Journal of Petrology Advance Access originally published online on April 15, 2005
Journal of Petrology 2005 46(9):1769-1803; doi:10.1093/petrology/egi033
© The Author 2005. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oupjournals.org
Are Arc Basalts Dry, Wet, or Both? Evidence from the Sumisu Caldera Volcano, IzuBonin Arc, Japan
Y. TAMURA1,*,
K. TANI1,
O. ISHIZUKA2,
Q. CHANG1,
H. SHUKUNO1 and
R. S. FISKE3
1 INSTITUTE FOR RESEARCH ON EARTH EVOLUTION (IFREE), JAPAN AGENCY FOR MARINEEARTH SCIENCE AND TECHNOLOGY (JAMSTEC), YOKOSUKA 237-0061, JAPAN
2 INSTITUTE OF GEOSCIENCE, GEOLOGICAL SURVEY OF JAPAN/AIST, TSUKUBA 305-8567, JAPAN
3 SMITHSONIAN INSTITUTION, WASHINGTON, DC 20560, USA
RECEIVED
APRIL 20, 2004;
ACCEPTED
MARCH 8, 2005
 |
ABSTRACT
|
|---|
Basaltbasaltic andesite (<55 wt % SiO
2) and daciterhyolite
(6674 wt % SiO
2) are the predominant eruptive products
in the Sumisu caldera volcano, IzuBonin arc, Japan. The
most magnesian basalt (8·5% MgO), as well as some of
the other basalts, has a low Zr content (2025 ppm), and
cannot yield basalts with higher Zr contents (2940 ppm)
through fractionation and/or assimilation. The high- and low-Zr
basalts have different phenocryst assemblages, olivine, plagioclase
and pyroxene phenocryst chemistries, REE (rare earth element)
patterns, and fluid-mobile element/immobile element ratios.
Estimated primary olivine compositions are more magnesian (>Fo
91)
in the low-Zr basalts compared with those in high-Zr basalts
(<Fo
89). The low-Zr basalts contain up to 11 vol. % augite,
but many high-Zr basalts are free of augite, which appears only
in their more differentiated products. The low-Zr basalts are
considered to be hydrous magmas in which olivine crystallizes
first followed by augite and plagioclase, whereas the high-Zr
basalts are dry. The low-Zr basalts have higher U/Th ratios
than the high-Zr basalts. We suggest that both dry and wet primary
basalts existed in the Sumisu magmatic system, each having different
trace element concentrations, mineral assemblages and mineral
chemistry. The lower contents of Zr and light REE and magnesian
primary olivines in the wet basalts could have resulted from
a higher degree of partial melting (

20%) of a hydrous source
mantle compared with

10% melting of a dry source mantle. The
Sr, Nd and Pb isotope compositions of the wet and dry basalts
are similar and are limited in range. These lines of evidence
indicate that a mantle diapir model might be applicable to satisfy
the configuration of such a mantle source region beneath a single
volcanic system such as Sumisu.
KEY WORDS: degree of melting; hot fingers; isotopes; mantle diapir; mantle wedge
 |
INTRODUCTION
|
|---|
Fluxing of water released from the subducted slab into the overlying
mantle wedge lowers the mantle solidus, triggering magma generation
that results in the formation of arc-front volcanoes. Thus we
expect arc basalts to be wet, and there are many examples containing
a few percent of water to support this (e.g. Sisson & Layne,
1993

; Roggensack
et al., 1996

, 1997

; Sobolev & Chaussidon,
1996

; Newman & Stolper, 2000

). For example, Sisson &
Layne (1993)

have undertaken an ion and electron microprobe
study of basalt and basaltic andesite glass inclusions in olivine
phenocrysts from Quaternary eruptions of arc volcanoes and directly
demonstrated that typical arc basalts and basaltic andesites
can have high H
2O contents, sometimes in excess of 6 wt % H
2O.
Roggensack
et al. (1997)

showed that Cerro Negro's 1992 magma
evidently retained high concentrations of volatiles (H
2O and
CO
2), up to 6·1 wt % H
2O and 1039 ppm CO
2, and erupted
explosively. On the other hand, substantial mantle upwelling
and pressure-release melting may also take place beneath arc-front
volcanoes. Sisson & Bronto (1998)

presented measurements
of the volatile content of primitive basalts from Galunggung
volcano in the Indonesian arc. Inclusions of mafic glass in
olivine phenocrysts in these basalts are characterized by uniformly
low H
2O concentrations (0·210·38 wt %)
but relatively high levels of CO
2 (up to 750 ppm), indicating
that the low H
2O concentrations are primary and not due to magma
degassing. Such nearly H
2O-free basaltic magmas, which are also
found along the Cascade arc, were considered to be derived by
pressure-release melting caused by upwelling in the sub-arc
mantle (Sisson & Bronto, 1998

). Similarly, some basalt magmas
at the front of the IzuBonin arc are either anhydrous
or nearly so (Aramaki & Fujii, 1988

; Fujii
et al., 1988

;
Nakano
et al., 1988

; Nakano & Yamamoto, 1991

). Significantly,
there is petrological evidence that plagioclase phenocrysts
accumulated in the upper parts of magma chambers beneath the
Izu-Oshima volcano in this arc. The plagioclase phenocryst content
in the Izu-Oshima lavas varies between 0 and 20% by volume and
mafic phenocrysts are rare, yet all of the lavas have similar
groundmass (melt) compositions. For plagioclase to float in
a basaltic liquid, the H
2O content must be less than 0·7%
(Aramaki & Fujii, 1988

