REE inverse modeling of HSDP2 basalts: Evidence for multiple sources in the Hawaiian plume



[1] The rare earth element (REE) concentrations of lavas from the Hawaiian Scientific Drilling Project (HSDP2) can be used to provide additional constraints on phase equilibria and the nature of the Hawaiian source. Major element analyses separate Mauna Kea lavas into two distinct populations, a high-silica and a low-silica suite. The low-silica samples can be separated stratigraphically into an upper low-silica alkalic series and a low-silica tholeiitic group that occurs deeper in the section. These contrasting groups could result from different extents of source partial fusion, or lithologically distinct source regions, or some combination of both factors. Petrologic modeling is performed to calculate that primary magma compositions contain about 20% MgO, and can be formed by 8–15% melting of a depleted mantle source for low-silica alkalic and high-silica lavas, respectively. The low-silica tholeiites could be generated by higher degrees of melting of a more fertile source. REE ratios and various isotopic systems reinforce the division of the low-SiO2 samples into the upper alkalic series, characterized by high Gd/Yb, and the deeper low-silica tholeiitic group, with low Gd/Yb. REE inverse modeling of fractionation-corrected basalts is consistent with lower degrees of melting to generate the late-stage alkalic lavas, with garnet present as a residual phase. The relatively constant Gd/Yb for low-silica tholeiites suggests that garnet is not an important residual phase during partial melting, implying higher extents of melting. The low-silica tholeiites are characterized by relatively enriched isotopic signatures that are consistent with contributions from a primitive source or from recycled subduction components. High 3He/4He associated with the low-silica lavas could derive from primitive mantle, mass transfer from the core, or from a refractory lithospheric contribution to a recycled subduction package. However, the combination of major element, REE and isotopic data suggests that the deeper low-silica suite is sampling the relatively fertile, interior part of the Hawaiian plume, whereas the high-silica lavas are extracted from the more depleted periphery; later alkalic lavas are generated from a depleted source as the volcano moves off the hot spot.

1. Introduction

[2] A major focus of the Hawaiian Scientific Drilling Project (HSDP) is to understand the long-term evolution of an individual volcano generated by the Hawaiian plume. A stratigraphic section in excess of 3000 m was collected in the second phase of the HSDP, comprising basalts from both Mauna Loa and Mauna Kea [see DePaolo et al., 1999, for details]. In this report, we use the rare earth element (REE) variations of Mauna Kea lavas to model partial melting of the Hawaiian plume.

[3] Previously, we performed REE modeling of Phase 1 HSDP basalts to constrain possible melting mechanisms and likely source mineralogies for the uppermost Mauna Kea lavas [Feigenson et al., 1996]. For the Phase 2 lavas, we utilize inverse modeling of individual suites of the recovered core to see if mineralogic variations in the calculated source are correlated with isotopic variations in the erupted lavas. Such correlations can be used to constrain possible source components, and may also help to infer their distribution within the plume.

2. Inverse Modeling

[4] The techniques employed here to model the Hawaiian source are based on years of development and modification [e.g., Shaw, 1970; Minster and Allègre, 1978; Hofmann and Feigenson, 1983; Hofmann et al., 1984; Albarède, 1995; Feigenson et al., 1996]. Basically, we use the variations of the REE in a cogenetic suite of lavas to model initial elemental concentrations and mineralogy of the basalt source. The REE are chosen because of their systematic geochemical behavior during partial melting and subsequent fractional crystallization.

[5] A major drawback of inverse modeling is the necessity of specifying a batch melting mechanism for the production of Hawaiian basalts. This simple model may perhaps not be consistent with current geophysical thought, which specifies that incremental processes, such as fractional melting or continuous melting, are more physically reasonable [e.g., McKenzie, 1985]. However, modeling performed on Phase 1 lavas [Feigenson et al., 1996] showed that pure, end-member incremental models give unrealistic source compositions, and require complete absence of garnet in the Hawaiian source, even for alkalic lavas. Better solutions are achieved by pooling fractional melts prior to extraction and eruption; these models closely approach the results calculated for batch processes. Because we cannot invert the nonlinear equations used for incremental melting processes, we use batch melting as a simpler analogue, noting that the real melting mechanism is probably some combination of end-member processes.

3. HSDP2 Sample Description

[6] Detailed sample descriptions are given in DePaolo et al. [1999] for 120 samples of the geochemical reference suite, covering over 3000 m of core from Mauna Loa and Mauna Kea volcanoes. The uppermost 245 m include mostly moderate to highly olivine-phyric, subaerially erupted, Mauna Loa basalts. Below this level, subaerial alkalic Mauna Kea basalts occur, ranging from aphyric to olivine- or plagioclase-phyric. Below about 300 m, the Mauna Kea lavas become dominantly olivine-phyric basalts. At about 1070 m, there is a transition from subaerial to submarine lavas. Samples from the submarine section include clasts from hyaloclastites, massive basalts, pillows, and intrusives. The entire reference suite is shown in Figure 1, subdivided into each descriptive type, illustrating variations in La and La/Yb with depth and age. Although there is a steady change in trace element concentration through the upper part of the core, and a large offset between Mauna Loa and Mauna Kea, there appears to be no correlation between geochemistry and sample type.

