Compositional Trends of Icelandic Basalts: Implications for Short–Length Scale Lithological Heterogeneity in Mantle Plumes

Article Volume 12, Number 11 18 November 2011 Q11008, doi:10.1029/2011GC003748 ISSN: 1525‐2027

Compositional trends of Icelandic basalts: Implications for short–length scale lithological heterogeneity in mantle plumes

O. Shorttle and J. Maclennan Department of Earth Sciences, University of Cambridge, Downing Street, Cambridge CB2 3EQ, UK ([email protected])

[1] Lithological variations in the mantle source regions under mid‐ocean ridges and ocean islands have been proposed to play a key role in controlling melt generation and basalt composition. Here we combine compositional observations from Icelandic basalts and modeling of melting of a bilithologic peridotite‐pyroxenite mantle to demonstrate that, while short–length scale major element variation is present in the mantle under Iceland, source heterogeneity does not make an important contribution to excess melt production. By identifying the major element characteristics of end‐member Icelandic melts, we find enriched melts to be characterized by low SiO2 and CaO, but high FeO. We quantitatively compare end‐member compositions to experimental partial melts generated from a range of lithologies, pressures and melt fractions. This comparison indicates that a single source composition cannot account for all the major element variation; depleted Icelandic melts can be produced by depleted peridotite melting, but the major element composition of enriched melts is best matched by melting of mantle sources that have been refertilized by the addition of up to 40% mid‐ocean ridge basalt. The enriched source beneath Iceland is more fusible than the source of depleted melts, and as such will be overrepresented in accumulated melts compared with its abundance in the source. Modeling of peridotite‐pyroxenite melting, combined with our observational constraints on the composition of the Icelandic mantle, indicates that crustal thickness variations in the North Atlantic must be primarily due to mantle temperature and flow field variations.

Components: 20,000 words, 14 figures, 7 tables. Keywords: Iceland; major elements; mantle heterogeneity; mantle plumes; melting model. Index Terms: 1009 Geochemistry: Geochemical modeling (3610, 8410); 1025 Geochemistry: Composition of the mantle; 1030 Geochemistry: Geochemical cycles (0330). Received 14 June 2011; Revised 11 October 2011; Accepted 12 October 2011; Published 18 November 2011.

Shorttle, O., and J. Maclennan (2011), Compositional trends of Icelandic basalts: Implications for short–length scale lithological heterogeneity in mantle plumes, Geochem. Geophys. Geosyst., 12, Q11008, doi:10.1029/2011GC003748.

1. Introduction last 15 years that a significant role for recycled oceanic crustal material in oceanic island and mid‐ [2] Subduction has long been identified as a gen- ocean ridge basalt (OIB and MORB) genesis erator of mantle isotopic and trace element het- has become widely proposed [e.g., Hauri, 1996; erogeneity [Hofmann and White, 1982; White and Hirschmann and Stolper, 1996; Niu et al., 2008; Hofmann, 1982]. However, it is only during the Sobolev et al., 2005; Herzberg, 2006; Prytulak and

Elliot, 2007; Sobolev et al., 2007; Dasgupta et al., demonstrated that small‐scale source variability is 2010; Herzberg, 2011]. In particular, these authors also present in the major elements. identify the recycled basaltic component to be [5] The aim of this work is to investigate the nature present as discrete pyroxenitic or eclogitic litholo- of lithological heterogeneity in the mantle under gies. Regions in the mantle of distinct mineralogy Iceland, and to constrain how such heterogeneity have important implications not only for models can influence melt flux in mid‐ocean ridge set- of melt generation, but also for mantle dynamics tings. In order to achieve this target, the study is and interpretation of geophysical data [e.g., Ita and divided into two parts. In the first part, we system- Stixrude, 1992; Van der Hilst and Kárason, 1999; atically assess major, trace element and isotopic Ritsema et al., 2009a, 2009b]. data from Icelandic basalts, to place constraints on [3] The identification of clear geochemical proxies the lithology of the sources producing the enriched for the involvement of varying lithologies in basalt and depleted melts. We find that while major ele- genesis remains a significant challenge. Part of the ment variability of the source beneath Iceland difficulty lies in the interpretation of basalt major is required, none of the geochemical observations element chemistry, which is both strongly affected necessitate the role of an olivine‐free lithology by low‐pressure crystallization, and the depth directly contributing to erupted melts. Instead, and extent of melting [Grove et al., 1992; Klein and variably enriched peridotites, refertilized by up to Langmuir, 1987; Langmuir et al., 1992]. As a result, 40% MORB material, are capable of generating all major element signatures inherited from the miner- the geochemical patterns observed, including the alogy of the source are difficult to uniquely dis- Ni content of olivines from four eruptions repre- tinguish. Therefore, while the trace element and sentative of the range of Icelandic melt composi- isotopic heterogeneity of the mantle has long tions. Enriched and depleted major element signals been recognized [Hart, 1971; Schilling, 1973], this exist within similar aged eruptions from a single diversity was for a long time incorporated into the volcanic system, indicating the major element classic peridotite model of Ringwood [1962]. variability in the Icelandic mantle source must be present on a short length scale. The nature of [4] Recently, Herzberg [2006, 2011] has inferred this source enrichment means the enriched lithol- the majority of the shield building phase on Hawaii ogy will be more fusible than the depleted source, to be composed of 100% pyroxenite melts, the aggregate melts will therefore be biased toward basis for this interpretation being the lower CaO melts from the deep melting, productive, enriched content of Hawaiian basalts than predicted from lithology. In the second part of this paper we partial melting of peridotite [Walter, 1998]. Simi- combine the constraints on mantle heterogeneity larly, the high Ni content of olivines found as from the first part, with the development of a new phenocrysts in Hawaiian and Icelandic basalts has model of melting a bilithological mantle, in order been taken by Sobolev et al. [2005, 2007, 2008] to to investigate the contribution of mantle composi- indicate the role of olivine‐free lithologies in their tional heterogeneity to excess melt generation mantle source regions. Focusing on Iceland, pre- around Iceland. Even allowing for melting of the vious workers have inferred the presence of recy- highly productive G2 pyroxenite [Pertermann and cled basalt in the source from incompatible trace Hirschmann, 2003b], oceanic crustal thickness element, radiogenic isotope and stable isotope variations observed cannot be produced by source compositions of Icelandic basalts [Fitton et al., enrichment alone. Mantle potential temperature (T ) 1997; Chauvel and Hémond, 2000; Thirlwall et al., p variations of at least +100°C are required, coupled 2004; McKenzie et al., 2004; Kokfelt et al., 2006; with a component of plume‐driven upwelling, to Peate et al., 2010]. The question remains, however, generate the thickest oceanic crust observed in to what extent these geochemical observations the North Atlantic. require a lithologically distinct basalt component in the mantle, as opposed to the material having [6] For 400 km of its ∼20,000 km length the Mid‐ reacted with and refertilized depleted peridotites. Atlantic Ridge becomes subaerial on Iceland. This Equally important is the scale of this heterogeneity; exposure provides a unique opportunity to gather trace elements and isotopes have been used to a high‐resolution spatial and temporal record of establish the presence of short–length scale het- mantle melting at a divergent plate boundary. We erogeneity in the Icelandic mantle [Zindler et al., have compiled a comprehensive major, trace and 1979; Slater et al., 2001; Stracke et al., 2003; isotopic data set from both onland and adjacent Maclennan, 2008a], however it remains to be submarine ridges, to characterize the geochemi-

2of32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 1. Map of Iceland showing the geographical distribution of data used in this study, with the neovolcanic zones marked in dark grey. The eruptions from which new olivine data are presented are marked: Thp, Theistareykir picrite; Gae, Gaesafjöll; Hal, Háleyjabunga; Stp, Stapafell. Inset shows Iceland in relation to the North Atlantic and the along‐ridge distribution of compiled literature data. Data sources listed in Table 6. cal variations associated with the plume and depleted melts erupting on Iceland, and determining explore their implications for lithological variation if nonperidotitic lithologies are necessary to match in the North Atlantic mantle. The list of sources the end‐member compositions. used in the Icelandic geochemical data set can be found in Appendix A, and in Figure 1 the sample distribution is mapped. 2. Characterizing the Icelandic End‐Member Melt Compositions [7] In Figure 2, the first‐order major element trends ‐ along the Mid Atlantic Ridge are displayed. Ice- 2.1. The Signal of Basalt Enrichment land lies at the extreme end of the array of global and Depletion in Iceland MORB compositions, having relatively low Na8, Si8 and high Fe8 contents (Figures 2a, 2c, and 2e, [8] Trace element and isotopic studies on Icelandic correction described in the figure caption). How- basalts have established the presence of distinct ever, there is an offset in Iceland’s position with depleted and enriched compositions [e.g., Elliott et al., respect to the array defined by most Atlantic MORB; 1991; Hémond et al., 1993; Chauvel and Hémond, with the onland average of Icelandic basalts lying 2000; Fitton et al., 2003; Thirlwall et al., 2004]. to slightly higher Na and lower Si than would be Trends of enrichment in Icelandic basalts show both expected given the elevation of the ridge axis. This incompatible trace element and isotopic enrichment offset hints at the combined role that melting correlating positively [e.g., Hémond et al., 1993; dynamics and source composition play in a plume‐ McKenzie et al., 2004; Stracke et al., 2003]. A influenced section of mid‐ocean ridge, in addition corollary of this is that trace element ratios can to Tp variations. The along‐ridge trends of Figures 2b, be used as a proxy for mantle source enrichment. 2d, and 2f show the specifics of geochemical var- This is demonstrated by Stracke et al. [2003], where iation on and away from Iceland; basalt Si content positive correlations are found between incompati- shows especially clear variation, dropping sharply ble trace element and isotopic compositions of in the volume‐averaged composition of onland basalts from Theistareykir. These relationships also basalts. These major element patterns raise the extend to the major elements, which covary with question of whether lithological heterogeneity is incompatible trace element ratios. Both extent of required in the mantle source beneath Iceland. We melting and the fusibility of the different lithologies address this possibility, first by identifying the are convolved in this pattern of joint isotopic and major element characteristics of the enriched and trace element enrichment, but despite this the iso-

3of32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 2. (a, c, and e) Icelandic major element chemistry in context with Atlantic MORB. Only samples with compositions between 7.5 and 8.5 wt% MgO have been selected so as to diminish the effects of fractional crystallization and provide an equivalent to the Klein and Langmuir [1987] global array. In calculating the onland Icelandic mean, samples have been weighted by the volume of the eruption they relate to. (b, d, and f) The major element chemistry along the plume‐influenced portion of Mid‐Atlantic Ridge, plotted as a function of radial distance from the plume center [Shorttle et al., 2010]. As before, only 7.5 to 8.5 wt% MgO samples have been selected, and onland data has been volume averaged every 50 km of ridge; vertical bars record the one sigma variation about the mean. Data sources listed in Table 6. topic data of Stracke et al. [2003] show the trend to Equally, Nb/Zr is resistant to change during frac- be consistent between different eruptions. tional crystallization; neither this process, nor mixing with partial melts of altered crust, dominate [9] The covariation of trace element and isotopic the Nb/Zr of moderately evolved basalts (MgO > enrichment in Icelandic basalts means we can use 5wt%), their compositions instead reflect an aver- Nb/Zr as a proxy for enrichment. The choice of this age of the melts derived from the mantle. There- ratio was partly pragmatic: while similar major and fore, at high MgO Nb/Zr preserves a signal of trace element relationships are obtained by taking mantle source enrichment. In this section we seek REE ratios, the availability of published major and to identify the major element variation associated trace element data on the same samples is greatest with high and low Nb/Zr of Icelandic basalts. for those elements accurately analyzed by XRF.

4of32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

2.2. Identifying Geochemical Variability as storage, mixing and crystallization average out From the Mantle the primary geochemical signals exiting the mantle. By 5 wt% MgO, mixing has almost completely [10] One method that has proved effective in cap- homogenized the basalts to a weighted mean of turing a broad spectrum of geochemical variability, ‐ the compositions input to the base of the crust is the analysis of olivine hosted melt inclusions [Maclennan, 2008b]. [Sobolev and Shimizu, 1993; Slater et al., 2001; Maclennan, 2008a]. However, while these inclu- [13] The relation of the trace element chemistry sions can faithfully record the isotopic and trace to the major elements is highlighted by the color element characteristics of the basalts in which they of data points in Figures 3 and 4. As expected, form, their preservation of major element signals is there is a positive correlation between incompati- strongly affected by post entrapment crystallization ble element concentrations (Ti, Na, K) and Nb/Zr and diffusion [Gaetani and Watson, 2000]. There- at fixed MgO. There is also a systematic relation- fore, major element compositions of melt inclusions ship between Nb/Zr and the SiO2, FeO and CaO may not give reliable constraints on variations in content of primitive melts. The trend in both the mantle source lithology. Instead, we take whole‐ NVZ and WVZ is for more enriched basalts to rock geochemical data from multiple eruptions have lower SiO2 and CaO contents, but with higher within the neovolcanic zones to capture the geo- FeO, than the depleted melts. This pattern of cou- chemical diversity of melts leaving the mantle. pled FeO and trace element enrichment has been Although a given eruption may only have limited described in Iceland by Langmuir et al. [1980], whole‐rock variability [Maclennan et al., 2003a], by Elliott et al. [1991], and Nicholson and Latin examining analyses from many eruptions, reflecting [1992], with correlations attributed to extent and melts mixed and fractionated to differing degrees, it depth of melting. However, our finding of low CaO is possible to recover a larger proportion of geo- in the most enriched melts is contrary to the find- chemical diversity. ings of previous authors [McKenzie et al., 2004]. ‐ The difference between our finding and that of [11] In constraining end member melt composi- McKenzie et al. [2004] can be attributed to con- tions it is necessary to be able to identify the role of current mixing and crystallization; if a basalt ini- mixing and crystallization processes, and isolate tially has low CaO and Nb/Zr (i.e., is depleted), those eruptions with more primitive chemistries. olivine crystallization will both lower MgO and We therefore construct plots of whole rock Nb/Zr raise CaO, while mixing with more enriched melts against MgO, and color the data points by a second will raise Nb/Zr (and to a lesser extent lower CaO). major element in order to identify major element The net effect is a trend of increasing CaO and Nb/ variation that is caused by mantle rather than Zr, purely a product of crustal processes. Presenting crustal processing. We apply this type of plot to the data as in Figures 3 and 4 controls for these track the effects of concurrent crystallization and processes, and allows the mantle derived major mixing processes, and to select melts reflecting the element signal to be identified. enriched and depleted end‐member compositions; enriched melts will have high Nb/Zr and depleted [14] Figures 3 and 4 display depleted and enriched melts low Nb/Zr, with this initial variability end‐member compositional characteristics qualita- decreasing as the melts evolve. The color of the tively, but in order to make a systematic compari- data points then demonstrates how the major ele- son with the results of experimental petrology we ment systematics relate to the enriched and depleted need to estimate compositions for each end‐member. mantle melts. We do this by taking an average of the ten percent of basalts with the highest and lowest Nb/Zr. The points 2.3. Major Element Composition of End‐ selected in this way are circled in black in Figures 3a Member Neovolcanic Zone Melts and 4a, and their means presented in Figures 3b and 4b and given in Table 1. While there is some [12] Separate plots of major and trace element data overlap between end‐member compositions, SiO2, have been constructed for the Northern and West- FeO, CaO, Na2O, K2O and TiO2 from both zones ern Volcanic Zones (NVZ & WVZ) and are shown show clear differences between the two populations in Figures 3 and 4. In each zone, basalts show of basalts selected (Table 1). evidence for concurrent mixing and crystallization, with basalts of >9.5 wt% MgO displaying the [15] A possible complication of using whole‐rock maximum diversity of Nb/Zr. Below 9.5 wt% data to compile these plots is that crystal accumu- MgO, variability of Nb/Zr decreases monotonically lation can shift whole‐rock compositions to higher