; Fujii
et al., 1988

; Nakano
et al.,
1988

; Nakano & Yamamoto, 1991

), and so the abundance of
phenocryst plagioclase and the rarity of mafic phenocrysts suggest
dry conditions at depth.
A fundamental question remains, therefore, as to whether arc basalts are dry, wet or both. What are the relationships between fluid-flux melting and pressure-release melting beneath arc volcanoes? The example we discuss here is the Sumisu caldera volcano in the IzuBonin arc. We present evidence that there are two kinds of basalt erupted from the volcano, which have petrographic, mineralogical and geochemical differences. These systematic and interrelated differences could have resulted from different water contents in the parental magmas and differing degrees of partial melting in the source mantle. We propose that the wet basalts were derived from a more depleted mantle source than the dry basalts, and resulted from different degrees of melting,
20% and
10%, respectively. We develop a model for a mantle diapir beneath the volcano, which could produce both dry and wet basalts simultaneously in the same magmatic system. Our findings may have relevance to magma genesis models in other subduction zones.
 |
GEOLOGICAL SETTING AND SAMPLE COLLECTION
|
|---|
The northern IzuBonin arc consists of 11 Quaternary volcanoes
and eight Quaternary submarine caldera volcanoes (
Fig. 1a and b).
Sumisu (31·5°N, 140°E) (
Fig. 2a), about 500
km south of Tokyo, is one of these arc-front volcanoes characterized
by a well-formed caldera, 8 km
x 9 km in diameter with 600700
m high inner walls (Maritime Safety Agency, 1997). The 136 m
high spire of Sumisu Island stands on the southern caldera rim
(
Fig. 2b). Some parts of the caldera rim and Sumisu Knoll No.
2, just west of the caldera, have flat tops about 100200
m below sea level, possibly reflecting wave planation during
periods of Pleistocene sea-level lowering. Thus, these flat-topped
bathymetric surfaces may be older than 20 ka
BP (Maritime Safety
Agency, 1997; Iwabuchi, 1999

). On the other hand, the summit
of the conical Shirane volcano is only about 7 m below sea level,
and there is no evidence of planation. Shirane is, therefore,
the newest volcanic feature in the Sumisu area. After Hakone
volcano, Sumisu is the largest volcanic complex in the IzuBonin
arc. We sampled the Sumisu area in SeptemberOctober and
December 2002 using JAMSTEC's manned submersible
Shinkai 2000 and the ROV
Dolphin 3K. Additional surveys included SeaBeam
mapping and single-channel seismic traverses.

View larger version (33K):
[in this window]
[in a new window]
|
Fig. 1. Location of the Sumisu caldera volcano. (a) Map of the IzuBoninMariana arc system (after Taylor, 1992 ). (b) Northern part of IzuBonin arc; the Sumisu caldera volcano is one of eight such structures in this arc segment. Numbered dots indicate sites drilled on the Philippine Sea plate during ODP Legs 125 and 126.
|
|

View larger version (76K):
[in this window]
[in a new window]
|
Fig. 2. (a) Detailed map of the Sumisu caldera volcano; this complex of centres includes the main caldera edifice and smaller knolls to the west and NW. Numbered arrows are sites (D1D8) dredged in 2002. (b) Photograph of the east side of Sumisu Island, which is 270 m long and 136 m high.
|
|
Between 27° and 34°N a series of narrow grabens, a few
tens of kilometres wide, rift the IzuBonin arc adjacent
to the active volcanic front edifices. Most of the grabens lie
west of the frontal arc volcanoes. The Sumisu Rift (
Fig. 1b)
is one of these grabens and provides a present-day example of
the initial stages of back-arc basin evolution (Ikeda &
Yuasa, 1989

; Fryer
et al., 1990

; Hochstaedter
et al., 1990

).
Ocean Drilling Program (ODP) Leg 126 drilled two sites 2·4
km apart in the Sumisu Rift (Sites 790 and 791) (Taylor, 1992

).
The basement basalts are similar to the younger Sumisu Rift
basalts, which are exposed on the Sumisu Rift sea floor (Gill
et al., 1992

).
Sumisu Island
The sharp spire of Sumisu Island is
270 m long,
100 m wide and rises 136 m above sea level. The island has no flat ground and is surrounded by a narrow boulder beach, where the surf pounds incessantly. Sumisu Island stands on a flat submarine pedestal, whose 2 km x 5 km area lies about 100 m below sea level. Only one chemical analysis (major element) of dolerite has previously been reported from this island (Ossaka et al., 1985
). We collected 12 new samples along the steep eastern side of the island (Fig. 2b); these include augiteplagioclase dolerite, olivineplagioclase basalt, and plagioclase basalt.
Submersible and dredge hauls
The ROV Dolphin 3K and the manned submersible Shinkai 2000 were used to obtain samples from the post-caldera central cones, caldera walls, and caldera floor. Dive tracks and locations of samples are shown in Fig. 3. Most samples were collected from lava breccias, jointed dykes and/or sills, and nearly all contained fresh glass.

