The temporal geochemical variations defined by lavas erupted throughout the growth of a single volcano provide important information for understanding how the Hawaiian plume works. The Hawaii Scientific Drilling Project (HSDP) sampled the shield of Mauna Kea volcano to a depth of 3100 meters below sea level during Phase 2 of the HSDP. Incompatible element abundance ratios, such as La/Yb, Sm/Yb, Nb/Zr, and Ti/Zr, in conjunction with SiO2 abundance and radiogenic isotopic ratios, especially He and Pb, in the reference sample suites of the Mauna Kea portion of cores from Phases 1 and 2 of the HSDP define three distinct geochemical groups. The upper 550 m of Mauna Kea lavas in the Phase 2 core include the Postshield Group with eruption ages of ∼200 ka to <370 ka. These lavas have relatively low SiO2 content, 3He/4He and 206Pb/204Pb, and they define a trend to relatively high La/Yb, Sm/Yb, and Nb/Zr. The eruption of these lavas coincides with migration of the Mauna Kea shield off the hot spot. As a result, extent of melting decreased, melt segregation occurred at greater depth, within the garnet stability field, and a geochemically distinct component associated with the periphery of the plume was sampled. Deeper in the Phase 2 core two other geochemical groups of lava are intercalated. One group has relatively low SiO2 abundance and high Nb/Zr Ti/Zr, 3He/4He and high 208Pb/204Pb at a given 206Pb/204Pb. These are distinctive geochemical characteristics of lavas erupted at Loihi seamount. Variations in incompatible element abundance ratios (e.g., Sm/Yb versus Nb/Zr and La/Yb versus Ti/Zr) define mixing trends between these low SiO2 lavas (Loihi-type) and lavas belonging to a high SiO2 group that are the dominant lava type in the shield part of the core (Kea-type). These two groups are presumed to reflect components intrinsic to the plume. Correlations of incompatible element abundance ratios, such as La/Nb, with radiogenic isotope ratios show that Hawaiian shields contain different proportions of geochemically distinctive components. The Koolau shield contains a recycled sedimentary component that is not present in the Mauna Kea shield. The anomalously high Ba/Th in Hawaiian lavas is inferred to be a source characteristic. Ba/Th is correlated with some radiogenic isotope ratios in Kilauea and Mauna Loa lavas, but there is no correlation in Mauna Kea lavas which range in Ba/Th by a factor of 2.6.
 In order to understand processes occurring within the earth's mantle, it is necessary to determine the role of mantle plumes in creating ocean island volcanoes. The strongest evidence for long-lived mantle plumes arises from volcanic chains that define age progressive hot spot tracks; the Hawaiian Ridge and Emperor Seamount Chain are well documented examples. An important approach for determining how a plume “works” is to study the life history of a single volcano as it approaches, passes over and moves away from the hot spot. This is a principal objective of the Hawaiian Scientific Drilling Project (HSDP) which is focused on sampling a large portion of the growth of Mauna Kea Volcano [Stolper et al., 1996].
 Core from the pilot hole of HSDP (designated Phase 1 in following text) drilled near Hilo, Hawaii recovered 280 m of lavas derived from Mauna Loa Volcano overlying 776 m of subaerially erupted Mauna Kea lavas ranging in age from <200 ka to ∼400 ka [Sharp et al., 1996]. Based on the success of this Phase 1 hole, Phase 2 of HSDP drilled and cored (95% recovery) Mauna Loa and Mauna Kea shield lavas to a depth of 3098 meters below sea level (mbsl) at a site ∼2 km south of the Phase I hole [DePaolo et al., 1999]. From top to bottom this core includes 246 m of subaerially erupted Mauna Loa lavas, 833 m (∼120 flow units) of subaerially erupted Mauna Kea lavas which at the depth of 1079 mbsl abruptly change to 905 m of submarine lavas, dominantly basaltic hyaloclastite debris flows but with a significant proportion of massive basalt with relatively low vesicularity. The lowermost 1114 m of Phase 2 core is ∼60% pillow lavas intercalated with volcaniclastic sediment. Also thirteen intrusive units occur in the deepest parts of the core forming ∼7% of the core from 2500 to 3100 mbsl. Both cores are dominantly tholeiitic basalt, but alkalic basalt occurs in the uppermost parts of the Mauna Kea section in both the Phase 1 (nine flow units [Yang et al., 1996]) and Phase 2 cores (two flow units (J. M. Rhodes and M. Vollinger, Geochemical stratigraphy of basaltic lavas from HSDP 2, manuscript submitted to Geochemistry Geophysical Geosystems, 2002, hereinafter referred to as Rhodes and Vollinger, submitted manuscript, 2002)). Eruption ages for Mauna Kea lavas in the Phase 2 core range from <230 ka to 630 ka (all ages mentioned in this text are measured, interpolated or extrapolated data from W. Sharp et al., The 40Ar/39Ar dating of lavas, hyaloclastics, and intrusions from HSDP 2 core, manuscript submitted to Geochemistry Geophysics Geosystems, 2002).
 Our research effort is part of a team effort to determine and interpret temporal geochemical variations of lavas forming the shield of Mauna Kea Volcano. We have determined abundances of trace elements in the reference suite of samples [DePaolo et al., 1999] from the Mauna Kea portion of the Phase 2 core (100 samples) by instrumental neutron activation (INAA) and inductively coupled plasma mass spectrometry (ICP-MS) (Table 1). We analyzed aliquots of the reference suite powders prepared by Rhodes and Vollinger (submitted manuscript, 2002). Details of the analytical procedures, estimates of accuracy and precision are discussed in Appendix A. Many trace elements were analyzed by ICP-MS, INAA and XRF (Rhodes and Vollinger, submitted manuscript, 2002), and the comparisons of the three data sets are in Appendix A. Figures in this paper show ICP-MS data unless noted otherwise in the figure captions. Our objective is to use these data along with major and trace element abundances determined by X-ray fluorescence (XRF) (Rhodes and Vollinger, submitted manuscript, 2002) and radiogenic isotopic ratios (J. Blichert-Toft et al., High resolution and high precision Hf and Pb isotope stratigraphy of Mauna Loa and Mauna Kea lavas from HSDP 2, manuscript submitted to Geochemistry Geophysics Geosystems, 2002; J. Bryce and D. DePaolo, High resolution Sr and Nd isotopic records of Mauna Kea volcano: Results from the second phase of Hawaii Scientific Drilling Project (HSDP 2), manuscript submitted to Geochemistry Geophysics Geosystems, 2002; M. D. Kurz et al., Helium isotope variations in HSDP lavas, manuscript submitted to Geochemistry, Geophysics, Geosystems, 2002) to understand how the geochemical characteristics of these lavas were affected by postmagmatic alteration, crystal fractionation within the crust, and mantle processes, such as extent of melting. To constrain the source components contributing to Hawaiian shield building lavas, we also evaluate correlations between abundance ratios of incompatible elements and radiogenic isotopic ratios in lavas from Mauna Kea and other Hawaiian shields.
Table 1. Trace Element Abundances in Mauna Kea Lavas Recovered From HSDP Phase 2 Drill Holea
Trace element abundances were determined by ICP-MS and INAA separately, and are reported in ppm, except for Na2O which is reported in %. n.d., not determined. For discussion of accuracy and precision see Appendix A.
Samples were powdered in WC shatterbox [Rhodes and Vollinger, 2002]; therefore, the Ta contents may be anomalously high. However, this effect appears to be minor (see text and Figure 12c).
 Abundances of incompatible elements that are not very sensitive to postmagmatic alteration, such as Nb, are inversely correlated with MgO content. This correlation is not linear because at a given MgO content the alkalic basalts are offset to higher Nb content (Figure 1). Within the HSDP Phase 2 reference suite of samples, the abundance range of incompatible elements varies from ∼3 for moderately incompatible elements such as heavy rare earth elements (REE) to ∼5 to 8 for highly incompatible elements, such as Ba, Th, U, Nb and light REE (Figure 2a). Generally this is also the order of increasing variability in abundance as indicated by the percent standard deviation (Figure 2b). For most incompatible elements the highest abundances are in the two alkalic lavas, SR 121 and SR 131 (Table 1, Figure 3). Note that the complete sample identification numbers are given in Table 1, but for brevity in the text we identify each reference sample only by SR number for Phase 2 and R number for Phase 1.