5of32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 3. Plots relating the major element chemistry of NVZ basalts to Nb/Zr. (a) Each plot shows the same data colored by a different major element. Plotting the data this way allows the processes controlling major element chemistry to be constrained. Magmas with more than 9.5 wt% MgO have experienced primarily olivine fractionation, and so their major and trace element chemistry remains largely controlled by source composition and melting processes. The distribution of grey data points, below 9.5 wt% MgO, shows predominant control by polyphase fractional crystallization and melt mixing. Circled points were selected to represent end‐member melt compositions. (b) Mean end‐ member melt compositions, with boxes marking the one sigma variation. The enriched flank zone melt composition was selected by taking the mean of the 10% of flank lavas with the highest Nb/Zr and MgO > 8.5 wt% and is shown for comparison. or lower MgO, while leaving Nb/Zr unaffected. higher pressures, than on‐axis melts [Kokfelt et al., The sensitivity of the major element patterns 2006; Peate et al., 2010]. Their aggregate compo- described in Figures 3 and 4 to crystal accumula- sitions are thus expected to be weighted toward tion is tested in Appendix B. The result is that, enriched fusible lithologies. It can be seen in while some of the variability in major element Figures 3b and 4b that the enriched flank zone compositions of high or low Nb/Zr basalts will be melts show the same systematics as the enriched due to accumulated crystals, the first‐order trends neovolcanic zone melts. We can thus conclude, that cannot be explained by the addition of olivine ± enriched basalts in Iceland are consistently char- clinopyroxene ± plagioclase. A further test of acterized by lower SiO2 and CaO, and higher FeO, whether the enriched melt signal is being correctly TiO2,Na2O and K2O than depleted melts. identified in the neovolcanic zone basalts is by [16] For the remainder of this contribution we comparison to flank zone melts. In the absence of consider only end‐member compositions of the well‐developed extensional tectonics and with a NVZ basalts when comparing to other data sets. As thicker lithosphere, flank zone melts are on average already discussed, the major element and trace thought to represent smaller degrees of melting, at

6of32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 4. Plots relating the whole‐rock major element chemistry of WVZ and REP basalts to Nb/Zr. Details as in Figure 3. element systematics between the zones are very match the range of major element compositions similar (Table 1), and so there is no significant observed between the end‐member melts, then this difference in taking either the WVZ, NVZ or an is evidence for the presence of lithological varia- average of the two. The choice of the NVZ end‐ tion in the mantle beneath Iceland. We then need member compositions was motivated by the greater to determine the nature of the lithological hetero- abundance of trace element data available from geneity this corresponds to, such as the presence NVZ basalts. The major element composition of of olivine‐free pyroxenite or clinopyroxene rich the NVZ end‐member melt is given in Table 1. peridotites.

[18] There has recently been discussion as to whether 3. Identifying the Lithologies Producing particular major element characteristics require End‐Member Icelandic Melts the presence of pyroxenite in the source [Herzberg, 2011; Putirka et al., 2011]. In particular, the low [17] With the major element characteristics of Ice- CaO content of some Hawaiian tholeiites and oli- landic depleted and enriched end‐member melts vines has been interpreted as their primary melts identified, the next step is to test whether this range being partly, or wholly, derived from pyroxenite of major element compositions is consistent with [Sobolev et al., 2007; Herzberg, 2006, 2011]. As melting of a single peridotitic source composi- found in the previous section, enriched Icelandic tion [Langmuir et al., 1980; Elliott et al., 1991; basalts are similarly distinguished by their lower Nicholson and Latin, 1992]. If varying depths CaO content than the depleted basalts. This raises and degrees of melting of a single source cannot the question of whether pyroxenite melting beneath

7of32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Table 1. Estimated Compositions of Depleted and Enriched and Flank Zone End‐Member Meltsa b depleted enriched Element Depleted 1s Enriched 1s pdepleted–enriched pNVZ–WVZ pNVZ–WVZ NVZ SiO2 48.7 0.6 47.8 0.9 0.004 0.261 0.218 TiO2 0.69 0.11 1.61 0.28 0.000 0.001 0.945 Al2O3 14.9 0.9 14.8 0.6 0.531 0.025 0.379 FeO 9.2 0.6 11.3 0.7 0.000 0.009 0.379 MgO 11.8 1.2 10.7 1.2 0.011 0.888 0.925 CaO 12.9 0.6 11.2 0.6 0.000 0.228 0.121 Na2O 1.7 0.2 2.02 0.2 0.000 0.088 0.075 K2O 0.04 0.01 0.23 0.09 0.000 0.000 0.002

WVZ SiO2 48.3 0.5 47.8 0.5 0.087 ‐‐ TiO2 0.56 0.13 1.63 0.14 0.000 ‐‐ Al2O3 15.7 0.9 14.3 0.9 0.002 ‐‐ FeO 8.8 0.6 11.2 0.2 0.000 ‐‐ MgO 11.8 1.7 11.1 1.6 0.489 ‐‐ CaO 13.1 0.6 11.4 0.5 0.000 ‐‐ Na2O 1.5 0.2 1.9 0.1 0.000 ‐‐ K2O 0.01 0.01 0.27 0.06 0.000 ‐‐

Flank Zone SiO2 ‐‐47.5 0.7 ‐‐‐ TiO2 ‐‐2.05 0.4 ‐‐‐ Al2O3 ‐‐14.2 1.0 ‐‐‐ FeO ‐‐10.3 0.6 ‐‐‐ MgO ‐‐11.1 1.8 ‐‐‐ CaO ‐‐11.4 0.3 ‐‐‐ Na2O ‐‐2.1 0.2 ‐‐‐ K2O ‐‐0.81 0.16 ‐‐‐ aFor both zones the p values indicate the significance of the compositional difference between the enriched and depleted end‐members. bThe p value significance estimate from a Kolmogorov‐Smirnov test, returning the probability that the two data sets being compared are drawn from the same distribution [Press et al., 1992, 757–1152].

Iceland should be invoked to explain the major [20] This type of analysis, comparing directly exper- element systematics of the enriched basalts. imental and natural melts, requires melts produced from a given lithology for each melting scenario to [19] For a pyroxenite lithology to be a plausible be compositionally alike, despite the difference in mantle source, its melts ought to provide a better melting style. If isobaric batch melting produces fit to the observed basalt chemistry than that melts radically different in composition from the offered by a lherzolite partial melt. Importantly, the polybaric fractional case, then to see through this pyroxenite partial melt must improve the total fit process difference to major element heterogeneity across all of the major elements, and not just CaO; in the source is hopeless. We address this issue in given the along‐ridge patterns of major elements in Appendix C, and conclude that for Iceland at least, Figure 2, and the differences between the enriched given the lithologies that provide the best fitting and depleted end‐members in Figure 3, it is evident melt compositions, the difference between experi- that suitability of source lithology needs to at least ment and nature should not be influencing our be assessed in the CaO‐FeO‐MgO‐Al O ‐SiO 2 3 2 identification of source heterogeneity. (CFMAS) system. The approach we take is therefore to systematically and quantitatively compare the end‐member melt compositions with the results 3.1. Method of Calculating Misfit Between of a wide range of partial melting experiments, for End‐Members and Experimental Melts each of the major element oxides. Quantification [21] We first compiled a data set of 377 partial of the misfit between experimental partial melts melting experiments with lherzolite, eclogite, pyrox- and the Icelandic end‐member compositions then enite and harzburgite starting materials, over a wide allows for selection of those lithologies with pressure range, and with hydrous and carbonated experimental melts closest to the composition of experiments incorporated. The full list of experi- basalts on Iceland. mental data sources is presented in Appendix A,

8of32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G with a description of how the data were filtered. 3.2. Comparison of End‐Member The misfit between melts in this data set and the Compositions With Experimental Partial Icelandic end‐member compositions is determined Melts for CFMAS and CFNKTMAS (CaO‐FeO‐Na2O‐ K O‐TiO ‐MgO‐Al O ‐SiO ) separately. Calcula- [27] The results of quantifying misfit between Ice- 2 2 2 3 2 ‐ tions proceed according to the following steps. landic end member compositions and melts from the experimental database are recorded in Table 2. [22] 1. An experimental partial melt is compared Only the melts with a summed misfit less than the oxide by oxide to the end‐member melt compo- total 2s propagated uncertainty are included. The sition. The absolute differences between oxides table also provides the experimental conditions for x (in wt%) of the experimental partial melt (w ) and each melt composition. Results are divided up first, 0 the end‐member composition (w ) are summed, according to whether comparison is to the neo- giving an initial total misfit (M) from equation (1), volcanic zone enriched or depleted end‐member, or X x 0 the flank zone enriched end‐member composition; M ¼ jwi wi j; ð1Þ i secondly, by whether the misfit calculated is for both major and minor elements (CFNKTMAS) or where the subscript i refers to a particular oxide. major elements alone (CFMAS). [23] 2. As it is likely that our estimated enriched ‐ [28] The most important result is the dominance of and depleted end member compositions have lherzolite lithologies in producing melts which experienced some olivine fractionation or accumu- match the Icelandic end‐member compositions lation, the next stage tests both incremental frac- (Table 2). Although melts from pyroxenite and tional crystallization and accumulation of olivine eclogite partial melting experiments are included in from/into the experimental melt to determine if the database, none of these proved adequate fits to the misfit is improved. Removal/addition of olivine the Icelandic depleted end‐member, enriched end‐ occurs in 0.05 wt% steps, calculating the equilibrium ’ member or flank zone compositions. In addition, composition according to Herzberg and O Hara apart from the one high degree melt of PERC (a [2002]. carbonated lherzolite [Dasgupta et al., 2007]) fitting [24] 3. After each increment of olivine removal or the flank zone composition, all other matching melts addition the misfit is recalculated as in step 1. are from anhydrous peridotite lithologies.

[25] 4. If, after a crystallisation/addition step, the [29] There are however important differences between experimental melt composition (modified by extrac- the starting materials melting to generate the depleted tion of olivine) is closer to the end‐member melt and enriched compositions. The depleted end‐member composition (has a lower misfit), then crystallisation/ is consistently matched by the 3 GPa melts from addition of olivine continues. Crystallisation/addition experiments on KR4003 by Walter [1998]. This of olivine is stopped when the misfit begins to peridotite represents a pyrolitic composition, close to increase again, or the mass fraction removed/added that of KLB‐1,withanMg#of89.2[Walter, 1998] reaches 50% of the original mass. The final misfit (Table 3). In contrast, the enriched end‐members are of the experimental melt is taken as the lowest matched by melts from KG1/KG2 [Kogiso et al., returned from the crystallization and accumulation 1998]; with these peridotites contributing the only steps. If instead, neither olivine removal nor addi- partial melts within uncertainty of the neovolcanic tion can improve the misfit, then the initial misfit zone enriched composition. The starting materials for calculated in step 1 is used. KG1 and KG2 are homogeneous mixtures of KLB‐1 and average MORB in ratios of 1:1 and 2:1, respec- [26] With the above procedure, each experimental tively [Kogiso et al., 1998]. They therefore represent melt is assigned a misfit with respect to depleted, highly fertile peridotites, with a lower bulk Mg# enriched and flank compositions. To assess the (81–85) and higher FeO, CaO, Na2OandTiO2 significance of the calculated misfit, we propagate concentrations than are present in KR4003 (Table 3). both the total error in the analyses of experimental The consequence of these major element differences melt compositions across the oxides considered and isformeltsofKG1andKG2tohavehigherFeO, the uncertainty of Icelandic end‐member compo- and lower SiO2 and CaO than melts of KR4003 at sitions, through the misfit calculation. An experi- equivalent pressures and melt fractions. mental melt having a summed misfit (M) less than the calculated 2s error is then considered a sig- [30] The effect on solidus temperatures is poorly nificant match to the end‐member composition. constrained by the limited pressure overlap between

9of32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Table 2. The Results of Determining the Misfit Between Experimental Partial Melts and Icelandic End‐Member Compositionsa Misfit (wt%) Material Lithology a/h/cb P (GPa) T (°) Fc (%) Mineralogyd Reference Depleted, Major Elements 1.18 KR4003 lrze a 3.0 1540 24.40 ol:opxf Walter [1998] 1.48 MPY 87 lrz a 1.0 1328 23.00 ol:opx Falloon et al. [2001] 1.99 KR4003 lrz a 3.0 1530 18.50 ol:opx:cpx Walter [1998] 2.46 KR4003 lrz a 4.0 1660 38.80 ol:opx Walter [1998] 2.52 TQ 40 lrz a 1.0 1495 36.40 ol:opx Falloon et al. [2001]

Depleted, Major and Minor Elements 2.31 KR4003 lrz a 3.0 1540 24.40 ol:opx Walter [1998] 3.13 KR4003 lrz a 3.0 1530 18.50 ol:opx:cpx Walter [1998]

Enriched, Major Elements 1.76 KG2 lrz a 3.0 1525 22.00 ol:cpx:gt Kogiso et al. [1998] 1.90 KG1 lrz a 3.0 1525 32.00 ol:cpx:gt Kogiso et al. [1998]

Enriched, Major and Minor Elements 3.66 KG1 lrz a 3.0 1525 32.00 ol:cpx:gt Kogiso et al. [1998] 3.80 KG2 lrz a 3.0 1525 22.00 ol:cpx:gt Kogiso et al. [1998]

Flank Zone, Major Elements 1.11 KG2 lrz a 3.0 1525 22.00 ol:cpx:gt Kogiso et al. [1998] 2.05 KG1 lrz a 3.0 1525 32.00 ol:cpx:gt Kogiso et al. [1998] 2.47 KR4003 lrz a 4.0 1610 12.90 ol:cpx:gt Walter [1998] 2.54 KR4003 lrz a 4.0 1660 38.80 ol:opx Walter [1998] 2.93 KR4003 lrz a 4.5 1650 37.20 ol:opx:gt Walter [1998] 3.48 PERC lrz c 3.0 1550 40.30 ol:opx Dasgupta et al. [2007]

Flank Zone, Major and Minor Elements 2.68 KG2 lrz a 3.0 1525 22.00 ol:cpx:gt Kogiso et al. [1998] 3.46 KG1 lrz a 3.0 1525 32.00 ol:cpx:gt Kogiso et al. [1998] 3.64 KR4003 lrz a 4.0 1610 12.90 ol:cpx:gt Walter [1998] aThe details of the partial melts lying within a 2s uncertainty of the end‐member compositions are shown here. Calculations were performed separately, comparing major elements only (CaO, FeO, MgO and SiO2), and major and minor elements together (also Na2O, K2O and TiO2) for each of the end‐member compositions considered. Details of the calculations can be found in the text. bAbbreviations: a, anhydrous; h, hydrous; c, carbonated experiment. cExtent of melting estimated from experimental charge. dResidual mineralogy after melting. eAbbreviations: lrz, lherzolite; px, pyroxenite. fAbbreviations: ol, olivine; opx, orthopyroxene; cpx, clinopyroxene; sp, spinel; gt, garnet. the experiments of Kogiso et al. [1998] and Walter in controlling the onset of melting [Hirschmann, [1998]. However, in the 2–3 GPa range where 2000], and Na in determining modal clinopyr- there are data available, KG1 and KG2 exhibit sol- oxene content, that KG1 and KG2 are likely to have idus temperatures lower by up to several tens of lower solidus temperatures and higher productiv- degrees. We can extrapolate from this, given the ities than KR4003 over a broad pressure interval importance of incompatible elements such as K

Table 3. Experimental Starting Materials Producing the Best Fitting Melts to Icelandic End‐Member Basalt Compositions a b Material SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2OK2OCr2O3 Mg# KR4003c 44.9 0.16 4.26 8.02 0.13 37.3 3.45 0.22 0.09 0.41 89.2 KG2d 46.22 0.57 7.69 9.22 ‐ 28.83 6.05 1.11 0.09 0.22 84.8 KG1d 46.97 0.78 9.75 9.77 ‐ 23.57 7.35 1.52 0.12 0.17 81.1 aFeO represents total Fe as FeO. bMg number calculated on molar (Mg/(Mg + Fe)) × 100. cWalter [1998]. dKogiso et al. [1998].