Figure 1.

Variations in La and La/Yb with depth and age; depth/age relations derived from DePaolo et al. [2001]. Lavas are differentiated with respect to their petrologic occurrence in the drill core: open triangles, Mauna Loa subaerial basalts; closed triangles, Mauna Kea subaerial basalts; open squares, MK hyaloclastites; closed squares, MK massive submarine flows; open diamonds, MK pillows; closed diamonds, MK intrusives. Although trace element concentrations vary throughout the core, there is no systematic variation with sample type.

4. Major Element Data

[7] Major elements for the reference suite (Table 1) have been obtained by J. M. Rhodes and M.J. Vollinger (Composition of Basaltic Lavas Sampled by the Hawaiian Scientific Drilling Project: Geochemical Stratigraphy and Magma Types, manuscript submitted to Geochemistry, Geophysics, Geosystems, 2002) and are required for the REE inverse modeling performed in this study. In addition, major element variation diagrams show distinct populations for HSDP2 samples that are best demonstrated in a plot of MgO versus SiO2 (Figure 2). Mauna Kea lavas can be subdivided on this diagram into two groups: a low-SiO2 group that includes some alkalic lavas, and a high-SiO2 group. Lavas from Mauna Loa overlap the Mauna Kea high-SiO2 group. A fundamental question to be addressed in this report is the relationship between these variable-silica groups: do they represent distinct mantle lithologies [e.g., Hauri, 1996] or do they result from different conditions of phase equilibria [e.g., Herzberg and O'Hara, 2002]? Trace element and isotopic systematics may be used to resolve these possibilities. Figures 3 and 4are a CMAS projection and an FeO versus MgO diagram illustrating the two lava suites, as well as calculated primary magmas.

Figure 2.

MgO-SiO2 for Mauna Kea lavas; data from Rhodes and Vollinger (submitted manuscript, 2002). Lavas can be divided into a high silica suite (red circles) and a low silica suite (blue squares) which includes alkali basalts. Geochemical variations within each suite are dominated by olivine fractionation, and these are modeled by small circles emanating from estimated primary magma compositions (crosses). Each suite has two primary magmas, the low MgO one estimated for accumulated fractional melting and the high MgO one for equilibrium melting. Melting mechanism has little effect on major element compositions. Variations in peridotite source composition also have little effect on primary magma estimates [Herzberg and O'Hara, 2002].

Figure 3.

CMAS projection of HSDP2 lavas from diopside [Herzberg and O'Hara, 2002]; symbols as in Figure 2. High and most low silica samples exhibit olivine control, whereas some alkalic samples show diverging fractionation trends. Residue assemblages, polybaric cotectics, and melt fraction contours are for equilibrium melting of a depleted peridotite source [Herzberg and O'Hara, 2002]. Primary magmas formed by equilibrium melting in each suite have highest MgO contents and project closest toward olivine. Note that high and most low silica primary magmas are similar to liquids that separated from a harzburgite residue [L + Ol + Opx], but the alkali basalt primary magmas are close to garnet-bearing residues.

Figure 4.

FeO-MgO diagram for HSDP2 basalts; symbols as in Figure 2. Total iron is separated as FeO and Fe2O3 for each lava with Fe3+/ΣFe = 0.1 [Herzberg and O'Hara, 2002]. Note the relatively constant MgO of the primary magmas for variable degrees of melting, but alkali basalts contain more FeO. Note also that high and most low silica primary magmas are similar to liquids that separated from a harzburgite residue [L + Ol + Opx]. Primary magmas are constrained by melt fractions unique to each suite in both this figure and in projection (Figure 3), and these range from about 0.08 for the alkali basalts to 0.16 for the high silica tholeiites.

Table 1. Major Element and REE Data for HSDP2 Basaltsa
  • a

    REE data determined by HR-ICP-MS at Rutgers University (see appendix A for details). Major elements from Rhodes and Vollinger, [2002].