 Based on compositional similarity, alkalic basalt SR 131 from the Phase 2 core is from the same flow as sample R 177 from the Phase 1 core (Table 2). Both R 177 and SR 131 have unusually low SiO2 contents and high alkalinity; they define the extremes in their respective cores (see Figure 1a of Yang et al.  and Figure 3 of Rhodes and Vollinger (submitted manuscript, 2002)), In addition these samples are from units with similar eruption ages of ∼240 ka (Table 2).
Table 2. Comparison of Alkalic Basalt From HSDP Phase 1 and 2 Cores
Argon isochron ages are for samples R177-6.0 and SR132-1.5 which are from the same units as the respective reference samples [Sharp et al., 1998, submitted manuscript, 2002].
Oxides in wt.%, trace elements in ppm. Phase 1 abundances data from Rhodes  and Yang et al. . The second column of major element data for R177-2.60 indicates result from a second (April, 2003) analysis; the agreement with the first analysis is excellent except for Na2O which was the only significant difference between R177-2.60 and SR131-6.92. The new Na2O value for R177-2.60 is in better agreement with the SR131-6.42 value. Phase 2 abundance data from Rhodes and Vollinger (submitted manuscript, 2002) and this paper. Na2O value in parentheses was determined by INAA.
 The systematic increase in abundance range from moderately to highly incompatible elements is consistent with control of incompatible element abundances (except Rb) by igneous processes involving mafic minerals such as olivine, pyroxenes, and perhaps garnet. Based on analyses of multiple samples from individual flow units in the Phase 2 core (Huang, Vollinger, Frey, and Rhodes, unpublished data, 2003), the extremely large range in Rb abundance (∼40, Figure 2a inset) reflects the effects of postmagmatic processes whereas Nb and Th are highly incompatible elements whose abundances were not affected by postmagmatic alteration. Abundance of Ba, Th and light REE are highly correlated, and in x-y plots the trends have near zero intercepts indicating that these elements were similarly incompatible (Figure 3a). The correlations are not as good for U, and especially Rb (Figure 3a); we infer that these poorer correlations reflect the effects of postmagmatic alteration [e.g., Kennedy et al., 1991; Cohen et al., 1996; Yang et al., 1996, Huang, Vollinger, Frey and Rhodes, unpublished data, 2003]. Although Sr, Sm and Hf abundances versus Th abundance are also highly correlated, the trends clearly define convex upwards trends; such convexity is even more apparent for Y and heavy rare earth element abundance (Figure 3b). Non-linearity in abundance variation plots involving elements of differing incompatibility is characteristic of melts derived by different extents of melting of a common source. This result arises from the mass balance equations
(where x and y are generic incompatible elements; Clx and Cox are melt and source concentrations, respectively; F = melt fraction; Dx and Dy are bulk solid/melt partition coefficients). When Ds of incompatible elements are much less than the extent of melting (F), the abundance ratio (Cx/Cy)l is constant (i.e., trends with constant slope intersecting the origin in Figure 3a); therefore (Cx/Cy)l is independent of F and is equal to the source ratio. In contrast when Ds of incompatible elements are comparable to F, the abundance ratio (Cx/Cy)l varies with extent of melting; i.e., the slope changes in a x-y variation plot [e.g., Hanson, 1989, Figure 3]. The convexity in Figure 3b trends indicates that the bulk solid/melt partition coefficient for Th was smaller than that for Sr, Sm, Hf, Y and Lu.
3. Discussion: HSDP Cores
3.1. Effects of Postmagmatic Alteration on Lava Compositions
 For subaerially erupted Hawaiian tholeiitic basalt, the abundance ratios K/Rb and K2O/P2O5 are sensitive indicators of postmagmatic alteration because K and especially Rb are preferentially leached from the basalts leading to unusually low K2O/P2O5 (<1) and high K/Rb (>1000) and Ba/Rb (>20) [e.g., Feigenson et al., 1983; Frey et al., 1991, 1994; Yang et al., 1996]. For the HSDP samples the effects of such alteration are obvious in a plot of Ba/Rb versus K2O/P2O5 where Ba/Rb, a ratio that is quite constant, 11.6 ± 0.2, in unaltered oceanic basalts [Hofmann and White, 1983], reaches values of >100 (Figure 4a). What is surprising about the HSDP trend in Figure 4a is that the subaerial and submarine lavas define sub-parallel trends with the submarine lavas slightly offset to lower Ba/Rb at a given K2O/P2O5. Submarine seawater alteration of MORB typically leads to increases in abundances of K and Rb [Staudigel et al., 1995], but submarine HSDP samples show no evidence for addition of K and Rb. Alkali loss is especially characteristic of the second pillow zone (2234 to 2470 mbsl, see Figure 4b) which is composed of low density, vesicular pillow lava [Moore, 2001]. We infer that this pillow zone is a region of fresh water migration. Additional geochemical effects of postmagmatic alteration are discussed by Chan and Frey  for lithium and M. J. Vollinger and J. M. Rhodes (Chemical alteration in basalts sampled by phase-2 of the Hawaii Scientific Drilling Project, manuscript submitted to Geochemistry, Geophysics, Geosystems, 2002) for major and trace elements.
3.2. Role of Crystal Fractionation
 Only about 20% of the HSDP cores contains phenocrysts of plagioclase or clinopyroxene, and only a few of these lavas have >1% plagioclase or clinopyroxene phenocrysts [Garcia et al., 1996; DePaolo et al., 1999]. The role of plagioclase fractionation can be evaluated using Al2O3 abundance, (Sr/Nd)PM and relative abundance of Eu (subscript PM indicates normalized to primitive mantle). Samples from Phase 1 and 2 cores define an inverse correlation between Al2O3 and MgO (Figure 5a). The increased scatter at the low MgO end of the trend reflects the tendency of alkalic basalt to have higher Al2O3; e.g., samples SR 121, SR 131 and R177; tholeiitic sample SR 129 is also offset to high Al2O3 and CaO (Figures 5a and 5b). Since the Al2O3 content of plagioclase exceeds that of coexisting equilibrium melts, segregation of mineral assemblages including significant plagioclase result in decreasing Al2O3 abundance with increasing differentiation. Consequently, the inverse Al2O3 versus MgO trend shows that plagioclase fractionation was not an important process. Because Sr and Eu are relatively compatible in plagioclase the variation of (Sr/Nd)PM and (Eu/Eu*) with MgO content in an individual shield can also be used to evaluate the role of plagioclase fractionation. In general, HSDP Phase 2 samples have (Sr/Nd)PM close to unity, ranging from 0.8 to 1.2, and (Sr/Nd)PM does not vary systematically with MgO content (Figure 5c). Sample SR 129 has an anomalously high (Sr/Nd)PM >1.4 (Figure 5c). Also it is the sample with the largest enrichment in Eu (Eu/Eu* = 1.06, Figure 5d). Sample SR129 is from Unit 47 which contains up to 13% plagioclase phenocrysts [DePaolo et al., 1999], and it is the only obvious plagioclase cumulate in the Phase 2 core.
 Sample SR125 has several anomalous geochemical characteristics. It has the lowest (Sr/Nd)PM (Figure 5c) and an unusually low Na2O content (1.27%) which is confirmed by similar values determined by XRF and INAA (Rhodes and Vollinger, submitted manuscript, 2002) (Table 1 of this paper). This sample has a relatively high Loss on Ignition (3.43%), low K2O/P2O5 (0.55) and high Ba/Rb (92) (Figure 4). It appears to have lost a plagioclase component during alteration; its CaO abundance, however, is not unusually low (Figure 5b), and Eu/Eu* is 0.98. In addition, it has an anomalously high Sc content (Figure 5e).
 Clinopyroxene is also a rare phenocryst in HSDP lavas [Garcia et al., 1996; DePaolo et al., 1999]. The role of clinopyroxene fractionation can be evaluated using CaO and Sc abundances. The CaO versus MgO trend is inverse and like the Al2O3 versus MgO trend, there is more scatter at low MgO; in this case the scatter is caused by the relatively low CaO content of alkalic basalt (Figure 5b), perhaps indicating a role for clinopyroxene fractionation in generating alkalic basalt. Consistent with this inference some of the alkalic lavas, such as SR 131 and R177, have relatively low Sc and Cr contents (Figure 5e and Table 1). The only Phase 2 tholeiitic basalt which may have experienced clinopyroxene fractionation is SR 850 which has high abundances of incompatible elements (Figure 3a) and a low MgO content (6.44%); for this MgO content, it has relatively low Sc abundance (Figure 5e). However, no clinopyroxene phenocrysts were observed in this sample [DePaolo et al., 1999].