10 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

[Hirschmann, 2000; Pickering‐Witter and Johnston, of small‐degree melts of peridotite are similar to 2000; Schwab and Johnston, 2001]. the enriched Icelandic end‐member (45 wt% and 10.5 wt%, respectively [Davis et al., 2009, 2011]), [31] The ability of experimental melts to match the the FeO content (9 wt%) remains too low. This specific oxide compositions of the Icelandic end‐ inability of small‐fraction melts of KLB‐1‐like members is presented graphically in Figure 5, with compositions to match the FeO content of Icelandic the best fitting melts from Table 2 highlighted. basalts is a key indication of source major element Considering only CFMAS, the matching experi- heterogeneity. mental melts fit the end‐member compositions with little systematic misfit. However, introducing [33] Allowing for discrepancies resulting from the minor elements reveals consistent differences batch melting versus fractional melting, the peri- between the experimental compositions and the dotites KG1/KG2 and KR4003 provide adequate Icelandic end‐members. Figure 5b shows that the fits to the major element systematics of Icelandic matching experimental melts have systematically basalts. Specifically, the FeO enrichments and low lower Na2O than the depleted Icelandic end‐ SiO2 and CaO contents of enriched basalts can be member, and almost every melt from the experi- generated by the high‐pressure melting (≥3 GPa) of mental database to have a higher K2O content. peridotite lithologies, made more fertile through the Similarly, the melts of KG1 and KG2 both over- addition of MORB‐like material. Direct melting of estimate the TiO2,Na2O and K2O concentrations of pyroxenitic lithologies however, proves unable to the enriched end‐member (Figure 5d). To an extent match the systematics of Icelandic end‐members. these misfits can be understood as reflecting the Despite both depleted and enriched lithologies different nature of melting in an experimental being matched by peridotite melting, it is important charge, isobaric batch melting, to that which pro- to note that the highly fertile nature of the KG1/ duced the melts erupted on Iceland, polybaric KG2 like enriched lithology beneath Iceland, makes fractional melting. This difference will have the it likely to have a lower solidus temperature and biggest impact on the incompatible elements, which increased productivity compared with the depleted will have been almost totally stripped from the lithology. source by the point at which the depleted melts form and separate. Hence, unless the starting 3.3. Comparison of End‐Member material experiences prior depletion, batch melting Compositions With Melts Mixed From Two experiments are not going to be able to reproduce Source Lithologies the low K2O contents of natural high degree melts. The higher concentration of incompatibles in KG1 [34] In the previous section we found that anhy- and KG2 melts compared with the enriched end‐ drous peridotite lithologies of varying fertility can member, possibly reflects mixing of the enriched match the major element patterns of depleted and Icelandic signature with more depleted melts dur- enriched Icelandic basalts. It has, however, been ing transport, a process exacerbated by the high suggested that melts from pyroxenite sources con- degrees of melting likely beneath Iceland Stracke tribute 20–40% by mass of the enriched melts et al. [2003]. That this same pattern of incom- erupting on Iceland, with the remaining fraction patible element overenrichment in KG1 and KG2 formed of peridotite partial melts [Sobolev et al., is not observed with respect to the flank zone 2007, 2008]. As the misfit calculations presented melts (which are thought to derive from smaller above looked only at the compositions of direct extents of melting [Kokfelt et al., 2006; Peate et al., melts from single lithologies, there is the possi- 2010], and therefore are presumably less likely to bility that the highly fertile KG1/KG2 melts, undergo mixing with depleted melts) supports this found to match the enriched end‐member, could hypothesis (Figure 5). be replicated by more refractory peridotite melts mixed with pyroxenite melts. To assess whether a [32] Very few of the experiments considered con- peridotite‐pyroxenite melt blend is consistent with tain melt fractions less than ∼10%. This leaves the major element patterns, we rerun the misfit cal- the possibility of a small‐degree partial melt of a culations allowing for mixing between peridotite peridotite in fact being able to match the enriched lithologies and the Px‐1 pyroxenite melts of Sobolev Icelandic composition, without the need for source et al. [2007]. enrichment. However, we know from the 3 GPa experiments of Davis et al. [2009, 2011] on a KLB‐1‐ [35] The procedure followed for the misfit calcu- like starting composition, that this is unlikely to be lations is the same as in the previous section, the case. Although the SiO2 and CaO content

11 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 5. The quality of fit of experimental melts to Icelandic end‐member melt compositions. Experimental melt compositions have been normalized to the depleted or enriched end‐member melt. The fits to major elements only, and fits to both the major and minor elements are shown separately. Drawn as black lines are the experimental melts from Table 2 that match, within uncertainty, Icelandic end‐member melt compositions. The average fits of partial melts from other lithologies are shown by the colored lines. (a and b) Experimental melts normalized to the depleted neovolcanic zone end‐member melt compositions, and (c and d) experimental melts normalized to neovolcanic zone enriched end‐member. (e and f) Experimental melts normalized to the enriched flank zone melt composition.

12 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Table 4. The Results of Misfit Calculations Between Peridotite Partial Melts and the Pyroxenite Melts of Px‐1 From Sobolev et al. [2007], to Determine if Mixing of Melts From a Peridotite and Pyroxenite Can Explain the Icelandic Enriched End‐Member Melt Composition a Misfit Materials Fpx P (GPa) T (°C) F (%) Residual Mineralogy Enriched, Major Elements 1.59 KG1 + Px‐1b 0.10 3.0/3.5 1525/1540 32.0/72.0 ol:cpx:gt/opx:cpx:gt 1.60 KG1 + Px‐1 0.09 3.0/3.5 1525/1575 32.0/79.0 ol:cpx:gt/opx:cpx:gt 1.61 KG1 + Px‐1 0.09 3.0/3.5 1525/1550 32.0/76.0 ol:cpx:gt/opx:cpx:gt 1.61 KG2 + Px‐1 0.05 3.0/3.5 1525/1575 22.0/79.0 ol:opx:gt/opx:cpx:gt 1.61 KG2 + Px‐1 0.05 3.0/3.5 1525/1550 22.0/76.0 ol:opx:gt/opx:cpx:gt

Enriched, Major and Minor Elements 3.05 KG1 + Px‐1 0.10 3.0/3.5 1525/1540 32.0/72.0 ol:cpx:gt/opx:cpx:gt 3.09 KG1 + Px‐1 0.09 3.0/3.5 1525/1575 32.0/79.0 ol:cpx:gt/opx:cpx:gt 3.10 KG1 + Px‐1 0.09 3.0/3.5 1525/1550 32.0/76.0 ol:cpx:gt/opx:cpx:gt 3.49 KG2 + Px‐1 0.05 3.0/3.5 1525/1575 22.0/79.0 ol:opx:gt/opx:cpx:gt 3.49 KG2 + Px‐1 0.05 3.0/3.5 1525/1550 22.0/76.0 ol:opx:gt/opx:cpx:gt aMass fraction of pyroxenite derived melt in the melt mixture. bKG1 and KG2 partial melts from Kogiso et al. [1998]. simply iterated to allow for varying proportions of mixes to the end‐member composition, but this peridotite‐pyroxenite melt mixtures (equation (2)), improvement is slight (compare with pure KG1/ X KG2 melts Table 2). The mass fraction of added M ¼ wn w0; ð2Þ n i i pyroxenite melt is also small: at most 10 wt% has i been added to the melts of KG1, and 5 wt% to KG2, to optimize their match to the enriched end‐ where Mn is the misfit for the particular mixing proportion n of peridotite and pyroxenite experi- member. A breakdown of the improvement in n fit for individual oxides is shown graphically in mental melts, and wi the oxide composition of that mixed melt. The returned misfit for a particular Figure 6. Compared to Figures 5a and 5b, there is melt combination is that of the best fitting from all little change in the pattern of misfit in CFMAS or the different peridotite:pyroxenite melt proportions CFNKTMAS. The offsets that do remain between KG1/KG2 melts and the Icelandic enriched end‐ cycled through (i.e., the smallest Mn). The results of these calculations in matching the enriched member are therefore not resolved by mixing with Icelandic end‐member are shown in Table 4. Only a direct pyroxenitic melt, and probably result from mixtures within uncertainty of the end‐member differences in the style of melting between exper- composition are displayed, but results have also iment and nature. been filtered on the basis of physical plausibility: [37] These observations indicate that mixing between the pyroxenite melt must represent a higher degree peridotite‐ and pyroxenite‐derived melts is not of melting and be formed at a higher temperature necessary to match the major and minor elements and pressure than the peridotite melt for the mix- of the Icelandic enriched end‐member. The clearest ture to be considered valid. Filtering the results like sign of this is that the peridotite component that this ensures that only melts which could have the pyroxenite melts are mixing with, is itself formed from the same upwelling column of mantle capable of melting to produce, within uncertainty, are combined; the lower solidus temperatures and the Icelandic end‐member composition. Px‐1 melts greater fusibility of pyroxenite lithologies mean cannot by themselves match the enriched Icelandic that they should always have melted to a higher composition, and any addition of pyroxenite‐ degree and at an average higher pressure than any derived melt in the quantities suggested by Sobolev surrounding peridotite. et al. [2007, 2008] generates a poorer fit than only [36] Calculations allowed mixing between Px‐1 adding small amounts. melts and melts from a wide range of peridotite [38] To summarize the main results of this section: lithologies. Table 4 shows that, KG1/KG2 are the The major and minor element systematics of preferred peridotite melt sources in models that enriched Icelandic basalts are explained by the ‐ match the enriched end member major element melting of a single fertile peridotite source, similar composition. The addition of melts from the garnet in composition to KG1/KG2. The depleted end‐ ‐ pyroxenite Px 1 has improved the fit of the melt member melts can be derived from melting of

13 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 6. Spider diagrams recording the fit of mixes of Px‐1 melts and experimental peridotite melt compositions to the enriched Icelandic end‐member. The five best fitting compositions from Table 4 are shown as black broken lines, with other permutations of Px‐1; peridotite melt mixes in grey for comparison. The mixed melt compositions have been normalized to the enriched end‐member melt.

KLB‐1‐like peridotites, but melts of a MORB‐ enriched end‐member and flank zone end‐member like pyroxenite cannot mix with these peridotite are very similar (Figure 7). Both zones show strong melts to produce the enriched end‐member. REE and incompatible element enrichment, and very similar compatible element abundances. The primary difference between the two end‐members is 4. Trace Element Constraints on Source in their incompatible element concentrations, with Lithology the flank zone melts systematically offset to higher abundances. That flank and neovolcanic zone [39] The trace element contents of basalts and of enriched basalts exhibit very similar trace element the olivines they crystallize have been used both to patterns, but with different absolute concentrations, support and refute the presence of pyroxenitic indicates that degree of melting is the most impor- lithologies in the source of oceanic island basalts tant control on the chemical variation between [Hirschmann and Stolper, 1996; Stracke et al., basalts from the two settings. However Pb, for 1999; Kokfelt et al., 2006; Sobolev et al., 2007; which the concentration in the flank zone melts Herzberg, 2011]. In this section we first demon- overlaps with the concentrations in the neovolcanic strate that the compatible and incompatible trace zone enriched melts, indicates the role of a sec- element characteristics of Iceland’s end‐member ondary process [e.g., Hofmann, 2003]. The REEs compositions, from neovolcanic zone depleted make evident the importance of residual garnet tholeiites to the highly enriched flank zone melts, during melting, generating the steep REE profiles of are adequately modeled by peridotite melting. Sec- the enriched and flank end‐member. While the ondly, presenting new olivine analyses from four likelihood of there being residual garnet in a purely Icelandic eruptions, which cover extremely depleted peridotitic MORB source has been questioned and enriched compositions, we demonstrate that the [Hirschmann and Stolper, 1996], both higher Tp lithological variation encompassed by KR4003 beneath Iceland and the trace amounts of water in the and KG2 source lithologies is enough to reproduce source [Nichols et al., 2002; Asimow and Langmuir, the measured range of olivine Ni contents; direct 2003], should allow even a depleted mantle to melting of an olivine‐free lithology is not required. begin melting by 3 GPa.