View larger version (43K):
[in this window]
[in a new window]
|
Fig. 3. Caldera map and detailed inserts showing the tracks of ROV Dolphin 3K and manned submersible Shinkai 2000 (labelled 3K and 2K, respectively). These maps also show the locations of samples listed in Table 1 (sample number = dive number + rock number). Bathymetric contours are shown in metres below sea level.
|
|
Eight dredge hauls (D1D8) were collected in the Sumisu
area in December 2002 (
Fig. 2a). Varying amounts of pumice were
collected from the flanks of the Sumisu caldera volcano, but
none was recovered from the caldera floor or walls. Thus, these
pumices pre-date the caldera or were erupted at the time caldera
formed. On the basis of the chemical composition and petrography
of the lava blocks and pumice, the samples fall into two groups.
Those produced during the Sumisu caldera-forming eruption (D1,
D2, D3, D5, D8) range from basalt, through andesite, to dacite
lavas and rhyolite pumice, whereas those from the flanks of
Sumisu Knoll No. 1 (D6) and Sumisu Knoll No. 2 (D4) and D7 are
mostly high-SiO
2 pumice. The Sumisu Knolls (No. 1, No. 2) and
other bathymetric highs west of Sumisu caldera probably postdate
the Sumisu caldera. Pumices from D4, D6, and D7 are white high-silica
rhyolite (

76% SiO
2) with phenocrysts in decreasing order of
abundance of plagioclase, quartz, orthopyroxene and opaque minerals.
In contrast, the pumices associated with the Sumisu caldera
are aphyric yellowish rhyolite (7073 wt % SiO
2).
 |
ANALYTICAL METHODS
|
|---|
After initial splitting and jaw crushing, all samples were pulverized
in an agate ball mill. Major and trace elements were determined
by X-ray fluorescence (XRF) at IFREE, JAMSTEC. Trace elements
were analysed on pressed powder discs, and major elements were
determined on fused glass discs. A mixture of

0·4 g of
each powdered sample and 4 g of anhydrous lithium tetraborate
(Li
2B
4O
7) was used; no matrix correction was applied because
of the high dilution involved. Rare earth elements (REE) were
determined by inductively coupled plasma mass spectrometry (ICP-MS)
using a VG Elemental® PQ3 instrument enhanced with a chicane
lens system, following the procedures described by Chang
et al. (2003)

. Whole-rock powder was digested with an acid mixture
of HFHClO
4HNO
3, finally dissolved in 2% HNO
3 and
spiked with
115In and
209Bi for internalization of signal in
ICP-MS measurements. The analytical precision of REE determinations,
as estimated from repeated analysis of well-established reference
standards (JB-2 and BHVO-1), is better than 2% (1

). Major and
trace element data and modal mineralogy for basaltic rocks from
the Sumisu caldera volcano are reported in
Table 1.
Sr, Nd and Pb isotope ratios were measured at the Geological
Survey of Japan/AIST on a seven-collector VG Sector 54 mass
spectrometer. About 250300 mg of hand-picked rock chips
with a grain size of 0·51 mm were leached with
6N HCl for 3045 min at 100°C prior to dissolution;
Sr and Pb were isolated using Sr resin (Eichrom Industries,
Darien, IL, USA) following the procedure described by Hoang
& Uto (2003)

. Procedural Pb blanks were <50 pg, and considered
negligible relative to the amount of sample analysed. For Nd
isotopic analysis, the REE were initially separated from major
elements and Ba by cation exchange, before isolation of Nd using
Ln resin (Eichrom Industries). Sr and Nd isotope ratios were
determined as the average of 200 ratios by measuring ion intensities
in multi-dynamic collection mode. Isotope ratios were normalized
to
86Sr/
88Sr = 0·1194 and
146Nd/
144Nd = 0·7219.
Measured values for NBS SRM-987 and JNdi-1 were
87Sr/
86Sr =
0·710276 ± 6 (2 SD,
n = 4) and
143Nd/
144Nd = 0·512104
± 12 (2 SD,
n = 4), respectively, during the measurement
period. Pb isotopic measurements were made in static mode using
four Faraday detectors. Natural (unspiked) run measurements
were made on 6575% of the collected Pb, giving beam intensities
of 0·51
x 10
11 A of
208Pb. The rest of
the collected Pb was spiked on a filament with the SouthamptonBrestLead
207204 spike (SBL74), such that the 204natural/204spike
is

0·1 (Ishizuka
et al., 2003

). The true Pb isotopic
compositions were obtained from the natural and mixture runs
by iterative calculation adopting a modified linear mass bias
correction (Johnson & Beard, 1999

). The reproducibility
of this Pb isotopic measurement (external error: 2 SD) by double
spike is <200 ppm for all
20xPb/
204Pb ratios. Internal errors
were estimated by propagating within-run 2 SE through a closed-form
linear double-spike deconvolution (Johnson & Beard, 1999

).
Measured values for NBS SRM-981 during the measurement period
were
206Pb/
204Pb = 16·9390 ± 19,
207Pb/
204Pb =
15·5013 ± 18 and
208Pb/
204Pb = 36·719 ±
4 (2 SD,
n = 8), respectively. Radiogenic isotope ratios of
representative basalts from the Sumisu caldera volcano and the
internal errors are reported in
Table 2.
Microprobe analyses were carried out on the JAMSTEC JEOL JXA-8900
Superprobe equipped with five wavelength-dispersive spectrometers
(WDS). Olivine analyses were made with a counting time of 100
s, using an accelerating voltage of 20 kV, a beam current of
25 nA and a probe diameter of 5 µm to ensure reliable
Ni values. Pyroxene and plagioclase analyses were made with
a counting time of 20 s, using an accelerating voltage of 15
kV and a beam current of 15 nA. Representative mineral compositions
in basalts and basaltic andesites from Sumisu caldera volcano
are given in
Table 3.
 |
ROCK TYPES
|
|---|
The northern part of the IzuBonin arc is characterized
by bimodal magmatism of basalt and daciterhyolite (Tamura
& Tatsumi, 2002

); the Sumisu caldera volcano is a good example
of this bimodal volcanism. All discussions in this paper refer
to analyses that have been normalized to 100% on a volatile-free
basis with total iron calculated as FeO.
Figure 4a shows the
frequency distribution of the SiO
2 content of the samples collected
by submersible and by dredging (D1, D2, D3, D5, and D8) (
Figs 2 and
3). Most samples collected west of Sumisu caldera (D4,
D6, and D7) are high-silica rhyolite (>75% SiO
2) and are
not included in this histogram.