 Olivine is the dominant phenocryst/xenocryst in most of the HSDP tholeiitic basalts and petrographic observation shows that in contrast to plagioclase and clinopyroxene fractionation, olivine accumulation and fractionation was an important process [e.g., Baker et al., 1996; Garcia et al., 1996; Rhodes and Vollinger, submitted manuscript, 2002]. A reflection of the role of olivine is the wide variation in MgO content (6% to 29%) and the positive correlation between MgO and Ni abundance (Figure 5f). In order to adjust for the effects of olivine fractionation and accumulation for each of the Phase 2 lavas, we added or subtracted olivine in 1% increments with KDFe/Mg = 0.3 until the calculated composition was in equilibrium with Fo90 olivine, typical of the most Mg-rich olivine phenocrysts in the core, or Fo85 which is closer to the average olivine composition [Baker et al., 1996; Garcia et al., 1996]. This adjustment greatly reduces the MgO variation (∼10 to 13% for Fo85 and ∼16.6 to 19.6% for Fo90), and decreases the abundance range of incompatible elements (Figure 2a). In summary, olivine addition and fractionation was a dominant process controlling the compositions of HSDP lavas (see also Rhodes and Vollinger, submitted manuscript, 2002). Although this process can explain much of the variation in abundance of incompatible elements (Figures 2 and 3), it cannot explain the convexity of trends in Figure 3b.
3.3. Geochemical Groups in HSDP Cores
3.3.1. Postshield Group
 As with Al2O3 and CaO versus MgO abundance (Figures 5a and 5b), the abundance of SiO2 and TiO2 are scattered at low MgO content because the eleven samples of alkalic basalt (nine in Phase 1 core and two in Phase 2 core) in the upper part of the cores have relatively low SiO2 and high TiO2 content (Figures 6a and 6b) and relatively high total iron contents [Rhodes, 1996; Rhodes and Vollinger, submitted manuscript, 2002]. The upper 152 m of the Mauna Kea section in the Phase 2 core (samples SR121 through SR175) includes intercalated alkalic and tholeiitic basalt which define a trend to relatively low SiO2 content (Figures 7c and 7f). In the range of 6 to 10% MgO, at a given MgO content these low SiO2 lavas within the upper 152 m have the highest abundance of Nb and other incompatible elements (Figure 3a), Al2O3 (Figure 5a) and TiO2 (Figure 6b). Although there is considerable scatter in downhole trends for incompatible element abundance ratios, such as Nb/Zr and La/Yb, these ratios are relatively high in the alkalic samples SR 121 and SR 131 near the top of the Mauna Kea section ( 7a and 7b). Moreover, within the uppermost Mauna Kea lavas, there is a trend to relatively high Nb/Zr and La/Yb with decreasing depth and age (Figures 7a, 7b, 7d, and 7e).
 The age range of samples SR121 to SR175, ∼200 ka to 330 ka (Figure 7f), overlaps with the oldest of the Hamakua Volcanics, ∼70 ka to 250 ka, which form the basaltic postshield stage that is exposed on the lower subaerial flanks of Mauna Kea volcano [Frey et al., 1990, 1991; Wolfe et al., 1997]. The Hamakua Volcanics also include intercalated tholeiitic and alkalic basalt. Yang et al. [1996, Figure 9] showed that the trend to high La/Yb at the top of the Phase 1 core is continuous with that defined by the Hamakua Volcanics. Therefore, we refer to the intercalated alkalic and tholeiitic basalt in the upper parts of the Phase 1 and 2 cores as the Postshield Group. The gradual increase of La/Yb and Nb/Zr with decreasing age shows that the transition from shield to postshield stage was gradual and prolonged over a span of ∼130 ka (Figures 7d and 7e). Also, during the shield to postshield transition there were systematic changes in radiogenic isotopic ratios. For example, beginning at ∼830 m 3He/4He decreases upward in the core (Figure 8a). The Postshield Group also has relatively low 206Pb/204Pb, 208Pb/204Pb and εHf, but relatively high εNd (Figures 8b, 8c, 8d, and 8e). Based on their relatively low SiO2 (Figures 7c and 7f), 3He/4He (Figure 8a) and 208Pb/204Pb at a given 206Pb/204Pb (Figure 8f), we include samples of SR232, SR276, SR328 and SR340 in this Postshield Group, thereby increasing the initiation of postshield volcanism to ∼370 ka (Figure 7f).
3.3.2. Low SiO2 Shield Group
 Lavas with relatively low SiO2 content (Figures 6a and 6c) are not restricted to the Postshield Group. In the Phase 2 core some of the deeper and older (>834 mbsl and >380 ka) lavas have relatively low SiO2 content (Figures 7c and 7f). We identify these as the Low SiO2 Shield Group (individual samples are labeled in Figure 7c). This group is characterized by relatively high 3He/4He and high 208Pb/204Pb at a given 206Pb/204Pb, Figures 8a and 8f). Based on these isotopic parameters, the youngest lava in the Low SiO2 Shield Group is subaerial sample SR354, and samples SR750, SR756 and SR762 also belong to this group (Figures 8a and 8f), despite their adjusted SiO2 contents of >48.8% (Figure 7c); that is we conclude that the arbitrary SiO2 discriminant does not perfectly identify this group. In part this difficulty arises from SiO2 mobility during postmagmatic alteration (Vollinger and Rhodes, submitted manuscript, 2002). In addition, as we emphasize in later discussion, there is within group variation in geochemical characteristics, e.g., La/Yb and Zr/Nb; that is some samples are transitional between the groups. Nevertheless 3He/4He and delta 208Pb/204Pb clearly distinguish the Low SiO2 Shield Group (Figure 8).
3.3.3. High SiO2 Shield Group
 Most of the lavas have adjusted SiO2 contents >48.8% (Figure 7c and 7f), and they define the main inverse trends in MgO variation plots (Figures 5 and 6). These samples form the High SiO2 Shield Group.
3.4. Comparison of Geochemical Groups Defined by Huang and Frey (this paper) With Those Defined by Rhodes and Vollinger (submitted manuscript, 2002)
 Rhodes and Vollinger (submitted manuscript, 2002) used adjusted SiO2 content and Zr/Nb as principle criteria for defining lava types in the Phase 2 core whereas we use adjusted SiO2 content, incompatible element abundance ratios, 3He/4He and Pb isotopic ratios to define geochemical groups. Consequently, the lava types defined by Rhodes and Vollinger (submitted manuscript, 2002) are generally similar to our groups, but there are some important differences.
 Rhodes and Vollinger (submitted manuscript, 2002) include only samples SR121 (246.2 mbsl) through SR152 (337 mbsl) in this type. We extend this grouping to include samples SR167, SR175, SR232, SR276, SR328 and SR340 (794 mbsl) that also have relatively low SiO2 (Figure 7f).
3.4.2. Type 1 of Rhodes and Vollinger (submitted manuscript, 2002)
 These are the dominant type of tholeiitic basalt in the core (61% of the Mauna Kea lavas analyzed by Rhodes and Vollinger (submitted manuscript, 2002)). They are present from 337 to 3098 mbsl and are equivalent to the High SiO2 Shield Group in Figures 7c and 7f.
3.4.3. Type 2 of Rhodes and Vollinger (submitted manuscript, 2002)
 This type includes low SiO2 lavas restricted to the upper part of the Mauna Kea section (above 1739 mbsl); these lavas are most abundant between 337 and 850 mbsl. This type includes tholeiitic lavas that resemble tholeiitic basalt occurring within the postshield group. Rhodes and Vollinger (submitted manuscript, 2002) include low SiO2 subaerial lavas SR276, SR328, SR340 and SR354, as well as the low SiO2 submarine lavas SR531, SR545, SR560 and SR675 in this type (Figure 7c). Because of their relatively high 3He/4He and delta 208Pb/204Pb (Figures 8a and 8f) we include samples SR354, SR531, SR545, SR560 and SR675 in our Low SiO2 Shield Group. These samples, however lack the high Nb/Zr that distinguishes older lavas in this group (Figures 7a and 7d). We agree that samples SR276, SR328 and SR340 are similar to the Postshield group and we include them in this group (Figures 7 and 8).