4.1. Trace Element Characteristics 4.2. Inversion of Trace Element ‐ of End Member Melts Distributions for Peridotite Melting

[40] The same basalts selected from Figure 3 to [41] To quantitatively model the trace element define the end‐member major element character- observations we applied the inversion code of istics, were used here to constrain the trace element McKenzie and O’Nions [1991], modified by White composition of the end‐member melts. Overall, the et al. [1992] and Gibson and Geist [2010]. The trace element patterns of the neovolcanic zone failure or success of this model in matching the

14 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 7. Elemental abundances in the mean Icelandic enriched neovolcanic zone and flank zone end‐member melt compositions. Concentrations have been normalized to the composition of the depleted end‐member melt, which is the horizontal line at unity in each diagram. One sigma variations about the mean for enriched and flank melts are indicated by vertical bars, and by a grey field for the depleted end‐member composition. Rare earth elements from the neovolcanic and flank zones have been separately inverted for, using the peridotite melting and inversion code of McKenzie and O’Nions [1991], with the model’s predictions of incompatible and compatible element abundances also shown. The success of the model in matching the data illustrates that the trace element patterns of Icelandic basalts do not require pyroxenite in the source. Percent uncertainties in the source compositions used in the inversion are given above each plot. trace element patterns, should indicate whether the broad range of compatible‐incompatible elements. trace elements require the involvement of a source Initial source compositions were selected to be other than peridotite. The inversion runs a forward primitive mantle (bulk earth equivalent) for mod- model, predicting the REE concentrations of an eling the enriched neovolcanic zone and flank end‐ aggregated polybaric fractional peridotite melt, members, and a depleted mantle composition for calculates the root mean square misfit between modeling of the depleted end‐member [McKenzie predicted and observed REE concentrations, and and O’Nions, 1991]. then uses Powell’s algorithm in adjusting the melt [42] The results of the inversion are shown in fraction with depth curve to minimize this misfit Figure 7. The model is successful in matching the [McKenzie and O’Nions, 1991; Press et al., 1992, concentrations of all the REE and also many of the pp. 757–1152]. As well as minimizing misfit with other incompatible and compatible elements con- respect to the REEs, model output also includes the sidered. Even where the model result deviates from non‐REE trace elements, allowing for the model’s exact matches to the concentrations, it generally success in reproducing the data to be tested across a remains close to the pattern of elemental abundance.

15 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Given the combined uncertainty in the end‐member Háleyjabunga and the Theistareykir picrite, are compositions, the starting source compositions and representative of the depleted Icelandic end‐ in the partition coefficients, the model result is in member, both having very depleted trace element most instances within uncertainty of the data. ratios (Nb/Zr = 0.036 and 0.032, respectively), isotopic values and overlapping olivine compositions [43] The only point of significant deviation between (Figure 8). The remaining two eruptions, Stapafell the inverse model and the data is in Pb con- and Gaesafjöll, more closely resemble the com- centrations. While model predictions of Pb content position of enriched Icelandic end‐member, each are within uncertainty of the enriched end‐member having enriched incompatible element (Nb/Zr = composition, they are predicted to be seven times 0.157 and 0.114, respectively) and isotopic sig- higher in the flank zone end‐member than is natures. The main difference between olivines observed, Figure 7. This is perhaps evidence for from Gaesafjöll and Stapafell eruptions is in for- the nature of the process generating the enriched sterite(Fo) content, with Gaesafjöll olivines offset end‐member. The KG1/KG2 sources were referti- from the Stapafell olivines to lower Fo values lized from a KLB‐1 starting composition by addi- (Fo = 83–84 for Gaesafjöll compared with mostly tion of average MORB. As noted by previous Fo = 84–87 for Stapafell). At a given Fo content authors [e.g., Zindler et al., 1982; Vidal, 1992; olivines from Gaesafjöll therefore have a higher Kokfelt et al., 2006], recycled MORB material is Ni content than olivines from Stapafell, by up to likely to have an increased U/Pb ratio due to pref- 500 ppm. erential Pb loss during the dehydration reactions experienced on subduction. It is probable therefore, [46] Using the melts from KR4003 and KG2 as that the relatively low Pb concentration of flank starting compositions, closest to the depleted and zone melts reflects the addition of low Pb MORB enriched end‐member, respectively (Table 2), we material to the mantle beneath Iceland. This mate- next model the Fo and Ni contents of olivines rial is now assimilated into the peridotite producing produced during fractional crystallization. The a fertile peridotitic lithology, enriched in all but the details of the calculations can be found in the caption most fluid mobile incompatible elements. Such an to Figure 8. The path of olivine compositions crys- interpretation might also explain the slight, but tallizing from the KR4003 melt (under olivine only consistent, offset of other fluid mobile elements in fractionation), marked as a thick line in Figure 8, Figure 7 (e.g., Cs, Rb, K and Ba) to lower con- generates an array through the Theistareykir picrite centrations than that predicted by the model. and Háleyjabunga olivine compositions. To model the olivines forming from enriched parental melts, [44] Despite the slight differences between the we take a partial melt from KG2 as an initial model result and the observed trace element pat- composition and model both olivine only and terns, the overriding feature of Figure 7 is the olivine + plagioclase + clinopyroxene fractionation quality of fit a peridotite melting model provides (thin and dashed lines, respectively, Figure 8). The to the trace element data. While this cannot prove Ni contents of Gaesafjöll olivines can mostly be the presence of an exclusively peridotitic source accounted for by olivine crystallization from a KG2 beneath Iceland, it at least indicates its viability as parental melt. However, allowing the removal of a a source lithology for the range of melt composi- gabbroic crystal assemblage after the liquid has tions observed. reached 9 wt% MgO explains the slight spread of data to higher Ni at a given Fo than is generated 4.3. Ni Content of Olivine as a Proxy by olivine only crystallization. The Stapafell oliv- for Source ine compositions lie at intermediate Ni contents [45] To understand what constraints the trace ele- between the array of olivines from a pure KG2 melt ment content of olivine places on the source crystallizing olivine only, and the KR4003 melt lithologies beneath Iceland, we analyzed olivines olivine array. from four eruptions, the locations of which are [47] It is worth emphasizing that there are large marked in Figure 1. The complete data set of uncertainties in both source Ni compositions, and electron microprobe analyses of the olivines and with the partition coefficients assumed during ini- details of the analytical conditions can be found ‐ 1 tial melting and low pressure fractionation steps. in the auxiliary material. Two of the eruptions, Therefore, while the role of gabbro fractionation in generating higher Ni contents of cocrystallizing 1Auxiliary materials are available at ftp://ftp.agu.org/apend/gc/ olivines is examined, the Gaesafjöll data fall within 2011gc003748.

16 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 8. Ni content of olivines from four Icelandic eruptions. The lines are modelled olivine compositions produced by fractionally crystallizing KG2 or KR4003 parental melts. The melts chosen for modeling were those with the best fit to Icelandic end‐member melt compositions from Table 2. Primary Ni contents of partial melts of KR4003 and KG2 were calculated following the same approach as Sobolev et al. [2007], in which the bulk solid composition, phase proportions and partition coefficients are used to solve for melt Ni content. The starting Ni composition of KR4003 was taken from Walter [1998]. KG2 Ni content was calculated knowing the 2:1 proportion in which KLB‐1 and average MORB were mixed, using the Ni content of KLB‐1 from Hirose and Kushiro [1993] and assuming average MORB contains 0.025 wt% NiO. Melts then had equilibrium olivine fractionally extracted from them according to the Fe‐Mg and Ni partitioning model of Beattie [1993]. A second model was run allowing olivine, clinopyroxene and plagioclase to crystallize in gabbroic proportions once the melt had reached 9 wt% MgO, the major element compositions of plagioclase and clinopyroxene were held fixed and taken as average Theistareykir pheno- cryst values. The paths followed by equilibrium olivine compositions crystallizing from these melts are drawn in black, with boxes indicating the assumed bulk Ni composition of the solid mantle source producing these primary melts. A representative 3s error bar on the olivine analyses is included. Data from Sobolev et al. [2007] are shown for comparison. the combined uncertainty from the errors on all of dotites, or takes place over long timescales in the the model parameters. solid state, we cannot constrain.

[48] From the results discussed above and presented in Figure 8, it seems likely that peridotitic 5. Isotopic Constraints on Source mantle sources can produce all the variability in Ni Lithology content observed in Icelandic olivines. This interpretation differs from that of Sobolev et al. [2007], [49] In Figures 9c–9f whole‐rock Nd isotopic which requires olivine‐free lithologies to be present composition is plotted against either CaO, SiO2, in order to match the observations. However, our FeO or Al2O3 for high MgO neovolcanic and flank model still places an importance on MORB in the zone basalts. The relationship between enrichment source region, as peridotites can only produce and major element composition is as identified melts crystallizing high‐Ni olivines if they have from the incompatible trace element relationships been refertilized by MORB addition. Olivine Ni of Figure 3: enriched (low Nd) basalts have higher contents therefore do require major element het- FeO, but lower CaO and SiO2 than the depleted erogeneity in the source. Whether this occurs in the basalts. For FeO in particular (Figure 9e), a mixture melt region beneath Iceland by melting and reac- of material lying between HIMU‐ and MORB‐like tion of MORB pyroxenites with surrounding peri- compositions explains the first order distribution of

17 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 9. (a and b) The Pb‐Nd‐Sr isotopic systematics of Icelandic basalts. (c–f) The major element compositions of high MgO basalts (MgO > 9.5 wt% for the neovolcanic zones, >8.5 wt% for flank zone melts) plotted with their corresponding whole‐rock Nd isotopes. Samples were selected to include only those with joint major and isotopic analyses. Vectors point to the global isotopic major element end‐member compositions identified by Jackson and Dasgupta [2008], of fractionation uncorrected basalts between 8 and 16 wt% MgO. The mixing of no two end‐ member components can fully account for the distribution of Icelandic basalts.

18 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

the data. However, considering also CaO, SiO2 ducing a fertile peridotite lithology much like the and Al2O3, a more complicated mix of HIMU and KG2 starting material of Kogiso et al. [1998]. EM‐like end‐members would be required to reproduce the enriched end of the array. The depleted end‐member in contrast, at least in terms 6. The Effect of a Fusible Lithology of major elements and Nd isotopes, is consistently on the Volume of Melt Generation matched by a source of similar lithology and isotopic composition to that producing MORB. [52] In the previous section we demonstrated that both enriched and depleted Icelandic basalt can be [50] The high FeO and low SiO2 contents of Ice- generated by melting of peridotite, albeit, in the landic enriched basalts are broadly consistent with case of enriched basalt, a peridotite of highly fertile what Jackson and Dasgupta [2008] identify as composition (KG2). A number of authors have HIMU‐like characteristics. However, unlike the suggested that melting anomalies around Iceland, HIMU of Jackson and Dasgupta [2008], enriched and large igneous provinces more generally [Sobolev Icelandic basalts tend toward low CaO. As the near et al., 2011], are controlled by compositional varia- solidus CaO contents of peridotite melts have been tion in the mantle [Foulger and Anderson,2005]. linked to CO2 content in the source [e.g., Brey, Here we develop a model of bilithological melting 1978; Hirose, 1997; Dasgupta et al., 2007], low that allows for assessment of the role of source CaO perhaps indicates that the enriched Icelandic heterogeneity in controlling overall melt produc- component is not especially volatile rich. This tion. In combination with our observational con- conclusion would be consistent with our analysis straints on the fraction of recycled material, this of experimental partial melts, in which volatile‐ model allows us to evaluate a likely upper bound present conditions generally produced melts with on excess melt production associated with compo- poor fits to the enriched end‐member composi- sitional heterogeneity under the North Atlantic. tion. Similarly, the absence of CO2 enrichments in [53] The significance of source enrichment for melt the source could account for why SiO2, although decreasing with enrichment (Figure 9b), does not production depends on how it changes: (1) the soli- show the extreme lows of the HIMU end‐member. dus temperature, and (2) the productivity (dF/dP), of An additional possibility is that the enriched end‐ the bulk matrix‐enriched streak mixture. As dis- member consists of a mix of EM and HIMU‐like cussed previously, it is likely that enriched litholo- components. Certainly, from the isotope systemat- gies will have both lower solidus temperatures and ics alone, HIMU is a poor choice of enriched end‐ greater productivities than depleted peridotites, and member. Despite this, the high FeO content of thus their presence in a source could tend to increase enriched Icelandic basalts would discount the EM‐ melt production. However, mantle temperature also like components [Jackson and Dasgupta, 2008] has a large effect on the volume of melt generated being a large proportion by mass of the source, and its average chemistry [McKenzie and Bickle, even if they are able to contribute significantly to 1988; Langmuir et al., 1992]. Given the progres- – trace element and isotopic variation. sive rise in Tp (from <1400°C 1500°C) as the center of the Iceland plume is approached [White et al., [51] For Iceland, none of the end‐member basalt 1995], the effect of Tp on melt production will also compositions identified by Jackson and Dasgupta need to be considered by the melting model. [2008] fully account for the isotopic and major element variation observed. This probably as [54] We develop a two‐lithology melting model much reflects the nonuniqueness of the global consisting of a peridotite‐pyroxenite dual source, end‐members, as it does demand a complex mix and vary both the proportion of pyroxenite () and between them to account for the distribution of the Tp to predict crustal thicknesses over a range of major element isotopic systematics. While the high conditions. This is coupled with a simple parame- time‐integrated U/Pb of Icelandic basalts (as recor- terization of melt SiO2 content for both peridotite ded by Pb isotopic compositions, Figures 9a and 9b) and pyroxenite, so that the effect of increasing indicates a role for recycled oceanic crustal material, pyroxenite fraction on melt composition can be ‐ nothing from the relationships in Figure 9 necessi- compared with along ridge SiO2 trends. tates that component to still be present as a distinct pyroxenitic lithology. In fact, the association of 6.1. Two‐Lithology Melting Model enrichment with low SiO and CaO, but high FeO, 2 [55] We base our model on the thermodynamic treat- points toward that precursor MORB material now ment of adiabatic decompression fractional melting having homogenized with adjacent peridotite, pro-

19 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G by Phipps Morgan [2001]. When coupled with scale of thermal equilibrium could be an order of parameterizations of the solidus and liquidus sur- magnitude smaller toward the base of the melt faces for each lithology, as well as a function region. The validity of the second assumption, of describing how productivity varies in the melting chemical isolation between sources and their melts, interval, the equations of Phipps Morgan [2001] will depend upon the melt extraction mechanism. calculate dT/dP and dF/dP for the two‐lithology Melts of the enriched lithology have potentially mixture. Each increment of melting is calculated by quite different major element compositions to melts following a reversible adiabatic path, with complete in equilibrium with the surrounding matrix, even melt extraction. There are in effect four separate when both are peridotitic (e.g., melts of KG2 and regimes that can be passed through by an upwelling KR4003 [Kogiso et al., 1998; Walter, 1998]). If parcel of two‐lithology mantle, depending upon these enriched melts infiltrate surrounding perido- the temperature of the mantle T, with respect to the tite they may react, altering both melt and matrix pyroxenite (superscript px) and peridotite (super- compositions [Yaxley and Green, 2000]. However, script pd) liquidus and solidus surfaces: partial chemical isolation of enriched melts is a px pd consequence of the channelized melt extraction [56] 1. T < T ,T Tsolidus,T Tsolidus,TTsolidus: G2, a MORB‐like pyroxenite [Pertermann and Pyroxenite and peridotite are melting. Productivity Hirschmann, 2003b], was chosen as an extreme of the pyroxenite drops as the peridotite ceases to case of enriched component fertility, having a lower thermally buffer the pyroxenite melting. dF/dP is solidus temperature (∼150°C) and much narrower controlled by both the peridotite and pyroxenite melting interval (by >100°C) than the peridotite of melting reactions. Katz et al. [2003] and of KG1/KG2 [Kogiso et al., px pd 1998]. It therefore represents the upper bound of [59] 4. T > T ,T>T : Peridotite‐only liquidus solidus what source enrichment can achieve in increasing melting, pyroxenite has been exhausted from melt volume, and as such offers the most favorable source. dF/dP and dT/dP reduce to the single scenario for those models invoking the presence lithology melting equations for the peridotite of recycled MORB in the mantle source to explain parameterization. This scenario occurs when T is p elevated crustal thicknesses. If the actual Icelandic high, or the mass fraction of pyroxenite in the source contains KG1/KG2‐like enriched peridotite, peridotite:pyroxenite mixture is low. As can be the potential increase in crustal thicknesses are seen from Figure 10a, the pyroxenite liquidus is likely to be much less [Kogiso et al., 1998]. Model almost intersected by 1 GPa if its mass fraction in parameters are listed in Table 5, and the details the source is infinitesimal. of the parameterizations for calculating the SiO2 [60] The key assumptions are that thermal equilib- contents of melts are given in Appendix D. rium is maintained between the two lithologies, and that sources and melts remain chemically iso- lated during and after melting. The former of these 6.2. Results of Melting Model assumptions is discussed by Phipps Morgan [2001]; [62] The most important result from the pyroxenite‐ the resulting relations imply that for the rates of peridotite melting calculations is Figure 10b, which upwelling generated by passive plate spreading shows that even with a purely pyroxenitic source, a beneath Iceland (half spreading rate ∼10 mm/yr temperature anomaly of >100° is required to match [DeMets et al., 1994]), material <3 km across should the crustal thicknesses observed in central Iceland. be in thermal equilibrium with its matrix. However, While this result indicates that source enrichment if plume‐driven upwelling is important, the length alone cannot provide the necessary melt volume to