View larger version (21K):
[in this window]
[in a new window]
|
Fig. 4. (a) Histogram of wt % SiO2 content of samples from the Sumisu area based on rocks collected by submersibles and dredging (D1, D2, D3, D5, and D8) (Figs 1 and 2). Most rocks collected west of Sumisu caldera (D4, D6, and D7) are high-silica rhyolites (>75% SiO2) and are not included in this histogram. (b) MgO vs SiO2 in volcanic rocks from Sumisu caldera volcano grouped according to magma type.
|
|
Three groups of rocks, basalts (<55 wt % SiO
2), andesites
(5661 wt % SiO
2) and dacitesrhyolites (>66
wt % SiO
2) occur in the Sumisu area, with a distinct SiO
2 gap
of

6 wt % between the andesites and the dacites (
Fig. 4b). The
relatively high MgO content of the andesites (28 wt %)
is not consistent with crystal fractionation from parental basalts,
and the SiO
2 gap between the andesites and the dacitesrhyolites
also discourages the use of crystal fractionation models to
explain these relationships. Plots of selected major and trace
elements and Y/Zr vs SiO
2 for the Sumisu caldera volcano are
shown in
Fig. 5. Data from the entire northern IzuBonin
arc (Tamura & Tatsumi 2002

) are also plotted in
Fig. 5a for comparison. The compositional trends in the Sumisu caldera
volcano are much the same as those along the entire IzuBonin
arc and emphasize the regional nature of this bimodal magmatism
(Tamura & Tatsumi, 2002

).
Here, we discuss the basalts of the Sumisu caldera volcano (data
presented in
Tables 1 and
2); the petrogenesis of the associated
andesites, dacites, and rhyolites will be discussed elsewhere.
Two types of basalts
Two types of basalt exist at the Sumisu caldera volcano, which have different Zr contents; these are defined here as low-Zr basalt and high-Zr basalt, respectively. These exhibit characteristic differences in phenocryst assemblage and phenocryst content, mineral chemistry, bulk-rock trace element contents and REE patterns.
Low-Zr and high-Zr basalts
Figure 6 shows the interpreted boundary between low-Zr basalts and high Zr basalts. The low-Zr basalts are represented by only four samples; these include samples with the lowest SiO2 (D3-R16) and the highest MgO contents (575R4). D3-R16 was dredged from the sea floor just west of Sumisu caldera (Fig. 2). Samples 575R4, 575R5 and 1391R4 are from the SE caldera wall, collected by ROV Dolphin 3K and Shinkai 2000, respectively (Fig. 3). These rocks probably came from the jointed dykes and sills that intrude (and thus post-date) the pre-caldera edifice. In contrast, all basalts from Sumisu Island are high-Zr basalts, together with rocks from the SE slope of the Sumisu caldera volcano (D1 and D2) (Fig. 2). ROV Dolphin 3K also collected high-Zr basalts from the same dive traversing the southeastern caldera wall (3K575); these samples are thought to be from brecciated lava flows cropping out in the caldera wall and thus represent samples of the pre-caldera edifice (Fig. 3). The available evidence therefore suggests the high-Zr basalts pre-date the low-Zr basalts.
 |
PETROGRAPHY AND MINERAL CHEMISTRY
|
|---|
Mineral assemblages
Low-Zr and less-evolved high-Zr basalts have different phenocryst
assemblages and phenocryst contents.
Figure 6 shows the mode
of phenocrysts, Y/Zr and U/Th ratios of low-Zr basalts and representative
high-Zr basalts. All the low-Zr basalts contain olivine + clinopyroxene
+ plagioclase, and one of these (575R4) contains more than 10
vol. % of clinopyroxene. Many high-Zr basalts, on the other
hand, contain only olivine + plagioclase. Augite appears only
in the more evolved lavas of the high-Zr basalt type. Evolved
high-Zr basalts contain clinopyroxene ± orthopyroxene
up to 1 vol. % (
Table 1), but these phenocryst contents are
dramatically less in volume compared with the low-Zr basalts.
The total volume of phenocrysts is also greater in the low-Zr
basalts (2338 vol. %) than most of the high-Zr basalts
(231 vol. %) (
Table 1). Because the low-Zr basalts are
more magnesian than the high-Zr basalts, the large percentages
of clinopyroxene in the more mafic rocks (low-Zr basalts) and
the virtual absence of clinopyroxene in the less mafic rocks
(high-Zr basalts) are not consistent with fractionation models.

View larger version (19K):
[in this window]
[in a new window]
|
Fig. 6. Zr ppm vs SiO2 (wt %) for basaltic rocks of the Sumisu caldera volcano. Low-Zr and high-Zr basalts have distinct phenocryst assemblages and modal proportions, and Y/Zr and U/Th ratios.
|
|
Representative mineral compositions in low-Zr and high-Zr basalts
and basaltic andesites are presented in
Table 3 and shown in
Figs 7
10.

