hand picked. Because all melt inclusions were crystalline, they were first rehomogenized by heating to 1245°C in a vertical 1 atm quench furnace, held at an oxygen fugacity one unit below the quartz–fayalite–magnetite buffer for at least 20 min. The samples were then hand polished using Buehler Alpha micropolish alumina, to expose the melt inclusions, and then coated with gold (ion probe) or carbon (electron probe). The abundances of La, Ce, Nd, Sm, Eu, Dy, Er and Yb were determined at Woods Hole Oceanographic Institution, using a Cameca IMS 3f ion probe and procedures described by Shimizu & Hart (1982) and Shimizu (1998). The samples were bombarded with a beam of negatively charged oxygen ions with a net energy of 12·5 kV and a spot diameter of 20 µm. Ti, V, Cr, Sr, Y and Zr were measured with the same instrument, but with a spot diameter of 5 µm. The energy filtering technique (Shimizu & Hart, 1982) was used to suppress molecular ion interferences by offsetting the secondary accelerating voltage by 60 V for the REE, and 90 V for the other elements. Elemental abundances were calculated by taking ratios with respect to 30Si for the inclusions. Nineteen repeated analyses, some at times separated by 2 years, were used to estimate the standard deviation of the REE concentrations measured by the ion probe. All error estimates that are listed and plotted correspond to 1 SD. The standard deviation depends on the concentration, and varies from 5% for Ce to 15% for Eu (see Table 2). The concentrations of SiO2, TiO2, Al2O3, FeO, MnO, MgO, CaO, Na2O, K2O and P2O5 in the melt inclusions and their host phenocrysts were determined at Cambridge using a Cameca SX50 electron microprobe in energy dispersive mode with a count time of 50 s. The beam was focused to a spot of 5 µm, using a current of 3 nA and 20 keV for the melt inclusions, and 10 nA and 20 keV for the phenocrysts. The compositions of the samples that host the melt inclusions (Fig. 1b) are listed in Table 1, and those of the inclusions themselves in Table 2. All the new analyses from postglacial lavas are listed in an electronic appendix, as are those of the melt inclusions. These data can be downloaded from http://www.petrology.oupjournals.org
|
|
CRUSTAL AVERAGES |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
Mantle upwelling beneath the Northern Volcanic Zone produces the melt required to form the Icelandic crust. The line of the refraction profile of Staples et al. (1997)
is shown in Fig. 1a, and passes across the southern part of Theistareykir (Fig. 1b), where the crustal thickness is 19 km. The mean composition of the melt by volume is dominated by the Storaviti lava shield, whose volume of 30 km3 is much larger than that of any of the other flows of this segment.
The inversion scheme of McKenzie & O’Nions (1991), with the modification described by White et al. (1992) to take account of decompression melting, was used to estimate the melt fraction as a function of depth, using the mean composition of the Storaviti lavas as input and a source generated by mixing 87% of the MORB and 13% of the primitive source of McKenzie & O’Nions (1991), to give 
Nd = 8·4 (Elliott et al., 1991). Figure 3a shows the fit of this one-stage model to the observations. The fraction of material lost by fractionation was estimated using the expression of McKenzie & O’Nions (1991, 1992), which uses the concentrations of FeO and MgO. The resulting crustal thickness is 24 km when corrected for fractionation of 19%. Although the mean misfit, of 0·29 between the calculated and observed concentrations, is within 1 SD of the data, Fig. 3a shows that the misfits are systematic. The observed concentrations of La and Ce define a steep curve that is not well matched by the one-stage model. Better agreement can be obtained by mixing 10% of a small melt fraction, obtained by melting 0·4% of the MORB source in the garnet stability field, with 90% of melt produced between 110 and 16 km along the decompression curve shown in Fig. 3d. Although the melt fraction involved in the generation of this small melt fraction is not well constrained, it needs to be small to generate the steep slope between La and Pr in Fig. 3a. The calculated crustal thickness is 18·6 km when corrected for 19% fractionation, which agrees with the value of Staples et al. (1997) from seismic refraction. The melt fraction as a function of depth (Fig. 3d) also agrees with that calculated for isentropic decompression when the potential temperature is 1480°C. The seismic and geochemical observations are therefore consistent with the generation of most of the crust by passive isentropic upwelling of the mantle with a potential temperature of 1480°C. This temperature is 200°C higher than that beneath a normal ridge axis, as a result of the presence of the plume.
The origin of the small melt fraction is less obvious. Variable fractions of such melt are required to model the compositions of other Icelandic magmas, such as the Tertiary plateau basalts from eastern Iceland (Wood, 1978), and those outside the rift zone reported by Hardarson & Fitton (1997). Melt of the necessary composition could be generated by a long tail to the melt distribution extending to depths of several hundred kilometres. A more likely alternative is that the small melt fraction is generated by active plume upwelling, as it is beneath Hawaii. As Ribe et al. (1995) showed, a thin hot plume centred on the ridge beneath Iceland generates more melt than can be accommodated by the Icelandic crustal thickness. They suggested that the plume needed to be thick and cool to account for the observed variations in crustal thickness. Ito et al. (1999) instead suggested that the upwelling velocity of mantle material is reduced as melting starts, because its viscosity is increased by the loss of water. In their model active upwelling occurs below 100 km, but becomes steadily less important as the viscosity increases at shallower depths where most melt is produced. The melting model in Fig. 3, proposed to account for the composition of the Storaviti lavas, is therefore consistent with geophysical observations and with models of melt generation, and the instantaneous melt compositions generated along the decompression path in Fig. 3d can be used to discover whether the observed variations in melt compositions can be produced by mixing the melt from different depths.
|
MELT COMPOSITIONS |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
The composition of individual lava flows, of melt inclusions within phenocrysts from an individual flow, and even of individual melt inclusions within a single phenocryst, is variable. The variability can be used to constrain the physical processes involved in melt generation.