Depth mbsl9.5034.0045.5053.4059.5068.7086.2091.2098.80125.40137.00149.90163.30177.80197.40209.10222.50233.70242.00246.20252.90256.50261.70267.50274.40281.30293.00305.80326.70353.00378.40398.10421.20443.60467.80490.90516.20542.10563.50589.20615.80636.00658.30678.60695.90724.10759.80793.60812.70833.90871.20888.40921.80948.90984.201012.401037.701061.201083.801098.201123.201229.601265.201311.901352.601394.901404.101435.401474.701497.701521.401548.201549.301581.201605.001636.301678.701705.501739.301763.201794.801823.301852.001883.601921.601933.801973.802009.802062.702098.602123.702157.402209.502218.202280.202300.202321.602357.002414.102467.302503.502525.302550.902581.802615.002654.102730.202759.302770.902789.902825.802837.602919.502961.002967.803009.203019.003058.003068.90

[8] Primary magmas are estimated (Table 2) with a hybrid forward and inverse model discussed in detail in Herzberg and O'Hara [2002]. The forward component identifies potential primary magma compositions as functions of melt fraction for an assumed peridotite source and melting mechanism, based on a parameterization of experimental data. These are displayed in diagrams of CMAS projections (Figure 3) and FeO-MgO (Figure 4). The inverse component selects a derivative liquid in an erupted lava suite for which olivine fractionation is computed, and this provides an array of possible primary magma compositions shown as dotted arrays in Figures 24. The arrays of potential primary magmas in both forward and inverse models are compared, and model melt fractions are used to seek a unique primary magma solution for each lava suite. Model solutions provide primary magma composition, average melt fraction, residuum mineralogy, and temperature of eruption and melt collection in the melting regime.

Table 2. Mauna Kea Primary Magmasa
 Hi SiHi SiLow SiLow SiAlkali BasaltAlkali Basalt
  • a

    EQ, Equilibrium Meling; APFM, Accumulated Perfect Fractional Melting; Melt Fractions for a Depleted Mantle Peridotite Source.

Mg# Olivine91.4190.891.390.890.9990.5
Melt Fraction0.

[9] Calculated primary magmas needed to generate the low-silica, high-silica and alkalic lava suites are given in Table 2 for both equilibrium melting and accumulated perfect fractional melting. Results presented in Figures 3 and 4 are for accumulated fractional melting of an assumed depleted peridotite source having 41.8% MgO, 8.1% FeO and 2.36% Al2O3 [Herzberg and O'Hara, 2002]; primary magmas contain about 20% MgO and are formed by 0.08 to 0.15 mass fractions of melting (Figure 4, Table 2). The effect of equilibrium melting is to raise the MgO content of the model primary magma by about 1 wt.% (Table 2). A fertile peridotite with 38.1% MgO, 8.0% FeO and 4.3% Al2O3 provides nearly identical primary magma compositions for accumulated fractional melting, but requires about 0.3 mass fractions of melting [Herzberg and O'Hara, 2002]. An assumed Fe-rich peridotite source with FeO raised from 8.0% to 9.0 wt.% yields primary magmas with the same FeO contents as given in Table 2, a result that may not be intuitively obvious, but with MgO contents that are lowered to about 17% [Herzberg and O'Hara, 2002].

[10] MgO in our primary magmas is restricted to 19–21%, similar to that offered by Albarède [1992] but higher than those in most models of Hawaiian volcanism [e.g., Hauri, 1996]. The amount of olivine added in the inverse model is regulated according to constraints imposed by forward melting models, and yields primary magmas that would precipitate olivines with forsterite contents of 90–91 (Table 2). These are similar to maximum forsterite contents of phenocrysts from Mauna Kea lavas [Baker et al., 1996; Garcia, 1996]. Most models compute the MgO content of a primary magma from the FeO content of a lava series and known olivine phenocryst compositions using the exchange coefficient KDOl/LFeO/MgO, which varies from 0.30 for basalts with 8.0% MgO to 0.36 for picrites with about 20% MgO [Herzberg and O'Hara, 2002]. Use of 0.30 instead of 0.36 for primary picritic magmas will result in an underestimation of MgO by about 2 weight%. Iron in Hawaiian lavas is typically about 10% ferric, and any computation procedure must consider only the FeO that is exchangeable with olivine. Finally, estimates of primary magma compositions are often based on direct observations of the MgO contents of erupted lava compositions, but these are typically derivative liquids having compositions that cannot be in equilibrium with any peridotite composition [Herzberg and O'Hara, 2002]. The problems associated with estimating the MgO contents of primary magmas for Hawaii and other occurrences are discussed in more detail in Herzberg and O'Hara [2002]. For the REE modeling that follows, the exact MgO content of the primary magma is not critical: all lavas are corrected for the effects of olivine fractionation, and if the incorrect MgO content is used for the parent, all daughter lavas will be offset to the same extent. Because the REE modeling traces relative variations in REE behavior, there is no additional error introduced by using a primary magma with incorrect MgO (assuming that olivine is the only liquidus phase).