3.4.4. Type 3 of Rhodes and Vollinger (submitted manuscript, 2002)
 These are low SiO2 lavas confined to the submarine portion of the core. They are equivalent to our Low SiO2 Shield Group. They are a distinctive group because low SiO2 is coupled with relatively high Nb/Zr, 3He/4He and delta 208Pb/204Pb (Figures 7a, 7d, 8a, and 8f). As noted by Rhodes and Vollinger (submitted manuscript, 2002) these are also characteristics of Loihi lavas. An important difference in our grouping is that sample SR354 is a Type 2 sample in the Rhodes and Vollinger classification because it has “normal” Nb/Zr, low SiO2 and is in the subaerial part of the core. However, it has relatively low SiO2, high 3He/4He and delta 208Pb/204Pb; these are characteristics of of our Low SiO2 Shield Group (Figures 7c, 7f, 8a, and 8f). This sample is important because it occurs high in the core within the subaerial section (834 mbsl with an inferred age of 380 ka). It is the uppermost sample of our Low SiO2 Shield Group.
3.4.5. Type 4 of Rhodes and Vollinger (submitted manuscript, 2002)
 This type includes only 4 samples (two are from unit 92) and is defined by relatively high Nb/Zr but they do not have relatively low SiO2 contents. We analyzed three of these samples (SR222, SR860 and SR956) and also find that they have relatively high Nb/Zr (Figure 7a). However, unlike Group 3 lavas, they do not have relatively high 3He/4He and delta 208Pb/204Pb (Figures 8a and 8f); therefore, we do not recognize them as a distinct group.
3.5. Petrogenesis of Geochemical Groups
3.5.1. Postshield Group
 The abundance ratios La/Sm, La/Yb, Sm/Yb and Nb/Zr are positively correlated with incompatible element content in the Postshield Group (e.g., Th in Figure 9), and the highest La/Yb and Nb/Zr ratios are in samples from the upper part of the core (Figures 7a, 7b, 7d, and 7e). Although radiogenic isotope ratios vary systematically with depth in this group (Figure 8), a plausible inference is that changes in proportions of source components were accompanied by variations in extent of melting with the alkalic lavas representing the lowest degree of melting [Yang et al., 1996]. To test this hypothesis we did forward modeling for partial melting of spinel- and garnet-peridotite (see Figure 10 and Appendix B). The relatively large variations of La/Yb, Nb/Zr and Sm/Yb within the Postshield Group (Figures 10a, 10b, and 10c) are consistent with variable extents of batch melting of garnet peridotite with La/Yb, Nb/Zr and Sm/Yb ratios slightly greater than estimates for primitive mantle (Table A4 of Appendix B). It is surprising that the range in La/Yb and Nb/Zr in the Postshield Group is similar to that for historical, ∼200 yr, Kilauea lavas (Figure 10e). These variations in Kilauea shield lavas have been explained by a factor of two change in extent of melting of garnet peridotite [Pietruszka and Garcia, 1999].
 Segregation of the primary magmas for the Postshield Group within the stability field of garnet is consistent with the inverse modeling of rare earth element abundances by M. D. Feigenson et al. (Partial melting and fractional crystallization of Mauna Kea lavas: Evidence from rare earth element concentrations of HSDP basalts, manuscript submitted to Geochemistry Geophysics Geosystems, 2003) who inferred that alkalic lavas in the upper part of the core formed by low degrees of melting of a garnet-bearing source. In addition, we suggest that the Ti/Zr ratio is useful for identifying a role for residual garnet because Dcpx/meltTi/Zr > 1 whereas Dgt/meltTi/Zr < 1, especially for Ca-rich garnet (Table A4). The Postshield Group trend of varying La/Yb at nearly constant Ti/Zr is clearly most consistent with partial melting of garnet peridotite (Figure 10d). Melt segregation of the postshield lavas at relatively high pressure is also consistent with their relatively low SiO2 content (Figure 7f) and high FeO content (Rhodes and Vollinger, submitted manuscript, 2002, Figure 6). Hence the petrogenetic inferences for the Postshield Group are consistent with the “overall trend of increasing pressure of melt segregation and decreasing extent of melting with decreasing eruption age” inferred by Yang et al.  for the HSDP 1 core.
 In addition to variable extents of melting, Mauna Kea postshield lavas experienced more extensive crustal processing than shield lavas; for example, samples showing the effects of plagioclase or clinopyroxene fractionation are dominantly confined to subaerially erupted lavas (Figure 5). Moreover, samples with relatively low (∼7%) MgO content are also abundant in the upper part of the Phase 2 core (Rhodes and Vollinger, submitted manuscript, 2002, Figure 5). The relatively low MgO content and a role for fractionation of plagioclase and clinopyroxene in the upper part of the core are consistent with longer residence times in crustal magma reservoirs as the magma supply from the mantle decreased during the transition from shield-stage to postshield-stage volcanism [e.g., Frey et al., 1990, 1991; DePaolo and Stolper, 1996].
3.5.2. Shield Groups
 The trends in Figure 10 for the High SiO2 Shield Group are not as clearly defined as those for the Postshield Group, but they may also reflect variations in extent of melting. However, variations in radiogenic isotopic ratios within the core and even within the defined lava groups (Figure 11) shows that the petrogenesis of each group is more complex than variable extents of melting of a geochemically homogeneous source. As an example, the inverse Sm/Yb-Nb/Zr trend (Figure 10b) and relatively constant La/Yb but variable Ti/Zr of the Low SiO2 Shield Group (Figure 10d) are inconsistent with variable extents of melting of a common peridotite source. Rhodes and Vollinger (submitted manuscript, 2002) noted that each of their major lava types ranges widely in MgO content. They argued that the near linear trends in MgO variation plots (e.g., Figures 5a, 5b, 6a, and 6b) reflect magma mixing in addition to olivine accumulation and fractionation. Although this complexity does not mask the compositional groups that we identify, the recognition of magma mixing as a significant process is important in explaining geochemical variations within these groups because the mixing end-members may be derived from geochemically distinct sources.
 An important observation is that the high Ti/Zr and Nb/Zr portion of the trends defined by the Low SiO2 Shield Group is defined by the low SiO2 lavas erupted in two distinct intervals (Figure 10e), the 5 samples from SR732 to SR762 (depth range of 1974 to 2124 mbsl and inferred age range of 503 ka to 520 ka) and the 7 samples from SR791 to SR842 (depth range of 2280 to 2504 mbsl and inferred age range of 537 ka to 561 ka). Also, a lower pillow basalt SR871 and intrusive sample SR940 are included in this subgroup. In addition to high Ti/Zr and Nb/Zr and relatively low SiO2 contents these samples have relatively high 3He/4He (Figure 8a), and they define the low 206Pb/204Pb end of the Kea high 8 Pb array of Figure 2 in J. Eisele et al. (The long-term Pb isotope evolution of Mauna Kea lavas from the HSDP-2 drill core, manuscript submitted to Geochemistry Geophysics Geosystems 2002). A plot of 3He/4He versus delta 208Pb/204Pb (see Kurz et al., submitted manuscript, 2002, Figure 7) shows that for He and Pb isotopes these lavas trend toward the field of lavas from Loihi Seamount (also see Figure 11a). Based on SiO2 content, Nb/Zr and 3He/4He, Rhodes  and Kurz and Curtice  were the first to note the similarity of these Mauna Kea lavas to lavas from Loihi Seamount. In addition to high Nb/Zr (Figure 10f), high Ti/Zr is typical of some Loihi lavas (Figure 10g). Therefore, we conclude that samples of the Low SiO2 Shield Group with the lowest 206Pb/204Pb and highest Nb/Zr and Ti/Zr contain the largest proportion of a Loihi-type component (Figures 10f, 10g, and 11a).
3.6. Abundance Ratios Involving Elements of Similar Incompatibility
 Each of the abundance ratios Nb/La, Nb/Ta, Nb/Th, Nb/U, Zr/Hf and Ce/Pb involve elements with similar geochemical characteristics. These ratios are not highly variable in oceanic basalt; typically they vary by less than a factor of two. However, numerous recent studies have shown that significant variations in these ratios are found in mid-ocean ridge and oceanic island basalt. These variations are inferred to reflect magmatic processes, specifically differences in partition coefficients, e.g., DHf > 2DZr for clinopyroxene/melt [Blundy et al., 1998]. However, these ratios in the Phase 2 core are typical for ocean island basalt and there are few significant differences between the Postshield and Shield groups (Figure 12).