The smallest-scale variations that have been observed occur between different melt inclusions within individual phenocrysts. Three examples are illustrated in Fig. 4, which shows sections through one olivine and two clinopyroxene phenocrysts. There is no obvious petrographic difference between the two melt inclusions in Fig. 4a with different light rare earth element (LREE) compositions. Inclusions 26·2 and 26·3 are the same size and shape and have different compositions, whereas 26·1 and 26·3 are of different sizes, yet have essentially the same composition. The difference in composition between k-1 and k-2 in Fig. 4c is even larger than that between 26·2 and 26·3. Because ion probe observations of trace element concentrations of a few parts per million in inclusions of diameters of a few hundred microns or less are difficult to measure accurately, part of the difference between the measured concentrations is due to errors. Similar variations have previously been reported (Nielsen et al., 1995) from the Gorda Ridge, and appear to be common. They are not easily observed unless the phenocryst contains many inclusions, because a section that is designed to expose one melt inclusion is not in general likely to pass through other inclusions that are large enough to be probed. Probably for this reason the largest range in the concentration of Ce in individual inclusions in a single phenocryst of olivine of 8·70–1·21 ppm, and of clinopyroxene, of 7·77–2·41 ppm, is slightly smaller than the range of the composition of inclusions within an individual flow. The total observed range in Ce concentration is 8·74–1·21 ppm for olivine-hosted and 7·77–1·21 ppm for cpx-hosted inclusions.
|
Rare earth elemental concentration ratios of all the inclusions are shown in Fig. 5, from the numbered locations in Fig. 1b. Table 2 gives estimates of the measurement errors from 19 replicate analyses of olivine-hosted inclusions, which are plotted in Fig. 4. The estimated standard errors from the measurements 
r can then be removed from the observed standard deviation 0 to estimate the true standard deviation of the population t, and the signal-to-noise ratio t/r, which is listed for eight REEs. The largest value of the ratio in Table 3 is for Ce, of 7·2 for inclusions in olivine, and shows that the observed variations in concentration are real.
|
|
There is no obvious relationship between the composition of the host lavas and that of the inclusions. The lowest concentration of La in inclusions are found in 9435 from Borgarhraun (0·30 ppm), 9394 from Langaviti (0·27 ppm) and 9390 from a picrite (0·24 ppm), whereas the picrites have the lowest average host rock La concentrations, of 0·76 ppm. The highest La concentrations in inclusions come from 9335 from Borgarhraun (3·37 ppm) and from another phenocryst in the same picrite sample, 9390 (3·46 ppm), that yielded the lowest bulk-rock value of La. The mean concentrations of the olivine-hosted inclusions are similar to those of the host lavas in Fig. 5a and b, but that of clinopyroxene-hosted inclusions is not. The obvious explanation of this difference is that some exchange of trace elements has occurred between the host crystal and the inclusion. However, such exchange is difficult to reconcile with the existence of inclusions such as 9409-k-2 (Fig. 4c) that have much greater concentrations of trace elements than do otherwise identical inclusions in the same crystal.
An important concept in discussions of variability within lava flows is that of the correlation length. If the correlation between the composition of samples taken at different separations is measured as a function of their separation, the correlation length is the smallest separation at which the compositions are uncorrelated. Unfortunately, the correlation length cannot be estimated from the compositional variations within single phenocrysts, because the spatial variation reflects the entrapment process, not the spatial variation within the melt. Nor can it yet be estimated from compositional variations between samples from the flows in Fig. 1b, because there are insufficient samples available.
It is, however, possible to obtain some information from the variations in composition. The Kolmogorov–Smirnov test (Press et al., 1992) can be used to discover whether the inclusions grouped by the mineral phase in which they occur could be samples of a single population. When this test was applied to the concentrations of La and Ce, measured in inclusions in olivine, clinopyroxene and plagioclase, the probability that the inclusions in different hosts were samples of the same population was <1·1%. This observation is consistent with the plots of concentration ratios shown in Fig. 5, which show that all but one of the melt inclusions in clinopyroxene have lower concentrations of all REE than the average of the olivine-hosted melt inclusions. These statistical results are not easy to interpret. Inclusions in olivine and in plagioclase are unlikely to have exchanged REE elements with their hosts (except Eu in plagioclase), and therefore the statistical differences between the populations of inclusions in these two minerals are likely to have resulted from the growth of olivine and plagioclase from melts of different composition. Because Mg and Fe exchange with the host olivine, and Al and Ca with the host plagioclase, it is not possible to test whether the major element concentrations of the melt inclusions also differ between the various types of phenocryst.