[11] There is little change in the MgO contents of primary magmas with increased melt fraction during decompression because adiabatic T-P paths are nearly coincident with olivine-liquid T-P saturation surfaces [see Herzberg and O'Hara, 2002]. SiO2 contents of computed primary magmas vary from about 45.5% for the alkali basalts to 47.7% for the high silica suite. High silica contents in some Mauna Kea lavas arise from primary magmas that develop high SiO2 owing to dissolution of orthopyroxene during decompression melting of residual harzburgite [Herzberg and O'Hara, 2002], and from the effects of olivine fractionation (Figure 2). Hauri [1996] suggested that primary magmas with 47–48% SiO2 are too silica rich to have formed from mantle peridotite, and suggested that the silica problem can be solved by the involvement of veins of recycled eclogitic crust. But the silica problem for Hawaii is an artifact of (1) the amount of olivine that is added to a lava in the inverse model (less olivine, more silica) and (2) a comparison of Hawaiian lavas with experimental liquids in equilibrium with lherzolite [Hauri, 1996], not harzburgite. Primary magmas formed by harzburgite melting are usually higher in SiO2 relative to those for lherzolite melting [Herzberg and O'Hara, 1998, 2002]. The exception is extremely low melt fractions of spinel lherzolite melting [Baker and Stolper, 1994], but these are restricted to low pressures and will contribute minimally to the total mass fraction of an aggregate melt.

5. Rare Earth Element Data

[12] Analyses of the REE contents of 120 basalts collected from HSDP Phase 2 are presented in Table 1; analytical methods are discussed in detail in Appendix A. The raw data are displayed in Figures 5a through 5f to show the variations in REE patterns with depth in the core. Most obviously, there is a sharp break between the relatively low LREE abundance lavas from Mauna Loa and the more LREE enriched lavas from the uppermost levels of Mauna Kea. This is also shown in Figure 1 and confirms the contrast between the volcanoes seen from the Phase 1 REE results [e.g., Feigenson et al., 1996]. Samples older than about 300 ky show lower La/Ce, La/Sm, La/Yb and Gd/Yb than the younger alkalic samples (Figures 6a6d). For the older samples, there appears to be a tendency for the low-silica lavas to exhibit higher La/Ce, La/Sm and La/Yb than the high-silica suite.

Figure 5.

Chondrite-normalized REE concentrations (normalizations from McDonough and Sun [1995]) for lavas from HSDP2. (a) Uppermost 19 Mauna Loa flows (triangles); 0–246 mbsl. (b) Mauna Kea lavas 121–4.40 to 240–3.30 (246–564 mbsl); samples are subdivided into low silica and alkalic samples (open squares) and high silica lavas (solid circles). (c) MK lavas 256–0.95 to 450–3.55 (589–1083 mbsl), symbols as in 5b. (d) MK lavas 455–7.40 to 675–6.90 (1098–1739 mbsl), symbols as in 5b. (e) MK lavas 683–5.75 to 826–20.60 (1763–2413 mbsl), symbols as in 5b. (f) MK lavas 836–5.80 to 967–2.75 (2467–3068 mbsl), symbols as in 5b.

Figure 6.

Ratios of REE versus age for MK lavas; (a) La/Ce, (b) La/Sm, (c) La/Yb, (d) Gd/Yb; symbols as in Figure 2. Note the strong increase in all ratios for low-silica alkalic lavas near the top of the section, and the tendency for low-silica lavas at greater depth toward slightly elevated La/Ce, La/Sm, La/Yb and slightly lower Gd/Yb compared to high-silica lavas.

[13] Variations in REE content with depth can be due to several factors, including changes in the degree of partial melting, differences in source mineralogy and source concentration, and different extents of subsequent fractional crystallization. The geochemical modeling performed below attempts first to correct for the effects of variable amounts of fractional crystallization, and then to solve for relative source differences in concentration and mineralogy, independent of the extent of melting. The ultimate aim is to see if the high- and low-silica suites represent variability in the mantle source with respect to initial REE concentration, source mineralogy, or extent of partial melting.

6. Inverse Modeling of HSDP Phase 2 Basalts

[14] As described above, the geochemical modeling used here follows that of Feigenson et al. [1996] and is discussed in detail in Appendix B. The procedure in brief is this: first, the lavas are corrected to primary magma compositions (Table 2) by using least squaress modeling [Bryan et al., 1969] of major elements (Rhodes and Vollinger, submitted manuscript, 2002); fractionation-corrected REE concentrations are plotted and the data regressed on “process identification” diagrams [see Minster and Allègre, 1978]; finally, various melt mineralogies are input to calculate source REE concentrations and bulk partition coefficients relative to the La concentration in the initial source. Uncertainty in the final calculations is introduced by not knowing the exact parental magma composition or its precise high pressure mineralogy, and by scatter in the process-identification diagrams (Figures 7a7e). The first two problems affect all lavas equally, but scatter in Figures 7a7e introduces error into the final calculations (Figures 8a8d).

Figure 7.