 The mean Nb/Th for the Phase 2 reference suite is 15.4 ± 1.4 (in this section the ± value indicates two standard deviations), very similar to that for the Phase 1 core (15.0 ± 3.6 and 16.2 ± 2.4 [Albarede, 1996; Hofmann and Jochum, 1996, respectively]), for postshield lavas on the subaerial slopes of Mauna Kea (14.3, [Kennedy et al., 1991]) and to the average (15.2 ± 1.7) reported for Hawaiian shield basalt by Hofmann . It is notable that the Nb/Th ratio is greater than the primitive mantle estimate (8.6 [Sun and McDonough, 1989]), but to date there is no indication that Nb/Th is a useful discriminant for the different components contributing to Hawaiian volcanism.
 The mean Nb/U for the two groups of shield lavas is 45 ± 10, well within the global average of 47 ± 10 for oceanic basalt [Hofmann, 1986], and similar to that for the Phase 1 core (48 ± 12 and 53 ± 14 in Albarede  and Hofmann and Jochum , respectively). The Postshield lavas range to higher Nb/U, but given the similarity of Nb/Th in all groups, the higher Nb/U in upper lavas probably reflects U loss during postmagmatic alteration as proposed by Kennedy et al.  for postshield lavas erupted on the high rainfall east flank of Mauna Kea. In the Phase 2 core, sample SR129 has the highest Nb/U (74), and its altered nature is indicated in Figure 4a.
 The average Nb/Ta of the shield lavas is 13 ± 2, similar to that reported for the Phase 1 core (12 ± 6 [Albarede, 1996]) and close to that for average MORB and OIB (16.7 ± 1.8 and 17.1 ± 2, respectively [Kamber and Collerson, 2000]). The highest Nb/Ta (up to 15.7 in SR125) occurs in the youngest postshield lavas. The slightly lower Nb/Ta of Phase 2 samples compared to OIB and MORB probably reflects Ta contamination arising from use of a WC shatterbox (see Rhodes  for description of sample preparation for Phase 1 and 2 reference suite). Surprisingly Phase 1 Mauna Kea samples analyzed by Albarede  range to much lower Nb/Ta of 2 to 3, hence the large standard deviation. Apparently, the samples of Albarede  were contaminated by Ta.
 There is a general trend, with considerable scatter, for Nb/La to decrease upwards in the HSDP core (Figure 12). The mean value of 1.19 ± 0.10 of the Phase 2 data overlaps with the Phase 1 mean (1.18 ± 0.24 and 1.19 ± 0.08 in Albarede  and Yang et al. , respectively). All of these values exceed that of the primitive mantle estimate of 1.04 [Sun and McDonough, 1989]. Like Nb/Zr, the Nb/La ratio of Hawaiian shield lavas is correlated with radiogenic isotopic ratios (see subsequent discussion in section 4).
 The Zr/Hf ratio does not vary systematically with depth in the core and the mean value, 39 ± 3, is within the range for primitive OIB ∼37 to 44 [David et al., 2000] and similar to that for the Phase 1 core (42 ± 2 [Yang et al., 1996]).
 Similarly, Ce/Pb does not vary with depth in the core and the mean 29 ± 8 (excluding two samples with anomalously high Pb content, see Figure 3) is similar to that typical of oceanic basalt (25 ± 5 [Hofmann et al., 1986]). The range to low Ce/Pb (<20 in a few HSDP samples may reflect Pb contamination, but the lowest Ce/Pb occur over the depth interval from ∼300 to 1300 m. Possibly this is a zone of Pb mobility, but it does not overlap with the uppermost lavas with high Nb/U. The Ce/Pb ratios in the Phase 2 core are lower than the 40 ± 4 reported by Kennedy et al.  for postshield Mauna Kea lavas but similar to the Phase 1 average (33 ± 8 and 40 ± 11 in Albarede  and Yang et al. , respectively).
3.7. Implications of Temporal Geochemical Trends in HSDP Cores
 In summary, the temporal geochemical trends defined by Mauna Kea lavas in the HSDP cores lead to the following conclusions.
 1. HSDP lavas show that at Mauna Kea volcano there was a gradual and prolonged transition from shield- to postshield-stage volcanism commencing at ∼370 ka (sample SR 340), and entering a phase of intercalated tholeiitic and alkalic basalt at ∼230 ka (SR 131, Table 2). Studies of subaerially exposed Mauna Kea lavas show that postshield basaltic volcanism (Hamakua Volcanics) continued until 70–65 ka and was followed by a hawaiitic substage (Laupahoehoe Volcanics) erupted from 65–4 ka [Frey et al., 1990, 1991; Wolfe et al., 1997]. Although the HSDP postshield lavas are not highly evolved (MgO > 6.3%), they, in general, are more evolved than the underlying shield lavas, thereby providing evidence for increased crystal fractionation resulting from a decreasing magma supply to the crust, an inference that is consistent with the long time interval represented by these lavas (Figure 7f).
 Within the intercalated tholeiitic and alkalic basalt of the postshield-stage, the alkalic basalt does not have distinctive isotopic characteristics (Figure 8); hence we infer that low SiO2 content and incompatible element enrichment of alkalic basalt was, at least in part, controlled by process. Specifically, we conclude that the uppermost HSDP lavas in the Mauna Kea section, the Postshield Group including intercalated alkalic and tholeiitic basalt, segregated at a lower extent of melting and higher pressure, within the stability field of garnet, than the underlying Mauna Kea lavas. Similar inferences were made from studies of the Phase 1 core [Yang et al., 1996] and onland sections of Mauna Kea basalt [Frey et al., 1991]. In addition the source of the Postshield Group had lower 3He/4He and less radiogenic Pb isotope ratios than older shield lavas. Similar results were obtained for the HSDP Phase 1 core [Kurz et al., 1996; Lassiter et al., 1996] and for other Hawaiian shields, such as Haleakala [Kurz et al., 1987; Chen et al., 1991]. Hence as a Hawaiian volcano migrates away from the hot spot, shield-stage volcanism wanes, extent of melting decreases, depth of melt segregation increases, magma supply to the crust decreases, crystal fractionation within the crust is enhanced, and there is an increased role for a depleted component with typically lower 3He/4He and 87Sr/86Sr, higher 143Nd/144Nd and 176Hf/177Hf and less radiogenic Pb. In detail the trend to lower εHf for the Postshield Group (Figure 8c) is anomalous.
Lassiter et al. [1996, Figure 9] interpreted results for the Phase I core as reflecting a thermally and geochemically zoned plume. As the volcano moves over the relatively cool and depleted plume margin, melting extent decreases, melt segregation occurs at relatively higher pressure and partial melts have isotopic characteristics reflecting a role for a depleted source; i.e., relatively low 3He/4He and non-radiogenic Pb. The geochemical data for the Postshield Group of the Phase 2 core are consistent with this interpretation.
 2. A new conclusion inferred from the lower part of the HSDP 2 core, deeper than 800 m, is that the Low SiO2 Shield Group sampled a source whose partial melting led to relatively low SiO2 content, relatively uniform La/Yb, relatively high Nb/Zr and Ti/Zr, high 3He/4He and high 208Pb/204Pb at a given 206Pb/204Pb (i.e., high delta 208Pb/204Pb). This group cannot be distinguished on the basis of Sr, Nd or Hf isotopes or specific incompatible element abundance ratios, such as La/Yb and Nb/Zr. However, distinctive trends are defined by this group in ratio-ratio plots, such as Sm/Yb versus Nb/Zr and La/Yb versus Ti/Zr (Figures 10e and 10g); moreover, these ratios are correlated with delta 208Pb/204Pb (Figure 11). This Low SiO2 Shield Group occurs throughout the core from ∼800 to 3000 mbsl forming clusters of samples (e.g., the 3 subgroups formed by SR531–SR545–SR560, and SR732–SR741–SR750–SR756–SR762, and SR791–SR796–SR800–SR814–SR826–SR836–SR842) which are intercalated with high SiO2 lavas (Figures 7 and 8). In addition to relatively low SiO2 content the two older subgroups (SR732 to SR762 and SR791 to SR842) are characterized by relatively high Nb/Zr and Ti/Zr (Figures 10d and 10e) and high 3He/4He and delta 208Pb/204Pb (Figures 8a and 8f). These are characteristics of lavas from Loihi seamount. This is a surprising result because it requires that volcanoes on both the Loa (Loihi) and Kea (Mauna Kea) trends [e.g., DePaolo et al., 2001] sampled similar source materials early in their evolution. If this Loihi component is associated with the core of the Hawaiian plume, the conclusion of Lassiter et al. [1996, Figure 10] that Mauna Kea volcano sampled only the relatively cool peripheral part of the plume thereby bypassing the central part of the hot spot (plume source) is incorrect. However, if sample SR354, a relatively young (380 ka) subaerially erupted sample, also sampled the Loihi component, as we infer by including it in the Low SiO2 Shield Group (Figure 8), this component contributed to the growth of Mauna Kea for 200 ka.