20 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure 10. Modeling of two‐lithology melting to constrain the contributions to crustal thickness and major element chemistry of having a more fusible component embedded in peridotite. (a) Decompression paths for the end‐member cases of pure peridotite melting, dark blue, parameterized from Katz et al. [2003], and pure pyroxenite melting, red, parameterized from Pertermann and Hirschmann [2003b], for a Tp of 1305°C. Solidus and liquidus locations are marked: px‐s, pyroxenite solidus, px‐l, pyroxenite liquidus, pd‐s, peridotite solidus, pd‐l, peridotite liquidus. (b) Crustal thicknesses generated by melting a peridotite source containing varying pyroxenite mass fraction ()ata range of Tp, with the top of the melt region set at 1 GPa. The open diamond represents the crustal thickness produced by melting pure peridotite, but incorporating 20× plume‐driven upwelling for the first 5.5% of peridotite px total melting. (c) The relative contribution of pyroxenite melting (tc ) to the overall crustal thickness (tc ). The dotted lines record the pyroxenite fraction in the source if it contributes 40% of the melt generation, for a given Tp. (d) Predicted SiO2 of pooled melts from mixed lithology melting, compared with the volume‐averaged SiO2 of Icelandic rift zone basalt. The silica content of peridotite partial melts has been parameterized as a function of P and F, and G2 pyroxenite partial melts as a function of F. match Icelandic crustal thicknesses, Figure 10b melt production under central Iceland, if there is a also shows that pure peridotite melting at a plau- 20 fold increase in upwelling rate at the base of sible Tp for Iceland of 1500°C [White et al., 1995], the melting region (open diamond, Figure 10), con- cannot produce crustal thicknesses in central Iceland sistent with the inferences of Ito et al. [1999], either. However, a peridotite source can reproduce Maclennan et al. [2001b], and Kokfelt et al. [2003].

Table 5. Parameters Used in Modeling Two‐Lithology Melting of a Peridotite From Katz et al. [2003], and the G2 Pyroxenite Lithology of Pertermann and Hirschmann [2003b] Parameter Peridotite Pyroxenite Units Referencea Mass fraction cpx 0.2 ‐ Heat capacity 1000 1000 Jkg−1K−1 1 Thermal expansivity, solid 40 × 10−6 40 × 10−6 K−1 1 Density, solid 3300 3550 kgm−3 1,2 Entropy of fusion 300 300 JK−1 1 aReferences: 1, Katz et al. [2003], peridotite and pyroxenite assumed to have the same properties; 2, Pertermann and Hirschmann [2003b].

21 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

The high productivity of G2 pyroxenite and its [65] The above analysis mapped onto Figure 10b lower solidus temperature, means that melting (vertical dotted line), constrains the maximum preferentially samples the enriched pyroxenitic part crustal contribution from pyroxenite derived melts to of the mantle source, biasing the average melt com- be ∼4 km (compare with the crustal thickness (tc) position with respect to bulk source (Figure 10c); estimates from central Iceland, where tc =35–40 km only 5–12% of solid pyroxenite is required before [Darbyshire et al., 2000; Allen et al., 2002; Kaban its melts are dominant by mass. et al., 2002; Li and Detrick, 2006]). The two‐ lithology melting model therefore, indicates that [63] The effect on melt SiO composition from 2 source enrichment alone is very unlikely to be the increasing pyroxenite fraction in the source is cause of excess melt production at Iceland. Without recorded in Figure 10d. MORB‐like pyroxenites a temperature anomaly >100°C, a source much such as G2 melt to produce high SiO melts 2 more fusible than even a pure MORB composition [Pertermann and Hirschmann, 2003b], with the would be required to generate 35 km of crustal simple consequence that mean melt SiO increases 2 thickness. It is likely that any source more fertile as the mass fraction of pyroxenite in the source than G2 would have wide ranging implications for increases. As the peridotite fraction is reduced there geochemical and geophysical observables around is the added effect of reducing the thermal insula- Iceland, of which we see no evidence. A tempera- tion it provides to the pyroxenite, reducing the ture anomaly beneath Iceland of >100°C, should pyroxenite’s overall degree of melting and further therefore be a requirement of any pyroxenite or increasing the average SiO of the pyroxenite 2 peridotite source model. Given our finding of vari- melts. The primary effect of higher T is to increase p ably enriched peridotite lithologies being able to the average pressure and extent of melting, which explain the geochemical systematics of Icelandic for both the pyroxenite and peridotite source lowers basalts (i.e., a highly productive MORB‐like lithol- the mean SiO content of the melts they produce. 2 ogy in the source is not an observational require- Compared to the volume‐averaged SiO of Ice- 2 ment), it is probable that both a temperature anomaly landic basalts (horizontal grey bars, Figure 10d), of >100°C and plume‐driven upwelling are required adding almost any pyroxenite of G2 composition to to match crustal thicknesses on Iceland. the source causes SiO2 to be higher than observed. Peridotite‐only melting, with and without plume‐ [66] The along‐ridge compositional trend of Figure 2f, driven upwelling at the base of the melt region, can describes a decrease in the mean SiO2 content of match the SiO2 contents of Icelandic basalts; basalts as Iceland is approached. The two‐lithology although the systematic difference between SiO2 of modeling results show that this is the opposite of the SNVZ and WVZ basalts requires a difference what would be predicted if the contribution of direct of either source, melting dynamics or crustal pro- melts of silica saturated (i.e., MORB‐like) pyroxe- cessing to be explained. nite was increasing toward Iceland (Figure 10d). The major elements alone therefore, refute the 6.3. Implications of the Two‐Phase Melting presence of recycled MORB lithologies as a pri- Model for Mantle Lithology mary source of melts in the Iceland plume(see section 3.2). Our earlier result however, indicated [64] Using the modeling results presented in that a mixed peridotite + MORB lithology provided Figure 10c it is possible to estimate the fraction of the closest fit to the Icelandic enriched end‐member, pyroxenite present in a source, provided the frac- so MORB is likely to have been added to the Ice- tional melt contribution of pyroxenite is known. landic mantle. Any added MORB must have reacted ∼ Sobolev et al. [2008] estimated up to a 40% with adjacent depleted peridotite (in the solid state pyroxenite contribution to Icelandic melts. There- or during pyroxenite melting) to produce a fertile, fore, taking the Sobolev et al. [2008] estimate as the but nonetheless peridotitic, KG2‐like lithology; upper limit of possible pyroxenite contribution to this reaction permits later generation of melts with Icelandic melts, gives a maximum ∼10% mass frac- sufficiently low SiO2 (and low CaO and high tion of pyroxenite in the Icelandic source (Figure 10c, FeO) for Icelandic enriched basalt compositions to dotted lines). This is consistent with our own be matched. observations: as KG2 (33:67 MORB:pyrolite mixture) matches the most enriched Icelandic major element compositions, no more than 33% recycled 7. Summary material (albeit not as a lithologically distinct component) is required in the source regions. [67] We have used a novel method to isolate the effects of mantle and crustal processes on the major

22 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G element systematics of basalts. This has confirmed increase in crustal thickness. Therefore, generation the strong relationship between the major and trace of the >35 km tc observed on Iceland requires element isotopic compositions of mantle melts. elevated mantle potential temperature of at least Isotopically and trace element enriched melts on 100°C, and if the source is predominantly perido- Iceland are characterized by high FeO, but low titic, also needs plume‐driven upwelling in the base SiO2 and CaO. of the melt region.

[68] Comparing a large database of experimental partial melt compositions to the enriched and Appendix A: Data Sources depleted melts from Iceland indicates that a single source composition cannot match the range of [73] For this study data from a large number of major element diversity observed. While KLB‐1‐ sources was compiled to provide the best available like peridotite lithologies produce melts similar to geochemical constraint on the composition of the the compositions of depleted Icelandic melts, Icelandic mantle. In Table A1 the references from enriched melts are best matched by melting of our database of Icelandic geochemistry are repro- KG1/KG2 [Kogiso et al., 1998], a peridotite refer- duced, with an indication of what type of data tilized with MORB material. Major element variation is therefore required in the Icelandic mantle Table A1. List of Icelandic Data Sources source. Data Types [69] The enriched source beneath Iceland is more Reference Majors Traces Isotopic fusible than the source contributing the depleted Brandon et al. [2007] 13 0 0 melts. Modeling shows that greater fusibility will Breddam [2002] 30 5 0 bias the average melt composition toward that of Breddam et al. [2000] 0 0 4 the enriched lithology. The extent of this effect will Chauvel and Hémond [2000] 1 0 3 Devey et al. [1994] 24 0 0 depend upon the solidus/liquidus slopes of the Eason and Sinton [2009] 10 0 0 depleted and enriched lithologies melting beneath Elliott et al. [1991] 5 0 6 Iceland. Furman et al. [1991] 0 29 0 Gee [1999] 194 0 0 [70] The variation in the mantle source’s major Haase et al. [2003] 25 40 0 element composition and fusibility is present on Hardarson and Fitton [1997] 109 10 0 small length scales. Within single zones, eruptions Hards et al. [2000] 0 235 0 separated by only several kilometers exhibit con- Hémond et al. [1993] 25 0 0 Jakobsson et al. [2008] 332 0 0 trasting major element and olivine Ni contents, Jónasson [2005] 12 6 0 characteristic of the depleted and enriched melts. Kempton et al. [2000] 0 0 8 Kokfelt et al. [2006] 35 46 9 [71] Both the depleted and enriched sources can be Kuritani et al. [2011] 1 4 1 olivine bearing peridotites, with the enriched Kurz et al. [1985] 2 0 0 source being derived by mixing ∼40% MORB with Lacasse et al. [2007] 0 24 0 KLB‐1 to create a fertile peridotite. This litholog- Maclennan et al. [2001a] 55 16 0 ical range in the source is consistent with the Ni Maclennan et al. [2003a] 108 49 0 Macpherson et al. [2005] 5 11 0 content of olivines crystallizing from basalts on Mattsson and Oskarsson 0100 Iceland. While the major element compositions of [2005] enriched Icelandic melts do not support direct origin Melson et al. [2002] 6 0 0 from a pyroxenite, we cannot distinguish between Meyer et al. [1985] 1 0 0 pyroxenite melting having infiltrated and referti- Murton et al. [2002] 176 98 0 ‐ Nicholson et al. [1991] 30 3 0 lized peridotite, versus a solid state re equilibration Peate et al. [2009] 23 26 4 between MORB and surrounding peridotite. Schiellerup [1995] 21 0 0 Schilling et al. [1983] 8 0 1 [72] Oceanic crustal thickness variation in the North Schilling et al. [1999] 18 0 2 Atlantic cannot result from changing the source Sigurdsson et al. [1978] 16 0 0 composition alone. Modeling results from the Sigurdsson [1981] 3 0 0 melting of a bilithologic peridotite‐pyroxenite Sinton et al. [2005] 258 0 0 mantle, combined with geochemical observations, Skovgaard et al. [2001] 20 15 8 Slater et al. [2001] 73 54 22 indicate no more than 10% recycled MORB is Stracke et al. [2003] 9 11 7 present in the bulk mantle, probably much less. Sun et al. [1979] 2 0 0 This 10% MORB could only account for a ∼4km Thirlwall et al. [2004] 0 0 22