View larger version (47K):
[in this window]
[in a new window]
|
Fig. 8. Frequency histograms of An content [100 x Ca/(Ca + Na)] of plagioclase phenocrysts in the low-Zr basalts, high-Zr basalts and basaltic andesites. Cores and rims are shown by dark and light bars, respectively.
|
|

View larger version (36K):
[in this window]
[in a new window]
|
Fig. 9. (a) Mg-values [100 x Mg/(Mg + Fe)] of phenocryst cores of coexisting augite and olivine in low-Zr and evolved high-Zr basalts. Olivine and augite in the low-Zr basalts have much wider compositional ranges than those in the more evolved high-Zr basalts. Average Mg-value indicated by a bar in each rectangle. (b) Magmatic temperatures estimated using the olivineaugite MgFe-exchange geothermometer of Loucks (1996) . Three temperatures are estimated for each rock using pairs of average, maximum, and minimum Mg-values of olivine and augite cores.
|
|

View larger version (16K):
[in this window]
[in a new window]
|
Fig. 10. Pyroxene quadrilateral showing the temperature contours for equilibration of coexisting opx and cpx in high-Zr andesitic rocks of Sumisu caldera volcano, following the method of Lindsley & Andersen (1983) .
|
|
Olivine
Olivine phenocrysts occur in both the low-Zr basalts and high-Zr
basalts. Plots of NiO vs Fo [100 Mg/(Mg + Fe)] for olivines
from eight representative samples of Sumisu basalts are shown
in
Fig. 7a and b. There are interesting differences between
the olivines in the low-Zr, clinopyroxene-bearing basalts and
those in the high-Zr, clinopyroxene-free basalts. Fo contents
in olivines in the high-Zr basalts are commonly around Fo
7579 (complete range Fo
7079;
Fig. 7a), whereas those in the
low Zr basalts vary widely from Fo
71 to Fo
91 (
Fig. 7b). Some
olivines from the low-Zr basalts overlap the distribution of
olivines from the high-Zr basalts. Olivines in the low-Zr basalts
have lower wt % NiO and/or higher Fo contents than those in
the high-Zr basalts at given Fo and at given NiO contents (
Fig. 7a and b).
We can consider the relationships between the host
basalts and phenocryst olivine following the approach of Tamura
et al. (2000)