 3. Despite the complexities arising from the role of geochemically distinct source components, the well-defined positive correlations between abundances of incompatible elements (Figure 3) show that these source components were not markedly different in abundance ratios involving highly incompatible elements (Figure 12).
4. Discussion: Comparisons of Hawaiian Shields
4.1. Intershield Differences in Abundance Ratios Involving Nb and Correlations With Radiogenic Isotope Ratios
 Like Mauna Kea samples in the HSDP core, lavas from different Hawaiian shields define positive trends between La/Yb and Nb/Zr, but the Koolau shield is offset to lower Nb/Zr (Figure 13a); this offset is much larger than Nb/Zr differences between groups within the Mauna Kea part of the HSDP cores. We infer that correlations between La/Yb and Nb/Zr reflect differences in extent of melting whereas intershield differences in these ratios reflect source heterogeneity. For example the La/Yb-Nb/Zr variation in HSDP postshield lavas is inferred to reflect varying extents of melting (Figure 10a), and relatively low extent of melting is also the likely explanation for the alkalic lavas from Loihi seamount that range to high La/Yb and Nb/Zr (Figure 13a). In contrast, the relatively low Nb/Zr and high La/Nb (a ratio that is less sensitive than Nb/Zr to variations in extent of melting) of Koolau shield lavas is inferred to reflect source differences (Figure 13). Although the Low SiO2 Shield Group at Mauna Kea is not distinct in La/Nb (Figure 12), we infer that their high Nb/Zr also reflects source control (Figures 10a and 10b).
 It is well established for Hawaiian shield lavas that incompatible element abundance ratios involving Nb correlate with radiogenic isotopic ratios [Frey and Rhodes, 1993; Roden et al., 1994; Yang et al., 1994; Rhodes and Hart, 1995, Chen et al., 1996]. Such correlations provide the opportunity to identify compositional characteristics of the different source components contributing to Hawaiian lavas. Two ratios which provide good correlations with isotopic ratios are Nb/Zr and La/Nb (Figures 14a and 14b). Clearly, lavas from Koolau and to a lesser extent Mauna Loa are deficient in Nb, and their lavas trend to relatively low 206Pb/204Pb, 143Nd/144Nd and high 87Sr/86Sr. Based on extrapolation of the intershield trends, the distinctive Koolau component has 206Pb/204Pb ∼17.8 and 87Sr/86Sr ∼0.7045 (Figure 14b). This constraint on 206Pb/204Pb is identical to that arising from the εHf versus 206Pb/204Pb trend defined by Hawaiian shield lavas [see Blichert-Toft et al., 1999, Figure 3]. In contrast, the Kea component defined by lavas from Mauna Kea and Kilauea has 87Sr/86Sr ∼0.7036, 143Nd/144Nd ∼0.5130 and 206Pb/204Pb ∼18.6 (but see Eisle et al.  for a discussion of Pb isotopic complexity in HSDP 2 lavas).
 It is also well known that abundance ratios involving Nb are strongly fractionated by the processes involved in arc magmatism. As a consequence, arc lavas and continental rocks have distinctly high La/Nb ratios [e.g., Hofmann, 1986]. Therefore, a plausible inference is that the relative deficiency of Nb in some Hawaiian shield lavas, such as Koolau lavas, reflects recycled source components derived from an arc tectonic setting. In addition to a Nb depletion, the Koolau component has high Pb/Hf, relatively high 18O/16O, 87Sr/86Sr, 187Os/188Os, high 176Hf/177Hf for a given 143Nd/144Nd and relatively low 143Nd/144Nd, 206Pb/204Pb, 207Pb/204Pb and 208Pb/204Pb [Roden et al., 1994; Eiler et al., 1996; Lassiter and Hauri, 1998; Blichert-Toft et al., 1999; Jackson et al., 1999]. All of these geochemical characteristics are consistent with the conclusion that the distinctive Koolau component was derived from ancient recycled oceanic crust including pelagic sediment with a continental signature, i.e., relative depletion in Nb and enrichment in Pb [Lassiter and Hauri, 1998; Blichert-Toft et al., 1999; Jackson et al., 1999]. Although subduction zone processing must modify subducted oceanic crust, it is evident that in some cases important geochemical characteristics of at least some parts of subducted plates are recycled into the mantle and returned to the surface via plume-related volcanism. Moreover, the distinctive major element characteristics of Koolau shield lavas, notably high SiO2/FeO and low CaO/Al2O3 [Frey et al., 1994], can be explained by partial melting of a peridotite source that was refertilized by melts derived from recycled oceanic crust occurring as eclogite within peridotite [Hauri, 1996; Green et al., 2001].
4.2. Abundance Ratios Involving Ba
 Abundance ratios involving Th, Ba, Nb and La are potentially useful in identifying components in the source of Hawaiian lava. These elements are highly incompatible in anhydrous phases; consequently, ratios such as Ba/Th, Ba/Nb and Ba/La in melts derived by >1% melting are equal to the source ratios. However, residues from partial melting will have ratios that differ significantly from those of the source. Also these ratios in melts are not very sensitive to fractional crystallization of anhydrous minerals, but cumulate phases, such as plagioclase, may have quite different ratios than the coexisting melts.
 High Ba/Th (>100) relative to that of primitive mantle is characteristic of most Hawaiian shield lavas (Figure 15) [Hofmann and Jochum, 1996], melt inclusions within olivine phenocrysts in Mauna Loa lavas [Sobolev et al., 2000] and even postshield- and rejuvenated-stage lavas [Yang et al., 2003]. As shown by Hofmann and Jochum , Ba/Th ratios in Hawaiian lavas are higher than those of MORB and most other ocean island basalt (Iceland is an exception [Chauvel and Hemond, 2000]). Hofmann and Jochum  concluded that relatively high Ba/Th is characteristic of the present-day Hawaiian source and not a result of melting or postmelting processes; they proposed that the high Ba/Th signature results from an eclogite component in the source that formed as plagioclase-rich cumulate gabbro. Such a component is apparently ubiquitous in the source of Hawaiian lavas; an inference that contrasts with the localized presence of the component creating relative Nb deficiencies.
 If Ra/Th ratio is a proxy for Ba/Th, further insight into Ba/Th fractionation during the melting process creating Hawaiian tholeiitic basalt is provided by measurements of (226Ra/230Th) disequilibria; e.g., Pietruszka et al.  found that the activity ratio (226Ra/230Th) is typically ∼1.1 in historic tholeiitic basalt from Kilauea Volcano. This 10% disequilibrium can arise from the melting process (a consequence of different partition coefficients for Ra and Th), and from radiogenic ingrowth of 226Ra during the melting process. Figure 7 of Pietruszka et al.  shows that at very low extents of melting (≪1%), ∼10% disequilibrium results from the difference in partition coefficients. However, at the 5 to 10% extent of melting generally proposed for Kilauea lavas [e.g., Pietruszka and Garcia, 1999] most of the (226Ra/230Th) disequilibria results from radiogenic ingrowth. We conclude that Ra/Th and Ba/Th were not controlled by the melting process; therefore anomalously high Ba/Th is a characteristic of the source for Hawaiian lavas.
