Spatial and Temporal Scale of Mantle Enrichment at the Endeavour Segment, Juan de Fuca Ridge

Abstract

Major ± trace element and Sr–Nd–Pb–Hf–He isotope data are presented for more than 300 geochemically diverse basalt samples collected by submersible from the Inflated Central Endeavour Segment of the Juan de Fuca Ridge. Seven chemically distinct basalt types are present, from depleted (D-) to enriched mid-ocean ridge basalt (E-MORB). By combining the geochemical data with high-resolution bathymetry and age determinations, the detailed spatial and temporal scale of on-axis mantle-derived basalt heterogeneity is determined. The basalts define binary mixing arrays in all isotope plots that are usual in their correlations, but unusual in the limited range of Sr–Nd–Hf isotope compositions for D- to E-MORB, and greater range in Pb isotopes. The basalts also define two different styles of enrichment of moderately incompatible elements. Geochemical enrichment began when the currently inflated axial ridge formed <10⁵ years ago. One enrichment style (the Inflated Ridge Trend) characterizes basalts erupted across the ∼5 km wide ridge from >10^000 to ∼4000 years ago, whereas the other enrichment style (the Graben Trend) characterizes most basalt types erupted within the axial graben after it formed ∼2300 years ago. We attribute the Inflated Ridge Trend to a relatively high proportion of pyroxenite (or melt derived therefrom) to enriched peridotite in the mantle during a phase of ridge inflation that lasted at least 6000 years. The Graben Trend reflects the reduced effect of pyroxenite after the axial graben formed. Because at least 14 different samplings of mantle components occurred within <1 km of ridge length and width during a time when <1 km of upwelling occurred, we infer that the scale of mantle heterogeneity far from a plume is < 1 km. The enriched mantle component at Endeavour is young with ²⁰⁶Pb/²⁰⁴Pb ∼19·0; Hf and He isotope ratios trend toward HIMU characteristics. These traits are regionally widespread and are shared with the next two ridge segments to the north (West Valley and Explorer).

INTRODUCTION

Mid-ocean ridge basalt (MORB) geochemistry at high spatial and temporal resolution bears on many current topics related to mid-ocean ridges, especially when the MORB have diverse trace element and isotope ratios. How do the periodicity of mantle melting events (‘melt supply’) and the duration of magma storage in the crust affect the chemical composition of erupted MORB (e.g. Rubin et al., 2009; O’Neill & Jenner, 2012)? Indeed, is magma mixing at ridges with melt lenses so ubiquitous that mixing erases information about mantle source heterogeneity unless primitive melts erupt or are trapped in melt inclusions? At how fine a scale do physical discontinuities in ridges or their melt lenses correlate with differences in the composition of overlying basalts (Carbotte et al., 2012)? That is, does magma rise vertically from the mantle and crustal reservoirs? How do melt supply and storage interact with the ambient stress of a ridge segment to influence the geological history, morphology, and hydrothermal systems of that segment (e.g. Perfit & Chadwick, 1998; Carbotte et al., 2006)? What is the nature and scale of chemical ± lithological heterogeneity of the mantle beneath ordinary ridges? Our ability to address these kinds of questions depends on the spatial and temporal scales of basalt sampling and analysis, which often are limiting factors (e.g. Rubin et al., 2009).

These questions are most tractable at intermediate spreading rate ridges such as the Juan de Fuca Ridge (JdFR) where rates of replenishment, eruption, and faulting are thought to fluctuate most rapidly. Relative to fast-spreading ridges such as the East Pacific Rise (EPR), magmas at the JdFR are thought to experience less frequent replenishment, erupt less frequently, differentiate at a deeper level, be more mafic on average on eruption (i.e. higher Mg#), result from slower mantle upwelling, and preserve more geochemical evidence of mantle heterogeneity (e.g. Langmuir et al., 1992; Rubin & Sinton, 2007). These parameters are thought to vary with time and space even within the JdFR and, currently, to be most like those of a slow-spreading ridge at the Endeavour Segment (Carbotte et al., 2008).

Mantle source heterogeneity is usually identified using coupled variations of element and isotope ratios such as K₂O/TiO₂, La/Sm and ⁸⁷Sr/⁸⁶Sr in basalts. The resulting diversity is expressed as depleted (D), normal (N), transitional (T), and enriched (E)-MORB along a compositional continuum. However, the spatial and temporal relationships of these basalt types, especially at ridge axes, are rarely known in sufficient detail to provide high-resolution answers to the kind of questions posed above. Most commonly the range in composition from D- to E-MORB occurs over tens to hundreds of kilometers of ridge near plumes (Schilling et al., 1983) or large transform faults (e.g. Ligi et al., 2005), is spread out along multiple segments of ridge (e.g. Cushman et al., 2004; Hoernle et al., 2011; Gale et al., 2013), or occurs in off-axis seamounts (e.g. Niu et al., 2002) or within transform faults (e.g. Sims et al., 2002). Well-mapped T- or E-MORB erupted far from plumes or transforms have been found within an axial graben, or within a few kilometers off-axis, at only a few ridge segments (e.g. along the northern EPR: Hekinian et al., 1989; Batiza & Niu, 1992; Reynolds et al., 1992; Clague et al., 2009; Waters et al., 2011). Nowhere before have basalts with K₂O/TiO₂ from <0·05 to > 0·3 been found in less than a kilometer within the axial graben of an intermediate- or fast-spreading ridge, and placed in a high-resolution spatial, temporal, and geochemical context.

The Endeavour segment of the JdFR is well suited to address these topics for several reasons. First, it was one of the former US-NSF RIDGE2000 Integrated Study Sites, which allows its basalt petrology to be integrated with extensive geophysical information about the mantle and crust, geological information about the ridge axis, and geochemical information about hydrothermal deposits and fluids (Kelley et al., 2012; Clague et al., 2014). Second, it includes abundant E-MORB and some N-MORB (Karsten et al., 1990). Third, its basalts have globally high levels of short-lived ²³⁰Th and ²³¹Pa that constrain the mantle upwelling velocity and that can be used for radiometric dating (Goldstein et al., 1992; Lundstrom et al., 1995). Finally, the segment is ordinary, with a globally average spreading rate and axial water depth, and it is far from a large transform or mantle plume.

Despite these characteristics and the pioneering work of Karsten et al. (1990), no chemical analysis including major elements and even rare earth elements (REE) and Sr–Nd isotope data for the same sample has ever been published for any Endeavour basalt. There are no previous inductively coupled plasma mass spectrometry (ICP-MS) trace element analyses or Pb or Hf isotope data. The few samples with published Sr, Nd, or U-series isotopes lack even major elements, and most samples are from long dredges and cannot be related to their geological or geophysical setting in detail.

To fill this gap, we present new major element analyses of 322 well-located and photographed Endeavour basalts collected in the axial graben and adjacent flanks between 2002 and 2011, mostly using remotely operated vehicles (ROVs). We also present internally consistent volatile, trace element, and multi-isotope analyses for subsets of the samples. This study summarizes the tectonic and geological context of the samples and discusses some of their implications for mantle sources. Companion (Clague et al., 2014) and future papers will address their relationship to the local geology, the last few thousand years of magmatic history, U-series disequilibria and numerical models of the mantle melting processes.

GEOLOGICAL AND GEOPHYSICAL SETTING

The Endeavour Segment of the 480 km long Juan de Fuca Ridge (JdFR) spreads at 52 mm a^–1 full rate (DeMets et al., 2010), the fastest along the JdFR. It extends ∼90 km along axis from 47·5°N to 48·3°N, between the Northern Symmetrical Segment to the south and the West Valley Segment to the north (Fig. 1). We use the name Inflated Central Endeavour Segment (ICES) for the 11 km long, shallowest portion of the segment from 47°54'N to 48°00'N shown in Fig. 2. This was the ‘bull’s eye’ of the Endeavour Integrated Study Site of the former NSF Ridge2000 program. It consists of two ∼2 km wide flank highs separated by a 0·4–1·2 km wide, 130–180 m deep axial graben that is the center of the spreading center. The graben deepens from 2100 m in the north between the Sasquatch and Salty Dawg hydrothermal fields to 2400 m south of the Mothra hydrothermal field. The flanks are asymmetric morphologically, with a shallower, wider, and less faulted west flank compared with the eastern flank. At a full spreading rate of 52 mm a^–1, the entire 5 km wide ICES should be ≤75^000 years old. Overviews of the ICES have been provided by Kelley et al. (2012) and Clague et al. (2014).

Fig. 1

Regional map showing the morphological features discussed in the text. Bathymetry is from Clague et al. (2014). The box in the inset shows the location of the larger figure. The ridge segment NW of that box is Explorer (‘Expl.’). The box on the Endeavour Segment shows the location of the ICES in Fig. 2.

Fig. 2

(a, b) Maps of the Inflated Central Endeavour Segment (ICES) showing sample locations and basalt types. (See Fig. 1 for the regional setting.) The location of Summit Seamount and the main hydrothermal vent fields are labeled. The base map from MBARI has sub-meter resolution, and the sample locations have been re-navigated where necessary to be consistent with the bathymetry (Clague et al., 2014). Dive tracks are numbered. Those from 2002 and 2004 have a prefix T; those from 2011 have a prefix D. Sample numbers are given inside circles in sequence for each dive, and the circle colors denote basalt types (see legend). Sample numbers in Supplementary Data combine dive and sequence. For example, the location of sample T737-27 is shown by the orange circle enclosing ‘27’ along the maroon T737 dive track in the upper part of the figure. Sample numbers of Alvin dives and the 2006 collection are the final three or four digits. The location of the two drill core samples collected using Jason are shown as ‘1’ and ‘2’ on the lower east and west flanks, respectively. RC indicates location of the four rock cores of the 2011 collection. Dredge locations are not shown because of their spatial uncertainty.

The Endeavour and Northern Symmetrical Segments have been ‘dueling propagators’ since 3·5 Ma, especially since ∼200 ka (Johnson et al., 1983). Endeavour also overlaps the West Valley Segment for about 10 km in the north (Karsten et al., 1986; Wilson, 1993). Only about a third of the overall Endeavour Segment is now free of some overlap but that includes all of the ICES.

The ICES has a typical oceanic crustal thickness and velocity structure with ∼0·5 km of volcanic extrusive rocks underlain by ∼1 km of sheeted dikes, such that the top of the gabbroic section (i.e. fossil magma chamber) is ∼1·5 km below the sea floor (Cudrak & Clowes, 1993; Van Ark et al., 2007). The crust is only ∼5·9 km thick but is separated from 7·8 km s^–1 (V_p) mantle by an ∼1 km thick lower velocity transition zone (Weekly et al., 2011). This typical 6–7 km crustal thickness conventionally implies a mantle potential temperature of ∼1350 °C (e.g. Asimow & Langmuir, 2003).

The ICES is underlain by a seismic reflector that indicates the presence of a current axial magma chamber (AMC) that is 2·2–3·3 km below the sea floor, 0·4–1·2 km wide, dipping 8–36° to the east, and segmented along-strike (Van Ark et al., 2007; Carbotte et al., 2012). It extends from the western boundary of the axial graben to beneath part of the east flank. It is continuous from the Sasquatch to Main Endeavour Field (MEF) hydrothermal fields, is somewhat disturbed between Main and Mothra, and is clearly offset downward above our southernmost samples near Stockwork, creating a fourth-order segmentation. (See Fig. 2 for locality names.)

Although no historical eruptions are known within the ICES, earthquake swarms in the area in 1999–2000 and again in 2005 were interpreted as magmatic in character. Within the ICES, earthquakes were concentrated within an ∼1 km thick band directly above the AMC (Weekly et al., 2013), and swarms coincided with changes in fluid chemistry in the hydrothermal vents (Lilley et al., 2003). Earthquakes may record an extended diking event north of the ICES itself that propagated southward toward the segment center (Bohnenstiehl et al., 2004; Weekly et al., 2013).

Because the ICES is more faulted than the EPR at 10°N, for example, it was long thought to be in an amagmatic phase (e.g. Karsten et al., 1990). Periods of increased melt supply were thought to build axial highs (as in the Vance Segment) whereas periods of reduced supply created axial grabens (e.g. Endeavour) (Kappel & Ryan, 1986). However, the presence of an AMC and the diking events show that the ICES is magmatically ‘active’. At least a dozen lavas are <2300 years old, including at least four that erupted during the two centuries between 300 and 500 CE (Clague et al., 2014). We will show that basalt compositions changed when the axial graben developed a few millennia ago. Therefore, the morphological difference between Endeavour and inflated segments such as Vance may reflect subtle differences in sources, melting processes, and local stress regimes (Carbotte et al., 2006), or in the geometry of melt ascent and storage, rather than melt supply rate alone.

Several seamount chains occur on the western side of the JdFR, and the short Heckle Seamount Chain projects toward the ICES (Fig. 1). The ICES sits on the eastern edge of an ∼30 km wide plateau that is ∼200 m shallower, and has crust ∼500 m thicker, than the adjacent sea floor. This has been attributed to excess melt production related to the Heckle Seamount Chain since ∼0·7 Ma (Karsten et al., 1986; Carbotte et al., 2008), although this chain is not considered to be a hotspot and its basalts are very depleted (Cousens et al., 1995), similar to the Lamont Seamounts west of the EPR (Fornari et al., 1988) or the Vance and President Jackson Seamounts west of the Vance and Northern Gorda Segments, respectively (Davis & Clague, 2000).

The recent volcanological and tectonic history of the ICES has been discussed by Jamieson et al. (2013) and Clague et al. (2014). All ICES lavas discussed in our study, from both the axial graben and flanks, are Holocene in age (<10 ka), except for the distal pillow mounds on the westernmost flank, which have ages of 10–30 ka (Scott, 2012). Most other basalts on both flanks are ∼11 to 4 kyr old, without apparent age progression, whereas basalts in the axial graben are <2·3 kyr old. The axial graben is proposed to have formed between about 4300 and 2300 years ago (Clague et al., 2014). Hydrothermal venting is widespread in the axial graben, but is rare and dormant on the flanks. Sheet and lobate lavas dominate within the axial graben and flow parallel to its axis, whereas the flanks have pillow lavas and mounds that are surrounded and underlain by sheet flows. Lavas of both morphologies flow away from the axis on the flanks.

In summary, the ICES is not near a plume or large transform fault, and its spreading rate and water depth lie midway between global extremes. It is, therefore, an excellent segment at which to evaluate how globally important parameters, including source heterogeneity and melting rate, vary with time and affect MORB compositions at an ordinary ridge segment. Within the overall JdFR, the ICES has been in a less magmatic, more tectonic phase for the last two millennia, but that may just indicate that eruptions occur every few centuries instead of every few decades. The width of the zone considered as being ‘on-axis’ may have varied from ∼1 km as now, to 3–4 km a few thousand years ago, including what are now flanks. Any lava that erupted in the axial graben after it formed flowed along axis, not off axis.

SAMPLE LOCATIONS and SAMPLING METHODS

We report analyses of 314 basalts from the ICES (i.e. from the axial graben or from the flanks within 2 km of the graben axis), plus eight basalts from 0·5–2·0 km beyond the flanks. All but nine of these samples were collected using an ROV or manned submersible so that the geological setting was documented and photographed, and most of the locations are known to within ∼10 m relative to other samples obtained during the same dive. Almost all samples were collected during eight ROV dives in 2002 (T463–471) funded by the Keck Foundation, five in 2004 (T737–741) funded by NURP, and three in 2011 (D264–266) funded by MBARI. All dives used the Monterey Bay Aquarium Research Institute’s R.V. Western Flyer and its ROVs Tiburon and Doc Ricketts. Four basalts are drill cores collected in 2003 using the ROV JasonII and the University of Washington’s R.V. Thomas Thompson. Twelve basalts were collected in 2006 and 2007 by D. Kelley (University of Washington) using DSRV Alvin. Ten basalts that were collected earlier using DSRV Alvin or by dredging were provided by D. Kelley and M. Rhodes (University of Massachusetts). Sample locations and dive tracks are shown in Fig. 2.

Although the relative locations of our ROV-collected samples are well known, integration of their location with the most recent bathymetric and side-scan sonar maps was challenging, as discussed by Clague et al. (2014). The six-digit latitude and longitudes for our samples in Supplementary Data (all Supplementary Data are available for downloading at http://www.petrology.oxfordjournals.org), and therefore their location with respect to the underlying bathymetry of Fig. 2, are thought to be accurate only to within 20–30 m. Samples from Tiburon dives T471 and T739, and the 2006 collections using Alvin, have the least accurate locations because of malfunctioning of the navigation systems during the dives.