.
Measured vs calculated olivine compositions
A series of equilibrium olivine and basalt compositions were calculated from an original basalt composition using the olivine fractionation model of Tamura et al. (2000)
, using estimated KD(Fe/Mg)ol/liq values in the range 0·300·31, and DNiol/liq from Kinzler et al. (1990)
. Equilibrium olivines in the low-Zr basalts have higher Fo contents than those in the high-Zr basalts at a given NiO content. These calculated olivines are similar to the actual olivine phenocrysts and, generally, the calculated olivines in equilibrium with both low-Zr basalts and high-Zr basalts also show similar compositional contrasts to the actual olivine phenocrysts; the former are more magnesian at a given NiO content than the latter. The relationships between actual olivine phenocrysts and calculated olivines are, however, different between the low-Zr and high-Zr basalts.
Figure 7 shows plots of NiO vs Fo for olivine phenocrysts, calculated equilibrium olivines and olivine fractionation trends for four high-Zr basalts (SI10, SI14, 575R7, and D2-R4; Fig. 7c) and four low-Zr basalts (1394R4, 575R4, 575R5 and D3-R16; Fig. 7d). High-Zr basalts contain olivine phenocrysts whose compositions overlap those of the calculated equilibrium olivines, and continuously extend towards more iron-rich compositions (Fig. 7c). On the other hand, the low-Zr basalts contain many olivine phenocrysts that are more iron-rich and/or more Ni-rich than the calculated equilibrium olivines, and also a few olivines that are more magnesian (
Fo90) and lower in NiO (
0·15 wt %) than the calculated olivines. In other words, the distribution of observed olivine compositions in the low-Zr basalts cuts across the calculated olivine fractionation trends (Fig. 7d).
Calculated primary olivines
Sato (1977)
pointed out that the first olivines crystallizing from primary magmas should have NiO contents of
0·4 wt %, because the olivines in upper-mantle lherzolites have uniform NiO contents of
0·4 wt %. Current understanding, however, suggests that peridotite olivines do not always have a regionally constant NiO content of 0·4 wt %. Abyssal peridotite olivines, for example, fall in the range of 0·220·4 wt % NiO (Dick, 1989
). It is, however, reasonable to assume that basalts from a single volcano, such as Sumisu, were derived from local mantle sources, which have similar and constant NiO contents in their constituent olivine. Thus we assume that olivine in the source mantle of the Sumisu basalts has a uniform NiO content of
0·4 wt %. On this basis a series of equilibrium olivine compositions were calculated from selected low-Zr and high-Zr basalts following the method of Tamura et al. (2000)
until the calculated equilibrium olivines had NiO contents of
0·4 wt %. We assume that the forsterite contents of these olivines are equal to that of olivines in the primary magmas and their source-mantle peridotites. Although these calculations ignore the effect of clinopyroxene fractionation, the effect of clinopyroxene fractionation in the low-Zr olcpx-bearing basalts is to increase the difference between the primary low-Zr and high-Zr basalts. Thus the differences between the primary low-Zr and high-Zr magmas based on olivine fractionation models represent minimum values of the actual differences between the respective mantle source peridotites.
The primary olivine compositions calculated for the low-Zr olcpx basalts (575R4, 575R5, 1391R4 and D3-R16) range from Fo88·7 to Fo91·9. In contrast, those calculated for the high-Zr ol basalts (SI10, SI14, 575R7 and D2-R4) range from Fo86·0 to Fo89·0. This suggests that the source mantle of the low-Zr basalts is more magnesian, and thus more depleted, than that of the high-Zr basalts.
Plagioclase
Plagioclase is the dominant phenocryst phase in both high-Zr and low-Zr basalts and basaltic andesites, ranging in volume from 14 to 29%, except for one aphyric andesite (D2-R2). Figure 8 shows frequency histograms of An content [100 x Ca/(Ca + Na)] in the low-Zr basalts and high-Zr basalts and basaltic andesites. Less evolved high-Zr basalts and low-Zr basalts have similar distribution patterns, which peak between An85 and An90. Most plagioclase rim compositions in these rocks extend to lower An contents than the core compositions (Fig. 8), indicating the weak zoning patterns in both basalt types. Plagioclases with An>95, however, appear only in the low Zr basalts (especially in D3-R16 and 575R4); those with An>90 are rare in the high-Zr basalts compared with those in low Zr basalts.
Evolved high-Zr rocks (D1-R9, 1018R2, 575R1 and 575R2) exhibit distinct normal zoning of their plagioclase phenocrysts (Fig. 8); cores have An contents ranging from An80 to An90, whereas the rims have lower An contents ranging from An70 to An80.
Augite
As shown in Fig. 6 and Table 1, augite phenocrysts appear at an earlier, more primitive stage in the low-Zr basalts and at a more evolved stage in the high-Zr basalts. Figure 9a shows the XMg [100 Mg/(Mg + Fe)] values of phenocryst cores of coexisting augite and olivine in both low-Zr and high-Zr basalts. Generally, augites have higher maximum XMg values than coexisting olivines, but the range of augite XMg values is greater than that of the olivines. Olivines and augites in the low-Zr basalts have much wider ranges in composition than those in the more evolved high-Zr basalts.
Olivineaugite geothermometry
Figure 9b shows magmatic temperatures estimated using the olivineaugite MgFe-exchange geothermometer of Loucks (1996)
. To use this geothermometer, we need an equilibrium pair of compositions of olivine and augite. Temperatures for each sample were estimated using pairs of average, maximum, and minimum Mg-values of olivine and augite cores. The high-Zr basalt samples yielded consistent temperatures of 11501200°C. In contrast, the three pairs of olivineaugite compositions from the low-Zr basalts show varying, inconsistent, temperatures in each sample. For example, the temperature estimated using the maximum Mg-values of olivine and augite cores in 575R4 is
1300°C, but the temperature estimated by using the average Mg-values is below 1100°C. Moreover, temperatures estimated by using minimum Mg-values (
1200°C) fall near the midpoint between the other two estimates. We therefore conclude that this geothermometer does not perform satisfactorily in the low-Zr basalts.
Two-pyroxene thermometry
Coexisting augitehypersthene phenocryst pairs are rare in the evolved high-Zr basalts, limiting the use of the two-pyroxene thermometer of Lindsley & Andersen (1983)
(Fig. 10). Pyroxenes in the evolved basalt 1018R2 indicate magmatic temperatures of 11001200°C, whereas those in andesite 575R1 clearly indicate a lower temperature of
1100°C. As olivines and augites in the same sample of 1018R2 indicate a temperature of
1200°C, the two-pyroxene thermometry is consistent with that estimated by the olivineaugite thermometer (Fig. 9).
Based on the temperatures indicated by the olivineaugite and two-pyroxene thermometers, we conclude that the high-Zr basalts and andesites had temperatures ranging from 1200°C to 1100°C, and that the andesites had slightly lower temperatures.
 |
FRACTIONATION MODELS
|
|---|
To understand the role of fractional crystallization in producing
the range of magma compositions observed at Sumisu, many parentdaughter
pairs were modelled using least-squares mass-balance calculations
and the observed phenocryst phases. Two low-Zr basalts, 575R4
(8·5 wt % MgO) and 575R5 (6·2 wt % MgO), were
chosen as parental magmas and the feasibility of deriving high-Zr
basalts from low-Zr parents was evaluated. In some cases, the
sum of the squares of residuals (
r2) becomes much lower if the
parent basalts were to assimilate small amounts of rhyolite,
together with fractional crystallization (AFC). For example,
removal of phenocryst olivine, augite and plagioclase, and the
addition of rhyolite can account for the major-element differences
between the low-Zr basalts (575R4 and 575R5) and basaltic andesite
(575R1).
Table 4 shows representative results of these calculations.
The low-Zr basalt 575R5 (6·2 wt % MgO) can be derived
from the low-Zr basalt 575R4 (8·5 wt % MgO) through crystal
fractionation. Generally, fractionation of phenocryst phases
from these low-Zr basalts can explain the major element differences
between the low-Zr basalts and many high-Zr basalts (
Table 4,
Fig. 11a). Zr and other trace element enrichments are, however,
not compatible with closed-system fractionation from the low-Zr
basalts (
Table 4 and
Fig. 11b). Although the fractionation models
need only 1030% removal of phenocrysts from the magnesian
low-Zr basalts to produce high-Zr basalts with lower MgO (36
wt % MgO), many high-Zr basalts have 1·52 times
more Zr than the low-Zr basalts over a small range of SiO
2 contents
(5153 wt %). The rhyolites in the Sumisu caldera contain
up to 127 ppm Zr (
Fig. 5b), which, however, could not produce
the steep trend in the ZrSiO
2 diagram if they had been
assimilated by mixing with the low-Zr basalts (
Table 4). In
conclusion, the high-Zr basalts in the Sumisu caldera volcano
cannot be produced from the low-Zr basalts by fractional crystallization
and/or assimilation (
Table 4). This is also consistent with
petrographic observations that the low-Zr basalts contain up
to 11 vol. % augite, but many of the high-Zr basalts are free
of augite (
Fig. 6).