23 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Table A2. List of Experimental Data Used in This Study were filtered from the data set. The logic behind ‐ Pressure this was that as the end member Icelandic com- Reference Lithology a/h/ca Range positions were basaltic, a good match between them and the high degree melts of basalts would Dasgupta et al. [2004] ecl c 3.0–4.1 Dasgupta et al. [2005] ecl c 3.0 likely follow. Yet such a match would not place Dasgupta et al. [2006] ecl c 1.0 particularly meaningful constraints on the lithology Dasgupta et al. [2007] lrz c 3.0 of the source. Falloon et al. [1999] lrz a 1.0–1.5 Falloon et al. [2001] lrz a 1.0 Gerbode and Dasgupta [2010] ecl c 2.9 Appendix B: Can Crystal Accumulation Hammouda [2003] ecl ch 5.0–6.5 Generate “Enriched” Melt Hirose [1997] lrz c 3.0 Kawamoto and Holloway lrz h 5.0 Characteristics? [1997] Keshav et al. [2004] px a 2.0–2.5 [75] To demonstrate that crystal accumulation is Kinzler [1997] px a 1.5–2.3 unlikely to be controlling the inferred end‐member Kogiso et al. [1998] lrz a 1.5–3.0 major element compositions, we have modeled the Kogiso and Hirschmann px a 1.0 effect of crystal accumulation on a starting magma [2001] ‐ Kogiso et al. [2003] px a 1.0–5.0 with a Nb/Zr similar to the enriched end member Kogiso and Hirschmann ecl a 3.0–5.0 composition. The results of adding different pro- [2006] portions of olivine, clinopyroxene and plagioclase Laporte et al. [2004] hrz a 1.0 to the high Nb/Zr starting composition are pre- Parman and Grove [2004] hrz a 1.5–2.0 sented in Figure B1. None of the permutations of Pertermann and Hirschmann ecl a 2.0–3.0 [2003a] an accumulated crystal cargo are able to move the Pickering‐Witter and Johnston lrz a 1.0 starting liquid toward the major element char- [2000] acteristics of the enriched (or depleted) melt. From Robinson et al. [1998] lrz a 1.5 this result we infer the variation of major element Schwab and Johnston [2001] lrz a 1.0 compositions in Figures 3 and 4 to reflect mantle Sobolev et al. [2007] px a 3.5 Spandler et al. [2008] ecl a 3.0–5.0 melting and/or source processes, and not resulting Takahashi [1986] lrz a 0.0–5.0 from accumulation. Takahashi et al. [1993] lrz a 0.0–6.5 Takahashi and Katsuji [2002] ecl a 2.0–2.7 Tsuruta and Takahashi [1998] px a 0.0–6.0 Appendix C: Relation of Accumulated Villager et al. [2004] lrz a 1.0 Polybaric Fractional Melts to Isobaric Walter [1998] lrz a 3.0–6.0 Wang and Takahashi [1999] ecl a 5.0 Batch Melts: Nature and Experiment Wasylenki et al. [2003] lrz a 1.0 Yasuda et al. [1994] ecl a 3.0–5.0 [76] The identification of lithological heterogene- Yaxley and Green [2000] ecl a 3.5 ity, by systematic comparison of natural melt Yaxley and Brey [2004] ecl c 2.5–5.0 compositions to an experimental database of partial – Yaxley and Sobolev [2007] ecl a 3.5 4.5 melts, raises the question of how the differences aAbbreviations: a, anhydrous; h, hydrous; c, carbonated. in melting style between nature and experiment may influence the compositional similarity of their was obtained from each. Where possible, analyses respective melts. In nature, melt generation beneath of the same sample by different authors have ridges and oceanic islands is envisaged to proceed been combined to provide a comprehensive set of by polybaric fractional melting, with melt extracted analyses. after a small threshold porosity has been reached. These melts may then aggregate during transport [74] Table A2 lists the references from which out of the mantle and storage in the crust, to erupt experimental partial melt compositions were used as accumulated fractional melts. Experiments in in finding matches to the Icelandic end‐member contrast, almost exclusively study batch melting, with compositions. Experiments which have subse- any evidence of chemical disequilibrium between quently been shown to be in disequilibrium are left coexisting phases rendering the experimental run out of the data set. However, even with a broader suspect. This difference in melting style between inclusion of experimental data, the overall conclu- the experimental and natural melts we are com- sions of the paper are not altered. In addition, high paring to, ought to be considered as a possible degree melts of basaltic starting compositions reason why melts of a single lithology cannot

24 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure B1. A test of the less compatible/more compatible element versus MgO plots as a means of identifying end‐ member melt compositions. Accumulation of olivine ± cpx ± plag in whole rock analyses of magmas will primarily move the data points to the right on Figures 3 and 4. To demonstrate that the observed major element systematics at high MgO are mostly controlled by source and melting process, the graphs above illustrate the effect of olivine only, olivine + clinopyroxene, and olivine + clinopyroxene + plagioclase addition to a starting melt composition chosen to have a similar Nb/Zr to the enriched Icelandic end‐member. Typical Krafla clinopyroxene and plagioclase compositions were used, and olivine compositions were calculated using the model of Herzberg and O’Hara [2002]. Crystals were added in 0.1 wt% increments until the magma contained ∼12 wt% MgO, or the composition deviated markedly from the enriched end‐member chemistry. match the range of melt compositions observed on The accumulated fractional melts are then com- Iceland. Specifically, we need to reject the possi- pared over their pressure interval of melting to all bility that polybaric fractional melts from a single the isobaric batch melts calculated, to find the best lithology could be best matched by isobaric batch fitting batch melt to the accumulated melt compo- melts of another lithology, simply because the sition, and therefore an inferred source composi- difference in melting style makes isobaric batch tion. The misfit routine used is the same as in melts of the same lithology a worse match. section 3.1. If the best fits to polybaric accumulated fractional melts of one lithology came from [77] To test whether major element heterogeneity in isobaric batch melts of a different lithology, then it the source is going to be misidentified because of would suggest that the difference between experi- the difference between experimental and natural mental and natural melting style is too great for a melting, we have used pMELTS [Ghiorso and direct comparison between the two to be valid for Sack, 1995; Asimow and Ghiorso, 1998; Ghiorso constraining source composition. et al., 2002] to calculate accumulated fractional melts, and produce a corresponding data set of [78] The results of the misfit comparisons are isobaric batch melts, for KR4003 and KG2 starting shown in Figure C2, the quality of fit of the best compositions [Walter, 1998; Kogiso et al., 1998]. fitting isobaric batch melt at each pressure incre- These lithologies were chosen because they melt to ment of accumulated fractional melting is plotted. produce basaltic compositions similar to the Ice- For KR4003, Figure C1a, its own isobaric batch landic enriched (KG2) and depleted (KR4003) end‐ melts most often fit its accumulated fractional melt members. The melting path of KR4003 and KG2 composition, apart from in a 0.1 GPa interval during decompression melting, and further details below 1.5 GPa. The result is similar for fractional of the melting conditions are given in Figure C1. melts of KG2, which most frequently look like

25 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure C1. Melting paths for KR4003 and KG2 [Walter, 1998; Kogiso et al., 1998] calculated using pMELTS [Ghiorso and Sack, 1995; Asimow and Ghiorso, 1998; Ghiorso et al., 2002]. The solutions were initiated at 4 GPa and 1500°C. The presence of Na2O and K2O in the bulk composition means that the stable solution always includes a liquid phase, producing the tail to high pressures. The higher alkali content of KG2 compared with KR4003 is responsible for the former’s high‐pressure melt tail being at a relatively high melt fraction (∼7%).

Figure C2. Calculated isobaric batch melts of KR4003 and KG2 compared with calculated accumulated polybaric fractional melts of (a) KR4003 and (b) KG2. Accumulated fractional melts of KR4003/KG2 have been selected at different pressures to the top of the melting column (Ptop) and compared to a synthetic data set, in which the same lithologies have been melted isobarically at a range of pressures and temperatures (0.5–4 GPa and 1300–1650°C). The misfit calculations were performed as described in section 3.1. The quality of fit (= [Mmax − Mi]/Mmax) of the best fitting isobaric batch melt to each of the accumulated fractional melts is plotted as a horizontal bar, which ranges between 0 and 1; Mmax is the maximum misfit of all the records found to fit the accumulated melt compositions, and Mi the misfit of the particular best fitting batch melt at a given pressure. The bars are colored according to which lithology has produced the best fitting isobaric batch melt. All melting calculations were performed using pMELTS [Ghiorso and Sack, 1995; Asimow and Ghiorso, 1998; Ghiorso et al., 2002].

26 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Figure C3. A comparison of the best fitting KR4003/KG2 isobaric batch melts to (a and b) the accumulated fractional melts of KR4003 and (c and d) the accumulated fractional melts of KG2. Cbatch is the oxide concentration in the batch melt. For plotting Cbatch has been normalized to the corresponding oxide concentration in the accumulated fractional melt (Cfrac) it is being compared to. Lines are colored according to the mean pressure of melting (Pfrac) of the accumulated fractional melt being compared to. batch melts of KG2, rather than KR4003 batch heterogeneity in the source because of the differ- melts. The quality of fit to each of the individual ences between experimental melting (isobaric batch), oxides is recorded in Figure C3; in each plot the and natural melt generation (polybaric fractional). best fitting isobaric batch melt of the specified Both KR4003 and KG2 produce isobaric and lithology is compared with the composition of the fractional melts more similar to themselves than to accumulated fractional melt it is trying to fit. Com- melts of the other lithology. However, this result paring Figures C3a and C3b, shows that KR4003 does not readily generalize to other lithologies; batch melts reliably fit KR4003 fractional melts KR4003 and KG2 systematically provide a good fit across CFMAS, but that KG2 melts tend to show a to the depleted and enriched end‐member Icelandic higher FeO content than KR4003 fractional melts. melts, respectively, and so demonstrating that batch The equivalent comparison between Figures C3c melts of one cannot look like the fractional melts and C3d for fractional melts of KG2, demonstrates of the other, is the burden of proof required in that in most instances KR4003 batch melts are a Iceland’s case for there to be convincing major poor fit to the KG2 fractional melts; they system- element heterogeneity in the source. In other set- atically over estimate the MgO content of the melt tings, with different lithologies appearing to pro- and to a lesser extent the Al2O3 content. KG2 batch vide closer matches to the melt compositions, it is melts in contrast, provide an overall better fit, suf- not a given that isobaric and fractional melts of fering less from the high MgO and Al2O3 of the lithologies will appear so compositionally alike. KR4003 melts. [80] This result is heavily model dependent, and as [79] We can conclude from Figures C2 and C3, that improved thermodynamic models of mantle melt- we are unlikely to falsely identify major element ing are produced the distinction between isobaric

27 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G batch and polybaric fractional melting should be References revisited, and thus the relation between experimental and natural melts made more certain. Allen, R. M., et al. (2002), Plume‐driven plumbing and crustal formation in Iceland, J. Geophys. Res., 107(B8), 2163, doi:10.1029/2001JB000584. Appendix D: Estimating SiO2 Content Asimow, P. D., and M. S. Ghiorso (1998), Algorithmic mod- ifications extending MELTS to calculate subsolidus phase of Peridotite and G2 Melts relations, Am. Mineral., 83, 1127–1131. Asimow, P. D., and C. H. Langmuir (2003), The importance of [81] To estimate the SiO2 content of melts pro- water to oceanic mantle melting regimes, Nature, 421,815–820. duced from peridotite and pyroxenite melting, we Beattie, P. (1993), Olivine‐melt and orthopyroxene‐melt equi- parameterized the experimental results of four libria, Contrib. Mineral. Petrol., 115, 103–111. Brandon, A. D., D. W. Graham, T. Waight, and B. Gautason peridotite melting experiments and the G2 melting 186 187 3 4 (2007), Os and Os enrichments and high He/ He experiment of Pertermann and Hirschmann [2003a]. sources in the Earth’s mantle: Evidence from Icelandic The pressure (in GPa), melt fraction (F) and SiO2 picrites, Geochim. Cosmochim. Acta, 71, 4570–4591. data of Walter [1998], Wasylenki et al. [2003], Breddam, K. (2002), Kistufell: Primitive melt from the Iceland Falloon et al. [2001], and Robinson et al. [1998] mantle plume, J. Petrol., 43(2), 345–373. were used to constrain a surface describing SiO Breddam, K., M. D. Kurz, and M. Storey (2000), Mapping out 2 the conduit of the Iceland mantle plume with helium iso- concentrations in peridotite melts between 1 and topes, Earth Planet. Sci. Lett., 176,45–55. 7 GPa and F = 0–1: this gave the function, Brey, G. (1978), Origin of olivine melilitites: Chemical and experimental constraints, J. Volcanol. Geotherm. Res., 3, pd ½SiO2 ¼ 51:13ðÞ0:04 1:88ðÞ0:77 P þ Fpd½2:769ðÞ3:5 61–88. 0:14ðÞ0:77 Pþ0:15ðÞ0:11 P2: ðD1Þ Chauvel, C., and C. Hémond (2000), Melting of a complete section of recycled oceanic crust: Trace element and Pb isotopic evidence from Iceland, Geochem. Geophys. Geosyst., 1(2), SiO2 in peridotite melts therefore increases with 1001, doi:10.1029/1999GC000002. increasing melt fraction, and for most of the pressure Darbyshire, F. A., R. S. White, and K. F. Priestley (2000), interval considered decreases with increasing pres- Structure of the crust and uppermost mantle of Iceland from sure. 95% Confidence ranges for each of the fitted a combined seismic and gravity study, Earth Planet. Sci. Lett., 181, 409–428. parameters are given in parentheses. For the G2 Dasgupta, R., M. M. Hirschmann, and A. C. Withers (2004), pyroxenite, the limited pressure interval over which Deep global cycling of carbon constrained by the solidus melting experiments were performed means a of anhydrous carbonated eclogite under upper mantle condi- parameterization of melt composition as a function tions, Earth Planet. Sci. Lett., 227,73–85. of pressure is not possible. However, melts from G2 Dasgupta, R., M. M. Hirschmann, and N. Dellas (2005), The effect of bulk composition on the solidus of carbonated eclo- do show a decrease in SiO2 with increasing melt gite from partial melting experiments at 3 GPa, Contrib. fraction, and it is this relationship that we have Mineral. Petrol., 149, 288–305, doi:10.1007/s00410-004- included to predict the SiO2 content of pyroxenite 0649-0. px melts. The equation for SiO2 obtained from a linear Dasgupta, R., M. M. Hirschmann, and K. Stalker (2006), regression of the G2 data is, Immiscible transition form carbonate‐rich to silicate‐rich melts in the 3 GPa melting interval of eclogite + CO2 and px ‐ ½SiO2 ¼ 56:19ðÞ1:4 7:31ðÞ2:7 Fpx: ðD2Þ genesis of silica undersaturated ocean island lavas, J. Petrol., 57(4), 647–671. Unlike the peridotite melt systematics, the G2 Dasgupta, R., M. M. Hirschmann, and N. D. Smith (2007), Partial melting experiments of peridotite + CO2 at 3 GPa pyroxenite melts show a decrease in SiO2 as the melt and genesis of alkalic ocean island basalts, J. Petrol., 48(11), fraction increases. Equations (D1) and (D2) were 2093–2124, doi:10.1093/petrology/egm053. then integrated over the melt region using equation Dasgupta, R., M. G. Jackson, and C.‐T. A. Lee (2010), Major 14 of McKenzie and Bickle [1988]. All integrations element chemistry of ocean island basalts–Conditions of were stopped at 1 GPa, which is approximately the mantle melt heterogeneity of mantle source, Earth Planet. Sci. Lett., 289, 377–392. pressure at the base of the crust beneath central Davis, F. A., M. M. Hirschmann, and M. Humayun (2009), Iceland. The composition of low‐degree partial melts of garnet peridotite at 3 GPa by modified iterative sandwich experiments Acknowledgments (MISE), Eos Trans. AGU, 90(52), Fall Meet. Suppl., Abstract V31F‐04. [82] The authors gratefully acknowledge the work of Annette Davis, F. A., M. M. Hirschmann, and M. Humayun (2011), Jezierska in collecting the olivine analyses used in this paper. The composition of the incipient partial melt of garnet peri- Constructive reviews from Godfrey Fitton and one anonymous dotite at 3 GPa and the origin of OIB, Earth Planet. Sci. Lett., 308, 380–390, doi:10.1016/j.epsl.2011.06.008. reviewer are gratefully acknowledged, as are comments on an DeMets, C., R. G. Gordon, D. F. Argus, and S. Stein (1994), early draft from Sally Gibson and David Neave. OS was sup- Effect of recent revisions to the geomagnetic reversal time ported by NERC grant NE/H2449/4.