Most samples are fresh, 0·25–1·0 kg sized lava fragments with glassy surfaces, collected using the ROV manipulators. Some are glass shards collected in wax-tipped pipes using the ROV arm, and seven were recovered as wax cores dropped from the surface ship. The four drill cores and 37 other samples are variably crystallized whole-rocks with insufficient glass for microprobe analysis.

In addition to our samples, some geochemical data for 40 other basalts from within the area of Fig. 2 are available from PetDB. Most are from Karsten et al. (1990), who reported major element and X-ray fluorescence (XRF) trace element data for basalts from 11 dredges near the ICES. They are mostly E-MORB from the flanks, but include N-MORB in two dredges up the west wall of the axial graben.

ANALYTICAL METHODS

This project involved sample preparation and analytical work by many personnel for almost a decade in different laboratories. We made considerable effort to standardize methods and calibrations, and then to normalize results to common reference values, and we present information below to help readers assess that process.

Sample preparation

Glass shards for microprobe analysis were selected under magnification after sonication in distilled water. Whole-rocks for XRF analysis were selected under magnification after removing altered surfaces, breaking into ∼1 cm chips, and sonicating in distilled water until no turbidity was visible. Crushing at University of California Santa Cruz (UCSC) or Washington State University (WSU) used agate or ceramic mills.

ICP-MS trace element and isotope analyses were performed on glass shards whenever possible or on minimally crystalline lava interiors. Unfortunately, the difference sometimes matters for Sr isotopes and U concentrations, and by requiring different analytical methods for major elements (microprobe for glass; XRF for whole-rocks). Both glasses and whole-rocks were sonicated in distilled water until no turbidity was visible, and broken to 0·5–1·0 mm using a steel percussion mortar. Between 0·2 and 1·0 g (the higher value when Pb and Hf isotopes were to be measured) of fresh material with no visible Mn crust or alteration was selected under magnification, and then leached in warm 2·5N HCl for 15 min. This protocol was found to remove Mn, Ni, and Co from Mn crusted glass shards, whereas only sonication in distilled water did not. Leaching in HBr instead of 2·5N HCl had no effect on Pb isotope ratios. Leaching in hot 6N HCl or H₂O₂ and oxalic acid had no further effect on trace element concentrations in glass within our external precision of 1–2% for separate digestions, even though the leached glass was visibly pitted. Nevertheless, leaching in 2·5N HCl and even H₂O₂ and oxalic acid did not always remove all secondary U, Pb, and Sr from fresh-appearing slightly crystalline rocks, as evidenced by ratios of Nb/U < 40 and Th/U <2, and slightly elevated ⁸⁷Sr/⁸⁶Sr and ²³⁴U/²³⁴U ratios. We attribute this to trace alteration minerals in vugs in slightly crystalline and vesicular rock. Sometimes sulfur was visibly released during HCl leaching of these samples. U and Pb concentrations are not reported when Ce/Pb and Nb/U ratios are <25 and <40, respectively.

Additional leaching experiments were made for Sr isotope analyses. For both glass and whole-rock, the 2·5M HCl leachate always had ⁸⁷Sr/⁸⁶Sr greater than the final leached material. Subsequent peroxide leachate had lower ⁸⁷Sr/⁸⁶Sr than the HCl leachate when leaching glass (although still ∼0·703), but even higher ratios when leaching whole-rocks, indicating more effective leaching of Sr from whole-rocks by H₂O₂ than by HCl. After HCl plus H₂O₂ treatments, the oxalic acid leachate had ⁸⁷Sr/⁸⁶Sr equal to the final leached material for glass but sometimes it was still slightly elevated for whole-rocks. Therefore, Sr isotope data for whole-rocks could be slightly high because of unremoved seawater alteration. In two test cases (738-40 and 740-27), even aggressively leached whole-rocks had ⁸⁷Sr/⁸⁶Sr 0·000020–0·000027 higher than a leached glass of the same sample. Several whole-rocks seem to be offset to higher ⁸⁷Sr/⁸⁶Sr by about this amount in T-MORB and E-MORB, and by as much as 0·00010 in some N-MORB whole-rocks and glasses.

All samples are currently archived at UCSC and available from the first author upon request.

Electron microprobe

Most of our major element data (88%) were obtained by microprobe and the rest by XRF. Because the microprobe analyses were obtained over a period of 10 years by different people on different instruments, we made a concerted effort to calibrate and normalize results in a consistent fashion. Samples collected in 2002–2004 were analyzed at the US Geological Survey (USGS)-Menlo Park by J. Kela following the methods of Davis & Clague (2000). A JEOL 8900 microprobe was used with an accelerating voltage of 15 kV, 25 nA current, and 10 μm beam diameter. Peak counting times were 40 s for Mg, Al, Na, K, and S; 30 s for Mn, P, and Cl; and 20 s for Si, Ti, Fe, and Ca. Overall precision estimated from 126 analyses of different chips of the same N-MORB glass (EPR ‘BBQ Flow’ 2392-9) during 10 analytical sessions over 2 years is 1–2 relative % for all major oxides except MnO, K₂O, and P₂O₅ for which the standard deviation was ≤0·008 wt % (Supplementary Data). Juan de Fuca glass standard VG2 was used to calibrate Si, Al, Fe, Mg, Ca, and Na using the United States National Museum (USNM) values in Supplementary Data (Jarosewich et al., 1980). The differences from the KKN93 preferred VG2 values of Niu et al. (1999), especially for MgO, should be noted. Corning glass standard GSC was used for K; Wilberforce apatite for P; TiO₂ for Ti; Mn₂O₃ for Mn; sodalite for Cl; and troilite for S. Drift was monitored using VG-2 but was <2% and inconsistent within analytical sessions, so no drift correction was applied. Variations within and between analytical sessions of up to 0·2–0·3 wt % for Al₂O₃, FeO, and CaO, and 0·1 wt % for Na₂O, are not resolvable for these data even though they might distinguish between subtly different liquid lines of descent amongst basalts of the same type. The K₂O/TiO₂ ratios of 2392-9 and VG2 also agree with accepted values when calibrated against USGS glasses, so the atypicality of Endeavour results is not an analytical artifact.

Results in Supplementary Data are the mean of 5–15 analyses of 3–5 glass shards per sample. The raw total for each sample is included; the average is 99·40% ± 0·20% 2σ. Each basalt type defines a separate positive correlation between raw total and SiO₂ with a slope of 0·5, confirming subtle differences in SiO₂ between basalt types.

To facilitate comparison with other laboratories, the data in Supplementary Data have been normalized to the values for EPR glass 2392-9 of Sims et al. (2002), for which assumed values, measured values and their standard deviations are given in Supplementary Data. The most important consequence of normalization is the 4·8% (relative) increase in MgO, which is about the difference between the USNM and Niu et al. (1999) values for VG2 noted above. (Even larger correction factors of 8–9% relative for MgO have been needed for data from other laboratories: Langmuir et al., 2006, table 1.) Normalization also increased the 2004 TiO₂, MnO, and P₂O₅ concentrations by about 0·05, 0·03, and 0·01 wt % respectively. Results are presented without additional normalization to 100%, even though slight differences in carbon coating of probe mounts can result in differences in count rate and, therefore, totals.

Table 1

Representative analyses of each ICES basalt type

Type:	D	N1	N2	T1	T2	T3	E
Sample:	D265-R12	T737-29	T738-23	T737-05	T737-15	T471-17	T740-38
SiO2	49·56	49·92	51·24	49·64	50·72	50·61	51·40
TiO2	0·99	1·56	1·25	1·95	1·63	1·71	1·71
Al2O3	16·68	15·66	14·67	15·95	14·93	14·85	14·81
FeO	9·26	9·13	9·96	9·48	9·53	9·52	9·78
MnO	0·15	0·17	0·22	0·19	0·20	0·21	0·18
MgO	9·15	8·17	8·28	7·52	7·80	7·40	7·09
CaO	12·56	12·10	12·35	11·01	11·86	11·77	11·40
Na2O	2·30	2·48	2·30	2·90	2·64	2·86	2·85
K2O	0·07	0·21	0·15	0·50	0·36	0·37	0·51
P2O5	0·10	0·15	0·14	0·32	0·22	0·24	0·28
Cl	0·034	0·0038	0·007	0·0168	0·0122	0·0146	0·0086
S	0·105	0·107	0·127	0·122
H2O		0·238	0·164	0·494	0·298	0·377	0·371
CO2 (ppm)		249	221	174	236	234	199
Total	100·95	99·90	100·85	100·09	100·19	99·92	100·39
Ni (ppm)	130·9	99·9	67·0	127·8	67·5	58·0	49·6
Cr	306	315	158	287	192	177	98
V	211	260	287	275	286	290	314
Sc	38·8	41·7	45·2	36·9	39·6	43·6	38·4
Co	38·5	42·4	44·3	40·2	41·4		41·2
Rb	0·7	3·62	2·75	9·21	5·98	5·60	7·07
Sr	86	148	101	225	179	231	231
Y	29·2	29·7	29·3	33·5	30·2	32·2	31·7
Zr	73	95	70	154	113	126	136
Nb	1·4	7·48	5·05	18·75	12·3	12·7	16·7
Cs		0·04	0·03	0·10	0·060	0·062	0·086
Ba	6·2	41	32	98	68	69	86
La	2·4	5·52	3·91	11·87	8·46	9·97	10·93
Ce	7·5	13·80	9·82	26·78	19·28	21·98	24·59
Pr	1·38	2·12	1·58	3·70	2·77	3·08	3·40
Nd	8·0	10·42	8·01	16·65	13·22	13·95	15·26
Sm	2·96	3·28	2·81	4·47	3·80	4·04	4·31
Eu	1·01	1·19	1·03	1·49	1·30	1·39	1·45
Gd	4·26	4·20	3·76	5·19	4·43	4·77	4·76
Tb	0·67	0·76	0·67	0·90	0·793	0·810	0·846
Dy	5·16	4·78	4·52	5·45	4·90	5·07	5·33
Ho	1·08	1·04	1·06	1·17	1·06	1·06	1·11
Er	3·33	2·91	2·93	3·23	2·94	3·03	3·26
Tm	0·51	0·42	0·43	0·46	0·42	0·45	0·45
Yb	3·07	2·68	2·86	2·95	2·70	2·90	2·85
Lu	0·47	0·39	0·43	0·42	0·40	0·46	0·43
Hf	2·28	2·33	1·99	3·44	2·69	3·07	3·07
Ta	0·10	0·46	0·33	1·16	0·74	0·82	1·00
Pb	0·24	0·47	0·35	0·84	0·62	1·03	0·77
Th	0·13	0·46	0·34	1·19	0·76	0·70	0·96
U	0·04	0·14	0·10	0·42	0·25	0·21	0·33
87Sr/86Sr	0·702470	0·70245	0·702469	0·702477	0·702484	0·702449	0·702484
143Nd/144Nd	0·513167	0·51315	0·513153	0·513137	0·513128	0·513146	0·513139
206Pb/204Pb		18·748	18·599	18·906	18·806	18·493	18·652
207Pb/204Pb		15·491	15·492	15·501	15·510	15·473	15·496
208Pb/204Pb		38·055	37·921	38·196	38·160	37·822	37·991
176Hf/177Hf			0·283173		0·283134	0·283167	0·283153

Samples collected in 2006 were analyzed at the USGS-Menlo Park in 2008 by Dave Clague using the same calibration standards and method as above but without analyzing 2392-9. Therefore, the only normalization correction applied to these data was to increase measured MgO by 4·8% relative. Samples collected in 2011 were analyzed at University of California Davis (UCD) in 2011 by Dave Clague using the same calibration standards and method, plus analyzing 2392-9 and USGS basaltic rock standards. Results for the USGS standards agree with recommended values, except for MgO contents, which are consistently ∼5% too low. Therefore, the only normalization correction applied to these data was to increase measured MgO by 5% relative, about the amount needed for normalization to 2392-9 or to calibrate against VG2 assuming MgO = 7·07 wt % (Niu et al., 2002).

Two samples (471-6 and 737-36) were re-analyzed 7 years apart on different microprobes, and two others (738-27 and 739-13) were compared with a different sample collected a decade later from what is thought to be the same flow unit. Most agreement is within long-term precision (1–2 relative %), although some 2011 analyses have up to 0·1 wt % higher TiO₂ and 0·07 wt % higher K₂O. Therefore, differences of these magnitudes between the 2002–2004 and the 2011 samples may have more analytical than geological significance.

For ∼100 samples, the Cl, F, S, K, and Ti contents were also measured simultaneously using long counting times and a high beam current with the four-spectrometer Cameca microprobe at the University of Tulsa. Glasses and standards were analyzed using a 15 kV beam and 80 nA beam current with a 20 μm beam diameter. Peak positions were counted for 200 s for each spot (160 s for K and 40 s for Ti), and background positions on either side of the peak were counted for 100 s each (80 s for K and 20 s for Ti). We verified that the intense beam did not alter the glass during the analyses and lead to compositional or count rate changes by checking several glasses for count rate variations over consecutive 20 s intervals. Reported analyses are the average of three or four single spots. The 2σ precision, based on multiple three-spot analyses of different chips of the same sample performed over several months or years, is similar to counting errors and is ± 0·0016% for Cl, ±0·004% for F, ±0·0016% for S, ±0·004 for K and ±0·01 for Ti. Standards were natural scapolite for Cl, synthetic F-phlogopite for F, pyrite for S, sanidine for K and ilmenite for Ti. Instrument drift was monitored and corrected for by using frequent analyses of a secondary standard: TR154 21D3 with 0·0404% Cl, 0·52% K, 0·107% S and 0·05% F (Michael & Schilling, 1989). Tulsa Cl analyses are about 10% higher on average than the less precise USGS ones for the same samples, probably a result of different calibrations between laboratories. Tulsa Cl analyses are about 4% lower than high-precision analyses by electron microprobe at the Australian National University, based on our unpublished data for 16 glasses from the Fonualei Spreading Center and those of Keller et al. (2008). Agreement between Tulsa and USGS for K and Ti is within 2%. The USGS values tend to be slightly lower for K₂O and higher for TiO₂, the latter reflecting the normalization of USGS values to 2392-9.

XRF analysis for major and trace elements

Aliquots (2·5 g) of all samples from 2002 (whole-rocks and glass), plus 40 other whole-rock samples for which there was not good quality glass, were analyzed for major and trace elements by XRF at the GeoAnalytical Laboratory of WSU. Original totals were 99·24 ± 0·58 wt % 2σ; LOI was not measured. The data are reported in Supplementary Data on an anhydrous basis. BIR was analyzed as an unknown; results are in Supplementary Data. There is no consistent difference between microprobe analyses of glass and XRF whole-rock analyses of the same sample, but XRF analyses frequently are higher in Al₂O₃, MgO, and CaO because of including some plagioclase and olivine crystals. In addition, the 2004 XRF K₂O contents are ∼0·05 wt % higher relative to MgO than microprobe analyses of glass in adjacent samples, even though XRF analyses of BIR agree with the accepted value. The XRF trace element data are reported only when ICP-MS data are unavailable. XRF results are within 5% of the UCSC ICP-MS values for similar samples.

H₂O and CO₂ by FTIR

Concentrations of H₂O and CO₂ dissolved in glasses were analyzed by Fourier transform infrared spectroscopy (FTIR) at the University of Tulsa using published methods and calibrations (Dixon et al., 1988, 1995) with slight modifications (Michael, 1995). Doubly polished glass wafers, 100–250 μm thick, were placed atop a 2 mm thick KBr pellet and analyzed using a NicPlan IR microscope equipped with a HgCdTe detector, attached to a Nicolet 520 FTIR system. Thickness was measured by two methods: first by digital micrometer, and second by focusing the calibrated z-axis of the FTIR microscope stage on the glass wafer and on the adjacent KBr disk using reflected light. Optically clear areas of known thickness (±1 μm), 80 μm × 80 μm, were analyzed with 256 scans per spot. Absorbance at the broad 3550 cm^–1 (combined OH and H₂O) and 1630 cm^–1 (molecular H₂O only) peaks was measured after subtraction of interpolated backgrounds. Density was assumed to be 2·8 g cm^–3. Molar absorption coefficients used for all glasses were 63 l mol^–1 cm^–1 for 3550 and 25 l mol^–1 cm^–1 for e1630. Analyses are the average of 3–4 spot determinations of 3550 cm^–1 on two separate wafers. Replicate analyses of different wafers from the same specimen were typically reproducible to ± 5%. All CO₂ is present as dissolved ${\text{CO}}_{\text{3}}^{\text{2}}$^–, which appears as an absorbance doublet at 1435 and 1515 cm^–1 (Fine & Stolper, 1986). Estimation of the complex background in this region is often the largest source of analytical error, so the spectrum of a volatile-free reference glass from Juan de Fuca Ridge (Dixon et al., 1988) was subtracted from each sample spectrum and the background of the resulting spectrum was determined using a spreadsheet designed for FTIR curve fitting of basalt glasses (S. Newman, personal communication). Concentrations are reproducible to about ±12 ppm CO₂.