If some reasonable assumptions are made, the standard deviation t of the elemental concentrations of the melt inclusion can also be used to estimate the volume of melt, Vi, whose composition is represented by the melt inclusions. If the host lava flows are produced by aggregating N melt volumes of volume Vi, the standard deviation between the mean compositions of the hosts, mt, is given by the standard relationship mt = t/• N. Values of t can be estimated from the ion probe measurements and are listed in Table 3. They are more reliable for the more incompatible elements such as La and Ce, where the signal-to-noise ratio is large, than for the heavy rare earth elements (HREEs) whose compositional variations are smaller. In the case of both the melt inclusions and host lavas, the true variability can be obtained from the observed variability by using repeats to estimate the contribution of measurement errors (Tables 1 and 2). For the purposes of such calculations systematic errors are of little importance, as it is the value of the standard deviation, not the mean, that is of interest. The best estimate of Vi is 0·06 km3, and is comparable with the volume of the smaller Theistareykir flows, such as Arnahvammurhraun and Höfudhreidharmuli, with volumes of 0·05 km3 or less. The same calculation was not carried out for the clinopyroxene-hosted inclusions, because the standard deviations are dominated by the concentrations of a single inclusion, k-2 in Fig. 4c.
How are such variations in the compositions of melt inclusions produced? They cannot be generated by fractional crystallization, because all the REEs except Eu are incompatible in olivine, clinopyroxene and plagioclase during crystallization. Their absolute concentrations, but not their ratios, can therefore be changed appreciably by fractionation. In contrast, the observed concentrations of La and Ce are much more variable than are those of the HREEs such as Er and Yb. The obvious explanation of this behaviour is that it results from fractional melting, as Nielsen et al. (1995), Gurenko & Chaussidon (1995), Sobolev (1996) and Shimizu (1998) have proposed. Those workers used plots of various concentration ratios to show that mixtures of instantaneous fractional melts from different depths could account for the observations. This proposal is also consistent with the average composition of the host lavas and melt inclusions shown in Fig. 5a and b, which are similar, with both showing higher concentration ratios of the LREEs La and Ce with respect to the heavier elements. The small difference in the mean compositions of the inclusions and the host lavas is not likely to be due to random errors in the ion probe measurements, because the measured concentrations of all elements are systematically lower in the melt inclusions than they are in the host lavas. A similar difference was reported by Sours-Page et al. (1999).
Table 4i shows the matrix R for the olivine-hosted melt inclusions, with those matrix elements whose values exceed 0·6 emphasized in bold face. La, Ce and Nd correlate well with each other, and with Zr. The middle rare earth elements (MREEs) Sm and Eu correlate with each other and with Ti, whereas the HREEs Dy, Er and Yb correlate well with each other but not with the LREEs and MREEs. The corresponding matrix for the host lavas is listed in Table 5, and shows similar but less extreme behaviour. It is clear from this pattern that there must be at least two independent effects that control the elemental concentrations in the melt inclusions, one that primarily affects the LREEs, the other the HREEs. The behaviour of Zr and Ti is similar to those rare earths with similar partition coefficients, and the other trace elements and major elements (not shown) show no obvious patterns.
|
|
The compositions of the olivine-hosted inclusions, ratioed with respect to the MORB source, are shown in Fig. 6, projected onto the plane containing their two largest principal components (Table 6i). The standard deviations estimated by projecting the repeat analyses onto this plane are shown as an ellipse labelled S.D. Figure 6d shows the contributions that individual elements make to eigenvectors 1 and 2 (see Table 6i). The observed standard deviation of second principal component is 1·69, compared with 0·68 from the repeat analyses. The true standard deviation is therefore 1·55 and the signal-to-noise ratio is 2·3. That this variability is real is also shown by the correlation coefficients for the HREEs in Table 4. It is less clear that the third, Fig. 6c, is significant, although the standard deviation of the points shown is 1·03, compared with 0·65 for the repeat analyses. These values suggest that the real standard deviation of the third eigenvector is 0·8 and that the variation it represents is a real feature of the analyses. Gurenko & Chaussidon (1995)
found similar variability in olivine-hosted melt inclusions from SW Iceland, which they argued resulted from mixing between two end-member compositions. Figure 6a shows that their proposal cannot account for the composition of the Theistareykir inclusions, because mixing between two end members must always produce a straight line in principal component space, irrespective of the composition of the end members. Therefore at least three end members are required to account for the distribution in Fig. 6a, and at least four if the variability in eigenvector 3 is real, as seems likely.
|
|
If the variability in the composition of melt inclusions does in fact result from incomplete mixing of fractional melts, infinitely many end members are present and it is not surprising that at least four are required by the ion probe data. A simple way to study such mixing is to plot the path traced out by the instantaneous melt compositions in the principal component plot. Figure 6a, b and f shows such a path, calculated from the melt generation curve for Storaviti in Fig. 3d. The shape of the curve is easily understood from the composition of the instantaneous melt, shown in the boxes in Fig. 6f. When mantle material first begins to melt at the bottom of the melting zone, the resulting melt, shown in the box for 132 km, contains high concentrations of La and other LREEs. This part of the path has large values of eigenvector 1, and lies far to the right of Fig. 6a and b. As melting continues, the incompatible elements become depleted and the path moves to the left. The boxes showing the compositions at depths of 132, 120 and 100 km clearly show this behaviour. However, until garnet breaks down to spinel, the HREE concentrations are buffered. When the last garnet has converted to spinel, these elements become incompatible and are removed by the melt. It is this effect that increases the concentration of the HREEs at a depth of 80 km, and that produces the steep part of the path as it approaches the origin. This behaviour cannot be modelled using Shaw’s (1970) expressions. The concentrations of all the rare earths fall rapidly as melting continues in the spinel stability field (see box for 66 km). The cross-hatched region in Fig. 6a, b and f shows those compositions that can be generated by mixing instantaneous melts from different depths. As all melt inclusions in olivine and clinopyroxene lie within this area, they can all be generated by this process.