View larger version (18K):
[in this window]
[in a new window]
|
Fig. 11. (a) MgO vs SiO2 (wt %) and (b) Zr (ppm) vs SiO2 (wt %) for basaltic rocks from Sumisu caldera volcano. Numbers refer to samples listed in Tables 1 and 4. (a) Removal of phenocrysts of olivine, augite and plagioclase (± a few percent addition of rhyolite) can account for the major-element variations between the low-Zr basalts and high-Zr basalts (Table 4). (b) Zr and other trace element enrichments are, however, not compatible with closed-system fractionation and/or fractionation coupled with assimilation from a low-Zr basalt parent.
|
|
 |
RARE EARTH ELEMENT (REE) PATTERNS
|
|---|
Figure 12 shows a C1 chondrite-normalized REE plot for the Sumisu
basalts. All the basalts are strongly depleted in the more incompatible
light rare earth elements (LREE) compared with middle and heavy
REE (MREE and HREE). Such a pattern of upward-convex curvature
is characteristic of mid-ocean ridge basalt (MORB) worldwide
and reflects the broad pattern of trace element concentrations
in the mantle source region. For example, Eocene MORB from the
West Philippine Basin (Pearce
et al., 1999

), also on the Philippine
Sea Plate, has a similar REE pattern to the Sumisu basalts.
Importantly, the low-Zr basalts are more LREE depleted than
the high-Zr basalts. Fractionation of augite from low-Zr basalts
would increase the abundances of LREE compared with MREE and
HREE. Thus it might be considered that the differing REE patterns
between the low- and high-Zr basalts could have resulted from
crystal fractionation of augite from the low-Zr basalts. This
scenario, however, is not consistent with petrographic observations
that these high-Zr basalts are free of augite phenocrysts; additionally,
mass-balance calculations are not consistent with crystal fractionation
from a low-Zr basalt parent to a high-Zr basalt. Thus, the more
LREE-depleted patterns in the Sumisu basalts may suggest higher
degrees of partial melting of the source mantle. Thus basalts
produced by different degrees of partial melting of the source
mantle may coexist in the same magmatic system. It is therefore
pertinent to determine whether similar REE patterns exist in
basalts from other Bonin arc volcanoes.
REE patterns of basalts from the IzuBonin arc
Taylor & Nesbitt (1988) presented extensive data for other
IzuBonin arc volcanoes, which include the five Quaternary
arc-front volcanoes of Oshima, Miyakejima, Hachijojima, Aogashima
and Torishima.
Figure 13 shows La/SmSm/Yb and La/SmY/Zr
plots for basalts from Sumisu caldera volcano (this study) and
the five other Quaternary volcanoes from the dataset of Taylor
& Nesbitt (1988). La/Sm is positively correlated with Sm/Yb
and is negatively correlated with Y/Zr, especially in the Sumisu,
Torishima and Hachijojima areas. La/Sm values at Sumisu range
from 0·7 to 1·05; the low-Zr basalts have lower
ratios ranging from 0·7 to 0·9. La/Sm values at
Hachijojima and Torishima range from 0·59 to 0·79
and from 0·57 to 0·88, respectively. As shown
by Taylor & Nesbitt (1988), variation in Nd isotopes along
the length of the frontal IzuBonin arc indicates that
heterogeneity existed within the mantle wedge prior to enrichment
by subduction fluids. Thus, comparison of La/Sm ratios among
the different frontal volcanoes may reflect the heterogeneity
in the mantle wedge and not different degrees of partial melting
of the source mantle. It is, however, important to note that
a range of La/Sm values exists within individual volcanoes.
Basalts from Hachijojima and Torishima have ranges of 0·2
and 0·31, respectively, similar to that of Sumisu (0·35).
This may suggest that the basalts of these individual volcanoes
might have resulted from differing degrees of partial melting
of the mantle source along the IzuBonin arc.

View larger version (25K):
[in this window]
[in a new window]
|
Fig. 13. (a) La/SmSm/Yb and (b) La/SmY/Zr plots for basalts from Sumisu caldera volcano (this study) and the five other Quaternary IzuBonin arc volcanoes (Taylor & Nesbitt, 1988). A range of La/Sm values exists within individual volcanoes, suggesting that the basalts of these volcanoes might have been the products of differing degrees of partial melting of the mantle source. Numbers refer to samples in Table 5.
|
|
Figure 14 shows a La/SmSm/Yb plot for the IzuBonin
arc basalt field based on the data from this study and Taylor
& Nesbitt (1998)

, the NE Japan arc basalt field after Shibata
& Nakamura (1997)

, SW Japan Daisen basalts from Tamura
et al. (2000)

, and the average enriched and normal MORB (E-MORB
and N-MORB) and ocean island basalt (OIB) compositions of Sun
& McDonough (1989)