Solution ICP-MS analysis for trace elements

Trace elements in all samples from the 2002 collection were analyzed by solution inductively coupled plasma optical emission spectroscopy (ICP-OES) at WSU using a Sciex Elan Model 250 instrument. The same element suite was analyzed at UCSC for 107 samples from the 2004 and later collections using an Element1 high-resolution ICP-MS system and the double-drift-correction methods of Ryder et al. (2006). The UCSC analyses used the same calibration solutions and methods throughout the study so that external error was 1–2% for concentrations >100 ppb based on replicate analyses of the same solution in different analytical sessions, and replicate digestions of the same powder in one session (Supplementary Data).

Some of the 2002 samples were analyzed at both WSU and UCSC and agreement is within the UCSC external error, except for La, which is ∼5% higher for WSU; Zr, which is 5–10% higher for UCSC; and Ta, Pb, Th, and U, all of which are 10–25% higher for UCSC. Because some trace ratios (e.g. La/Yb and Th/La) can differ by 5–25% between laboratories, only UCSC data are plotted in our figures. Isotope dilution thermal ionization mass spectrometry (ID-TIMS) Th and U concentrations from New Mexico State University (NMSU), which will be published separately, generally agree with both UCSC and WSU for U, and lie between them for Th. Nb/U ratios are 50 ± 5 for both WSU and UCSC for most samples but extend to lower values at UCSC for some whole-rock samples from which seawater U was not completely removed by leaching only in dilute HCl.

LA-ICP-MS analysis for trace elements

Supplementary Data contains trace element data for nine samples from the 2011 collection that were collected in 2012 at UCSC using the laser ablation (LA)-ICP-MS methods summarized by Dreyer et al. (2013). Because the LA data are less precise than solution data at concentrations <1 ppm, only solution data are reported in Supplementary Data for six additional samples for which both solution and LA data were obtained. In most instances, the LA results were within 5% of the solution results. In two cases, the LA-ICP-MS ratios are offset to lower values outside the error of the solution data. Consequently, we did not plot LA-ICP-MS results, except for D-MORB for which we have no solution ICP-MS results.

Sr–Nd–Hf–Pb isotopes

Sr, Nd, Pb, and Hf isotope ratios were analyzed at UCSC, Central Washington University (CWU) and NMSU over a 6 year period. Details are given below for each element. After normalization, external precision (between analytical sessions in one laboratory) for processed samples was <20 × 10^–6 for Sr, Nd, and Hf isotope ratios during this project, and <0·01 for Pb ratios. Normalized results for duplicate analyses of samples between the UCSC, CWU, and NMSU laboratories agree to within that external precision, and usually to within 0·000010. Results for USGS reference materials BCR2 and BHVO2 are given in Supplementary Data and agree with those of Weis et al. (2006) to within our external precision after normalization to common values.

All Sr measurements are by TIMS following the methods of Ryder et al. (2006) at UCSC and Ramos at CWU and NMSU. NBS 987 results at UCSC throughout the period of most of the analyses were ⁸⁷Sr/⁸⁶Sr =0·710266 ± 9 (2σ; n = 15), and results at NMSU were similar. Older results at UCSC and CWU were ∼0·71024 ± 1. All results have been normalized to 0·710270 for NBS987. Internal standard errors are ≤0·00001 2σ. Normalized results for samples that were analyzed at both UCSC and NMSU agree to within 0·000005 in most cases.

Most Nd measurements were made at UCSC by multicollector (MC)-ICP-MS (Neptune) following the methods of Tollstrup et al. (2010). La Jolla and JNdi results during the analytical period were ¹⁴³Nd/¹⁴⁴Nd = 0·511821 ± 7 and 0·512078 ± 1, respectively. A few of the measurements in Supplementary Data were made at CWU and NMSU by TIMS, for which JNdi ¹⁴³Nd/¹⁴⁴Nd = 0·512090. All results have been normalized to 0·512115 for JNdi (= La Jolla = 0·511858). Most internal errors are ≤00·000005 2σ. Normalized results for 22 samples that were analyzed at both UCSC and NMSU agree to within 0·000025, although the normalized NMSU TIMS results are systematically higher by 0·000015 on average.

All Hf measurements were made at UCSC by MC-ICP-MS (Neptune) following the methods of Tollstrup & Gill (2005). Results for JMC475 during the analytical period were ¹⁷⁶Hf/¹⁷⁷Hf JMC 475 = 0·282147 ± 5 (n = 5). Results are normalized to 0·282160 for JMC475. Most internal errors are ≤00·000005 2σ.

All Pb measurements were made at UCSC by MC-ICP-MS (Neptune) following the methods of Ryder et al. (2006) and using an NBS997 Tl-spike to correct for mass fractionation, assuming ²⁰³Tl/²⁰⁵Tl = 0·418911. Results for NBS981 during the analytical period were ²⁰⁶Pb/²⁰⁴Pb = 16·9373 ± 0·0002, ²⁰⁷Pb/²⁰⁴Pb = 15·4873 ± 0·0006, and ²⁰⁸Pb/²⁰⁴Pb = 36·6883 ± 0·0007. Results are normalized to the slightly more radiogenic triple-spiked values of Abouchami et al. (2000): ²⁰⁶Pb/²⁰⁴Pb = 16·9405, ²⁰⁷Pb/²⁰⁴Pb = 15·4963, and ²⁰⁸Pb/²⁰⁴Pb = 36·7219. Most internal errors are ≤00·002, 0·001, and <0·001 for ²⁰⁶Pb/²⁰⁴Pb, ²⁰⁷Pb/²⁰⁴Pb, and ²⁰⁸Pb/²⁰⁴Pb, respectively.

He isotopes

Vesicle ³He/⁴He and He contents were measured using a Nu Instruments noble gas mass spectrometer (Noblesse model) at Oregon State University; CO₂ concentrations in the same sample aliquots were determined by capacitance manometry. Details of the methods have been described by Graham et al. (2014). The Helium Standard of Japan (HESJ; Matsuda et al., 2002) was employed as a running standard to calibrate instrumental mass discrimination. The standard gas used at OSU has ³He/⁴He =  20·4 R_a determined through calibration to marine air collected in 2007. Over the course of this study, variation in the HESJ in the range of sample sizes was 0·30% (1σ, n = 53).

ANALYTICAL RESULTS

We made more than 7000 analytical measurements for our 322 basalts over a 10 year period. All samples have International Geo Sample Numbers (IGSN) and major element analyses, 120 have complete ICP-MS trace element analyses, 16 have only XRF trace element analyses, and 60 have isotope analyses. Our Pb and Hf isotope data are the first for the Juan de Fuca Ridge. Representative data are given in Table 1. Full results are given in Supplementary Data with footnotes providing information about sample materials and the data normalizations that allow comparability across time, laboratories, and personnel. He isotope plus vesicle He and CO₂ concentration data are given in Supplementary Data. Pb isotope data for sulfides are given in Supplementary Data. Results for standard reference materials are in Supplementary Data. All data and metadata for the samples are in the IEDA EarthChem Library doi:10.1594/IEDA/100408.

Because our goal is to compare and interpret high-precision, internally consistent results for well-located samples, all major element figures in this paper use only our microprobe data for ROV-collected samples and exclude our XRF data, data for legacy samples, and legacy data. (Our XRF and microprobe results agree well enough to assign samples to basalt types, but not enough to define subtle differentiation paths.) For trace elements we plot only solution ICP-MS data for ROV-collected samples except for one mafic D-MORB for which we have only LA-ICP-MS results. Most figures also exclude off-axis samples.

PETROLOGY AND GEOCHEMISTRY OF BASALT TYPES AT THE ICES

ICES basalt glasses are moderately fractionated olivine tholeiites ranging from ∼8 to ∼6 wt % MgO (mostly ∼7 wt %), with Mg# from 68 to 50 assuming Fe²⁺ = 0·83 × FeO_Total (Cottrell & Kelley, 2011). The median Mg# of 60 is typical of the level of differentiation at intermediate spreading rate ridges (Rubin & Sinton, 2007).

Almost all samples are aphyric, which removes uncertainties associated with crystal accumulation. About 10% have <1% isolated >0·1 mm plagioclase crystals. We found small olivine crystals in only 10 samples despite careful hand specimen inspection, and none have melt inclusions large enough to analyze (D. Wanless, personal communication) The only coarsely porphyritic samples are T463-2, a 7·4 wt % MgO E-MORB talus block from Summit Seamount, and R1079-0522, D265-R1, and D265-R5, which represent ∼8·2 wt % MgO N-MORB sheet flows from the southernmost (R1079-0522) and northernmost (D265-R1 and D265-R5) axial graben. All four samples have >10% large plagioclase crystals. The E-MORB has plagioclase with An_85–88 cores and An_71–77 rims, minor olivine crystals with Fo_85–86 cores and Fo₈₄ rims, and small Wo₄₄En₄₈Fs₈ clinopyroxene crystals with Al₂O₃ = 3 wt % and Cr₂O₃ = 0·6 wt %. The mafic minerals and plagioclase rims are in near-equilibrium with their host, but the plagioclase cores are too calcic.

Major elements: seven MORB types

We first divide the basalts into four main groups based on their K₂O/TiO₂ ratios with D-MORB ≤0·07, N-MORB 0·10–0·15, T-MORB 0·15–0·25, and E-MORB >0·25 (Fig. 3j). For brevity, we refer to these Endeavour groups as D, N, T, and E in italics throughout this paper. These compositional boundaries are similar to those of Hekinian et al. (1989) but slightly higher than those most commonly used (K₂O/TiO₂ = 0·12 < T < 0·20; e.g. Reynolds & Langmuir, 2000; Detrick et al., 2002). Segment-specific definitions of MORB types have precedent (e.g. Asimow & Langmuir, 2003, Fig. 3). Our definition separates groups with different major element, trace element, and isotopic characteristics, as discussed below. Nevertheless, 94% of basalts at the ICES would be classified as T- or E-MORB at most ridges; that is, with K₂O/TiO₂ > 0·12 and La/Sm_N > 1·0.

Fig. 3

(a–j) Major element variations. Only normalized non-legacy microprobe data for glasses are plotted (Supplementary Data). Each basalt type (D to E) is shown by a different color (see legend) as defined in (j) using K₂O/TiO₂. A summary of major element criteria for distinguishing sub-types (e.g. N1 vs N2) is given in Table 2. Comparison data for basalt glasses from Axial Seamount of the southern JdFR have been normalized to the same values for standards (Dreyer et al., 2013).

We further distinguish between two types of N (N1, N2), and three of T (T1, T2, T3) based on systematic differences in several major and trace elements. Major element data are shown in Fig. 3. Table 2 summarizes the major element characteristics of our seven basalt types at 7·5 wt % MgO based on separate linear regressions of the microprobe analyses of glasses for each type. We chose 7·5 wt % because most T1, T2, and E have 7–8 wt % MgO; most D and N have 8–9 wt %, whereas most T3 are below 7·5 wt %. Most differences in oxide–MgO slopes are statistically insignificant, and do not affect the distinction between basalt types. Differences in SiO₂, although subtle, are analytically robust because they are preserved when assigning all analytical error to SiO₂; that is, by regressing SiO₂ versus raw total in the microprobe analyses for each type and comparing their SiO₂ values at 100·0 wt % (Fig. 3i). Almost 60% of ICES basalts are E even by our restrictive criterion, about 20% are T2, and 10% T3.

Table 2

Major element oxide characteristics of each ICES basalt type at 7·5 wt % MgO, regressed from Fig. 3

	Graben Trend					Inflated Ridge Trend
	D	N1	N2	T1	T2	T3	E	(±)	Slope
Si7·5	50·5	50·4	51	49·5	50·4	50·8	50·8	0·2–0·3	(–0·23–0·4)
Ti7·5	1·36	1·78	1·47	1·98	1·72	1·65	1·58	0·05–0·10	(–0·2–0·4)
Al7·5	15	14·8	15	16	15	14·9	14·7	0·2	(+0·4–1·0)
Fe7·5	10·8	10·4	10·6	9·65	9·82	9·43	9·44	0·10–0·25	(–0·9–1·2)
Ca7·5	12	11·7	12	10·9	11·9	11·7	12	0·1–0·2	(+0·5–1·0)
Na7·5	2·39	2·79	2·4	2·91	2·7	2·89	2·61	0·1	(–0·25–0·33)
K7·5	0·04	0·2	0·19	0·49	0·35	0·35	0·45	0·01–0·03
P7·5	0·07	0·22	0·14	0·31	0·22	0·24	0·21	0·01–0·03
H7·5	n.a.	0·25	0·2	0·5	0·39	0·38	0·4	0·05
K2O/TiO2	0·05	0·10–0·15	0·10–0·15	0·25	0·16–0·25	0·20–0·23	0·25–0·33

Oxides in boldface are the most diagnostic for classification. Italics indicate the least tight regression. The slopes are the range for separate regressions for each type. ± is the average variance at 7·5% MgO for the oxide. n.a., not analyzed.

The principal major element differences between basalt types are shown in boldface in Table 2 and are as follows at comparable MgO. D is the most depleted in K₂O and P₂O₅, but not Na₂O. T at the ICES is not intermediate between N and E because all T have higher TiO₂ than N2 or E at comparable MgO, and all but T2 have higher Na₂O as well. That is, T is intermediate between N and E in K₂O but not TiO₂. T3 and E have higher SiO₂ and lower FeO at comparable MgO (i.e. higher Si7·5 and lower Fe7·5) than N1, T1, and T2. N1 and N2 parallel the difference between T3 and T2, respectively, in FeO and Na₂O, but show the opposite trend in SiO₂. T1 has the lowest SiO₂ and CaO, and the highest TiO₂, Al₂O₃, Na₂O, and P₂O₅. As at many other mid-ocean ridges, D and N are generally more mafic (usually ∼8 wt % MgO) than T or E.

The colors used to distinguish our seven basalt types (E, T1, T2, T3, N1, N2, and D) are shown in the legends of Figs 2 and 3, and are used consistently throughout. Each sample is assigned to one of these types in Supplementary Data Electronic Appendix 1, and that assignment is used when showing the sample location in Fig. 2. Assignment is ambiguous in a few cases, especially for XRF analyses.

Of the seven basalt types, only D and N2 have La/Sm_N < 1·0, and only they would be considered N-MORB by that criterion. We could, therefore, refer to N1 as T-MORB, and could define T1, T2, T3, and E as four types of E-MORB instead. However, our T-MORB types are transitional between local N- and E-MORB in K/Ti ratios, and differ from our E-MORB in several respects. All ICES basalts except D have one or more enriched trace element or isotope characteristic that we discuss using our seven categories.

There is no strict correlation between basalt chemical type and lava morphology: T and E occur as lobate to sheet flows, and as pillow lavas. However, most pillow lavas are E apart from the T3 pillow mounds on the west flank. Conversely, most D, N1, N2, T1, and T2 lavas are hackly, sheet, or lobate, suggesting that they erupted at higher rates (Chadwick et al., 2013). The overall proportion of pillow lava is low relative to the spreading rate, especially in the axial graben (Clague et al., 2014).

As shown in Fig. 4, the composition of N2 conforms most closely to what is expected on the basis of water depth (Langmuir et al., 1992; Niu & O’Hara, 2008) and spreading rate (Rubin & Sinton, 2007). In contrast, N1 and all T types deviate to higher Na8 and lower Fe8, calculated following Castillo et al. (2000). Consequently, they define a typical Pacific ‘local trend’ in which Na8 and Fe8 correlate negatively, as is also true for MORB globally (Langmuir et al., 1992; Niu & O’Hara, 2008). As elsewhere, E is slightly offset toward lower Na8 and Fe8, which has been attributed to more hydrous conditions of magma generation (Asimow & Langmuir, 2003).

Fig. 4

Total Fe as FeO (Fe8) and Na₂O (Na8) concentrations at MgO = 8 wt %, by basalt type. Data sources and symbols are as in Fig. 3. Analyses were regressed following Castillo et al. (2000). The star and error bar show the expected values for the average ICES water depth, and the grey field encloses the average values for global ridge segments (Langmuir et al., 1992, fig. 15). The N2 type is typical for the water depth and spreading rate of Endeavour. The higher Na8 and lower F8 in T types, and the offset of E samples toward the origin, are common globally.