28 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

scale on estimates of current plate motions, Geophys. Res. and J. M. Sinton, pp. 281–310, AGU, Washington, D. C., Lett., 21(20), 2191–2194. doi:10.1029/GM071p0281. Devey, C. W., C. D. Garbeshonberg, P. Stoffers, C. Chauvel, Haase, K. M., C. W. Devey, and M. Wieneke (2003), Mag- and D. F. Mertz (1994), Geochemical effects of dynamic matic processes and mantle heterogeneity beneath the melting beneath ridges: Reconciling major and trace‐element slow‐spreading northern Kolbeinsey ridge segment, north variations in Kolbeinsey (and global) mid‐ocean ridge basalt, atlantic, Contrib. Mineral. Petrol., 144, 428–448. J. Geophys. Res., 99(B5), 9077–9095. Hammouda, T. (2003), High‐pressure melting of carbonated Eason, D. E., and J. M. Sinton (2009), Lava shields and fissure eclogite and experimental constraints on carbon recycling eruptions of the Western Volcanic Zone, Iceland: Evidence and storage in the mantle, Earth Planet. Sci. Lett., 214, for magma chambers and crustal interaction, J. Volcanol. 357–368. Geotherm. Res., 186, 331–348. Hardarson, B. S., and J. G. Fitton (1997), Mechanisms of Elliott, T. R., C. J. Hawkesworth, and K. Gronvöld (1991), crustal accretion in Iceland, Geology, 25, 1043–1046. Dynamic melting of the Iceland plume, Nature, 351, 201–206. Hards, V., P. Kempton, R. Thompson, and P. B. Greenwood Falloon, T. J., D. H. Green, L. V. Danyushevsky, and U. H. (2000), The magmatic evolution of the Snæfell volcanic Faul (1999), Peridotite melting at 1.0 and 1.5 GPa: An exper- centre; An example of volcanism during incipient rifting imental evaluation of techniques using diamond aggregated in Iceland, J. Volcanol. Geotherm. Res., 99,97–121. and mineral mixes for determination of near‐solidus melts, Hart, S. R. (1971), K, Rb, Cs, Sr and Ba contents and Sr J. Petrol., 40(9), 1343–1375. isotope ratios of ocean floor basalts, Philos.Trans.R.Soc. Falloon, T. J., L. V. Danyushevsky, and D. H. Green (2001), London A, 268(1192), 573–587. Peridotite melting at 1 GPa: Reversal experiments on partial Hauri, E. H. (1996), Major‐element variability in the Hawaiian melt compositions produced by peridotite‐basalt sandwich mantle plume, Nature, 382, 415–419. experiments, J. Petrol., 42(12), 2363–2390. Hémond, C., N. Arndt, U. Litchenstein, A. Hofmann, Fitton, G. J., A. D. Saunders, M. J. Norry, B. S. Hardarson, N. Oskarsson, and S. Steinthorsson (1993), The heteroge- and R. N. Taylor (1997), Thermal and chemical structure neous Iceland plume: Nd‐Sr‐O isotopes and trace element of the Iceland plume, Earth Planet. Sci. Lett., 153, 197–208. constraints, J. Geophys. Res., 98, 15,833–15,850. Fitton, G. J., A. D. Saunders, P. D. Kempton, and B. S. Herzberg, C. (2006), Petrology and thermal structure of the Hardarson (2003), Does depleted mantle form an intrinsic part Hawaiian plume from Mauna Kea volcano, Nature, 444, of the iceland plume?, Geochem. Geophys. Geosyst., 4(3), 605–609. 1032, doi:10.1029/2002GC000424. Herzberg, C. (2011), Identification of source lithology in the Foulger, G. R., and D. L. Anderson (2005), A cool model for Hawaiian and Canary Islands: Implications for origins, the Iceland hotspot, J. Volcanol. Geotherm. Res., 141,1–22. J. Petrol., 52(1), 113–146. Furman, P., F. Frey, and K. Park (1991), Chemical constraints Herzberg,C.,andM.J.O’Hara (2002), Plume‐associated on the petrogenesis of mildly alkaline lavas from vestman- ultramafic magmas of Phanerozoic age, J. Petrol., 43(10), naeyjar, Iceland: The eldfell (1973) and surtsey (1963–1967) 1857–1883. eruptions, Contrib. Mineral. Petrol., 109,19–37. Hirose, K. (1997), Partial melt compositions of carbonated Gaetani, G. A., and E. B. Watson (2000), Open system behav- peridotite at 3 GPa and role of CO2 in alkali‐basalt magma iour of olivine‐hosted melt inclusions, Earth Planet. Sci. generation, Geophys. Res. Lett., 24(22), 2837–2840. Lett., 183,27–41. Hirose, K., and I. Kushiro (1993), Partial melting of dry peri- Gee, M. A. M. (1999), Volcanology and geochemistry of the dotites at high pressures: Determination of compositions of Reykjanes Peninsula: Plume‐mid‐ocean ridge interaction, melts segregated from peridotite using aggregates of dia- Ph.D. thesis, Univ. of London, London. mond, Earth Planet. Sci. Lett., 114, 477–489. Gerbode, C., and R. Dasgupta (2010), Carbonate‐fluxed melt- Hirschmann, M. M. (2000), Mantle solidus: Experimental con- ing of MORB‐like pyroxenite at 2.9 GPa and genesis of straints and effects of peridotite composition, Geochem. HIMU ocean island basalts, J. Petrol., 51(10), 2067–2088, Geophys. Geosyst., 1, 1042, doi:10.1029/2000GC000070. doi:10.1093/petrology/egq049. Hirschmann, M. M., and E. M. Stolper (1996), A possible role Ghiorso, M. S., and R. O. Sack (1995), Chemical mass trasfer for garnet pyroxenite in the origin of the “garnet signature” in magmatic processes IV. A revised and internally consis- in MORB, Contrib. Mineral. Petrol., 124, 185–208. tent thermodynamic model for the interpolation and extrapo- Hofmann, A. W. (2003), Sampling mantle heterogeneity lation of liquid‐solid equilibria in magmatic systems at through oceanic basalts: Isotopes and trace elements, in elevated temperatures and pressures, Contrib. Mineral. Petrol., Treatise on Geochemistry: The Mantle and Core, edited by 119, 197–212. R. W. Carlson, H. D. Holland, and K. K. Turekian, chap. Ghiorso, M. S., M. M. Hirschmann, P. W. Reiners, and V. C. 2.03, pp. 61–101, Elsevier, New York. Kress III (2002), The pMELTS: A revision of MELTS Hofmann, A. W., and W. M. White (1982), Mantle plumes for improved calculation of phase relations and major element from ancient oceanic crust, Earth Planet. Sci. Lett., 57, partitioning related to partial melting of the mantle to 3 GPa, 421–436. Geochem. Geophys. Geosyst., 3(5), 1030, doi:10.1029/ Ita, J., and L. Stixrude (1992), Petrology, elasticity, and com- 2001GC000217. position of the mantle transition zone, J. Geophys. Res., Gibson, S. A., and D. Geist (2010), Geochemical and geophys- 97(B5), 6849–6866. ical estimates of lithospheric thickness variation beneath Ito, G., Y. Shen, G. Hirth, and C. J. Wolfe (1999), Mantle Galápagos, Earth Planet. Sci. Lett., 300, 275–286, flow, melting, and dehydration of the Iceland mantle plume, doi:10.1016/j.epsl.2010.10.002. Earth Planet. Sci. Lett., 165,81–96. Grove, T. L., R. J. Kinzler, and W. B. Bryan (1992), Fraction- Jackson, M. G., and R. Dasgupta (2008), Compositions of ation of mid‐ocean ridge basalt (MORB), in Mantle Flow HIMU, EM1 and EM2 from global trends between radio- and Melt Generation at Mid‐Ocean Ridges, Geophys. genic isotopes and major elements in ocean island basalts, Monogr. Ser., vol. 71, edited by J. P. Morgan, D. K. Blackman, Earth Planet. Sci. Lett., 276, 175–186.

29 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Jakobsson, S., K. Jónasson, and I. Sigurdsson (2008), The lavas from Askja volcano, Iceland: Implications for mechan- three igneous rock series of Iceland, Jökull, 58, 117–138. isms and timescales of magmatic differentiation, Geochim. Jónasson, K. (2005), Magmatic evolution of the Heiðarsporður at Cosmochim. Acta, 75, 570–587. ridge, NE‐Iceland, J. Volcanol. Geotherm. Res., 147, Kurz, M. D., P. S. Meyer, and H. Sigurdsson (1985), Helium 109–124. isotopic systematics within the neovolcanic zones of Iceland, Kaban, M. K., O. G. Flóvenz, and G. Pálmason (2002), Nature Earth Planet. Sci. Lett., 74, 291–305. of the crust‐mantle transition zone and the thermal state of Lacasse, C., H. Sigurdsson, S. N. Carey, H. Jóhannesson, L. E. the upper mantle beneath Iceland from gravity modelling, Thomas, and N. W. Rogers (2007), Bimodal volcanism at Geophys. J. Int., 149, 281–299. the Katla subglacial caldera, Iceland: Insight into the geo- Katz, R. F., M. Spiegelman, and C. H. Langmuir (2003), A chemistry and petrogenesis of rhyolitic magmas, Bull. Volcanol., new parametrization of hydrous mantle melting, Geochem. 69,373–399, doi:10.1007/s00445-006-0082-5. Geophys. Geosyst., 4(9), 1073, doi:10.1029/2002GC000433. Langmuir, C. H., G. N. Hanson, and M. J. O’Hara (1980), An Kawamoto, T., and J. R. Holloway (1997), Melting tempera- evaluation of major element heterogeneity in the mantle ture and partial melt chemistry of H2O‐saturated mantle peri- sources of basalts, Philos. Trans. R. Soc. London A, 297(1431), dotite to 11 GPa, Science, 276, 241–243. 383–407. Kelemen, P. B., G. Hirth, N. Shimizu, M. Spiegelman, and Langmuir, C. H., E. M. Klein, and T. Plank (1992), Petrolog- H. J. B. Dick (1997), A review of melt migration processes ical systematics of mid‐ocean ridge basalts: Constraints on in the adiabatically upwelling mantle beneath oceanic melt generation beneath ocean ridges, in Mantle Flow and spreading ridges, Philos. Trans. R. Soc. London A, 355, Melt Generation at Mid‐Ocean Ridges, Geophys. Monogr. 283–318. Ser., vol. 71, edited by J. P. Morgan, D. K. Blackman, and Kempton, P. D., J. G. Fitton, A. D. Saunders, G. M. Nowell, J. M. Sinton, pp. 183–280, AGU, Washington, D. C., R. N. Taylor, B. S. Hardarson, and G. Pearson (2000), The doi:10.1029/GM071p0183. Iceland plume in space and time: A Sr‐Nd‐Pb‐Hf study of Laporte, D., M. J. Toplis, M. Seyler, and J.‐L. Devidal (2004), the North Atlantic rifted margin, Earth Planet. Sci. Lett., A new experimental technique for extracting liquids from 177, 255–271. peridotite at very low degrees of melting: Application to par- Keshav, S., G. H. Gudfinnsson, G. Sen, and Y. Fei (2004), tial melting of depleted peridotite, Contrib. Mineral. Petrol., High‐pressure melting experiments on garnet clinopyroxe- 146, 463–484, doi:10.1007/s00410-003-0509-3. nite and the alkalic to tholeiitic transition in ocean‐island Li, A., and R. S. Detrick (2006), Seismic structure of Iceland basalts, Earth Planet. Sci. Lett., 223, 365–379. from Rayleigh wave inversions and geodynamic implica- Kinzler, R. J. (1997), Melting of mantle peridotite at pressures tions, Earth Planet. Sci. Lett., 241, 901–912, doi:10.1016/j. approaching the spinel to garnet transition: Application to epsl.2005.10.031. mid‐ocean ridge basalt petrogenesis, J. Geophys. Res., 102(B1), Maclennan, J. (2008a), Lead isotope variability in olivine‐ 853–874. hosted melt inclusions from Iceland, Geochim. Cosmochim. Klein, E. M., and C. H. Langmuir (1987), Global correlations Acta, 72, 4159–4176. of ocean ridge basalt chemistry with axial depth and crustal Maclennan, J. (2008b), Concurrent mixing and cooling of thickness, J. Geophys. Res., 92(B8), 8089–8115. melts under Iceland, J. Petrol., 49(11), 1931–1953, Kogiso, T., and M. M. Hirschmann (2001), Experimental doi:10.1093/petrology/egn052. study of clinopyroxenite partial melting and the origin of Maclennan, J., D. McKenzie, K. Grönvold, and L. Slater ultra‐calcic melt inclusions, Contrib. Mineral. Petrol., 142, (2001a), Crustal accretion under northern Iceland, Earth 347–360, doi:10.1007/s004100100295. Planet. Sci. Lett., 191, 295–310. Kogiso, T., and M. M. Hirschmann (2006), Partial melting Maclennan, J., D. McKenzie, and K. Grönvold (2001b), experiments of bimineralic eclogite and the role of recycled Plume‐driven upwelling under central Iceland, Earth Planet. mafic oceanic crust in the genesis of ocean island basalts, Sci. Lett., 194,67–82. Earth Planet. Sci. Lett., 249, 188–199, doi:10.1016/j. Maclennan, J., D. McKenzie, F. Hilton, K. Gronvöld, and epsl.2006.07.016. N. Shimizu (2003a), Geochemical variability in a single flow Kogiso, T., K. Hirose, and E. Takahashi (1998), Melting from northern Iceland, J. Geophys. Res., 108(B1), 2007, experiments on homogeneous mixtures of peridotite and doi:10.1029/2000JB000142. basalt: Application to the genesis of ocean island basalts, Macpherson, C. G., D. R. Hilton, J. M. D. Day, D. Lowry, and Earth Planet. Sci. Lett., 162,45–61. K. Gronvöld (2005), High‐3He/4He, depleted mantle and Kogiso, T., M. M. Hirschmann, and D. J. Frost (2003), High‐ low‐d18O, recycled oceanic lithosphere in the source of central pressure partial melting of garnet pyroxenite: Possible mafic Iceland magmatism, Earth Planet. Sci. Lett., 233, 411–427. lithologies in the source of ocean island basalts, Earth Mattsson, H., and N. Oskarsson (2005), Petrogenesis of alka- Planet. Sci. Lett., 216, 603–617, doi:10.1016/S0012-821X line basalts at the tip of a propagating rift: Evidence from (03)00538-7. the heimaey volcanic centre, south Iceland, J. Volcanol. Kokfelt, T. F., K. Hoernle, and F. Hauff (2003), Upwelling and Geotherm. Res., 147, 245–267. melting of the Iceland plume from radial variation of McKenzie, D., and M. J. Bickle (1988), The volume and com- 238U‐230Th disequilibria in postglacial volcanic rocks, Earth position of melt generated by extension of the lithosphere, Planet. Sci. Lett., 214, 167–186. J. Petrol., 29, 625–679. Kokfelt, T., K. Hoernle, F. Hauff, J. Fiebig, R. Werner, and McKenzie, D., and R. K. O’Nions (1991), Partial melt distribu- D. Garbe‐Schönberg (2006), Combined trace element and tions from inversion of rare earth element concentrations, Pb‐Nd‐Sr‐O isotope evidence for recycled oceanic crust J. Petrol., 32(5), 1021–1091. (upper and lower) in the Iceland mantle plume, J. Petrol., McKenzie, D., A. Stracke, J. Blichert‐Toft, F. Albaréde, 47(9), 1705–1749. K. Grönvold, and R. K. O’Nions (2004), Source enrichment Kuritani, T., T. Yokoyama, H. Kitagawa, K. Kobayashi, and processes responsible for isotopic anomalies in oceanic island E. Nakamura (2011), Geochemical evolution of historical basalts, Geochim. Cosmochim. Acta, 68(12), 2699–2724.