One further test can be applied, which is concerned with the abundances of inclusions with different compositions. The observed values of the cross- and auto-correlations between the samples in the plane of the two largest principal components can be calculated from equation (6), and compared with the values obtained from the melting model using equation (8). The ellipse that represents the calculated correlation is shown by the continuous curve in Fig. 6e, and those from the observations by a dotted ellipses in Fig. 6c and e. As the orientation of the principal axes is constrained by the data, the principal axes of the dotted ellipses must lie along the principal components, as is indeed the case. But there is no similar constraint on the orientation of the calculated ellipse. As Fig. 6e shows, its principal axes in fact lie along the principal components of the data, and the ratio of its principal axes is 4·9, compared with an observed ratio of 3·1. This agreement is therefore consistent with a mixing model. However, the observed variability is a factor of between 2·7 and 4·2 less than that calculated. This difference shows that the melts trapped in melt inclusions are not themselves instantaneous melts, but could be mixtures of such melts from different depths. The failure of the melt inclusions to plot on the instantaneous melting curve also suggests that mixing has occurred. The value of the mixing parameter M, given by equation (10), is 0·76 for the melt inclusions in Fig. 6a.
The variation in the compositions of the lavas hosting the melt inclusions (Fig. 7) is similar to, although somewhat smaller than, that of the inclusions themselves. The matrix of cross-correlation coefficients (Table 5) is also similar, with high correlations between individual LREEs and HREEs, but smaller values between the light and heavy rare earths. This pattern leads to similar principal component vectors to those of the melt inclusions (Table 6iii). The first component is again controlled by the total melt fraction and the second by garnet. The principal components from the olivine-hosted inclusions, rather than those from the host lavas themselves, were used for the projections shown in Fig. 8. As with the melt inclusions, the cross-hatched region shows that the observed variations can be produced by mixing instantaneous melts resulting from decompression melting. The value of the mixing parameter M is 0·82 for the lavas hosting the melt inclusions, and is 0·78 for all the postglacial lavas.
|
|
There is no obvious way in which the geochemical observations can be used to constrain the physical processes that generate the lava compositions from those in the instantaneous melts. A simple way to model the mixing of instantaneous melts is to require the melt fraction present in the source to exceed some fraction 
0 before it can separate from its residue. The effect of such a melt fraction on the bulk partition coefficient is easily modelled by including a modal fraction 0 of a phase whose partition coefficient is unity for all elements. Although such a mixing scheme is easy to implement, it is not likely to be a good model of the physical processes that occur within the source regions. In the geological problem mixing must be controlled by the fluid dynamics of two-phase flow. There is therefore no reason to believe that a batch of melt will prefer to mix with that produced from an adjacent region, as such a calculation implies. Figure 9a and b shows the resulting paths, projected onto the plane containing the two largest principal components of the melt inclusions. The values of 0 were varied until the calculated and observed distributions, represented by the variability ellipses, were approximately the same size. As expected, the host lava compositions require a larger value of 0, and therefore more mixing, than do the melt inclusions. It is surprising that the observed and calculated ellipses are so similar, because the real process that mixes melts from different depths is not likely to be well modelled by this simple calculation.
|
The mean and standard deviation of the principal components of the melt inclusions and their host basalts can be used to estimate the volume of the melt batches from which the inclusions were taken, in exactly the same way as before. The results are shown in Table 3. The estimate of 0·04 km3 from component 1 is likely to be more reliable than that from any of the elements used individually.
One final issue concerns self-consistency: are the volumes required in the mixing models consistent with decompression melting? The standard deviation of the first principal component for the melt inclusions is about a factor of four less than that calculated for the instantaneous melt. As the volume of the batches of melt being trapped by the inclusions is estimated to be 0·04 km3, and their standard deviation is about 1/4 that of the instantaneous melt, the volume of the batches of instantaneous melt that must be mixed must be 0·04/16 km3, or 3 x 10-3 km3. If this volume is produced by melting within a cylinder of mantle of height r and radius r, its volume is 
r3. Decompression melting generates 0·3% melt for every kilometre of decompression. The average melt fraction within the cylinder is therefore 1·5 x 10-3r if r is measured in kilometres. An estimate for r can therefore be obtained from
 |
1 km. The melt composition extracted from this small volume will be close to the instantaneous melt composition for elements whose partition coefficients are 3 x 10-3 or more. Those for the rare earths satisfy this condition. The simple conclusion is that all the observed melt compositions can be produced by mixing instantaneous melts generated by fractional decompression melting along the path shown in Fig. 3d. Although this model is certainly nonunique, it can account for the elemental concentrations in the melt inclusions, and their host lavas, as well as the mean composition of the lavas from Theistareykir.
|
MORB MELTING |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
The melting process beneath a normal ridge axis is likely to be similar to that beneath Iceland, because both generate a magnesium-rich basalt by isentropic decompression of mantle peridotite. Therefore any model proposed to account for melting beneath Theistareykir should be able to describe melting beneath a normal ridge axis. Figure 10a, c and e shows melt compositions from olivine-hosted melt inclusions from the FAMOUS area (Shimizu, 1998
) of the Mid-Atlantic Ridge and plagioclase-hosted inclusions from the Gorda and Juan de Fuca Ridges in the NE Pacific (Nielsen et al., 1995; Sours-Page et al., 1999). The principal components of these distributions are listed in Table 6iv–vi, and are similar to those of the Theistareykir inclusions. Like them, the principal components from the FAMOUS inclusions are little affected by whether or not the Zr and Ti concentrations are included. In Fig. 10a the path is the melting path of R25, from White et al. (1992), obtained from the composition of lavas drilled in the FAMOUS area, whereas that used in Fig. 10c and e is from a ridge segment with a normal crustal thickness (Lawson, 1996). The compositions of the inclusions can be generated by mixing instantaneous melts, as before. However, the variability of the melt inclusions from FAMOUS is not well reproduced by a simple mixing model that retains a melt fraction 0 in the source region. Figure 10b illustrates an attempt to do so, and shows that the observed and calculated ellipses do not have the same orientation, although their shapes are similar. This disagreement is not surprising. What is more important is that Fig. 10 shows that the compositions of the inclusions from all three regions can be generated by mixing instantaneous melts. As for Theistareykir, the size of the second eigenvalue is larger than the estimated standard deviation of the measurements, so at least three end members are required.