Al₂O₃ contents and CaO/Al₂O₃ ratios correlate positively with MgO at < 8 wt % MgO in most basalt types. CaO/Al₂O₃ reaches a maximum value of ∼0·88 at 8 wt % MgO in N2, ∼0·82 in most types, and 0·68 in T1 (Fig. 3g). Correlations between all oxides and MgO are broadly consistent with fractionation of olivine + plagioclase + clinopyroxene at 1–4 kbar in the presence of ∼0·1–0·3 wt % H₂O, as discussed in detail elsewhere. This contrasts with the negative correlation between CaO/Al₂O₃ and MgO, and shallower clinopyroxene-free fractionation, at Axial Seamount (Fig. 3g;Dreyer et al., 2013).

Six E, two T2, and one T3 glasses from the ICES were analyzed for Fe³⁺/ΣFe ratios by Cottrell & Kelley (2011). All types have Fe³⁺/ΣFe = 0·168 ± 0·003 (or QFM + 0·2, where QFM is quartz–fayalite–magnetite buffer) although two E glasses are slightly more oxidized at 0·173. Consequently, differences in trace element and isotope ratios between these basalt types (see below) are not accompanied by differences in oxidation state. Moreover, their average redox state is slightly more oxidized than for global N-MORB, even though trace element enrichment is usually associated with more reduced Fe (Cottrell & Kelley, 2013).

Volatiles

Our new measurements of H₂O and CO₂ concentrations in ICES glasses are consistent with previous values (Dixon et al., 1988), but our data are combined with geological and other geochemical information about the samples. Of the volatile elements, only CO₂ is significantly degassed. Maxima of 200–300 ppm CO₂ occur in sheet flows of all basalt types in the axial graben. The minimum of ∼100 ppm is as expected for equilibrium at the axial eruption depth of ∼2200 m, and characterizes most pillow lavas.

H₂O contents at 7·5% MgO range from 0·15 wt % in N2 to 0·50 wt % in T1 (Table 2; Fig. 5a), and all are well below saturation levels for their eruption depths. Consequently, H₂O increases with decreasing MgO at the same rate as K₂O (Fig. 5a). H₂O/Ce is 175 ± 15 in all types, which is typical of Pacific MORB (Michael, 1995). However, H₂O/K₂O is lower at the ICES (∼0·8 for E and ∼1·0 for the other types) than in most other enriched MORB (e.g. Cushman et al., 2004). Consequently, basalts with the highest K/Ti are not the wettest relative to other elements.

Fig. 5

Volatile elements. All data are for concentrations in glass. Symbols as in Fig. 3. (a) H₂O. H₂O concentrations follow LREE. No results are yet available for mafic D (MgO ∼8 wt %). (Note the offset of N2 and the evolved D to lower H₂O, and T1 to higher H₂O). (b) S. The positive correlation between Fe and S in all basalt types is typical and demonstrates minimal S degassing and buffering by saturation of immiscible sulfide liquid. (c, d) Only Cl data from Tulsa obtained using long count times are shown, for which the 2σ error is ∼15 ppm. In (c), most samples define a band with negative correlation and a width of ∼100 ppm Cl. Samples within 15 ppm of the base of that band are considered pristine, and those above them are thought to have assimilated ≤100 ppm Cl from seawater-derived materials, probably in the magma chamber or during eruption. This assimilation contributes to the spread of Cl/K relative to K₂O/TiO₂ in (d) where the bold continuous line shows our proposed ratio for Endeavour basalts free from seawater assimilation. It is based on the lowest Cl/K observed (0·035) within analytical error. The fine lines show our 2σ analytical uncertainty. The concave-down curve is based on the expected incompatibility Cl > K > Ti, and observed values for depleted and ultra-depleted MORB from the Pacific (Saal et al., 2002). The highly Cl-contaminated T738-40 D sample is omitted to show the narrow range of the rest.

S contents are strongly correlated with FeO, and range from about 1100 ppm at 9 wt % FeO to almost 1700 ppm at 12·5 wt % FeO (Fig. 5b). The trend is similar to that of most MORB glasses globally and reflects the saturation of an immiscible sulfide liquid (Czemanski & Moore, 1977). The positive correlation of Cu and MgO (not shown) also reflects the segregation of sulfide during basalt crystallization.

Cl concentrations are uniformly low (50–250 ppm) whether measured using high-precision microprobe methods at the University of Tulsa or conventional microprobe methods at the USGS and UCD. However, to avoid methodological differences, we discuss and plot only the Tulsa data. Like H₂O, Cl behaves incompatibly in all Endeavour basalt types, defining a cluster of data less than 100 ppm wide at a given MgO (Fig. 5c). Only one sample lies well above this trend: T738-40, a very fractionated, slightly off-axis D. Almost all of the samples have Cl/K between 0·032 and 0·055, and show no trend with source enrichment (K₂O/TiO₂: Fig. 5d). Each group has a moderate range of Cl/K ratios, but there are some subtle differences. T1 and T3 tend to have higher Cl/K than T2 or E, and N and D extend to the highest Cl/K reflecting the greater susceptibility of more depleted glasses to seawater contamination (Michael & Cornell, 1998).

F concentrations range from 200 to 500 ppm. They are negatively correlated with MgO and positively correlated with K/Ti and La/Sm (not shown), confirming the incompatible behavior of F. With increasing K/Ti or La/Sm, F/Ce, F/Nd and F/P remain roughly constant at 15·0 ± 1·9, 22·7 ± 3·3, and 0·30 ± 0·04 respectively, consistent with ratios and trends for MORB from other regions (e.g. Schilling et al., 1980).

Trace elements

The seven types of MORB defined using major elements also differ in trace elements. Figure 6a shows REE patterns for averages of comparably mafic basalts, and Fig. 6b shows normalized La/Sm and Dy/Yb ratios for all samples analyzed at UCSC. REE from Eu to Lu are similar in D and N2 to those in global average N-MORB (Arevalo & McDonough, 2010). However, there is greater depletion in La to Sm in D, and relative enrichment of La to Nd in N2. Enrichment in La to Nd increases in all other types in the order N1 < E < T2 < T3 < T1. The heavy REE (HREE) increase in much the same order except they are lowest in E and highest in T3 such that REE patterns of both types cross the others. The T3 pattern is elevated because the average is more fractionated than that of the other types. E have lower concentrations of all REE as well as TiO₂ (Fig. 3b) than all T types at comparable MgO contents, even though their relative light REE (LREE) fractionation (e.g. La/Sm) is similar to or greater than in T1 and T2 (Fig. 6b). HREE patterns in all T and E types are parallel and have steeper negative slopes (i.e. higher Dy/Yb) than in D and N types. This contrasts with the EPR at 9–10°N where LREE fractionation between N- and E-MORB is similar to that at the ICES but there is little change in HREE fractionation (e.g. Waters et al., 2011). At the ICES, there is a six-fold variation in La/Yb in basalts <2300 years old and <300 m apart in the axial graben.

Fig. 6

Rare earth elements (REE). Only solution ICP-MS analyses from UCSC are shown except for D, for which only LA-ICP-MS data are available, which explains the more irregular shape of the HREE pattern. (a) Average of mafic samples for each basalt type normalized to Primitive Mantle (McDonough & Sun, 1995). MgO = 7·4–8·3 wt % for all types except T3, which are slightly more differentiated. The global average N-MORB (La/Sm_N < 1·0: Arevalo & McDonough, 2010), and an enriched near-axis E-MORB from 9·5°N on the EPR (2903-1: Waters et al., 2011), are shown for comparison. [Note the lower HREE (Dy–Lu) at the ICES than for the EPR E-MORB.] (b) Details of the shape of the REE pattern. La/Sm_N and Dy/Yb_N ratios are normalized to chondrites so that values of 1·00 reflect flat LREE and HREE patterns, respectively. Except for D and N2, all ICES types are LREE enriched, and are more HREE depleted than global N-MORB and EPR E-MORB [data from (a)]. The degree of HREE depletion is similar in all T and E types. E samples with Dy_N/Yb_N ratios <1·15 show no spatial or temporal pattern, so the results may be an analytical artifact. Only UCSC data are shown to avoid inter-laboratory differences. The error bar shows the 2σ variation (±2%) from duplicate digestions and calibrations for BIR and an ICES basalt.

Extended trace element patterns are shown in Fig. 7. Although D is more depleted than global average N-MORB, it has the same relative depletions, including Sr and Pb troughs. N2 is similar to global N-MORB in elements more compatible than Nd but it is more enriched in the more incompatible elements (Cs to Sr). Fresh glasses of all basalt types have typical MORB Nb/U and Ce/Pb ratios of 47 ± 5 and 28 ± 5, respectively (Hofmann, 1997). Deviations to lower Nb/U in some crystalline samples are accompanied by Th/U < 2·9 and, when measured, (²³⁴U/²³⁸U) > 1·01 (Scott, 2012), and are attributed to slight additions of seawater U. Rb and Ba are over-enriched relative to Cs–Th–U–Nb at the ICES, resulting in Rb_N > Ba_N in all basalt types and Ba_N > Th_N in the most enriched types. All types have positive Nb–Ta anomalies relative to LREE (Nb/La = 1·3–1·6), especially in E and T1. Small positive Zr–Hf anomalies also are greatest in E and T1, absent in D and N, and present even without Nb anomalies in T3. The enrichment pattern is generally in the same order as for LREE enrichment (E ∼ T1 > T2 > T3 > N1 > N2 > D). T3 is exceptional in having greater enrichment in Zr and Hf (especially Zr) than middle REE (MREE), and less enrichment in Nb, Ba, and Th. This results in T3 having higher Zr/Nb relative to La/Yb, and being more like enriched basalts elsewhere in this respect. Despite the range in REE patterns, isotope dilution Th/U ratios are relatively constant and high at 2·99 ± 0·15 in all types (Scott, 2012). Typical N-MORB Th/U (∼2·6) is found in only one off-axis N sample.

Fig. 7

Extended trace element diagram for the same mafic averages as in Fig. 6a, again normalized to Primitive Mantle (McDonough & Sun, 1995). Symbols as in Fig. 6a. T3 is hidden behind EPR-E. Rb and Ba are over-enriched relative to Cs and Th.

Trace and minor elements in E differ from those in most ridge segments in two respects. First, TiO₂ and REE concentrations in E are lower relative to MgO than in T (Figs 3d, 6a and 8a). Second, E is more enriched in K₂O than in most incompatible elements including TiO₂ (Fig. 8) and H₂O. Both are examples of T not being just intermediate between N and E. Figure 8b shows the positive correlation between K₂O/TiO₂ and La/Sm in D to T1 that is typical of global MORB, but a decoupled increase in K₂O/TiO₂ in E-MORB without commensurate change in La/Sm. This results in two separate enrichment trends of some incompatible element ratios. K₂O/TiO₂ is higher relative to ratios such as Nb/Zr and Ba/Ti in T3 and E than in N1, T2, and T1 (Fig. 8c). The same two separate trends also stand out relative to Pb and Hf isotopes (see below).

Fig. 8

K₂O/TiO₂ ratio systematics. The grey field is for EPR from Niu et al. (1999, 2002) and Waters et al. (2011), omitting samples with MgO <6 wt %. Axial Seamount data are as in Fig. 3. (a) K₂O/TiO₂ vs TiO₂. Only samples with microprobe analyses of glass are shown. It should be noted that most E have lower TiO₂ than T1, despite higher K₂O/TiO₂, in contrast to the EPR. This is also evident in Fig. 3b and shows that T is not merely a mixture of N and E. The high TiO₂ in one D reflects its low MgO. The N1 basalts with higher TiO₂ are from the southernmost Stockwork area in Fig. 2 above an offset in the AMC. (b) K₂O/TiO₂ vs La/Sm_N. Only UCSC data are shown. The highest K₂O/TiO₂ in E is not accompanied by higher La/Sm_N as at the EPR. (c) K₂O/TiO₂ vs Nb/Zr. Only UCSC data are shown. Two trends are evident, with T3 and E (the Inflated Ridge Trend) having higher K₂O/TiO₂ relative to Nb/Zr than N1–N2–T1–T2 (the Graben Trend). The former is more like data for the EPR and Axial Seamount whereas the latter is more like West Valley (Cousens et al., 1995; not shown). (d) K₂O/TiO₂ vs ¹⁴³Nd/¹⁴⁴Nd. ICES data define a negative correlation, as in other ridges, suggestive of binary mixing. It should be noted that enriched ICES basalt types have higher ¹⁴³Nd/¹⁴⁴Nd than similarly enriched basalts at the EPR. Off-axis samples are omitted for clarity. La/Sm normalization factors are from McDonough & Sun (1995).

There are also some differences in compatible transition elements between basalt types. T1 has the highest Ni and lowest Sc, and E the lowest Ni and Cr, at comparable MgO. N2 has the highest V. These differences may be related to the relative proportion of olivine to clinopyroxene in the source or during differentiation (e.g. less olivine and more pyroxene in the source of T1 than in that of E; least pyroxene for N2). All basalt types at Endeavour have less Cr and Sc relative to MgO than at Axial Seamount where shallow crystallization delayed clinopyroxene saturation (Dreyer et al., 2013).

Sr–Nd–Pb–Hf–He isotopes

Differences in Sr, Nd, and Hf isotope ratios are small (<1 εNd and <2 εHf units, and only about twice our external analytical error) despite the six-fold difference in K/Ti and La/Yb ratios (Fig. 9a and b). Differences in Pb are greater (Fig. 10). Isotopes are generally consistent within basalt types, although there are exceptions. Moreover, despite the small range, all isotopes are correlated with each other, and with incompatible trace element ratios (Fig. 8d), within the ICES as in MORB and OIB globally, suggesting binary mixing of typical sources. ICES basalts just have less variation in Sr–Nd–Hf isotopes relative to variation in trace element ratios than elsewhere, such that their isotope compositions occupy a smaller portion of the global range than in most ridge segments that contain E-MORB, such as the EPR (Fig. 8d). Indeed, all ICES basalt types have Sr–Nd–Hf isotope compositions within the range of N-MORB elsewhere, except that the Nd–Hf isotope correlation is steeper than the Terrestrial Array of Vervoort et al. (1999), deviating below it and EPR MORB at the enriched end in the direction of HIMU (Nebel et al., 2013). Sr isotope ratios in glass are also low relative to those of Nd and Hf. There are rough positive correlations between parent–daughter trace element ratios and Sr, Nd, and Hf isotopes but not Pb isotopes (not shown). In subsequent discussion, we refer to the combination of higher Sr and Pb isotope ratios, lower Nd and Hf isotope ratios, and higher ratios of more-incompatible to less-incompatible trace elements (e.g. Figs 9–11) as ‘enriched’, and the opposite as ‘depleted’.

Fig. 9

Sr–Nd–Hf–Pb isotopes. For consistency with trace element figures, only data for ROV-collected samples with ICP-MS trace element analyses are shown. (a) ⁸⁷Sr/⁸⁶Sr vs ¹⁴³Nd/¹⁴⁴Nd. All data are normalized to 0·710270 for NBS987 and 0·512115 for JNdi. Internal errors are <0·000010 2σ for both ratios. The two N1 basalts with the lowest ⁸⁷Sr/⁸⁶Sr ratios are from the southernmost Stockwork area in Fig. 2 above an offset in the AMC. Samples shown by half-filled symbols are slightly crystalline and show some chemical evidence of having assimilated seawater: excess U (Nb/U < 40) or (²³⁴U/²³⁸U) > 1·005. The rest define a small but otherwise typical inverse correlation as illustrated by the ‘Pacific Reference Line’ (PRL) for EPR basalts from Vlastélic et al. (1999). The altered N1 and N2 with high ¹⁴³Nd/¹⁴⁴Nd are from off-axis; the E with high ¹⁴³Nd/¹⁴⁴Nd may be an analytical artifact. (b) ¹⁴³Nd/¹⁴⁴Nd vs ¹⁷⁶Hf/¹⁷⁷Hf. All Hf data are normalized to 0·282160 for JMC475 with internal errors <0·000005 2σ. One of the two anomalously low ¹⁷⁶Hf/¹⁷⁷Hf ratios for E has been duplicated. The other, and the high ratio for one T2, have not. In addition to these data, there are five N with ¹⁴³Nd/¹⁴⁴Nd > 0·51317 but without Hf isotope data. The grey field is for the EPR (Salters et al., 2011; Class & Lehnert, 2012). It includes the most enriched EPR E-MORB from Waters et al. (2011), which is shown as a half-filled square, and the Terrestrial Array of Vervoort et al. (1999). In contrast, T1 and T2 lie below the EPR field in the direction of HIMU (Nebel et al., 2013). (c) ²⁰⁶Pb/²⁰⁴Pb vs ¹⁴³Nd/¹⁴⁴Nd. EPR data (grey field) as in (b) with E-MORB extending to even lower ¹⁴³Nd/¹⁴⁴Nd than shown. Axial Seamount data lie within and to the high ²⁰⁶Pb/²⁰⁴Pb side of the EPR field (Chadwick et al., 2005). Although ¹⁴³Nd/¹⁴⁴Nd data for ICES basalts are toward the depleted end for the EPR, Hf and especially Pb are more isotopically enriched at the ICES.