Process-identification regression plots for the REE [after Minster and Allègre, 1978]; slopes and intercepts of regressions with 1σ errors are given in each graph. The information obtained from these figures is input into inverse modeling equations to derive REE patterns and partitioning behavior in the source region (see text and Appendix B for details). The entire suite of samples is subdivided in two different ways; first, by stratigraphic position (Figures 7a and 7b), and second, by variable silica (Figures 7c and 7d). (a) REE regressions for all Mauna Kea lavas above 284–1.75 (inclusive). (b) Regressions for MK lavas below 294–7.65 (inclusive). (c) Regressions for all high silica MK samples from the entire section (see Figure 2 for sample classification). (d) Regressions for low silica and alkalic lavas from the entire suite. (e) Low silica samples from lower levels in the core; regression information cannot be obtained because all data plot as a single point.

Figure 8.

(a) Calculated source REE concentrations (Ci) and bulk partition coefficients (Di), relative to CLa (source) = 1, for uppermost MK lavas (SR0284-1.75 and above). Spectrum of calculated source parameters derived from variable input melt norms (Pi) including 0–20% garnet in the melt. Calculations based on regressions from Figure 7a and equations given in the text and Appendix B. Best fit (broad similarity between Pi and Di) is obtained for about 5–15% garnet entering the melt. Calculated source REE patterns are roughly chondritic (La/Yb = 1). (b) Calculated source parameters as in Figure 8a, but for MK samples from SR0294-7.65 and below. Compared to the uppermost lavas, the calculated source requires less garnet (0–5%) but all source REE patterns are enriched (La/Yb > 1). (c) Source REE patterns for high-silica samples from throughout the core, parameters as in Figure 8a. Results are similar to Figure 8b, in that the calculated source is LREE-enriched, but contains little (0–5%) garnet. (d) Calculated source parameters for low-silica and alkalic MK lavas; all successful results require significant garnet entering the melt (5–15%). Note the much steeper slope of Di for the heavy REE compared to Figure 8c.

[15] Calculated primary magma compositions, following the method of Herzberg and O'Hara [2002], are shown in Figures 24. Similar results are obtained whether the melting process is one of aggregated fractional melts or bulk equilibrium melts. All primary magmas contain between 19% and 21% MgO, which remains relatively constant, independent of the degree of melting.

[16] The fractionation-corrected REE data for the Mauna Kea basalts are listed in Table 3 and source parameters calculated by inverse modeling are shown in Figures 8a8d. The reference suite has been subdivided two different ways: first, by separating lavas based on stratigraphy (upper MK versus lower MK, with the division between 284–1.75 and 294–7.65); and second, based on silica content (high silica versus low silica). For the separation based on stratigraphy, the uppermost MK lavas appear to be derived from a less enriched source than the lower MK lavas (the calculated La/Yb of the source is lower for the upper lavas), and garnet is a more important residual phase (compare Figures 8a and 8b). However, the division based on stratigraphy alone may blend together lavas that might be derived from different sources, leading to less clear-cut distinctions in the final calculations.

Table 3. Fractionation-Corrected Major Element and REE Data for Mauna Kea Basalts From HSDP2a
  • a

    Samples corrected to primary magma compositions from Table 2.

Depth in meters2462532562622672742822933063273533783984214444684915165425645896166366586796967247607938138348718899229499841012103810611083109811231229126513121352139514041435147414971521154815491574158116361678170517391763179418231851188319211933197320092062209821232157220922182280230023212356241324672503252525502581261426532730275927702791282528372919296029673008301930573068

[17] Lavas regressed separately based on silica content (Figures 7c and 7d) lead to sharper distinctions in the source parameters calculated by inverse modeling (Figures 8c and 8d). The low silica lavas (dominated by the uppermost alkalic lavas) show a strong influence of residual garnet (calculated D values increase steadily from LREE to HREE) and a somewhat less enriched source (La/Ybcn < 1 for several different input P values). The high silica lavas require a source with slight LREE enrichment (La/Ybcn > 1) but with a minimal garnet signature (less than a few percent). No direct distinction can be made between the two suites regarding relative degrees of melting, because the inverse procedure removes that parameter from the modeling. However, lower Gd/Yb for the high-silica suite (Figure 9) implies higher melt fraction than for the alkalic lavas. The deeper low-silica lavas have the lowest Gd/Yb of any MK samples, implying still higher degrees of melting and absence (or near absence) of residual garnet.

Figure 9.

Gd/Yb versus 3He/4He (M. D. Kurz, J. Curtice, D. Lott, and A. Solow, Rapid helium isotopic variability in Mauna Kea shield lavas from the Hawaiian Scientific Drilling Project, manuscript submitted to Geochemistry, Geophysics, Geosystems, 2002) for HSDP2 lavas; symbols as in Figure 5. Uppermost alkalic lavas have high Gd/Yb but low 3He/4He, indicating low degree melting of a depleted mantle source with residual garnet. Low-silica lavas from deep in the core have low Gd/Yb but high 3He/4He, resulting from moderately high degrees of melting of a relatively primitive source (but see text for an alternative model). Mauna Loa lavas are distinct from all MK lavas by their lower Gd/Yb.