|
The cross-correlation coefficients listed in Table 4 for the olivine-hosted inclusions from Theistareykir and the FAMOUS area show no correlation between the HREEs and LREEs, unlike those in plagioclase from the Gorda and Juan de Fuca Ridges. A possible explanation for this difference is that the depth at which melting starts is greater in plumes than it is beneath normal ridges. As Fig. 6f shows, the LREEs and HREEs are efficiently fractionated from each other by plume melting. This effect is less important for melting beneath normal ridges, because less melt is generated in the garnet stability field. Alternatively, the difference may instead result from the difference in host phenocryst, for which the melting models provide no obvious explanation.
In the case of the Juan de Fuca Ridge the conclusion that the composition of the melt inclusions can be generated by mixing instantaneous melts disagrees with that of Sours-Page et al. (1999). They attempted to fit their observations with a number of different models, but were unable to generate the composition of the inclusions in NMORB and EMORB plagioclase phenocrysts by fractional melting of a single homogeneous source. Their approach did not allow them to model mixing as easily as does that used here, and perhaps for this reason they did not discover that mixing could also account for their observations. Although preferential melting of clinopyroxene veins can produce the enriched melt compositions, such a model is not required by their measurements.
Although no samples of residual mantle have yet been discovered in Iceland, many abyssal peridotites have been dredged from ridges. Johnson et al. (1990) carried out a detailed study of the rare element concentrations in clinopyroxenes in such abyssal peridotites, as well as measuring their modes. Their samples came from fracture zones on the Southwest Indian Ridge and on the Atlantic–Antarctic Ridge. For the purposes of modelling, their data were divided into three groups: those with the highest, average and lowest LREE concentrations. The fits are shown in Fig. 11e–g and calculated and observed modes in Fig. 11b–d. As Fig. 11a shows, the melt models from the REE concentrations in clinopyroxenes from abyssal peridotites agree well with the deeper part of the melting model obtained by inversion from a collection of basalt analyses from the Southwest Indian Ridge (P. Janney, personal communication, 1999). As Fig. 11f shows, the observed REE concentrations, especially those of Ce, Nd and Sm, strongly constrain 0. As Johnson et al. (1990) remarked, to produce the strong depletion of La without removing large fractions of Er and Yb requires the value of 0 to be <1% (Fig. 11f), and the values estimated here are even smaller.
|
The instantaneous melt compositions that are in equilibrium with the three groups of abyssal peridotite are shown in Fig. 10d, together with the path from the melting model in Fig. 11a. Unlike the composition of the melt inclusions, those of melts in equilibrium with the clinopyroxenes from the abyssal peridotites closely resemble those of instantaneous melts, and therefore plot close to the melting path in Fig. 10d. For the same reason they constrain
0, the melt fraction present during fractional melting, to be small. Figure 11f shows that the value of 0 of 5% required to model the compositional variations of the melt inclusions is not compatible with the observed compositions of abyssal clinopyroxenes. This result requires the mixing to occur after the instantaneous melts have ceased to be in equilibrium with their residues. |
TEMPERATURE AND DEPTH OF INCLUSION FORMATION |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
Melt inclusions within phenocrysts are believed to form from skeletal overgrowth, followed by textural equilibration. As is commonly the case, in Theistareykir textural equilibration after entrapment has produced elliptical inclusions, illustrated in Fig. 4. Their shapes are therefore of little help in understanding how entrapment occurs. However, the melt inclusions in plagioclase plotted in Fig. 10c show compositional variations related to zoning in the plagioclase crystals (Nielsen et al., 1995
), and therefore formed from melts of variable composition that came in contact with a single crystal. It is straightforward to estimate the temperatures and pressures at which the Theistareykir inclusions became trapped from the clinopyroxene compositions and that of their hosts if they are in chemical equilibrium. The difficulty is that the variable composition of the inclusions in a single phenocryst shows that it cannot have been in equilibrium with all the melts that it trapped. It is therefore unclear whether such temperature and pressure estimates have any meaning. None the less, they are calculated here because at present they provide the only estimates of the depth of entrapment.