30 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Melson, W. G., T. O’Hearn, and E. Jarosewich (2002), A data Ringwood, A. E. (1962), A model for the upper mantle, brief on the smithsonian abyssal volcanic glass data file, J. Geophys. Res., 67(2), 857–867. Geochem. Geophys. Geosyst., 3(4), 1023, doi:10.1029/ Ritsema, J., W. Xu, L. Stixrude, and C. Lithgow‐Bertelloni 2001GC000249. (2009a), Estimates of the transition zone temperature in a Meyer, P., H. Sigurdsson, and J. Schilling (1985), Petrological mechanically mixed upper mantle, Earth Planet. Sci. Lett., and geochemical variations along Iceland’s neovolcanic 277, 244–252. zones, Geophys. Res., 90(B12), 223–226. Ritsema, J., P. Cupillard, B. Tauzin, W. Xu, L. Stixrude, and Murton, B. J., R. N. Taylor, and M. N. Thirlwall (2002), C. Lithgow‐Bertelloni (2009b), Joint mineral physics and Plume‐ridge interaction: A geochemical perspective from seismic wave traveltime analysis of upper mantle tempera- the Reykjanes Ridge, J. Petrol., 43(11), 1987–2012. ture, Geology, 37(4), 363–366. Nichols, A. R. L., M. R. Carroll, and A. Höskuldsson (2002), Robinson, J. A. C., B. J. Wook, and J. D. Blundy (1998), The Is the Iceland hot spot also wet? Evidence from the water beginning of melting fertile and depleted peridotite at contents of undegassed submarine and subglacial pillow 1.5 GPa, Earth Planet. Sci. Lett., 155,97–111. basalts, Earth Planet. Sci. Lett., 202,77–87. Schiellerup, H. (1995), Generation and equilibration of olivine Nicholson, H., and D. Latin (1992), Olivine tholeiites from tholeiites in the northern rift zone of Iceland. A petrogenetic Krafla, Iceland: Evidence for variations in melt fraction study of Bláfjall table mountain, J. Volcanol. Geotherm. within a plume, J. Petrol., 33(5), 1105–1124. Res., 65, 161–179. Nicholson, H., M. Condomines, J. G. Fitton, A. E. Fallick, Schilling, J. G. (1973), Iceland mantle plume: Geochemical K. Gronvöld, and G. Rogers (1991), Geochemical and study of reykjanes ridge, Nature, 242, 565–571. isotopic evidence for crustal assimilation beneath krafla, Schilling, J. G., M. Zajac, R. Evans, T. Johnston, W. White, Iceland, J. Petrol., 32(5), 1005–1020. J. D. Devine, and R. Kingsley (1983), Petrology and geo- Niu, Y., K. D. Collerson, R. Batiza, J. I. Wendt, and M. Regelous chemical variations along the Mid‐Atlantic Ridge from (2008), Origin of enriched‐type mid‐ocean ridge basalt at 29°N to 73°N, Am. J. Sci., 283, 510–586. ridges far from mantle plumes: The East Pacific Rise at Schilling, J. G., R. Kingsley, D. Fontignie, R. Poreda, and 11°20′N, J. Petrol., 49(4), 633–664. S. Xue (1999), Dispersion of the jan mayen and Iceland Parman, S. W., and T. L. Grove (2004), Harzburgite melting mantle plumes in the arctic: A He‐Pb‐Nd‐Sr isotope with and without H2O: Experimental data and predictive tracer study of basalts from the Kolbeinsey, mohns, and modeling, J. Geophys. Res., 109, B02201, doi:10.1029/ knipovich ridges, J. Geophys. Res., 104(B5), 10,543–10,569. 2003JB002566. Schwab, B. E., and A. D. Johnston (2001), Melting systemat- Peate, D. W., J. A. Baker, S. P. Jakobsson, T. E. Waight, A. J. R. ics of modally variable, compositionally intermediate perido- Kent, N. V. Grassineau, and A. C. Skovgaard (2009), Historic tites and the effects of mineral fertility, J. Petrol., 42(10), magmatism on the Reykjanes Peninsula, Iceland: A snap‐shot 1789–1811. of melt generation at a ridge segment, Contrib. Mineral. Shorttle, O., J. Maclennan, and S. M. Jones (2010), Control of Petrol., 157(3), 359–382, doi:10.1007/s00410-008-0339-4. the symmetry of plume ridge interaction by spreading‐ridge Peate,D.W.,K.Breddam,J.A.Baker,M.D.Kurz,A.K. geometry, Geochem. Geophys. Geosyst., 11, Q0AC05, Barker, T. Prestvik, N. Grassineau, and A. C. Skovgaard doi:10.1029/2009GC002986. (2010), Compositional characteristics and spatial distribution Sigurdsson, H. (1981), First‐order major element variation in of enriched Icelandic mantle components, J. Petrol., 51(7), basalt glasses from the Mid‐Atlantic Ridge: 29°N to 73°N, 1447–1475, doi:10.1093/petrology/egq025. J. Geophys. Res., 86(B10), 9483–9502. Pertermann, M., and M. M. Hirschmann (2003a), Anhydrous Sigurdsson, H., J. Schilling, and P. Meyer (1978), Skagi and partial melting experiments on morb‐like eclogite: Phase langjokull volcanic zones in Iceland: 1. Petrology and struc- relations, phase compositions and mineral‐melt partitioning ture, J. Geophys. Res., 83(B8), 3983–4002. of major elements at 2–3 GPa, J. Petrol., 44(12), 2173–2201. Sinton, J., K. Grönvold, and K. Sæmundsson (2005), Postgla- Pertermann, M., and M. M. Hirschmann (2003b), Partial melt- cial eruptive history of the Western Volcanic Zone, Iceland, ing experiments on a MORB‐like pyroxenite between 2 and Geochem. Geophys. Geosyst., 6, Q12009, doi:10.1029/ 3 GPa: Constraints on the presence of pyroxenite in the 2005GC001021. basalt source regions from solidus location and melting rate, Skovgaard, A. C., M. Storey, J. Baker, J. Blusztajn, and S. R. J. Geophys. Res., 108(B2), 2125, doi:10.1029/2000JB000118. Hart (2001), Osmium‐oxygen isotopic evidence for a Phipps Morgan, J. (2001), Thermodynamics of pressure recycled and strongly depleted component in the Iceland release melting of a veined plum pudding mantle, Geochem. mantle plume, Earth Planet. Sci. Lett., 194, 259–275. Geophys. Geosyst., 2(4), 1001, doi:10.1029/2000GC000049. Slater, L., D. McKenzie, K. Gronvöld, and N. Shimizu (2001), Pickering‐Witter, J., and A. D. Johnston (2000), The effects of Melt generation and movement beneath Theistareykir, NE variable bulk composition on the melting systematics of fer- Iceland, J. Petrol., 42(2), 321–354. tile peridotitic assemblages, Contrib. Mineral. Petrol., 140, Sobolev, A. V., and N. Shimizu (1993), Ultra‐depleted pri- 190–211. mary melt included in an olivine from the Mid‐Atlantic Press, W. H., S. A. Teukolsky, W. T. Vetterling, and B. P. Ridge, Nature, 363, 151–154. Flannery (1992), Numerical Recipes in FORTRAN: The Art Sobolev, A. V., A. W. Hofmann, S. V. Sobolev, and I. K. of Scientific Computing, 2nd Ed., Cambridge Univ. Press, Nikogosian (2005), An olivine‐free mantle source of Hawai- Cambridge, U. K. ian shield basalts, Nature, 434, 412–417. Prytulak, J., and T. Elliot (2007), Tio2 enrichment in ocean Sobolev, A. V., et al. (2007), The amount of recycled crust in island basalts, Earth Planet. Sci. Lett., 263, 388–403, sources of mantle derived melts, Science, 316, 590–597. doi:10.1016/j.epsl.2007.09.015. Sobolev,A.V.,A.W.Hofmann,G.Brügmann,V.G. Putirka, K., F. J. Ryerson, M. Perfit, and W. I. Ridley (2011), Batanova, and D. V. Kuzmin (2008), A quantitative link Mineralogy and composition of the oceanic mantle, J. Petrol., between recycling and osmium isotopes, Science, 321(5888), 52(2), 279–313, doi:10.1093/petrology/egq080. 536.

31 of 32 Geochemistry Geophysics 3 SHORTTLE AND MACLENNAN: COMPOSITIONAL TREND OF ICELANDIC BASALTS 10.1029/2011GC003748 Geosystems G

Sobolev,S.V.,A.V.Sobolev,D.V.Kuzmin,N.A.Krivolutskaya, crystallization–An experimental study at 1.0 GPa, J. Petrol., A. G. Petrunin, N. T. Arndt, V. A. Radko, and Y. R. Vasiliev 45(12), 2369–2388, doi:10.1093/petrology/egh042. (2011), Linking mantle plumes and, large igneous provinces Walter, M. J. (1998), Melting of garnet peridotite and the ori- and environmental catastrophes, Nature, 477,312–316, gin of komatiite and depleted lithosphere, J. Petrol., 39(1), doi:10.1038/nature10385. 29–60. Spandler, C., G. Yaxley, D. H. Green, and A. Rosenthal Wang, W., and E. Takahashi (1999), Subsolidus and melting (2008), Phase relations and melting of anhydrous K‐bearing experiments of a K‐rich basaltic composition to 27 GPa: eclogite from 1200 to 1600°C and 3 to 5 GPa, J. Petrol., 49, Implications for the behaviour of potassium in the mantle, 771–795, doi:10.1093/petrology/egm039. Am. Mineral., 84, 357–361. Stracke,A.,V.J.M.Salters,andK.W.W.Sims(1999), Wasylenki, L. E., M. B. Baker, A. J. R. Kent, and E. M. Stolper Assessing the presence of garnet‐pyroxenite in the mantle (2003), Near‐solidus melting of the shallow upper mantle: sources of basalts through combined hafnium‐neodymium‐ Partial melting experiments on depleted peridotite, J. Petrol., thorium isotope systematics, Geochem. Geophys. Geosyst., 44(7), 1163–1191. 1(1), 1006, doi:10.1029/1999GC000013. White, R. S., D. McKenzie, and K. O’Nions (1992), Oceanic Stracke,A.,A.Zindler,V.J.M.Salters,D.McKenzie,J.Blichert‐ crustal thickness from seismic measurements and rare earth Toft, F. Albarède, and K. Grönvold (2003), Theistareykir element inversions, J. Geophys. Res., 97(B13), 19,683–19,715. revisited, Geochem. Geophys. Geosyst., 4(2), 8507, White, R. S., J. W. Bown, and J. R. Smallwood (1995), The doi:10.1029/2001GC000201. temperature of the Iceland plume and origin of outward‐ Sun, S.‐S., R. W. Nesbitt, and A. Y. Sharaskin (1979), Geo- propagating V‐shaped ridges, J. Geol. Soc. London, 152, chemical characteristics of mid‐ocean ridge basalts, Earth 1039–1045. Planet. Sci. Lett., 44, 119–138. White, W. M., and A. Hofmann (1982), Sr and Nd isotope Takahashi, E. (1986), Melting of a dry peridotite KLB‐1upto geochemistry of oceanic basalts and mantle evolution, 14 GPa: Implications on the origin of peridotitic upper man- Nature, 296, 821–825. tle, J. Geophys. Res., 91(B9), 9367–9382. Yasuda, A., T. Fujii, and K. Kurita (1994), Melting phase rela- Takahashi, E., and N. Katsuji (2002), Melting process in the tions of an anhydrous mid‐ocean ridge basalt from 3 to Hawaiian plume: An experimental study, in Hawaiian Vol- 20 GPa: Implications for the behaviour of subducted oceanic canoes: Deep Underwater Perspectives, Geophys. Monogr. crust in the mantle, J. Geophys. Res., 99(B5), 9401–9414. Ser., vol. 128, edited by E. Takahashi et al., pp. 403–418, Yaxley, G. M., and G. P. Brey (2004), Phase relations of car- AGU, Washington, D. C. bonate bearing eclogite assemblages from 2.5 to 5.5 GPa: Takahashi, E., T. Shimazaki, Y. Tsuzaki, and H. Yoshida Implications for petrogenesis of carbonatites, Contrib. Mineral. (1993), Melting study of a peridotite KLB‐1to6.5GPa, Petrol., 146,606–619, doi:10.1007/s00410-003-0517-3. and the origin of basaltic magmas, Philos. Trans. R. Soc. Yaxley, G. M., and D. H. Green (2000), Reactions between London A, 342(1663), 105–120. eclogite and peridotite: mantle refertilisation by subduction Thirlwall, M. F., M. A. M. Gee, R. N. Taylor, and B. J. Murton of oceanic crust, Contrib. Mineral. Petrol., 139, 326–338. (2004), Mantle components in Iceland and adjacent ridges Yaxley, G. M., and A. V. Sobolev (2007), High‐pressure par- investigated using double‐spike Pb isotope ratios, Geochim. tial melting of gabbro and its role in the Hawaiian magma Cosmochim. Acta, 68(2), 361–386. source, Contrib. Mineral. Petrol., 154,371–383, doi:10.1007/ Tsuruta, K., and E. Takahashi (1998), Melting study of an s00410-007-0198-4. alkali basalt JB‐1 up to 12.5 GPa: Behaviour of potassium Zindler, A., S. R. Hart, and F. A. Frey (1979), Nd and Sr iso- in the deep mantle, Phys. Earth Planet. Inter., 107, 119–130. tope ratios and rare earth element abundances in Reykjanes Van der Hilst, R. D., and H. Kárason (1999), Compositional het- Peninsula basalts evidence for mantle heterogeneity beneath erogeneity in the bottom 1000 kilometers of earth’s mantle: Iceland, Earth Planet. Sci. Lett., 45, 249–262. Toward a hybrid convection model, Science, 283, 1885–1888. Zindler, A., E. Jagoutz, and S. Goldstein (1982), Nd, Sr, and Vidal, P. (1992), Mantle: More HIMU in the future, Geochim. Pb isotopic systematics in a three‐component mantle: A Cosmochim. Acta, 56, 4295–4299. new perspective, Nature, 298, 519–523. Villager, S., P. Ulmer, O. Müntener, and A. B. Thompson (2004), The liquid line of descent of anhydrous, mantle‐ derived, tholeiitic liquids by fractional and equilibrium

32 of 32