7. Correlations With Isotopes

[18] Isotopic data (obtained by other researchers) amplifies some of the source distinctions noted in the major elements and REEs. Figure 9 shows the strong correlation between Gd/Yb and 3He/4He [from Kurz, 2000]. Low silica samples show two distinct populations: late-stage alkalic lavas are quite depleted in 3He/4He, whereas the deeper low silica samples show the greatest enrichment in 3He/4He; high silica lavas are intermediate. The uppermost low silica, alkalic samples are easily explained as small degree melts of a source that is increasingly infused with asthenospheric mantle [e.g., Chen and Frey, 1983; Feigenson, 1984]. The deeper low silica samples are more difficult to explain: inverse modeling cannot be performed on them, because the entire group plots as a single cluster of points on the process identification diagrams (Figure 7e). However, the low and approximately constant ratio of Gd/Yb for these samples implies melt fractions high enough such that garnet is no longer a residual phase during partial melting. This group of lavas tends to be isotopically enriched compared to the high silica suite for a number of isotopic systems (e.g., low 143Nd/144Nd, high 87Sr/86Sr, elevated 208Pb/204Pb, low 176Hf/177Hf; Figures 1012; data from J.G. Bryce and D.J. DePaolo (High resolution Sr and Nd isotopic records of Mauna Kea volcano: Results from the second phase of the Hawaii Scientific Drilling Project (HSDP-2), manuscript submitted to Geochemistry, Geophysics, Geosystems, 2002) and Blichert-Toft et al. [2002]). Yet 207Pb/204Pb is not distinguishable from the high silica suite, pointing to a long-term fractionation of U from Th in the source of the low silica samples.

Figure 10.

208Pb/204Pb versus 206Pb/204Pb for all lavas from HSDP2 (data from Blichert-Toft et al. [2002]); symbols as in Figure 5. High-silica MK lavas and all Mauna Loa samples lie on a single line between DM and HIMU; deep, low-silica MK lavas fall off the trend in the direction of EMII (pelagic sediment, or recycled continental crust). The low-silica samples are not distinct in their 207Pb/204Pb characteristics, suggesting that their source has been fractionated in U/Th compared to normal mantle reservoirs.

Figure 11.

3He/4He versus epsilon Nd (from Bryce and DePaolo, submitted manuscript, 2002) for MK lavas; symbols as in Figure 5. Note the strong negative correlation, with upper level alkalic samples showing depleted mantle (DM) characteristics (high εNd and low 3He/4He), and deep low-silica lavas displaying relatively enriched-mantle influence.

Figure 12.

Sr-Nd isotope correlation diagram for MK lavas; symbols as in Figure 5. The high-silica samples show almost no correlation, but an overall negative trend is displayed by the low-silica lavas. Late-stage alkalic basalts are derived from a relatively depleted source, whereas the deep low-silica samples have the most enriched signatures.

[19] It appears then that the two populations of lavas from Mauna Kea that are contrasted by their major element compositions are also distinct in isotopic and trace element characteristics. Therefore, the different lava types are extracted from distinct mantle domains within the Hawaiian plume.

8. Discussion

[20] The combination of major element, REE and isotopic data for Mauna Kea basalts produces some interesting contrasts among the erupted lavas. Based on major element variations, the MK lavas are separable into high- and low-silica suites. The low-silica suite can be further subdivided based on stratigraphy: samples near the top of the MK section are alkalic and formed by low degrees of melting of a garnet-bearing, isotopically depleted source; deeper low-silica samples are derived from an isotopically enriched source relative to the high-silica suite. There exist a number of possibilities to account for the low-silica tholeiites, including: (1) direct melting of recycled, siliceous pelagic sediment [e.g., Hauri, 1996]; (2) melting of a recycled package, including pelagic sediments and residual lithosphere; or (3) derivation from primitive mantle. Additionally, mass transfer from the outer core [e.g., Porcelli and Halliday, 2001] may play a role. These scenarios are addressed as follows:

8.1. Isotopically Enriched Components of the Hawaiian Plume Derive From Direct Melting of Recycled Pelagic Sedimentary Components

[21] This model accounts for the isotopic enrichment of low-silica tholeiites, except for elevated 3He/4He. Inconsistency between 207Pb and 208Pb for these lavas can be addressed by U/Th fractionation during hydrospheric involvement. The He isotopic enrichment could be generated by other sources, such as core outgassing. However, direct melting of pelagic sediment would lead to high-silica lavas, the opposite of what is observed for HSDP2. Moreover, such melting of a siliceous component to generate basaltic magma is not likely; rather any partial fusion is expected to refertilize peridotite [e.g., Herzberg and O'Hara, 2002; Yaxley, 2000]. In addition, there is no reason that enriched 3He/4He should be correlated with other isotopes, as a separate source is involved.