Figure 12a shows pressure and temperature estimates from the clinopyroxene compositions and that of their host lavas using the method proposed by Putirka et al. (1996). The pressures are larger than those in the crust, and the temperatures are close to the solidus. The temperature estimates are lower than those expected for isentropic decompression with TP = 1480°C, and therefore some heat must have been lost, either by solid-state conduction or by hydrothermal circulation. That some heat loss has occurred is also suggested by the presence of the phenocrysts themselves. The pressure estimates suggest that the large range in melt composition seen in the inclusions is present at depths as shallow as 30 km. The mixing model also provides a constraint on the depth at which melt inclusions can form. The most depleted instantaneous melt compositions are produced only at depths shallower than 70 km (Fig. 6), and such compositions are required as one of the end members of the mixing model. Melt inclusions must therefore form at shallower depths, as melt must always move upwards. The depth estimates from the clinopyroxene phenocrysts satisfy this requirement. Figure 12b shows the depth estimates plotted on a section constructed from seismic refraction profiles (Staples et al., 1997). It suggests that the phenocrysts form in the mantle, a location that is consistent with their high values of mg-number (Table 2).
|
|
DISCUSSION AND CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
This study of the volumes and compositions of the Theistareykir magmas, and of their phenocrysts and melt inclusions, has provided a considerable amount of information about the processes of melt generation and transport. The average melt composition is dominated by that of Storaviti, the large shield volcano with a volume of
30 km3, and the crustal thickness is 19 km. The mean composition and the crustal thickness are those expected if melt is generated by isentropic decompression of mantle whose potential temperature is 1480°C. The composition of the lavas produced by this single ridge segment is rather variable, and that of the melt inclusions in the olivine, clinopyroxene and plagioclase phenocrysts is even more so. This difference in variability can be exploited to estimate the mean volume of the batches of melt represented by the melt inclusions, which is 0·04 km3. The variability of the melt inclusions is a factor of about four less than that calculated for the instantaneous melt, generated by fractional melting with no residual porosity. Principal component analysis shows that the observed compositions of the melt inclusions can be generated by mixing instantaneous melts. Such mixing probably occurs after the melts separate from their residues, because various arguments suggest that the melt fraction present during melting is less than 0·6%. The first argument depends on the composition of the clinopyroxenes in abyssal peridotites, and is discussed above. The second argument involves the observed disequilibrium between various parent and daughter isotopes of the 238U and 235U decay series. The last argument depends on the fluid dynamics of compaction, which suggests that melt fractions that exceed a few tenths of a percent will be rapidly expelled from a partially melted rock beneath a ridge.
The surprising and striking feature of melt inclusions is that their compositions are so variable. It is difficult to understand how such variability could exist if melt moved upward by porous flow, which would cause melts from different depths to mix. Furthermore, the percolating melt would consist of melt from different parts of the melting column, and therefore its composition would differ from that in equilibrium with the clinopyroxenes from abyssal peridotites. These difficulties are avoided if the melt moves upwards in channels. Because the movement is fast, and because few channels are required, little interaction with the residue occurs. Furthermore, channels whose width is a few millimetres can transport melt sufficiently fast to satisfy the U-series disequilibrium observations, even if 210Pb, with a half life of 22 a, is out of equilibrium with 226Ra (Oversby & Gast, 1968; Sigmarsson, 1996).
Perhaps the most important result to come from this study concerns the importance of magma mixing. Despite the small compositional variations present in MORBs, especially those from fast-spreading ridges, it has been suspected for some time that melting occurs by fractional melting. To achieve such uniformity, starting from fractional melts with very variable compositions, requires the mixing to be universal and to be extremely efficient. In Theistareykir some basic flows, such as Borgarhraun, were not well mixed when they were erupted. Their variability is, however, less than that of the melt inclusions, which is in turn less than that expected for instantaneous melts. It is therefore clear that any approach to modelling these variations needs to be concerned with mixing, rather than with the evolution of closed systems in thermodynamic equilibrium. Indeed, there is little indication that equilibrium processes control the composition of the lavas from Theistareykir. The consequences of mixing are easily discussed in principal component space, which can provide a simple intuitive representation of the observations, the models, and perhaps most importantly, the consequences of mixing.
This study shows that all the variability in the melt inclusions from Theistareykir, and from three other suites of samples, from FAMOUS, the Gorda and the Juan de Fuca Ridges, can be modelled in the same way, by melting a homogeneous source and mixing the resulting instantaneous melts. Therefore the elemental concentrations do not require any variations in source composition. That the mantle is not in fact homogeneous is clear from the observed correlation between trace element concentrations and isotopic ratios (Elliott et al., 1991; Hémond et al., 1993). However, this study shows that only the isotopic observations require such heterogeneity, and that the correlation between isotopic ratios and elemental concentrations is at present the only way of extracting information about variations in source composition. Extensive measurements of the isotopic ratios of many parent–daughter systems are at present being carried out by a number of groups on the samples discussed above.
|
ACKNOWLEDGEMENTS |
|---|
We would like to thank T. Elliott, G. Fitton, P. Janney, R. Nielsen, K. Putirka and A. Stracke for their help. The REE concentrations were obtained using the Natural Environment Research Council ICP-MS facility at the Centre for Analytical Research in the Environment, Silwood Park, Ascot. This research was supported by the Natural Environment Research Council, the Royal Society, and a Shell studentship to L. Slater. This paper is Earth Sciences Contribution 5998.
|
FOOTNOTES |
|---|
*Extended dataset can be found at: http://www.petrology.oupjournals.org
Present address: Amerada Hess Ltd, 33 Grosvenor Place, London SW1X 7HY, IK.
Corresponding author. Fax: +44-1223-360779. E-mail: mckenzie{at}esc.cam.ac.uk
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONUSEFUL THEORETICAL IDEASANALYTICAL METHODSCRUSTAL AVERAGESMELT COMPOSITIONSMORB MELTINGTEMPERATURE AND DEPTH OF...DISCUSSION AND CONCLUSIONSREFERENCES |
|---|
Aharonov, E., Whitehead, J. A., Kelemen, P. B. & Spiegelman, M. (1995). Channelling instability of upwelling melt in the mantle. Journal of Geophysical Research 100, 20433–20450.