Fig. 10

Pb isotopes. Symbols as in Fig. 9. (a) ²⁰⁶Pb/²⁰⁴Pb vs ²⁰⁷Pb/²⁰⁴Pb and (b) ²⁰⁶Pb/²⁰⁴Pb vs ²⁰⁸Pb/²⁰⁴Pb. All data are normalized to NBS981: ²⁰⁶Pb/²⁰⁴Pb = 16·9405, ²⁰⁷Pb/²⁰⁴Pb = 15·4963, ²⁰⁸Pb/²⁰⁴Pb = 36·7219. Errors are <0·002 2σ. The N2 with the least radiogenic Pb is from off-axis. ICES samples define shallower slopes than do samples from the EPR, which are shown by the grey field (Class & Lehnert, 2012) and straddle the Northern Hemisphere Reference Line (NHRL) of Hart (1984). Data for the most enriched EPR E-MORB from Waters et al. (2011) are shown as half-filled squares.

Fig. 11

Element concentration ratios vs isotope ratios showing separate trends. (a) and (b) show ²⁰⁶Pb/²⁰⁴Pb vs K₂O/TiO₂ and La/Sm_N, respectively. T3 and E form one trend (Inflated Ridge Trend) whereas N1, T2, and T1 form another (the Graben Trend). Depending on the element ratio, N2 ± D plot with the Graben Trend or at the intersection of both trends. Simple binary mixing is precluded by the existence of two trends. The color-coded pentagon symbols in (b) show results for representative numerical models that are discussed in the text and presented in Table 4. The Inflated Ridge Trend can be explained by having a higher fraction of melt from pyroxenite relative to enriched mantle. (c) and (d) show the same element ratios vs ¹⁷⁶Hf/¹⁷⁷Hf isotope ratios. Apart from the three anomalous samples noted in Fig. 9b, the same two trends are evident. Most West Valley data (dark grey field: Cousens et al., 1995) lie along the Graben Trend whereas most EPR data (light grey field) lie along the Inflated Ridge Trend. (Note the more restricted range of ¹⁷⁶Hf/¹⁷⁷Hf relative to element ratios at Endeavour than at the EPR.) Data sources for EPR and West Valley as in Fig. 8. Only UCSC data are shown.

Pb isotopes are tightly correlated, providing the strongest evidence of binary source mixing (Fig. 10). Pb isotopes also are more enriched relative to Nd–Hf–Sr isotopes than is true of EPR E-MORB, for example (Fig. 9c). Both Pb isotope correlations are shallower than the Northern Hemisphere Reference Line (NHRL) or at the EPR, especially for ²⁰⁸Pb/²⁰⁴Pb. The latter is surprising in light of the higher Th/U at Endeavour than in global N-MORB, and requires a recent increase in Th/U in the ICES source. The shallow Pb enrichment trends point toward a young HIMU component (Nebel et al., 2013). The closest isotopic analogues in ocean island basalt (OIB) are from the eastern Galapagos and Cook–Austral Islands; the closest MORB examples are from Segment 2 of the southern MAR at 8–9°S (Hoernle et al., 2011).

Isotope correlations are binary, although T3, E, and N2 may be slightly lower in Pb isotopes relative to Sr–Nd–Hf isotopes than in the other types (Fig. 9c). For Sr–Nd–Hf, N1, N2, and T3 are the most depleted types whereas T1, T2, and E are the most enriched, with some exceptions. For Pb, the on-axis enrichment pattern is somewhat different, increasing in the order T3 < (N2, E) < N1 < (T1, T2). Once again, the T types are not intermediate between N and E. The least radiogenic Pb occurs in off-axis samples.

In detail, nine samples are offset from the negative Sr–Nd isotope correlation defined by most ICES glasses and Pacific MORB (half-filled symbols in Fig. 9a). They have 50–100 ppm higher ⁸⁷Sr/⁸⁶Sr ratios. Most of these samples are slightly crystalline instead of isotropic glass. Three of them have (²³⁴U/²³⁸U) ratios of 1·01–1·06 even after aggressive leaching (Scott, 2012), four have the highest Cl/K of their basalt type, and three have such anomalously low Nb/U ratios (<35) that we do not report their U contents. Because these slightly high Sr isotope ratios are associated with sample lithology, we attribute them to minor alteration during or after eruption rather than to mantle source variability or magma chamber processes. Omitting these suspect samples tightens the negative correlation of the Endeavour glasses.

Fifteen samples from all MORB types except D were analyzed for He isotopes (Supplementary Data). All lie within the range of N-MORB and are similar to previous data for the Endeavour area (Lupton et al., 1993). All except T3 have ratios between 8·1 and 8·3 R_A, and any subtle difference between types lies within analytical error. These values are lower than in N-MORB at the EPR 11–12°N (Hahm et al., 2009). In contrast, both T3 samples have higher ratios of 8·4–8·5 R_A. Therefore, the enriched component at Endeavour has lower, not higher, ³He/⁴He ratios, as has been recognized elsewhere (e.g. Graham et al., 2014).

Off-axis samples

We studied three depleted and five enriched basalts from east and west of the flanks of the ICES. The three depleted samples are slightly crystalline rather than isotropic glass and show geochemical evidence of slight alteration. They are N or D types with La/Sm_N < 1·0, flat HREE patterns, and depleted isotope compositions, including the least radiogenic Pb at Endeavour, but they have slightly elevated concentrations of the most incompatible elements. One sample (T465-3) is from the first constructional ridge east of the ICES, called the East Endeavour Ridge by Karsten et al. (1990). U–Th disequilibria in a basalt from the ridge gave a model age of ∼180 ka that is ∼100 kyr younger than expected from a uniform half-spreading rate of 26 mm a^–1 (Goldstein et al., 1992), and U–Th data for T465-3 are consistent with that result (Scott, 2012). The relatively young ages indicate previous off-axis magmatism, and a mid-crustal reflector, possibly indicating a melt lens, is currently present beneath the ridge (Wells et al., 2011). The other two samples come from just west of the ICES. Sample T738-40 is from an ∼40 m high pillow mound a few hundred meters west of the base of the ICES west flank. Sample T466-3 is from an elongate pillow tube ∼1 km further west. Both have less ²³⁰Th enrichment than basalts from the ICES itself, and are in ²²⁶Ra–²³⁰Th equilibrium (Scott, 2012). Therefore, they predate the surface rocks of the ICES flanks.

The five enriched off-axis samples are E that could have erupted on-axis, then flowed off. The one from the west (TT175-59-1) is from a dredge that ended on the flank and it is similar in composition to nearby flank lavas. The four on the east (T465-1, T738-1, T740-1, D266-19) are from the lightly sedimented sheet flow that fills the depression between the ICES and East Endeavour Ridge. They are somewhat differentiated (5·2–6·7 wt % MgO) and, based on bathymetry, may have erupted on-axis and flowed widely off axis (Clague et al., 2014).

Therefore, a first-order observation is that we found no evidence of significant enrichment >10⁵ years before the ICES formed, although the N-MORB were already slightly enriched in the most incompatible trace elements.

Summary of results

Although almost all ICES basalts are enriched relative to global N-MORB in elements more incompatible than Nd, the major element compositions of D and N2 are close to expectation for the local crustal thickness, water depth, and spreading rate. That is, most aspects of D and N2 are ‘normal’ in a global perspective, but they form <4% of our samples and are restricted to young flows in the axial graben, and older off-axis locations east and west of the ICES.

One important aspect of the ICES is the abundance of basalts that have enriched incompatible element concentrations, especially K₂O, but not enriched Sr–Nd–Hf isotope ratios. Although the range in incompatible trace element ratios is similar to that in many other enriched ridge segments, the isotope ratios vary less (Fig. 8c). They lie within the range of global N-MORB and show normal correlations between isotopes.

As in other ridge segments, there is apparently simple binary mixing between D and N on the one hand compared with T and E on the other. This is true of many incompatible element and isotope ratios (Figs 8–10). However, the second important aspect of the ICES is the presence of two mixing trends for some moderately incompatible element ratios relative to each other (Fig. 8c) and relative to isotopes (Figs 11 and 12). The continuous trend lines in Figs 8c and 11 are informal fits to data for samples of the designated basalt types. This separation is especially evident for Pb and Hf isotopes relative to moderately incompatible element ratios including K₂O/TiO₂, La/Sm, Nb/(Th, La, Zr), Nd/Hf, and Sr/(Sm, Y, HREE). The separation is less obvious for Sr and Nd isotopes, and does not apply to highly incompatible element ratios such as Ba/(Th, La) or Th/(U, La). The more incompatible elements are most enriched relative to isotopes in the T3 and E types (e.g. their K₂O/TiO₂ and La/Sm ratios are higher at a given ²⁰⁶Pb/²⁰⁴Pb). We call this the Inflated Ridge Trend, in which the relative order of enrichment is T3 < E. We call the other group the Graben Trend, in which the relative order of enrichment is N2 < N1 < T2 < T1, and isotopic enrichment reaches its maximum. D can belong to one trend or both. This separation into two binary mixing trends is why T is not simply intermediate between N and E at Endeavour. T1 has the highest concentrations of TiO₂, Na₂O, P₂O₅, H₂O, all REE, Nb, Zr, Th, U, Ba, Sr, and Pb relative to MgO, and higher LREE/HREE (e.g. La/Yb), MREE/HREE (e.g. Dy/Yb), HFSE/REE, Nb/Zr, Zr/Hf, and Th/Ba ratios despite lower K₂O/TiO₂ than in E (Fig. 8). All types of the Graben Trend except N2 have lower SiO₂ and higher FeO and TiO₂ at comparable MgO than the Inflated Ridge Trend.

Fig. 12

La/Yb vs Zr/Nb. Most Endeavour basalts are more enriched in Nb relative to Zr and LREE than at the EPR. This is least true for D (two additional D samples lie within the EPR field at higher Zr/Nb and lower La/Yb) and T3. The N1 and N2 that approach the EPR field are from off-axis and the southernmost Stockwork area in Fig. 2, above an offset in the AMC. The relative Nb enrichment at Endeavour also characterizes the West Valley and Explorer Segments. Data sources for EPR (light grey field) and West Valley (dark grey field) as in Fig. 8.

The third important observation is that these two trends have geological significance, hence their names. Most basalts of the Inflated Ridge Trend erupted 30^000 to 4000 years ago when the ridge was magmatically robust, its axis was several kilometers wide, lava flows were large, and lava compositions were relatively evolved and uniform (6·8 ± 0·4 wt % MgO in E-MORB). During this time, basalt compositions progressed from T3 to E. In contrast, basalts of the Graben Trend are confined to the kilometer-wide axial graben in the last 2300 years. They are more mafic, and formed smaller lava flows at higher eruption rates, especially near the western wall of the graben near the edge of the current axial magma chamber. There is no obvious age progression within the Graben Trend, although T1 may be the oldest. At ∼400 CE, N1, E, and T2 erupted in that order within one century and within 2 km of one another along-strike (ages from Clague et al., 2014).

This breakdown of simple binary mixing is not uncommon because moderately incompatible element ratios are more sensitive to melting processes whereas the range in isotope and highly incompatible element ratios reflects source heterogeneity (Rudge et al., 2013). However, to our knowledge this is the first discovery of such systematics in such a small area at a mid-ocean ridge, and with such a clear tectonic link. Some element ratios for N- and T-MORB at the Lucky Strike Segment of the Mid-Atlantic Ridge are offset from Sr and Nd isotope ratios for other nearby ridge segments (Gale et al., 2011). However, the offset is less consistent than for Endeavour, does not apply to Pb isotopes, and is not present within a single segment. Some lavas at the FAMOUS Segment are offset from binary mixing trends between moderately incompatible trace element and isotope ratios, but this has been attributed to crustal-level magma mixing (Gale et al., 2013). Both of these segments are closer to a plume and large transform than is Endeavour.

Within each trend, the isotopically more enriched end-member also has more enriched trace element ratios. Apart from the mafic D lavas that we discovered too late for full characterization, neither trend extends to compositions as depleted as global N-MORB in elements more incompatible than Nd (Fig. 7). The T3 pillow mounds on the westernmost flank have the lowest Pb and highest He isotope ratios within the ICES, and the least enrichment in Nb relative to LREE and Zr. In this sense, they originated from the most depleted source. They are the oldest basalts of the ICES, and most similar in composition to the still older off-axis basalts to the east and west. Enrichment in the subsequent E basalts is characterized by high K relative to LREE, H₂O, and HFSE. In contrast, the younger T1 and T2 types of the Graben Trend are the most enriched and most like HIMU isotopically with ²⁰⁶Pb/²⁰⁴Pb ∼19·0, flatter slopes in Pb isotope diagrams, low ¹⁷⁶Hf/¹⁷⁷Hf relative to ¹⁴³Nd/¹⁴⁴Nd, lower ³He/⁴He than in T3, and maximum enrichments in Nb and Ta.

SPATIAL AND TEMPORAL DISTRIBUTION OF BASALT TYPES

The spatial distribution of the seven ICES basalt types is shown in Fig. 2. All types occur within the axial graben where about half of our samples are E and half are T. N1, N2, T1, and T2 are found only there, whereas T3 also occurs on the lower western flank and on the axial graben wall east of Mothra. The greatest geochemical diversity is near the base of the western wall where all basalt types except T3 erupted within <1 km of one another. Because our samples were collected using ROVs, we know that they are not talus. This wall marks a fault that is directly above the western terminus of the current AMC and is the main structural control for the southern ICES hydrothermal fields (Delaney et al., 1992; Glickson et al., 2007). Apparently magmas as well as fluids rise along this fault. Because magma types are so diverse, either they bypassed the AMC, or AMCs are short-lived, or they are emptied and refilled within centuries by magma from slightly different combinations of sources.

The oldest basalts at Endeavour, a few kilometers away from the inflated ICES, are D to N with little hint of the enrichment observed in younger samples. Basalts of the Inflated Ridge Trend predate the axial graben, although E continued to erupt in it. Basalts erupted on the west flank before 11 ka were T3. All basalts erupted on both flanks from about 11 to 4 ka were E and most were more differentiated and more uniform in composition (6·8 ± 0·4 wt % MgO; Mg# = 58·3 ± 2·5) than later E in the northern axial graben. These E are the most likely to have come from a steady-state magma chamber in which recharge and cooling were balanced. In contrast, basalts of the Graben Trend are restricted to the axial graben, are recent (age <2·3 ka), and are relatively mafic (>7·5 wt % MgO). They ascended and were emplaced rapidly enough to retain up to 200 ppm excess CO₂ in hackly, sheet, and lobate lava flows.

Because our samples were collected by ROVs and are co-registered with high-resolution bathymetry, we can subdivide them even further. Basalts of the same type (e.g. T2) that seem to lie on a common liquid line of descent and are similar enough in location, morphology, depth, and age that they could have been one lava flow or at least one ‘magma batch’ are grouped into a ‘chemostratigraphic unit’ (CSU). We have defined at least 14 CSUs that will be discussed in detail elsewhere. There are 11 CSUs within the axial graben and therefore <2300 years old, so we infer at least one eruption per 200 years since the graben formed and probably more. CSUs differ enough in major elements and in trace element and isotope ratios that they represent separate mantle melting and crustal magma storage episodes. Some may represent open magma systems but most appear closed. The CSU to which each sample belongs is given in Supplementary Data.

N1 south of Mothra, in the area called Stockwork, lie above the part of the AMC that is offset to greater depth (Fig. 2). Although we have only a few samples, all basalts from Stockwork are N1 that have lower K, Rb, Ba, and Th contents relative to MgO, higher TiO₂ (Fig. 8a), less Nb enrichment (Fig. 12), lower ²⁰⁶Pb/²⁰⁴Pb, and lower ⁸⁷Sr/⁸⁶Sr relative to ¹⁴³Nd/¹⁴⁴Nd (Fig. 9a), than N1 north of the Main Endeavour Field where the AMC is continuous. This indicates a subtle difference in magma composition between fourth-order offsets of the current AMC, and that the distinctive Endeavour enrichment is largely absent from the southern area.