8.2. Low-Silica Tholeiites are Derived From Recycled Crust/Residual Lithosphere

[22] In this model, the source of isotopic enrichment for the low-silica tholeiites is mainly generated from recycled crust. In contrast to the first model, 3He/4He enrichment derives from the incorporation of residual lithosphere as part of a recycled package. It is possible that incomplete extraction of He combined with complete removal of U and Th during oceanic crust formation would lead to pockets of primordial He in the residual lithosphere [Albarède, 1998]. U/Th fractionation in sediments included in the subduction package can explain the Pb isotope systematics. This model implies that isotopically enriched heterogeneities will be perhaps randomly distributed throughout the plume.

[23] Although this model can adequately account for the isotopic systematics of the low-silica tholeiites, there is an inconsistency with major element and REE variations. In particular, melting of a residual lithospheric component would necessarily lead to smaller melt fractions, which would be more likely to preserve residual garnet during melting. In fact, garnet control is absent for these samples, compared to the strong role of residual garnet shown for the uppermost low-silica alkalic lavas.

8.3. Low-Silica Tholeiites are Derived From a Relatively Fertile, Primitive Mantle Source, Possibly Combined With Mass Transfer From the Core

[24] In this model, the relatively enriched isotopic signatures of the low-silica tholeiites, perhaps including high 3He/4He, are derived from a primitive mantle source. Relatively high degrees of melting of an undepleted, fertile peridotite source are consistent with the major element chemistry of these lavas. As previously stated, the constant Gd/Yb for this suite suggests the absence of residual garnet, which is consistent with high degrees of melting. The only significant negative aspect of this model is the Pb isotopic variations, which seem to require U/Th fractionation in the source. Although this is unlikely to occur in any pure mantle reservoir, it is possible that mass transfer from the outer core takes place [e.g., Porcelli and Halliday, 2001] and could account for the disparate behavior of 207Pb and 208Pb. Moreover, in this case, significant 3He could be contributed to the plume from the core. Physically, this model implies that the Hawaiian plume is zoned, with a relatively fertile, isotopically enriched, hotter interior (yielding low-silica, high degree melts), that grades outward toward a more depleted mantle that generates high-silica lavas. When the volcano migrates off the hot spot and melting intensity decreases, late-stage, low-silica alkali basalts are produced.

[25] In summary, the Hawaiian plume is isotopically heterogeneous, and these heterogeneities correlate with petrologic controls on melting. High-silica and low-silica lavas are generated from isotopically discrete source regions: high-silica magmas are derived by melting of depleted peridotite, whereas low-silica magmas form in two different environments. Late-stage, low-silica alkalic lavas are generated by small degree melts of a depleted peridotite source, similar to the source of high-silica magmas. Low-silica tholeiites, by comparison, are created by higher degrees of melting of more fertile peridotite. This source is also isotopically enriched compared to the source of the high-silica magmas, and may receive a contribution from the Earth's core. Although recycled crustal/lithospheric components may be present in the plume, the genesis of Hawaiian basalts is governed by peridotite melting.

Appendix A.: Analytical Procedures

[26] Rock powders were digested in Savillex screw-top Teflon vials using a HF/8N HNO3 mixture, followed by two fluxes of 8N HNO3 to ensure the removal of any fluoride complexes. Before each new flux, samples were sonicated for a minimum of 30 min to segregate any sample grains thus minimizing undigested material. Following each flux, the samples were cooked down to dryness. The final digested product was taken up in 4N HNO3. Samples were further diluted for instrumental analysis in 3% HNO3. Along with each batch of samples (10 to 12) a digestion blank or the rock standard BHVO-1 was also digested to check any possible contamination as well as the precision and accuracy of the entire analytical process.

[27] Rare earth elements were analyzed on a Finnigan MAT Element, a high resolution inductively coupled plasma mass spectrometer (HR-ICP-MS), following the method of Field and Sherrell [1998]. In short, concentrations were determined by isotope dilution, using an enriched 145Nd and 171Yb spike. The Nd spike was used to calculate REE from 139La to 157Gd, and the Yb was used to calculate from 158Gd to 175Lu. Both a light and heavy isotopic spike are needed due to the magnet jump between 157Gd and 158Gd. Sample introduction was preformed using a Cetac MCN6000, which heats the sample to minimize oxide production. The argon and nitrogen additional gas flows on the MCN6000 were tuned to maximize sensitivity and stability while minimizing oxides (LaO/La = 0.008%) and doubly charged ions (Ba++/Ba = 5%). The SEM voltage was tuned to obtain flat top peaks, while the focus offset was optimized to get the proper isotopic ratios of the REE. All other tuning parameters were optimized for sensitivity and stability.

[28] The standard, BHVO-1, was digested a total of six times. Each of these digestions were diluted and run on the HR-ICP-MS five times throughout the run. The average of the BHVO-1s is listed in the table below along with the standard deviation between each of the six BHVO-1 digestions. This average BHVO-1 digestion is compared to BHVO-1 values collected by A. W. Hofmann (personal communication, 1998) using a thermal ionization mass spectrometer (TIMS). Also listed below are average values of the five digestion blanks and twenty instrumental blanks (3% HNO3) run. Note there is not much difference between the digestion and instrumental blanks.