 Within the Phase 2 core, Ba/La, Ba/Nb and Ba/Th are quite variable but strongly correlated; these ratios vary by a factor of ∼2.6 with one standard deviation of ∼15% (Figure 16). In contrast to these ratios involving Ba, other ratios involving only highly incompatible elements, such as La/Nb, are much less variable, especially within a shield (Figures 15c and 16). Even the variability of several ratios involving elements with significant differences in their incompatibility (e.g., La/Sm, Sm/Yb and Nb/Zr) is less than that of Ba/(La, Nb, Th) ratios; in fact, the ranges for Ba/La, Ba/Nb and Ba/Th are similar to that found for Nb/Y and La/Yb, ratios involving elements of very different incompatability (Figure 16).
 If high Ba/Th is a source characteristic, we expect this ratio to correlate with radiogenic isotope ratios. Historic Kilauea lavas define an inverse trend for Ba/Th versus 206Pb/204Pb and Mauna Loa lavas (<30 ka) show correlations of Ba/Th with Sr, Nd and Pb isotopic ratios (Figure 14c). However, in older and more altered lavas, e.g., from the HSDP cores and Koolau shield, Ba/Th ratios are considerably more variable than in lavas from Kilauea and Mauna Loa (Figure 15). Analyses of 101 Phase 1 HSDP Mauna Kea samples (not the reference suite) by Albarede  showed that for the lower parts of the Phase 1 core (689 to 1041 m) ratios of Rb/Th, Cs/Th and to a lesser extent Ba/Th decrease upwards in the core. A Principal Component Analysis of the data showed that a component, accounting for 5.1% of the variance, has maximum loading on Rb and Cs, moderate loading on Ba and marginal loading on U [Albarede, 1996]. Given the well-documented postmagmatic mobility of Rb, Cs, and U, a plausible inference is that relatively low Ba/Th reflects Ba loss and that the variable Ba/Th in the upper part of the Phase 1 core reflects lateral fluid (H2O) advection [see Albarede, 1996, Figure 3]. In the Phase 2 core Ba/Th ranges from 70 to 185, but there is no correlation with depth or compositional group (see inset in Figure 14c). To determine the effects of postmagmatic processes on Ba abundance in the Phase 2 core, Huang, Vollinger, Frey and Rhodes are determining compositional variations within four flow units. Based on data for 6 to 7 samples from these units, we find that Ba is not correlated with K. This result is consistent with the lack of correlation between K2O/P2O5 and Ba/Th in the reference suite of samples. Nevertheless, variable Ba abundance at nearly constant Th content within individual units leads us to conclude that some of the Ba/Th variability in the Phase 2 core arises from postmagmatic alteration. Because there is evidence for both loss and addition of Ba, we infer that the extremes of Ba/Th (<100 and >150) may reflect alteration.
 In summary the data arrays in Figure 14c could be explained by mixing of components with distinct combinations of Ba/Th and isotopic ratios. Particularly important are the diverse Ba/Th versus 87Sr/86Sr and 143Nd/144Nd trends defined by Koolau and Mauna Loa lavas. If these trends reflect magmatic characteristics, they convey important information about source components. However, Koolau lavas (∼2 to 3 Ma) are variably altered [Frey et al., 1994], and it is necessary to determine if their systematic Ba/Th-isotope ratio trends in Figure 14c are a fortuitous combination of source heterogeneity (isotopic ratios) and alteration (Ba/Th).
 Based on study of Mauna Kea lavas in reference sample suites from the two HSDP cores we identify 3 major types of geochemically distinct magmas.
 1. Lavas within the upper 150 m (<200 ka to ∼330 ka) of the Mauna Kea section are low SiO2 basalt with intercalated alkalic and tholeiitic basalt in the upper 50 m; these lavas form the Postshield Group. Their relatively low SiO2 is coupled with high incompatible element content, and high La/Yb and Sm/Yb; hence, relative to underlying lavas they were derived by lower extents of melting and segregated at higher pressure within the stability field of garnet. Relative to underlying lavas they also contain more of a component with relatively low 3He/4He, high εNd and non-radiogenic Pb. These changes in process and source are interpreted to have occurred as the Mauna Kea shield passed over the periphery of the thermally and geochemically zoned plume.
 2. Deeper in the Phase 2 core, >800 mbsl, two other groups, presumably intrinsic to the plume, are intercalated. These shield lavas are distinguished by different SiO2 contents at a given MgO content. Most of the HSDP Mauna Kea lavas belong to the relatively high SiO2 group (∼49–51% SiO2 after adjusting lava composition to be in equilibrium with Fo85 olivine). This group defines the Kea component.
 3. Like the Postshield Group, the other group is distinguished by a relatively low SiO2 content (<48.8% after adjusting lava compositions to be in equilibrium with Fo85 olivine). Unlike the Postshield Group, this shield group has relatively high Nb/Zr, Ti/Zr and 3He/4He and high 208Pb/204Pb at a given 206Pb/204Pb. These are the characteristics of tholeiitic lavas erupted at Loihi seamount. This geochemically distinctive component contributes to early shield volcanism at both Kea and Loa trend volcanoes. Based on abundance ratios such as La/Yb, Sm/Yb and Nb/Zr this group did not segregate within the garnet stability field.
 4. Although in detail the HSDP cores contain distinctive geochemical groups of lavas, within the context of all Hawaiian shields Mauna Kea shield lavas are not highly heterogeneous in isotopic ratios of Sr, Nd and Hf or in La/Nb and Nb/Zr. Ratios involving Nb reflect the Koolau component (recycled oceanic crust including pelagic sediment), and there is no evidence for this component in Mauna Kea lavas. Like other Hawaiian lavas, Mauna Kea lavas have high Ba/Th (>100), but this ratio is highly variable, possibly a result of alteration, and it is not correlated with radiogenic isotope ratios.
A2. Inductively Coupled Plasma Mass Spectrometry (ICP-MS)
A2.1. Sample Dissolution
 Samples were dissolved using a HF-HNO3 multistage digestion procedure in Teflon Savillex beakers. Rock powder, 50 to 100 mg, was weighed into a precleaned Teflon Savillex beaker; then 2 mL concentrated HF and 0.5 mL 14M HNO3 were added. The mixture of acids and rock powder in uncapped beakers was dried at 350°F to form a moist sample cake. Then 3.5 mL concentrated HF and 0.5 mL of 7M HNO3 were added to this sample cake. The capped beaker was heated at 250°F for 3 to 5 days. Then the sample was dried at 350°F to form a thick liquid gel, and 5 mL of 7M HNO3 were added. Again the sample was dried at 350°F to form a moist cake, and 5 mL of 7M HNO3 were added. Then the capped beaker was heated at 250°F overnight. Approximately 10% of the samples were not totally dissolved at this stage. These samples were transferred to a Teflon bomb and the same procedure outlined above was followed. The bomb treatment was repeated until the sample was totally dissolved. The samples were then diluted by a factor of 2000 to 8000 using 1M HNO3 and spiked with a solution, containing three isotopes, 77Se, 115In and 209Bi, with known concentrations prior to aspiration into a Fissons (VG) Plasmaquad 2+S ICP-MS.
A2.2 Analytical Procedure
 Our procedure typically involves sequential analysis of 37 sample solutions over a ∼6 hour time span. One to three isotopes are selected for each element based on high natural abundances and no or minimal isobaric and oxide interferences (Table A1a). Data are collected and reported as count rates (counts per second or “CPS”) for these isotopes. We use a standard calibration curve to convert these ion count rates into elemental abundances. Several USGS standard samples, BHVO-1, BCR-1 and AGV-1, are used to obtain the standard calibration curve (Table A2). The sensitivity of the instrument at a given mass is not stable with time. In addition, this temporal fluctuation in signal strength varies with isotope mass. These mass- and time-dependent variations are described as “sensitivity drift” (Figure A1).
Abbreviations are as follows: EDM, external drift monitor solution; STD, standard sample solution; BLANK, blank solution; UNKNOWN, unknown solution.
Table A2. Trace Element Abundances (ppm) in 3 USGS Standard Samples Used in ICP-MS at MIT
 In order to correct for sensitivity drift, both internal and external drift monitors are used. Three spiked isotopes, 77Se, 115In and 209Bi, are used as internal drift monitors (IDM). They record the sensitivity drift for these three isotopes in every analyzed solution. Beginning with the first solution in the analytical procedure, every sixth solution is an identical solution, such as BHVO-1, known as the external drift monitor (EDM). We use these EDMs to record the sensitivity drift in all the analyzed isotopes. Table A1b shows the typical sequence of analysis: external drift monitor, standard sample solution, 4 unknown solutions, etc.