All basalt types are most mafic (MgO = 8·1 ± 0·2 wt %) in the shallowest part of the axial graben in the north. Therefore, the hottest (most MgO-rich) magmas and the principal melt focusing for the ICES are at ∼48·0°N, adjacent to Summit Seamount. Lavas of all types except T3 extend southward within the axial graben for 2–3 km from this shallowest point, and they generally drop to < 7 wt % MgO between the High Rise and Main Endeavour hydrothermal fields, where several of our east–west sampling transects lie. The lack of phenocrysts in the evolved lavas suggests that they are products of increased fractional crystallization within subsurface melt lenses along-strike. The preserved diversity in basalt type and in MgO along-strike suggests rapid, near-vertical ascent from mantle sources and crustal melt lenses that are heterogeneous at small scales.

REGIONAL COMPARISONS

Comparison with the northern EPR and southern JdFR

T- and E-MORB as defined here (i.e. with K₂O/TiO₂ >0·15 and La/Sm_N >1·0) occur sporadically along the EPR axis from at least the equator to 23°N, and along the southern segments of the JdFR, but are uncommon. Most examples are from off-axis seamounts, and few have been thoroughly enough analyzed for detailed comparison with ICES basalts. Indeed, the only well-analyzed on-axis basalts with K₂O/TiO₂ >0·25 (our E) were collected in a single dredge at 11·43°N (DR7-1,3: Hekinian et al., 1989; Prinzhofer et al., 1989). Another well-analyzed E-MORB was collected 4 km off-axis at 9·55°N (ALV2703-1: Waters et al., 2011). These EPR E-MORB have similar or less trace element fractionation compared with ICES T and E (e.g. La/Sm_N ∼1·5), more enriched Sr and Nd isotopes (>0·7027 and <0·5130; Fig. 8d), and less enriched ²⁰⁶Pb/²⁰⁴Pb (≤18·6) but relatively higher ²⁰⁷Pb/²⁰⁴Pb and ²⁰⁸Pb/²⁰⁴Pb (Fig. 10). One kind of T-MORB near the EPR has lower ²⁰⁶Pb/²⁰⁴Pb than N-MORB (Niu et al., 1999; Castillo et al. 2000; Waters et al., 2011), and that kind of enrichment is absent at the ICES. EPR E-MORB lie closer to the earlier Inflated Ridge Trend at the ICES than to the younger Graben Trend (Figs 8, 11 and 12). Specifically, T3 is most similar to EPR T-MORB in having greater enrichment in Zr and less in Nb relative to LREE than other ICES T-types, and the least radiogenic Sr and Pb isotope compositions. Consequently, in some respects, the Inflated Ridge Trend could originate from a two-component source like that at the EPR. However, the enriched component at the ICES differs from that of the EPR in Pb and Hf isotopes (Figs 9 and 10). EPR E-MORB also has less enrichment in Th to Ta relative to LREE, more enrichment in Ba, and a less fractionated Zr/Hf ratio (Waters et al., 2011).

The most studied example of somewhat enriched basalt associated with increased crustal thickness on the JdFR is at Axial Seamount (Chadwick et al., 2005; Carbotte et al., 2006; Dreyer et al., 2013). Consequently, data fields for Axial Seamount are shown in several figures. Although all Axial basalts are N-MORB as defined in this study, for the last 600 years they have had higher K₂O/TiO₂, been more differentiated (∼7 wt % MgO), and been aphyric compared with the preceding 300 years. They have similar Fe8 values to the ICES N2 type, confirming that type’s typicality for the JdFR spreading rate. Many differences in other major elements between Axial Seamount and the other ICES basalt types (e.g. negative slope for CaO/Al₂O₃ vs MgO, flatter slopes for TiO₂, Na₂O, and incompatible trace elements, and offset to lower FeO*; Fig. 3) reflect the shallower differentiation at Axial Seamount with less clinopyroxene fractionation (Dreyer et al., 2013). Higher Ni but lower V and Sc relative to MgO at Endeavour may reflect the same thing, or a more pyroxenitic source. Axial basalts lie between the ICES D and N1 types in level of incompatible element enrichment (more depleted than N2 in elements more incompatible than La, as enriched as N1 in Na and Sr, and more enriched in HREE than anything at Endeavour except N2). Axial basalts, especially the youngest aphyric ones, have slightly more enriched Nd–Sr isotope compositions, and higher ²⁰⁸Pb/²⁰⁴Pb relative to ²⁰⁶Pb/²⁰⁴Pb, even when measured in the same laboratory [compare Dreyer et al. (2013) and this study]. Consequently, our previous conclusions about differences between the enriched component at the ICES and the EPR probably also apply relative to the southern JdFR.

Comparison with the Heckle Seamounts

Basalts from the Heckle Seamounts (Fig. 1) have been dredged at three sites >50 km west of the ICES (Hegner & Tatsumoto, 1989; Leybourne & van Wagoner, 1991; Cousens et al., 1995). All are D- or N-MORB but only sample 7115-10-3 has enough information for direct comparison with this study (Cousens et al., 1995). This basalt is similar to N2 in several characteristics including K₂O and Sr–Nd isotopes, but it is more like D in incompatible trace elements. Its Pb isotope compositions are less radiogenic than any basalts at the ICES. Basalts from the intersection of the Heckle trend with the next older ridge west of the ICES [dredge site ENV1986-001, referred to as 8601 by van Wagoner & Leybourne (1991)] range from depleted to enriched in their K₂O/TiO₂ and La/Sm ratios. The enriched sample is most like our E type, but the attribution is uncertain without re-analysis and isotopic constraints. Therefore, although enhanced magmatism related to the Heckle Seamounts might be responsible for the thickened crust between the ICES and the seamounts (Carbotte et al., 2008), the source of basalts erupted at the ICES itself includes a more enriched component.

Comparison with the West Valley and Explorer Segments

West Valley and Explorer are the next two ridge segments north of Endeavour (Fig. 1). Analyses are available for about two dozen basalts that were dredged from the southern end of the West Valley Segment in the 1970s and 1980s, and analyzed as whole-rocks by XRF, instrumental neutron activation analysis (INAA), ICP-MS, and TIMS (van Wagoner & Leybourne, 1991; Cousens et al., 1995). The sampling sites are only about 25 km NW of the ICES. All samples from the axial graben are N-MORB as defined in this study, but highly depleted to highly enriched MORB are present on the flanks and on adjacent seamounts. The West Valley and Explorer E-MORB are most similar to the ICES T1 and T2 types in major and trace elements and Sr–Nd–Pb isotopes, although their K₂O and K₂O/TiO₂ extend to values as high as or higher than in ICES E-MORB. Indeed, basalts collected from Southwest Seamount, at the intersection of the West Valley axial graben and the trace of the Heck Seamount Chain, are more extreme examples of T1 than anything yet discovered at the ICES. In contrast, the West Valley and Explorer N-MORB are similar to N2, but with up to 9 wt % MgO and a negative correlation between CaO and MgO, reflecting temperatures higher than for clinopyroxene saturation.

The Explorer Ridge Segment is offset ∼130 km to the west of the West Valley Segment by the Sovanco Fracture Zone transform fault. Like Endeavour, it contains a wide range of basalt types, with on-axis E-MORB as defined in this study mostly along the northern 30 km of the ridge (Cousens et al., 1984; Michael et al., 1989). Jenner & O’Neill (2012) re-analyzed one of the E-MORB using LA-ICP-MS. Explorer basalts appear to be similar to those from West Valley, although again most trace element measurements are by XRF and INAA, and no single sample from the spreading center has been analyzed for major and trace elements plus isotopes. The depleted basalts at Explorer are like D and N2 at the ICES. The enriched basalts are like T1 at the ICES in terms of most major and trace elements and Pb isotopes, but with higher K₂O/TiO₂ (Fig. 11).

All basalts from the West Valley, Explorer, and Endeavour ridge segments appear to be more enriched in Nb relative to the LREE and Zr than are enriched basalts from the EPR and Axial Seamount, although there is uncertainty because of the different analytical methods employed. Consequently, although differences in Zr/Nb between N-, T-, and E-MORB are more subtle at these northern Pacific ridges, Zr/Nb ratios are lower relative to K/Ti and La/(Sm, Yb) in all of these ridges than at the EPR (Fig. 12). Similarly, Pb isotope ratios are higher relative to K/Ti and La/Sm in these ridges than at the EPR (Fig. 11), as well as being more HIMU-like. Therefore, current information suggests that the older Inflated Ridge Trend is like that of enrichments at the EPR and southern JdFR in some respects, whereas the younger Graben Trend is more like that of the West Valley and Explorer Segments. If so, then the first appearance of enrichment at the ICES accompanied a period of ridge inflation and reflects a more ‘typical’ mantle process, whereas its current manifestation is more atypical. The enriched mantle may now extend to beneath the northern ICES, or enriched magma may be entering the Endeavour Segment from the north through southward propagating intrusions such as those in 1999–2005 (Weekly et al., 2013).

DISCUSSION

Evidence that crustal assimilation is minor

As noted above, only two of the 322 samples appear to have assimilated substantial Cl (i.e. have Cl/K > 0·2). All remaining T and E have Cl/K well below the mantle limit (Cl/K = 0·07) of Michael & Cornell (1998), and their Cl/Nb ratios are 9 ± 5, which was considered uncontaminated by Le Voyer et al. (2015). Cl/K does not increase as MgO decreases, as sometimes happens at fast-spreading ridges (Michael & Cornell, 1998), arguing against widespread assimilation at Endeavour. Nevertheless, we believe that many of our samples have tens of ppm of secondary Cl. The tendency of some D and N samples to have higher Cl/K is the opposite of what would be expected for mantle-derived basalts based on the order of incompatibility: Cl > K > Ti (Fig. 5d). Instead, their higher values are consistent with the Cl/K ratios of the more depleted, lower-K basalts being more sensitive to small amounts of assimilation of seawater-influenced components. We propose that the mantle-derived Cl/K ratio of the Endeavour Segment is ≤0·035, and the mantle-derived Cl/Nb ratio is ≤7, close to the lowest values observed for E and T. Higher ratios may reflect small amounts of Cl contamination owing to assimilation. The ‘excess’ Cl, over and above the mantle contribution, ranges up to about 100 ppm, but is usually <50 ppm and much lower than in many other ridge segments.

Additional evidence for seawater assimilation in some samples comes from U concentrations and isotope ratios, and from Sr isotope ratios. Most of our samples—even those with 30–60 ppm excess Cl—have typical MORB Nb/U and Ce/Pb ratios (47 and 25, respectively; Hofmann, 1997) after leaching in dilute HCl. Exceptions are slightly crystalline basalts with Nb/U < 35 and Th/U < 2·3 for which the acid leaching was insufficient to remove the effects of secondary assimilation. Three of these samples also have (²³⁴U/²³⁸U) ratios >1·005 even after more aggressive leaching, indicative of secondary seawater U (Scott, 2012). Finally, as noted above, 12 samples have ⁸⁷Sr/⁸⁶Sr ratios slightly (50–100 ppm) higher than expected from their Nd isotopes (Fig. 9a). Most of these have a slightly crystalline groundmass, or evidence for secondary U, or both. Only one has >30 ppm excess Cl. Some of the samples with atypically high U contents evolved sulfur during HCl leaching, presumably caused by breakdown of sulfides in micro-vesicles. Whatever the cause, it is uncommon at Endeavour and we exclude these samples when considering Sr isotope and Th/U ratios, to focus attention on mantle processes.

The small amount of assimilation, especially of Cl, is surprising because the vigorous hydrothermal systems, abundant faults, active seismicity just above the AMC, and phase separation in hydrothermal fluids at Endeavour all suggest the potential for hydrothermally altered, Cl-rich roof rocks and brine above the current melt lens. Instead, the Cl contents of ICES basalts are well below those at faster spreading ridges with shallower melt lenses such as the EPR (LeRoux et al., 2006) and less than at Axial Seamount (Dreyer et al., 2013), although ICES Cl concentrations are higher than those of Siqueros melt inclusions (Saal et al., 2002). Cl/K ratios at Endeavour are also significantly lower than those at Explorer Ridge, which is spreading at a similar rate and also erupts E-MORB. The evidence of minimal assimilation is consistent with most of the magma differentiation at Endeavour occurring deeper in the ocean crust than the current melt lens (>2–3 km), or even in the mantle.

Parental or primitive compositions

Our use of 7·5 wt % MgO to discriminate between basalt types in Table 2 may hide important differences between their more primitive parents. MgO = 7·5 wt % corresponds to an Mg# of only 61 for D and N2, and 63 for the more enriched basalt types, whereas melts in equilibrium with mantle peridotite with Fo₉₀ olivine have Mg# ∼72 and MgO ≥10 wt %. There is no consensus about the best way to calculate such primitive melt compositions, or about the composition of primitive melts at mid-ocean ridges (see Kushiro, 2001; Falloon et al., 2008). As an example, Table 3 lists the average primitive melt compositions for each basalt type calculated following Niu & O’Hara (2008), who assumed multiple saturation with clinopyroxene, plagioclase, and olivine. Because the same equations were applied to all ICES basalt types, and the equations are based on generic dry N-MORB liquid lines of descent with Fe³⁺/ΣFe =  0·10 (so that more correction is needed to reach Mg# 72) and early plagioclase saturation, they probably do not strictly apply to the enriched magma types at Endeavour (T and E) or capture the potential diversity in differentiation paths. However, they allow global comparisons that illustrate two important points.

Table 3

Major element compositions of possible primitive magma compositions for each ICES basalt type

	Av. D	Av. N1	Av. N2	Av. T1	Av. T2	Av. T3	Av. E
Si72	49·07	49·33	50·11	49·02	49·66	50·12	50·08
Ti72	0·80	1·20	0·89	1·54	1·28	1·31	1·18
Al72	16·44	16·29	15·90	17·21	16·31	16·26	16·28
Fe72	8·69	8·20	8·41	7·77	7·90	7·65	7·78
Mg72	10·67	10·25	10·43	9·97	10·05	9·87	9·95
Ca72	11·73	11·18	11·59	10·18	11·14	10·90	11·12
Na72	2·21	2·48	2·16	2·91	2·53	2·78	2·44
K72	0·04	0·19	0·15	0·46	0·33	0·36	0·45
P72	0·06	0·16	0·10	0·30	0·19	0·23	0·18
Sum	99·72	99·29	99·73	99·37	99·37	99·48	99·45
CaO/Al2O3_72	0·71	0·69	0·73	0·59	0·68	0·67	0·68
K2O/TiO2_72	0·05	0·16	0·16	0·30	0·26	0·28	0·38
Po (GPa)*	1·86	1·56	1·69	1·30	1·38	1·23	1·30
To (°C)*	1392	1353	1370	1319	1329	1310	1319
Tp (°C)*	1359	1325	1339	1295	1304	1287	1296

Oxide concentrations for melts with Mg#72 (in equilibrium with Fo90 olivine) were calculated for each ICES glass analysis following Niu & O’Hara (2008). Only samples with Mg > 7·0 wt % were used in calculating the average. There is some systematic offset from values in Niu & O’Hara (2008) because we used Fe³⁺/ΣFe = 0·17 whereas they used 0·10, resulting in lower Mg# in glasses and, therefore, more correction to Mg#72. This offset is greatest for Fe72, resulting in almost 1 wt % higher values in our averages.

*Pressures and temperatures of mantle melting were calculated for these primitive compositions following Langmuir et al. (1992); these may be systematically too low because of using Fe72 instead of Fe8.

First, much of the global range of major element variations in MORB (e.g. CaO/Al₂O₃, SiO₂, Na₂O, FeO at Mg# 72) is present just within the axial graben of the ICES. Types D and N2 are typical for their water depth and spreading rate but the more enriched types are not, especially T1. The enriched types have higher Na₂O and lower FeO that are usually found in deeper or slower spreading ridge segments. This ‘local trend’ is characteristic of ridge segments affected by plumes, or it has been attributed to small-scale heterogeneity in mantle sources at the fast-spreading EPR (e.g. Langmuir et al., 1992). The range of MORB at Endeavour, with its intermediate spreading rate and currently reduced magma flux relative to other parts of the JdFR, is like the range at the faster spreading EPR. Endeavour is unusual only because its geochemical diversity occurs in such a small area on-axis and in such a short time, and because of its large trace element enrichment (e.g. higher mean La/Yb) but small isotopic variation for its spreading rate.