Table A1. BHVO-1 Analysis Comparison
 Average of 6 Digestions of BHVO-1 Measured by HR-ICP-MS, ppmPrecision of Digestions and DilutionsBHVO-1 Values From Hofmann Measured by TIMS, ppmAccuracy Compared to TIMS DataAverage of 5 digestion Blanks, ppmAverage of 20 Instrumental Acid Blanks, ppm
Pr5.383%  0.0040.002
Tb0.9152%  0.0020.002
Ho0.9722%  0.0010.001
Tm0.3282%  0.0010.001

Appendix B.: Inverse Modeling

[29] Modeling the source and mechanism of basalt genesis is notoriously difficult because of the large number of assumptions that have to be made. The important uncertain factors are the mineralogy of the source, the trace element concentrations in the source, the degree of melting to generate basalt, and even the melting process itself. In geochemical modeling, the methodology often adopted is to assume starting compositions and source trace element concentrations, and then to calculate degrees of melting to generate the observed lavas. However, such modeling never produces unique results, and rarely even places constraints on the melting process because of the initial assumptions. For any melting mechanism, the other three unknowns (source mineralogy, initial element concentration, degree of melting) are linked, and if any two were known, the third could be determined. Unfortunately, none of these parameters is known for certain.

[30] It is possible to reduce the uncertainty in the modeling somewhat by considering the variations of the REE for a suite of cogenetic lavas, where the assumption is that all the lavas are derived from an equivalent source. This assumption can be tested by observing isotopic variations in the lavas, but it can never be proven unequivocally. The REE are chosen because of their consistent geochemical behavior in normal igneous systems. Moreover, by calculating relative changes in the REE patterns of cogenetic basalts, some information about source REE concentrations, and even source mineralogy, can be obtained. As discussed in the text, the only melting mechanism that can be addressed by inverse modeling is equilibrium batch melting. This mechanism was shown to produce internally consistent results for lavas from the HSDP pilot hole [Feigenson et al., 1996], and yields results similar to those for aggregated fractional melting.

[31] The basic equation for equilibrium melting is from Shaw [1970]:

display math

where is Cli the concentration of element i in the melt, Coi is the concentration of that element in the initial source, Doi is the bulk distribution coefficient of i in the source, Pi is the sum of partition coefficients of phases in the proportions that they enter the melt, and F is the degree of melting [Allègre and Minster, 1978]. Albarède [1995] discusses the theoretical development of inverse modeling; the method employed here is based on modifications by Feigenson and Carr [1993]. By comparing the REE to each other, the dependence of concentration on F in A1 can be eliminated, thereby reducing the number of unknown parameters from three to two. Following Minster and Allègre [1978] and Hofmann and Feigenson [1983], equation A1 can be adjusted by normalizing concentrations to the initial source abundance of the most incompatible of the REE:

display math

where Ch refers to the concentration of a highly incompatible element, Si and Ii are the slope and intercept of the first equation, and can be evaluated from a graph of Ch/Ci versus Ch. This procedure removes the dependence on F and allows comparison of shapes of REE patterns; however, no explicit information on the absolute degree of melting can be obtained. The overconstraint of fixed Pi is eliminated, but the relative melt partitioning behavior is maintained, by inputting a number of different melt combinations.

[32] We use La for Ch, as it is the most incompatible of the REE. In this way, all calculated source values for Ci and Di can be compared relative to an assumed source concentration of La. Si and Ii values are obtained from the “process identification diagrams” (Ch/Ci versus Ch) of Minster and Allègre [1978]. As will become evident in the inverse procedure below, the increasing scatter in the correlations from light to heavy REE causes an increase in the error of the reconstructed source patterns and partition coefficients for the HREE.

[33] As shown by Hofmann and Feigenson [1983], source concentration and source partitioning behavior can be calculated from Si and Ii determined from linear regressions calculated from Ch/Ci versus Ch combined with various values of the input parameter Pi:

display math

A large range of mineral combinations are used to calculated Pi with the only restrictions that Pi must be less than one if all intercepts are positive, and that the shape of Pi must bear some resemblance to Doi (because the minerals that enter the melt must have been initially present in the source). This broad range of possibilities translates into the acceptable fields for source concentration and bulk partition coefficients.


[34] We thank Mike Rhodes, Mark Kurz, Julie Bryce, Janne Blichert-Toft and Dominique Weis for providing access to their geochemical analyses of the HSDP2 basalts, and to Julie Bryce and John Lassiter for extensive and thoughtful reviews of an early version of the manuscript. Additional comments by Don DePaolo and Bill White and discussions with Mark Kurz were very helpful. This research was supported by a subgrant from NSF Continental Dynamics to Don DePaolo, Ed Stolper and Don Thomas.