Figure A1 shows the sensitivity drift in the three spiked isotopes, 77Se, 115In and 209Bi, in all solutions during one analytical procedure (∼6 hours). The first drift correction, the internal drift correction, is based on these 3 spiked isotopes. Since the concentrations of the three spiked isotopes are known in every solution, CPS for these three isotopes are first normalized by their isotope concentrations; then are normalized to CPS of the first EDM. Hence, for each of the three isotopes we have a correction factor for every solution. We call this the internal drift correction factor (IDF). We assign an IDF to each analyzed isotope mass using a linear interpolation (extrapolation for 232Th, 238U and 45Sc).
 In most situations, the internal drift correction does not adequately correct the sensitivity drift for all analyzed isotopes; therefore, we also correct for sensitivity drift using the external drift monitors. Figure A2 shows the sensitivity drift, after internal drift correction, in all the analyzed isotopes in EDMs during one procedure (∼6 hours). Clearly, the range in internal drift corrected CPS from 0.92 to 1.08 (Figure A2) shows that internal drift correction has not adequately corrected the sensitivity drift. Therefore, we also apply an external drift correction. We normalize the CPS in all EDMs to that in the first one; hence, we have a correction factor for each of the analyzed isotopes in every external drift monitor. We call this the external drift correction factor (EDF). Because an external drift monitor is analyzed as every sixth solution in the analytical sequence (Table A1b), we must interpolate to determine the EDF for each intervening solutions, i.e. the 4 unknown and one standard solutions. This interpolation is done using a cubic spline fit. After application of the EDF, the reproducibility of all analyzed elements is better than ±6%, and better than ±3% for most of these elements. The only “cost” is that seven EDMs are analyzed (Table A1b), which increases machine time by 20%.
A3. Estimation of Precision and Accuracy
 Most of the uncertainty in ICP-MS data are caused by sensitivity drift during the analysis. Therefore, the sensitivity drift correction factors are very important in controlling the precision and accuracy of the data determined by ICP-MS. In order to evaluate our sensitivity drift correction procedure, we repeatedly analyzed a BHVO-1 solution using our typical analytical procedure, i.e., in this procedure all the unknown and standard sample solutions and EDMs are a BHVO-1 solution. In this experiment, after sensitivity drift corrections (IDF and EDF), all the solutions should have the same signal strength, i.e., the same CPS. For comparison, we normalized the CPS of all the solutions to that of the first solution. Figure A3 shows the normalized CPS values for each of the 25 analyzed elements. Overall, the normalized CPS values are within ±6% for all the solutions, most are within ±3%. The uncertainties, one standard deviation in percent, of these 25 elements are ∼1–2%.
 In order to evaluate accuracy, we analyzed, BHVO-2, a newly prepared USGS standard sample. The average values and standard deviations of 11 replicate analyses obtained over ∼5 months are reported in Table A3. For comparison, Table A3 also shows the trace element abundances in BHVO-2 reported by Ila and Frey , Lin et al. , Raczek et al.  and the USGS recommended values. Except for Tb and Yb, the trace element abundances in BHVO-2 analyzed by ICP-MS at MIT are within ±5% of the INAA values reported by Ila and Frey . Values of Tb and Yb in BHVO-2 analyzed by MIT ICP-MS are about 8% higher than reported by Ila and Frey . Our reported values for BHVO-2 are also within ±5% of the values reported by Raczek et al. , and for most of the elements the differences between the two reported values are within analytical uncertainties (Table A3). Taking the uncertainty into account, except for Gd and Y, our results are also in agreement with that reported by Lin et al. . Compared with the USGS recommended values, the difference between our reported values for BHVO-2 and the recommended values are within the reported standard deviations (Table A3).
Table A3. Trace Element Abundances in BHVO-2 Analyzed by MIT ICP-MS and Other Techniquesa
All element abundances were determined by ID-TIMS, except for Pr, Tb, Ho and Tm which were determined by MIC-SSMS.
A4. Comparison With Other Techniques (XRF and INAA) and Another ICP-MS Facility
 Trace element abundances for HSDP Phase 2 reference samples have also been determined by X-ray fluorescence (Rhodes and Vollinger, submitted manuscript, 2002), INAA at MIT (this paper) and ICP-MS at Rutgers (Feigenson et al., submitted manuscript, 2002). These multiple data sets for the same samples provide the opportunity for comparison of results obtained by different techniques or the same technique (ICP-MS) in different facilities.
A4.1 ICP-MS (MIT) Compared to XRF (University of Massachusetts)
 In Figure A4, we compare results obtained by ICP-MS (MIT) and XRF (Univ. Massachusetts). The linear correlation coefficients between the two data sets for Ba, Sr, Zr, Nb, Rb and Y are greater than 0.95 (Figure 4). These results show that the precision of both techniques is comparable, ∼3–6%, and the slopes are near unity for each element (except Y), thereby indicating that both analytical techniques yield similar abundances. However, for Y abundance there is a systematic difference between the two techniques with the Y abundance determined by ICP-MS approximately 12% higher than that determined by XRF.
A4.2. ICP-MS (MIT Compared to Rutgers)
 In Figure A5 we compare ICP-MS results for REE abundances obtained at MIT and Rutgers. For La, Ce, Pr, Nd, Sm and Eu the correlation coefficients are 0.98 to 0.99 and systematic differences are <4%. Correlation coefficients for Gd, Tb, Dy, Ho, Er, Tm, Yb and Lu are lower 0.93 to 0.96 and systematic differences are ≤5%, except for Tm, Yb and Lu whose abundances determined at MIT are 6 to 7% higher than those determined at Rutgers.
A4.3 ICP-MS (MIT) Compared to INAA (MIT)
 In Figure A5, we also compare results obtained by ICP-MS and INAA for REE, Hf and Th. Except for Tb and Th the linear correlation coefficients between the results from ICP-MS and INAA are greater than 0.96. The poorer correlation for Tb and Th reflects the poorer precision of INAA data for these elements (see precision estimates in Yang et al. [1996, Table 1]). Also there is a systematic discrepancy between the two techniques for Tb, Yb and Lu. The INAA results are ∼8% lower than the ICP-MS data; this same discrepancy was found for analyses of BHVO-2 (Table A3).
 In order to evaluate the relationships between HSDP 2 lavas, we calculated La/Yb-Nb/Zr, Sm/Yb-Nb/Zr, Sm/Yb-La/Yb and La/Yb-Ti/Zr trends for variable extents of batch melting of garnet- and spinel- peridotite (Figure 10). Source abundances and mineral proportions and melting reactions are in Table A4. The partition coefficients are assumed to be zero for olivine/melt and spinel/melt. For clinopyroxene and garnet the partition coefficients vary with mineral composition and the values used are in Table A4. For garnet peridotite we used data for high pressure clinopyroxene and garnet (TM 694-6) from Salters and Longhi . For spinel peridotite we used data for a lower pressure clinopyroxene (TM 1094-9) from Salters and Longhi . For orthopyroxene, we estimated partition coefficients from measured Dopx/cpx ratios [Eggins et al., 1998]. For more detailed information, see the footnote of Table A4.
Since KdTiGrt/CPX ranges from 0.4 to 2 [Harte and Kirkley, 1997], two values are used for KdTiGrt/melt. In model “garnet peridotite a”, KdTiGrt/melt = 0.2; in model “garnet peridotite b”, KdTiGrt/melt = 0.03. These yield (Ti/Zr)grt/(Ti/Zr)melt = 0.07 and 0.49; the latter value is similar to the ratio (0.3 to 0.7) found by van Westrenen et al. .
 We thank D. DePaolo, E. Stolper and D. Thomas for their leadership of the HSDP, M. Garcia, C. Seaman and the logging crew for their on site efforts, and our colleagues studying these cores who shared their data. We thank J. M. Rhodes for numerous discussions and an informal review. P. Ila is thanked for her supervision of the MIT neutron activation facility. A. Pietruszka, K. Sims, and H. Zou provided helpful constructive reviews. This research was supported by National Science Foundation Grant EAR-9528594. We thank M. Schmitz and B. Grant for their assistance in ICP-MS analysis.