Second, for comparative purposes, temperatures and pressures of melting were calculated for the primitive compositions in Table 3 using equation (29) of Langmuir et al. (1992). Because these compositions apply to Mg# 72 rather than MgO = 8 wt %, the absolute values are too high, but the relative results are informative. They indicate higher pressures, temperatures, and potential temperatures for the D and N2 types than for the more enriched types, and lowest of all for the T3 and E types of the Inflated Ridge trend. The apparent 72 °C range in potential temperature at one place and time is clearly unrealistic, and it would be odd for the most alkalic basalts to result from the shallowest melting of peridotite. Nonetheless, the major element compositions of the potentially primary D and N2 types in Table 3 are similar to those of ∼10% experimental melts of depleted spinel lherzolite with Mg# 90 at low pressures and temperatures (∼1·0–1·5 GPa), and to N-MORB with ∼10 wt % MgO (e.g. Kushiro, 2001). Compositions of the more enriched types, especially T1, are shifted toward that of experimental peridotite melts formed at higher pressure and lower degrees of melting, which is opposite to the calculated temperatures and pressures in Table 3 but is common for MORB from near hotspots (Langmuir et al., 1992). They also are shifted toward the composition of melts expected from more pyroxenitic sources (e.g. Hirschmann & Stolper, 1996; Dasgupta et al., 2010). However, this conclusion depends on the uncertain differentiation paths of the relatively hydrous and evolved enriched basalts because Table 3 assumes uniform paths for all types. For all of these reasons, we conclude that the primary differences between basalt types are inherited from sources that are not homogeneous peridotite.

Origin of enriched and depleted basalts at Endeavour

Background

Six aspects of the enrichment at the ICES require explanation. First, even though there is only modest Sr–Nd–Hf isotopic variability, the range in trace element ratios amongst basalt types is too great to be explained just by varying the per cent of melting of a homogeneous source. Differences between N2 and E require a three- to four-fold decrease in the degree of melting for the most abundant type, E, and this is inconsistent with the observed typical crustal thickness. Even the differences between D and N2 in the most incompatible elements would require more than a two-fold difference in the degree of melting. Enriched mantle source(s) are required.

Second, MORB geochemical diversity at Endeavour is typical in having positive correlations between element ratios such as K₂O/TiO₂ and La/Sm (Fig. 8b), and between Sr, Nd, Hf, and Pb isotope ratios (Figs 9 and 10). However, Endeavour is unusual in having two different correlations between element and isotope ratios in which one correlation (the Inflated Ridge Trend) predominates for tens of thousands of years and over a broad region of ridge inflation (∼5 km wide), and the other correlation (the Graben Trend) is restricted to the narrow axial graben (≤1 km) during ridge deflation and tectonism in the last 2300 years.

Third, the enrichment at the northern end of the JdFR (Endeavour, West Valley, and Explorer Segments) occurs without proximity to a mantle plume. Although the Explorer Segment might be affected by the Kodiak–Bowie hotspot track, geological relations along that track are complex at best so that the enriched component may not be present near-axis (Dalrymple et al., 1987). The short Heck and Heckle seamount chains west of the West Valley and Endeavour Segments are even less likely plume candidates. Although a 130 km long transform fault separates the West Valley and Explorer Segments, enriched basalts in both segments are most common near the end farthest from the transform. Therefore, the enrichments probably reflect small-scale mantle heterogeneities. Whether two separate trends of enrichment are present in these northern ridge segments is not known, but denser sampling and more extensive analyses might discover them. Available information for these segments is most like the Graben Trend of enrichment at the ICES (Figs 11 and 12).

Fourth, the enrichment at Endeavour is spatially and temporally restricted, as foreseen by Karsten et al. (1990). Spatially, it is largely absent from the seafloor immediately east and west of the ICES and is absent from the Heckle Seamounts. Our basalts from even a few kilometers off-axis are D or N types. The oldest samples at the ICES itself, the >10 ka T3 pillow mounds on the lower west flank, have the lowest ²⁰⁶Pb/²⁰⁴Pb and Nb/Zr ratios, and the highest ³He/⁴He ratios. T3 also are found on the faulted walls of the axial graben near Mothra (i.e. they pre-date the graben), and the N1 basalts south of Mothra are more depleted isotopically than N1 basalts farther north. This spatial distribution suggests less isotopic enrichment in basalts prior to graben formation and in the south, but the southern extent of enrichment awaits future study of the North Symmetric Segment.

Temporally, the isotopically most enriched lavas (T1 and T2) are <2300 years old and were erupted within a few tens of meters of one another early in the history of the axial graben. The current magma lens in the area is about 6 km long, and mantle would have upwelled only about 500 m during the 10^000 years of most of the volcanism discussed here. At least 14 separate episodes of mantle melting, differentiation, and eruption occurred in that time to give the different CSUs. This spatial and temporal restriction requires less than kilometer-scale heterogeneity in the mantle, so that different melting episodes can tap different combinations of mantle sources. Although similar conclusions have been inferred for other ridge segments (e.g. Prinzhofer et al., 1989; Cousens, 1996), the ICES is the clearest example yet that the melting history of ordinary mid-ocean ridges varies at similar spatial and temporal scales to those at mantle plumes (e.g. Iceland: Peate et al., 2008; Sims et al., 2013; Hawaii: Marske et al., 2007).

Fifth, despite the presence of an axial magma chamber currently, evidence of these mantle heterogeneities was not erased by magma mixing in such chambers at this intermediate spreading rate segment during its graben-forming phase. Instead, lavas with MgO down to 6 wt % (not just olivine-hosted melt inclusions) preserve substantial mantle-derived differences.

Sixth, relative to their elemental enrichment (e.g. high K₂O/TiO₂), basalt types at the ICES have less enrichment in Sr–Nd–Hf isotopes, more enrichment in ²³⁰Th, ²³¹Pa, and ²⁰⁶Pb/²⁰⁴Pb, and greater HREE depletion than at the EPR. In addition, Pb isotope compositions are displaced below the NHRL, and Hf isotope compositions lie below the Hf–Nd Terrestrial Array, especially for T1 and T2. These characteristics are associated with low SiO₂, high Nb–Ta relative to Th, LREE and Zr, and lower ³He/⁴He than in T3. Consequently, the ICES enriched component is a HIMU variety (Nebel et al., 2013), along the PC4 vector in the review by Stracke (2012). It is similar to that of the A2 Segment of the southern Mid-Atlantic Ridge at 8–9°S, south of Ascension Island (Hoernle et al., 2011). It is young enough that its primary isotopic expression has high ²⁰⁶Pb/²⁰⁴Pb ratios. A geochemically similar enriched component is known from late-stage alkalic rocks on seamounts approaching the Aleutian Trench (Chadwick et al., 2014), so that the enrichment has been present in the NE Pacific west of the JdFR for at least ∼30 Myr. In summary, the enriched component at Endeavour is a young HIMU variety, regionally widespread, and small in size.

It is hard to say much about the depleted mantle component at Endeavour without more analyses of the rare D samples. Although N2 has normal major element and Sr–Nd–Hf isotope compositions in a global context, it is enriched in elements more incompatible than Nd relative to global N-MORB, has ratios of highly incompatible elements (e.g. Th/U) similar to those of the more enriched basalt types, has higher ²⁰⁶Pb/²⁰⁴Pb than in regional D-MORB, and has much lower Nd and Hf isotope ratios than anything at the Gorda Ridge to the south (Salters et al., 2011). Therefore, even the N2 source includes some enriched component.

Two alternative explanations for these aspects of enrichment at Endeavour seem viable, both of which invoke derivation of the most N-MORB-like basalts of each trend (T3 and N2) from the most depleted mantle, and the most enriched basalts (E and T1) from more enriched mantle. One alternative attributes the differences between the two trends to crustal magma mixing, and the other attributes them to mantle melting processes. Both require numerical models that are beyond the scope of this study because of the space needed to explain all necessary assumptions, and because we cannot yet explain all aspects of all basalt types in an internally consistent way. Nevertheless, some qualitative conclusions can be made as follows.

Crustal processes

We previously ruled out extensive crustal assimilation. However, the presence of shallow axial magma chambers beneath fast-spreading ridges makes it likely that mixing in those chambers will homogenize mantle-derived melts (e.g. Rubin et al., 2009), and increase the relative enrichment of more incompatible elements (O’Neill & Jenner, 2012; Coogan & O’Hara, 2015). Such a process might explain the higher ratios of more incompatible to less incompatible elements (e.g. K/Ti and La/Sm) relative to isotope ratios in the Inflated Ridge Trend (Fig. 11). Basalts of that trend are more differentiated than the same basalt type in the axial graben; E remained fairly homogeneous in composition for ∼5000 years (i.e. much longer than the time over which a melt lens could stay melt-rich without frequent replenishment), and most samples are aphyric. This explanation is consistent with the presence of plagioclase-hosted melt inclusions that range from N-MORB to E-MORB in E-MORB host-lavas that were dredged from somewhere near the ICES, but not in N-MORB host lavas (Sours-Paige et al., 1999). We will explore this alternative in a separate publication, but currently we consider it unlikely because element concentrations and ratios do not vary enough relative to MgO within the T3 and E basalts of the Inflated Ridge Trend.

Mantle sources and melting processes

The alternative explanation of the two trends attributes both the variation within each trend and the difference between them to mantle source heterogeneity and melting processes. At least one depleted and two enriched components are required. Variations within trends can be explained by different proportions of depleted and enriched sources, whereas differences between trends can be explained by different proportions of melts from enriched components.

Many attempts have been made to numerically model basalt genesis from a heterogeneous source by invoking depleted to enriched peridotite ± pyroxenite ± a low-degree melt of peridotite (e.g. Hirschman & Stolper, 1996; Spandler et al., 2008; Stracke & Bourdon, 2009; Gale et al., 2011; Waters et al., 2011). These ideas have been applied qualitatively to Endeavour before (Karsten et al., 1990; Cousens, 1996). Model results vary because of the particular suites being studied, which element concentrations and ratios and which isotope ratios are matched, the closeness of match accepted between model and observation, the specifics of the melting model (e.g. solidi, melt productivity, source mineralogy, partition coefficients, whether melts from different lithologies mix in the mantle or the enriched melt reacts with the depleted host before it melts, etc.), the compositions of the source peridotites and pyroxenite, and the physical constraints such as potential temperature (T_p) and degree of melting (F).

As one example, we present here some modeling results for Endeavour basalt types using the Ocean Basalt Simulator (OBS) code of Kimura & Kawabata (2015), to which readers are referred for model details. Our results depend on particular aspects of the OBS code including the experimentally based phase proportions during pyroxenite melting, the pMELTS-based phase proportions during peridotite melting, and especially the assumption that melts of both lithologies remain in equilibrium with their residue throughout the same range of adiabatic decompression (Kimura & Kawabata, 2015). However, the general principle of needing two enriched components with similar isotopes but different trace element concentrations seems robust.

We explored three source variables: depleted mantle peridotite (DMM of Workman & Hart, 2005); Primitive Upper Mantle peridotite (PUM of McDonough & Sun, 1995); and pyroxenite. Because it is not possible with OBS to explain any E-MORB compositions using pyroxenite that has the composition of altered ocean crust (Kimura & Kawabata, 2015), we used instead a modified HIMU-type basalt composition for the pyroxenite. Specifically, we chose a Tubuai basanite (Chauvel et al., 1992), and increased its Nb, Ta, and K concentrations to best explain the ICES enriched basalts. Isotope ratios are arbitrary. For DMM, we used ratios slightly more depleted than observed at Endeavour, and for PUM values were slightly more enriched. For simplicity, we used the same enriched isotope ratios for both PUM and pyroxenite. We used a potential temperature and total degree of melting (T_p = 1290 ± 10 °C; F = 0·10 ± 0·02) that are common to all models and consistent with ICES crustal thickness. We assumed 300 ppm H₂O in the enriched peridotite, which is appropriate for the H₂O/Ce ∼175 of Endeavour.

Many aspects of the Graben Trend can be explained by melting mostly peridotite that is 35% PUM–65%DMM for N2, 50% PUM–50%DMM for N1, 80%PUM–20%DMM for T2, and 100% PUM for T1. In all cases, melting of peridotite starts at 100 km and stops at 30–40 km depth. Little pyroxenite is needed to explain N2 or N1 basalts, but ∼4% pyroxenite is needed for T1 and T2 if pyroxenite melts constantly mix with peridotite melts, or < 0·5% if pyroxenite melts metasomatize the peridotite before it melts.

In contrast, more pyroxenite relative to PUM is needed to explain the Inflated Ridge Trend. T3 needs only ∼15% PUM and ∼3·5% pyroxenite (or < 0·5% for metasomatism) to explain its low ²⁰⁶Pb/²⁰⁴Pb and high Nb/Zr ratios, and only 20% PUM plus a similar amount of pyroxenite is needed to explain E. The pyroxenite to PUM ratio is 0·15–0·25 for T3 and E but <0·05 for the Graben Trend. Melting starts at the same depth as for the Graben Trend because T_p is the same, but melting stops deeper, at 50–60 km, resulting in more garnet in the average peridotite residue, lower per cent melting of peridotite, and a larger fraction of pyroxenite melt.

Within each trend, the positive correlation between ratios of more to less incompatible elements on the one hand, and between isotope ratios on the other, depends mostly on the ratio of PUM to DMM in the peridotite source, and secondarily on the amount of pyroxenite and the mean pressure of melting. In contrast, the higher value of more incompatible to less incompatible element ratios at any isotope ratio in the Inflated Ridge Trend reflects the larger proportion of pyroxenite relative to PUM in the source, and the higher mean pressure of melting. We attribute the distinctively high K₂O relative to TiO₂, H₂O, Rb, Ba, and Th in E ± T3, and the high Th/U of everything, to the high pyroxenite to PUM ratio in the Inflated Ridge Trend. The higher SiO₂ and lower FeO in E and T3, their relatively low Ni and Cr, and their relatively high Sc and V may reflect the same thing.

Although these specific ideas depend on the OBS model, the fact that none of the T types lie between N and E in many respects and the presence of two mixing trends separated by time suggest the need for two different enriched components with similar isotope compositions, one of which involves a melt or lithology with greater and different trace element enrichment compared with DMM. This conclusion has precedent: two different, though related, enriched components were also needed to explain varieties of enriched MORB near the Azores plume (Gale et al., 2011). Whether one or both enriched components at Endeavour are melts versus solids, or pyroxenite versus enriched peridotite in lithology, or originated at the same or different depths, depend on the model and have not yet been determined definitively for Endeavour. We refer to one enriched component as ‘pyroxenite’ only for simplicity. The unusually low H₂O/K₂O at Endeavour, especially in E, suggests the presence of phlogopite in this component.

These ideas suggest that when enriched mantle entered the melting region beneath Endeavour within the last 50^000 years, small pyroxenite regions in mostly depleted peridotite were the first to melt. This kind of pyroxenite occurs elsewhere in the NE Pacific (Cousens, 1996; Chadwick et al., 2014). The resulting basalt is the most similar to that which had characterized the Endeavour region previously and characterizes most of the JdFR to the south, specifically the T3 pillow mounds on the west flank. Subsequent melting extended to more fertile peridotite and produced the widespread E of both flanks, but still with a relatively large proportion of melt from pyroxenite (20%). The relatively large amount of pyroxenite melt may have led to ridge inflation that lasted at least 7500 years during which time <500 m of mantle upwelling occurred. Throughout this time, melting was unusually deep, contributing to the relatively steep HREE patterns for E. The unusually high Th and Pa excesses in N2 also are consistent with unusually deep melting, and ascent of melt from 60 km entirely during the Holocene (Lundstrom et al., 1995). The relatively Si-rich, Fe-poor character of the T3 and E basalts of this stage are inconsistent with deeper melting of fertile peridotite, so must reflect the more depleted nature of the peridotite host or the character of the pyroxenite. This stage was followed by the Graben Trend in which the percentage of PUM continued to increase, and melting extended to a shallower depth more typical for ocean ridges (∼30 km). The most globally ‘typical’ basalt type, N2, erupted during this time and requires ∼35% PUM and little or no pyroxenite.

Although we assumed a PUM composition for enriched mantle, the source is more likely to be metasomatized than primitive. Our model is consistent with the possibility that HIMU-type melts reacted with depleted mantle hundreds of millions of years ago, leaving pyroxenite dikes surrounded by variably enriched and metasomatized peridotite having a similar isotopic composition. Such enrichment is known from elsewhere in the NE Pacific (Chadwick et al., 2014). Melting started in the more distal dikes surrounded by the least metasomatized peridotite host, and progressed into more enriched hosts. When the dikes were exhausted, magma approached global normality in the more depleted basalt types of the Graben Trend. This may explain why early T3, with the least PUM mantle (Table 4), has the lowest ²⁰⁶Pb/²⁰⁴Pb and Nb/Zr, and the highest ³He/⁴He. A simpler though less well-constrained temporal cycle toward more depleted MORB has been inferred elsewhere (e.g. Choi et al., 2013), so might be common. From the spatial and temporal scale of basalt variability at Endeavour, we infer that the enriched areas are sub-kilometer in scale, that melting is episodic at that scale, and that the products of separate melting events can migrate to the surface in semi-isolation.