Albarède, F. (1995). Introduction to Geochemical Modelling. Cambridge: Cambridge University Press.
Albarède, F. (1996). High-resolution geochemical stratigraphy of Mauna Kea flows from Hawaii Scientific Drilling Project core. Journal of Geophysical Research 101, 11841–11853.
Allègre, C. J., Hamelin, B., Provost, A. & Dupré, B. (1987). Topology in isotopic multispace and origin of mantle chemical heterogeneities. Earth and Planetary Science Letters 81, 319–337.
Allègre, C. J., Schiano, P. & Lewin, E. (1995). Differences between oceanic basalts by multitrace element ratio topology. Earth and Planetary Science Letters 129, 1–12.[Web of Science]
Barcilon, V. & Lovera, O. (1989). Solitary waves in magma dynamics. Journal of Fluid Mechanics 204, 121–133.[Web of Science]
Barcilon, V. & Richter, F. M. (1986). Non-linear waves in compacting media. Journal of Fluid Mechanics 164, 429–448.[Web of Science]
Brandsdottir, B., Menke, W., Einarsson, P., White, R. S. & Staples, R. K. (1997). Faroe–Iceland Ridge experiment 2. Crustal structure of the Krafla central volcano. Journal of Geophysical Research 102, 7867–7886.
Elliott, T. R., Hawkesworth, C. J. & Grönvold, K. (1991). Dynamic melting of the Iceland plume. Nature 351, 201–206.
Fitton, J. G., Saunders, A. D., Larsen, L. M., Hardarson, B. S. & Norry, M. J. (1998). Volcanic rocks from the southeast Greenland margin at 63°N: composition, petrogenesis and mantle sources. In: Saunders, A. D., Larsen, H. C. & Wise, S. H. (eds) Proceedings of the Ocean Drilling Program, Scientific Results, 152. College Station, TX: Ocean Drilling Program, pp. 331–350.
Gurenko, A. A. & Chaussidon, M. (1995). Enriched and depleted primitive melts included in olivine from Icelandic tholeiites: origin by continuous melting of a single mantle column. Geochimica et Cosmochimica Acta 59, 2905–2917.[Web of Science]
Hardarson, B. S. & Fitton, J. G. (1997). Mechanisms of crustal accretion in Iceland. Geology 25, 1043–1046.
Hémond, C., Arndt, N. T., Lichtenstein, U. & Hofmann, W. (1993). The heterogeneous Icelandic Plume Nd–Sr–O isotopes and trace element constraints. Journal of Geophysical Research 98, 15833–15850.
Ito, G., Shen, Y., Hirth, G. & Wolfe, C. (1999). Mantle flow, melting, and dehydration of the Iceland mantle plume. Earth and Planetary Science Letters 165, 81–96.
Johnson, K. T., Dick, H. J. B. & Shimizu, N. (1990). Melting in the oceanic upper mantle: an ion microprobe study of diopsides in abyssal peridotites. Journal of Geophysical Research 95, 2661–2678.
Jull, M. G. & McKenzie, D. (1996). The effect of deglaciation on mantle melting beneath Iceland. Journal of Geophysical Research 101, 21815–21828.
Kelemen, P. B., Hirth, G., Shimizu, N., Spiegelman, M. & Dick, H. J. B. (1997). A review of melt migration processes in the adiabatically upwelling mantle beneath oceanic spreading ridges. Philosophical Transactions of the Royal Society of London, Series A 355, 283–318.
Langmuir, C. H., Klein, E. M. & Plank, T. (1992). Petrological systematics of mid-ocean ridge basalts: constraints on melt generation beneath ocean ridges. In: Phipps Morgan, J., Blackman, D. K. & Sinton, J. M. (eds) Mantle Flow and Melt Generation at Mid-Ocean Ridges. Geophysical Monograph, American Geophysical Union 71, 183–280.
Lawson, N. K. (1996). Crustal accretion near ridge–transform intersections: Kane Fracture Zone, Mid-Atlantic Ridge. Ph.D. thesis, University of Durham.
Le Maitre, R. W. (1968). Chemical variation within and between volcanic rock series—a statistical approach. Journal of Petrology 9, 220–252.
McKenzie, D. (2000). Constraints on melt generation and transport from U-series activity ratios. Chemical Geology 162, 81–94.
McKenzie, D. & O’Nions, R. K. (1991). Partial melt distributions from inversion of rare earth element concentrations. Journal of Petrology 32, 1021–1091.
McKenzie, D. & O’Nions, R. K. (1992). Correction to fractionation expression. Journal of Petrology 33, 1453.
McKenzie, D. & O’Nions, R. K. (1995). The source regions of ocean island basalts. Journal of Petrology 36, 133–159.
Nielsen, R. L., Crum, J., Bourgeois, R., Hascall, K., Forsythe, L. M., Fisk, M. R. & Christie, D. M. (1995). Melt inclusions in high-An plagioclase from the Gorda Ridge: an example of local diversity of MORB parent magmas. Contributions to Mineralogy and Petrology 122, 34–50.
O’Nions, R. K., Pankhurst, R. J. & Grönvold, K. (1976). Nature and development of basalt magma sources beneath Iceland and the Reykjanes Ridge. Journal of Petrology 17, 315–338.
Oversby, V. M. & Gast, P. W. (1968). Lead isotope compositions and uranium decay series disequilibrium in Recent volcanic rocks. Earth and Planetary Science Letters 5, 199–206.
Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. (1992). Numerical Recipes. Cambridge: Cambridge University Press.
Putirka, K., Johnson, M., Kinzler, R., Longhi, J. & Walker, D. (1996). Thermobarometry of mafic igneous rocks based on clinopyroxene–liquid equilibria, 0–30 kbar. Contributions to Mineralogy and Petrology 123, 92–108.[Web of Science]
Ribe, N., Christensen, U. R. & Theissing, J. (1995). The dynamics of plume–ridge interaction 1: ridge-centered plumes. Earth and Planetary Science Letters 134, 155–168.
Richardson, C. N. & McKenzie, D. (1994). Radioactive disequilibria from two-dimensional models of melt generation by plumes and ridges. Earth and Planetary Science Letters 128, 425–437.
Shaw, D. M. (1970). Trace element fractionation during anatexis. Geochimica et Cosmochimica Acta 34, 237–243.[Web of Science]
Shimizu, N. (1998). The geochemistry of olivine-hosted melt inclusions in a FAMOUS basalt ALV519-4-1. Physics of the Earth and Planetary Interiors 107, 183–201.
Shimizu, N. & Hart, S. R. (1982). Applications of the ion microprobe to geochemistry and cosmochemistry. Annual Review of Earth and Planetary Sciences 10, 483–526.[Web of Science]
Sigmarsson, O. (1996). Short magma chamber residence time at an Icelandic volcano inferred from U-series disequilibria. Nature 382, 440–442.
Slater, L. (1996). Melt generation beneath Iceland. Ph.D. thesis, Cambridge University.
Sobolev, A. V. (1996). Melt inclusions in minerals as a source of principle petrographic information. Petrology 4, 209–220.
Sours-Page, R., Johnson, K. T. M., Nielsen, R. L. & Karsten, K. L. (1999). Local and regional variation of MORB parent magmas: evidence from melt inclusions from the Endeavour Segment of the Juan de Fuca Ridge. Contributions to Mineralogy and Petrology 134, 342–363.
Staples, R. K., White, R. S., Brandsdottir, B., Menke, W., Maguire, P. K. H. & McBride, J. H. (1997). Faroe–Iceland Ridge experiment 1. Crustal structure of northeastern Iceland. Journal of Geophysical Research 102, 7849–7866.
White, R. S., McKenzie, D. & O’Nions, R. K. (1992). Oceanic crustal thickness from seismic measurements and rare earth element inversions. Journal of Geophysical Research 97, 19683–19715.
Wiggins, C. & Spiegelman, M. (1995). Magma migration and magmatic solitary waves in 3-D. Geophysical Research Letters 22, 1289–1292.
Wood, D. A. (1978). Major and trace element variations in the Tertiary lavas of eastern Iceland and their significance with respect to the Iceland geochemical anomaly. Journal of Petrology 19, 393–436.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  L. M. Larsen and A. K. Pedersen Petrology of the Paleocene Picrites and Flood Basalts on Disko and Nuussuaq, West Greenland J. Petrology, September 1, 2009; 50(9): 1667 - 1711. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Maclennan Concurrent Mixing and Cooling of Melts under Iceland J. Petrology, November 1, 2008; 49(11): 1931 - 1953. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. F. Kelley and M. Barton Pressures of Crystallization of Icelandic Magmas J. Petrology, March 1, 2008; 49(3): 465 - 492. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. J.R. Kent Melt Inclusions in Basaltic and Related Volcanic Rocks Reviews in Mineralogy and Geochemistry, January 1, 2008; 69(1): 273 - 331. [Full Text] [PDF]
|
|
|
|
 |
|
 E. ALDANMAZ, M. K. YALINIZ, A. GUCTEKIN, and M. C. GONCUOGLU Geochemical characteristics of mafic lavas from the Neotethyan ophiolites in western Turkey: implications for heterogeneous source contribution during variable stages of ocean crust generation Geological Magazine, January 1, 2008; 145(1): 37 - 54. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. J. Falloon, D. H. Green, and L. V. Danyushevsky Crystallization temperatures of tholeiite parental liquids: Implications for the existence of thermally driven mantle plumes Geological Society of America Special Papers, January 1, 2007; 430(0): 235 - 260. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. H. Natland {Delta}Nb and the role of magma mixing at the East Pacific Rise and Iceland Geological Society of America Special Papers, January 1, 2007; 430(0): 413 - 449. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. HERZBERG Partial Crystallization of Mid-Ocean Ridge Basalts in the Crust and Mantle J. Petrology, December 1, 2004; 45(12): 2389 - 2405. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
X. LIU and H. ST. C. O'NEILL The Effect of Cr2O3 on the Partial Melting of Spinel Lherzolite in the System CaO-MgO-Al2O3-SiO2-Cr2O3 at 1{middle dot}1 GPa J. Petrology, November 1, 2004; 45(11): 2261 - 2286. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. W. SCHMIDT, D. H. GREEN, and W. O. HIBBERSON Ultra-calcic Magmas Generated from Ca-depleted Mantle: an Experimental Study on the Origin of Ankaramites J. Petrology, March 1, 2004; 45(3): 531 - 554. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. HERZBERG and M. J. O'HARA Plume-Associated Ultramafic Magmas of Phanerozoic Age J. Petrology, October 1, 2002; 43(10): 1857 - 1883. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. BREDDAM Kistufell: Primitive Melt from the Iceland Mantle Plume J. Petrology, February 1, 2002; 43(2): 345 - 373. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||