Table 4

OBS melting model results for representative ICES basalts

	Graben Trend
	N2 model	N2 target	N1 model	N2 target	T2 model	T2 target	T1 model	T1 target
XPUM	0·35	737-34	0·5	739-21	0·8	737-17	1	737-38
Xpxite	0		0·01		0·04		0·043
Tp	1280		1290		1290		1290
Pmt	1·2		1·1		1		1·3
Tmt	1274		1275		1267		1279
Total-Fmt	0·086		0·118		0·149		0·117
Tmean	1290		1297		1294		1300
Pmean	1·69		1·70		1·72		1·94
Fmean	0·079		0·110		0·136		0·105
Fmt(Per)	0·086		0·114		0·132		0·100
Ol	0·575		0·600		0·617		0·578
Gar	0·000		0·000		0·000		0·000
Opx	0·266		0·270		0·266		0·254
Cpx	0·159		0·130		0·118		0·168
Fmt(Px)	0·000		0·423		0·410		0·383
FPx/TotF	0·00		0·03		0·11		0·15
Rb	2·95	3·2	3·45	3·5	5·70	7·1	8·30	9·15
Ba	32·59	38	40	41	73	83	103	97
Th	0·40	0·4	0·52	0·45	0·96	0·91	1·34	1·21
U	0·11	0·12	0·13	0·14	0·24	0·3	0·33	0·35
Nb	4·02	6·25	6·14	7	14·20	14	18·20	18·6
Ta	0·24	0·39	0·38	0·43	0·92	0·85	1·16	1·14
K	1617	1494	1747	1328	2856	3320	3871	4067
La	4·27	4·7	5·09	5·2	9·33	9·19	12·40	12·32
Ce	11·37	11·78	12·35	13	19·90	20·5	27·00	27·66
Pb	0·46	0·41	0·49	0·48	0·77	0·68	1·08	0·84
Pr	1·88	1·84	1·94	2·01	2·85	2·83	3·83	3·8
Sr	140	112	139	146	189	196	261	224
Nd	9·30	9·35	9·51	9·91	13·44	13	17·67	16·96
Sm	3·05	3·17	2·99	3·15	3·67	3·69	4·61	4·56
Zr	73·2	84·0	72·7	92·0	92·3	118·0	119·9	153·0
Hf	2·13	2·2	2·04	2·2	2·42	2·47	3·14	3·4
Eu	1·12	1·14	1·09	1·14	1·27	1·28	1·56	1·5
Gd	3·92	4·22	3·78	4·08	4·26	4·49	5·14	5·31
Tb	0·69	0·77	0·66	0·74	0·72	0·8	0·85	0·92
Dy	4·68	5·18	4·44	4·68	4·65	4·94	5·38	5·56
Y	25·3	32·8	24·8	29·1	27·1	29·6	31·7	33·9
Ho	0·99	1·12	0·94	1·01	0·97	1·03	1·11	1·18
Er	2·85	3·25	2·71	2·86	2·77	2·87	3·12	3·28
Tm	0·43	0·48	0·41	0·41	0·41	0·41	0·46	0·47
Yb	2·83	3·06	2·68	2·63	2·71	2·63	3·03	2·96
Lu	0·43	0·46	0·41	0·39	0·40	0·38	0·45	0·44
87Sr/86Sr	0·70244	0·70247	0·70246	0·70245	0·70249	0·70248	0·70250	0·70249
143Nd/144Nd	0·51316	0·51315	0·51314	0·51315	0·51312	0·51313	0·51312	0·51314
176Hf/177Hf	0·28316	0·28317	0·28315	0·28315	0·28313	0·28313	0·28313	0·28312
206Pb/204Pb	18·58	18·61	18·73	18·74	18·87	18·86	18·91	18·90
207Pb/204Pb	15·48	15·49	15·49	15·50	15·50	15·51	15·50	15·50
208Pb/204Pb	37·93	37·93	38·05	38·06	38·17	38·16	38·20	38·19

	Inflated Ridge Trend
	T3 model	T3 target	E model	E target	Pyroxenite
XPUM	0·15	471-23	0·2	740-16
Xpxite	0·035		0·033
Tp	1280		1290
Pmt	1·4		1·7
Tmt	1281		1298
Total-Fmt	0·076		0·058
Tmean	1295		1309
Pmean	1·95		2·21
Fmean	0·067		0·051
Fmt(Per)	0·063		0·046
Ol	0·567		0·569
Gar	0·000		0·000
Opx	0·256		0·236
Cpx	0·177		0·194
Fmt(Px)	0·364		0·369
FPx/TotF	0·17		0·21
Rb	4·01	5·3	5·29	7	19·8
Ba	54	64	68	81	321
Th	0·75	0·63	0·91	0·95	4·67
U	0·20	0·19	0·24	0·32	1·13
Nb	13·30	11·5	14·50	15·5	100*
Ta	0·88	0·71	0·95	0·96	6·8*
K	2881	3112	3577	4150	13000*
La	8·95	9·29	10·35	10·19	50
Ce	19·30	21	22·88	23·5	83·8
Pb	0·69	0·82	0·89	0·77	2·7
Pr	2·89	3	3·38	3·22	10·1
Sr	187	228	225	228	514
Nd	13·53	13·81	15·50	15·5	44·3
Sm	3·76	4·04	4·20	3·95	8·47
Zr	89·7	128	102·1	137·0	202
Hf	2·49	3·11	2·86	2·95	4·7
Eu	1·30	1·4	1·43	1·38	2·7
Gd	4·35	4·8	4·74	4·69	8·1
Tb	0·74	0·823	0·80	0·83	1·25
Dy	4·88	5·18	5·18	5·31	5·85
Y	25·3	32·7	26·8	31·6	37
Ho	1·01	1·07	1·06	1·12	1·03
Er	2·89	3·1	3·01	3·13	2·74
Tm	0·44	0·453	0·45	0·45	0·32
Yb	2·83	2·94	2·92	2·85	2·03
Lu	0·44	0·467	0·45	0·41	0·29
87Sr/86Sr	0·70243	0·70243	0·70243	0·70248	0·70250
143Nd/144Nd	0·51316	0·05132	0·51316	0·51314	0·51312
176Hf/177Hf	0·28317	0·28316	0·28317	0·28315	0·28312
206Pb/204Pb	18·53	18·45	18·53	18·65	18·91
207Pb/204Pb	15·47	15·48	15·47	14·49	15·50
208Pb/204Pb	37·89	37·80	37·89	37·94	38·20

The first four rows are the independent variables: XPUM is the mass fraction of Primitive Upper Mantle in peridotite, the balance is DMM; Xpxite is the mass fraction pyroxenite in the mantle source, the balance is peridotite; Tp is potential temperature in °C; Pmt is the pressure in GPa at which melting stops. The following rows in italics are calculated parameters [see Kimura & Kawabata (2015) for detailed explanations]. Tmt is the temperature at Pmt. Total-Fmt is the total fraction of melt from both peridotite and pyroxenite at Pmt. Tmean, Pmean, and Fmean are the average temperature, pressure, and per cent melting during decompression melting. Fmt(Per) and Fmt(Px) are the degree of melting of peridotite and pyroxenite, respectively, at Pmt, and FPx/TotF is the fraction of total melt from pyroxenite. The mineral modes are the residue of peridotite melting at Pmt. The element concentrations and isotope ratios in the Model columns give the overall melt composition at Pmt. The pyroxenite composition is a Tubuai basanite (Chauvel et al., 1992) except for increased Nb, Ta, and K contents (*) to achieve a better fit to the Endeavour targets. Results of these models are plotted in Fig. 11b and d.

RELATIONSHIP BETWEEN BASALTS AND HYDROTHERMAL SYSTEMS

The preceding discussion links the past tens of thousands of years of mantle upwelling to the diversity of basalts erupted on the sea floor. However, Endeavour is most famous for the hydrothermal systems built on those basalts (Kelley et al., 2012). Studies of the hydrothermal fluids have emphasized the need for a sedimentary fluid source despite the absence of surface sediment within the axial graben (Lilley et al., 2003), thereby creating doubt about the location of recharge zones.

The involvement of a sediment source is confirmed at the adjacent Middle Valley Segment because Pb isotopes in its sulfides lie between those of the underlying basalts and sediments (Cousens et al., 2002). However, at Endeavour, the mean of conventional TIMS analyses of Pb in sulfide minerals from each hydrothermal field lies on the same trend as the basalts (Fig. 13). We interpret this collinearity to imply that the hydrothermal Pb at Endeavour, unlike Middle Valley, is derived entirely from basalt without contribution from sediment. Alternatively, if the slope of the Endeavour sulfide Pb isotope ratios is emphasized rather than the mean values, then a few per cent sediment contribution is required as noted by Yao et al. (2009) for their data. However, these slopes are similar to those of TIMS mass fractionation during analysis, and some of the TIMS results lie below our Tl-spiked basalt trend for which the only explanation is under-correction for mass fractionation. The reader is referred to Kimura et al. (2015) for a thorough evaluation of a similar example. Therefore, no sediment is required to explain the origin of Pb in Endeavour hydrothermal sulfides. The mantle-derived Pb isotopes of the basalts are passed on directly to the hydrothermal systems.

Fig. 13

Unspiked TIMS Pb isotope data for sulfide samples from hydrothermal vent fields at Endeavour in comparison with Tl-spiked ICP-MS data for Endeavour basalts (crosses) from Fig. 10. The sulfide data are from UCSC (Supplementary Data; Cook, 1994) and LeHuray et al. (1988), shown by filled circles and diamond, respectively, and from Yao et al. (2009), shown by half-filled circles. Colors indicate vent fields that are listed from north to south in the legend. The sulfide data parallel the mass fractionation vector of 1·02 per a.m.u., but the spread in ²⁰⁶Pb/²⁰⁴Pb is likely to reflect robust differences between the vent fields.

Similarly, the unusually large amount of barite in Endeavour chimneys (Reyes et al., 1995; Jamieson et al., 2013) and the distinctive Ba/Rb and Cs/Rb ratios of the fluids (Butterfield et al., 1994) can be explained by the high Ba content and Ba/Rb and Cs/Rb ratios of the enriched basalts. Therefore, the only remaining evidence for a non-basaltic fluid source is the high CH₄ and NH₃ content of the fluid (Lilley et al., 2003). Just as there is little contamination of basalt magma from brine or altered roof rocks, so also there is little or no sediment in the source of the hydrothermal metals.

We attribute the greater uniformity in ²⁰⁶Pb/²⁰⁴Pb in hydrothermal sulfides than in basalts to the homogenizing effect of fluids circulating within the basalt lavas. Because most of the sulfides have slightly higher ²⁰⁶Pb/²⁰⁴Pb (∼18·63–18·73) than most of the basalt types, the recharge zones in the axial graben (i.e. most of the crust) may be dominated by the T1 and T2 basalt types. In addition, there are differences in ²⁰⁶Pb/²⁰⁴Pb ratios between sulfides from different Endeavour hydrothermal fields (Fig. 13 and Labonté et al., 2006). Sulfides in the southern fields (MEF and Mothra) have the most radiogenic Pb and are built on T2 basalt lavas that have the most radiogenic Pb. Sulfides at High Rise, with more N nearby, have less radiogenic Pb, as do the basalts. Consequently, the geochemical diversity of basalt at the ICES might be used to constrain the hydrothermal recharge zones.

CONCLUSIONS

It has been known for a quarter of a century that the Endeavour Segment of the JdFR has abundant E-MORB (Karsten et al., 1990). We now add thorough geochemical analyses of > 300 Holocene basalts collected by ROV from the Inflated Central Endeavour Segment (ICES) that are associated with high-resolution maps and dates (Clague et al., 2014) and stratigraphic control. We divided the basalts into seven geochemical types, ranging from depleted (D) to highly enriched (E) compositions. Basalt type N2 has major element compositions typical for the local water depth and spreading rate, whereas the more enriched types differ in ways generally typical of Pacific E-MORB. Endeavour basalts define binary mixing arrays in all isotopes that are usual in their correlations, but unusual in the limited range of Sr–Nd–Hf isotope compositions for D to E yet greater range of Pb isotopes. They also define two different styles of enrichment of moderately incompatible elements. We show the following: (1) the geochemical enrichment began when the currently inflated axial ridge formed <10⁵ years ago; (2) one enrichment style (the Inflated Ridge Trend) characterizes basalts erupted across the ∼5 km wide ridge from >10^000 to ∼4000 years ago; (3) the other enrichment style (the Graben Trend) characterizes basalts erupted within the axial graben after it formed ∼2300 years ago.

As a result, we can now return to the questions posed at the outset. How do melt supply and storage interact with the ambient stress of a ridge segment to influence the geological history, morphology, and hydrothermal systems of that segment? At the ICES, magmas are most homogeneous and most enriched in incompatible elements, especially K₂O, when the ridge was most inflated. After the axial graben formed, basalts have included the complete global range of MORB types, from depleted to very enriched, all erupted at high rates. At least some of the diversity in basalt geochemistry is passed on to the overlying hydrothermal deposits with enough fidelity to perhaps define their recharge zones.

How do the periodicity of mantle melting events (‘melt supply’) and the duration of magma storage in the crust affect the chemical composition of erupted MORB? At the ICES there have been at least 14 episodes of mantle melting, crustal magma storage, and eruption within a few hundred meters of one another in the axial graben in the last 2300 years, each having different major and trace element and isotope ratios. Consequently, even though an axial melt lens is present today, melt residence within such lenses is likely to be ephemeral so that the lenses have not erased evidence of source heterogeneity at this intermediate spreading rate ridge segment.

At how fine a scale do physical discontinuities in ridges or their melt lenses correlate with differences in the composition of overlying basalts? All of the basalt types in the axial graben erupted above a currently continuous axial magma chamber, and we cannot tell which type is from the current chamber. However, all basalt types become more evolved, though still aphyric, as water depth increases along axis. The clearest offset in the AMC, south of Mothra, correlates with subtle differences in Sr and Pb isotopes and the most incompatible trace elements in one type of N-MORB. We infer, therefore, that the distinctive Endeavour enrichment is restricted to the northern ICES, and that magmas rise vertically from the mantle and crustal chambers within and between fourth-order ridge segments.

What is the nature and scale of chemical ± lithological heterogeneity of the mantle beneath ordinary ridges? We attribute the Inflated Ridge Trend to a relatively high proportion of pyroxenite (or melt derived therefrom) to enriched peridotite in the mantle source during a phase of ridge inflation that has lasted at least 6000 years. The Graben Trend reflects a reduced effect of pyroxenite after the axial graben formed. Because multiple random samplings of these components occurred within <1 km during a time when <1 km of upwelling occurred, we infer that the scale of mantle heterogeneity far from a plume is < 1 km. The enriched component at Endeavour is a HIMU-type that is regionally widespread and shared with the next two ridge segments to the north (West Valley and Explorer) but not segments to the south (e.g. Axial) or seamounts to the west (Heckle), or even the next older Endeavour ridge segment 7 km to the east.

Although the ICES may seem atypical in some respects, that may just reflect the atypicality of such dense sampling, analysis, and geological constraints. In any event, the current phase of tectonics and source enrichment at the ICES provides effective tools with which to address these fundamental questions.

ACKNOWLEDGEMENTS

This decade-long project continues the legacy of Jill Karsten, who provided initial advice, encouragement, and samples. Part of this work was the MS thesis of J.W. Deb Kelley provided a crash course in Endeavour science and samples, and encouragement. Mike Perfit was helpful and inspirational throughout. Brian Cousens and Mark Hannington shared information about Pb isotopes in sulfides. Caroline Harris and undergraduates Dominic Papia and Ryan Anderson helped with sample preparation and analysis. Emily Peterman and Jugdeep Aggarwal did the Sr isotope TIMS spectrometry; Darren Tollstrup and Frank Tepley did most of the Nd, Hf, and Pb MC-ICP-MS spectrometry; Zenon Palacz analyzed Pb in sulfides while at UCSC. Sarah Roeske and Brian Joy assisted with glass analyses at UCD. Mike Rhodes provided three powders and their XRF analyses. Jenny Paduan and Julie Martin helped with the maps. The RIDGE2000 program provided timely intellectual support while it lasted. B. Cousens, Y. Niu, and an anonymous reviewer caught some mistakes and provided helpful comments. Marjorie Wilson and Alastair Lumsden helped greatly with the submission. We thank all of these people and programs because a project of this scale and duration would have been impossible without them.

FUNDING

The Keck Foundation, a NOAA-NURP award to J.B.G., and MBARI funded the sample collecting. NSF-OCE awards 0623161 and 1043274 to J.G., P.M., F.R., and D.C. funded post-cruise work. The MBARI ship support and post-cruise analytical work were supported by a grant from the David and Lucile Packard Foundation.

SUPPLEMENTARY DATA

Supplementary Data for this paper are available at Journal of Petrology online.