. IzuBonin arc basalts have similar
(or more depleted) REE patterns than average N-MORB.
 |
DEGREE OF MELTING
|
|---|
Differences in olivine chemistry, incompatible trace element
concentrations, and REE patterns cannot be explained by crystal
fractionation models from the same primary magma. On the other
hand, the combination of depleted olivine chemistry, low content
of incompatible elements, and LREE-depleted pattern of the low-Zr
basalts from the Sumisu caldera volcano suggest that higher
degrees of melting of the source mantle could have produced
these magmas. However, we need a way to quantify the melt fraction.
Stolper & Newman (1994)
presented a widely cited fluid-fluxed melting model and concluded that for a given set of solid/melt partition coefficients (D values), the best-fit degree of melting by which Mariana trough basalts are generated is positively and approximately linearly correlated with the amount of H2O in the mantle source. One possible approach would be to apply the Stolper & Newman fluid-fluxed melting model to see if it reproduces the trace element patterns observed in the Sumisu high- and low-Zr basalts. There are, however, three assumptions in their model that we find questionable and must be commented upon. First, they used a given set of solid/melt D values for batch melting of a source comprising 60% olivine, 30% orthopyroxene and 10% clinopyroxene. Generally, this batch modal melting cannot adapt to the typically higher degrees of melting required to produce arc lavas, in which the residual modal mineralogy of the source is expected to change drastically. The degrees of melting required to produce the Mariana arc magmas, for example, are much higher than those required to generate the Mariana trough basalts (Stolper & Newman, 1994
). Second, all of their compositions were reconstructed from actual glass or lava compositions by addition of olivine until the liquidus olivine is Fo90; the effect of fractionation of high-calcium clinopyroxene was neglected (Stolper & Newman, 1994
). For the Sumisu caldera volcano, as shown in Fig. 7 and discussed below, we concluded that the low-Zr basalts were derived from a more magnesian, and thus more depleted mantle source than the high-Zr basalts. Thus the assumption that the sources of the high- and low-Zr basalts have the same fertility is not valid. Third, the smallest degree of partial melting of the source of the Mariana trough magmas was assumed to be
5% (partial melting of N-MORB source). This is acceptable for the MORB-like Mariana trough basalts, but to assume such a small degree of melting for arc magmas, such as at Sumisu, seems artificial. Thus we do not apply the Stolper & Newman (1994)
fluid-fluxed melting model.
The degree of partial melting of the source of the primary magmas can be estimated using the ratios of trace element concentrations in lavas with different compositions (Maaløe, 1994
; Maaløe & Pedersen, 2003
). To accomplish this these workers assumed (1) the likely modal mineralogy of the source mantle, and (2) the relative proportions of the mineral constituents that enter the melt. Based on these assumptions, the estimated degree of melting depends on the trace element ratios in two selected lava compositions and not on their absolute values.
As shown in Fig. 14, the IzuBonin arc basalts are among the most depleted on Earth, and the two assumptions of Maaløe (1994)
and Maaløe & Pedersen (2003)
would be difficult to justify. In particular, the clinopyroxene content in the source mantle is unknown. Instead, we use the bulk distribution coefficient between the primary magma and the residual solid as a variable. Thus, the mode of the residual mantle, consisting of olivine, orthopyroxene and clinopyroxene, will change according to the degree of partial melting. As we do not make the two assumptions listed above, degrees of melting cannot be obtained uniquely. To constrain the degree of melting, however, we use the three ratios La/Sm, Sm/Yb, and Zr/Y.
We develop this line of reasoning as follows. In the context of batch partial melting, the variation in the concentration of a trace element, CL, in the melt is given by
where
C0 is the source concentration,
F the degree of melting,
D the
bulk distribution coefficient between the primary magma and
the remaining solid (residue). The original source mineralogy
and the changes that it undergoes during partial melting are
immaterial. Moreover, this equation is applicable for non-modal
melting, and therefore we do not assume that the source/residue
mode and
D values are constant. Sometimes, however, this equation
is erroneously thought to be applicable only to modal batch
melting, in which case the residual mineral mode, and therefore
D, do not change with the degree of melting. Thus it is useful
to repeat the derivation of the basic equation (e.g. Albarède,
1995

). We consider a partially molten multi-mineral assemblage,
consisting of
n 1 residual mineral phases and a liquid
phase. Let
Cj be the concentration of an element in a residual
phase and
Xj the mass-fraction of each phase
j relative to the
partially molten source (and not to the residue). Mass balance
requires that
Let
fj be the mass fraction
of phase
j relative to the residue and
Kj be the partition coefficient
between melt and phase
j.
Xj and
Cj are given by
Then
The bulk distribution
coefficient between liquid and remaining solid (residue) is
given by
This leads to the relationship
known as the equilibrium, partial, or batch-melting equation
(e.g. Albarède, 1995

):
Let the
concentration of La ppm, Sm ppm, Yb ppm, Y ppm and Zr ppm in
two lavas (1 and 2) and the source be given by (La
1, Sm
1, Yb
1,
Y
1, Zr
1), (La
2, Sm
2, Yb
2, Y
2, Zr
2), and (La
0, Sm
0, Yb
0, Y
0,
Zr
0) respectively. Let the bulk distribution between the primary
magmas (1, 2) and their residues and the degree of partial melting
be given as (

,

,

,

,

) and (

,

,

,

,

), and
F1 and
F2,
respectively;
Then
where
If the likely modes of the residual solids of the two primary magmas are initially assumed, then the degree of melting of one primary magma, F2, will be a function of that of another magma, F1. Following the development of equations for LaSm (as above), additional relationships can be considered for SmYb and YZr. The modes of the residual solids can then be varied until the three sets of equations intersect. The point of intersection indicates the degree of partial melting.
Although, generally, we cannot obtain unique solutions, some consistent values are obtained, as shown in Table 5. The degrees of partial melting required to produce the primary magmas of the low-Zr basalts are
2022%, whereas those of the high-Zr basalts are 1012%. Residual clinopyroxene plays a major role in fractionating La from Sm and Zr from Y. Thus, to produce lower La/Sm and higher Y/Zr in the primary low-Zr basalts than in the high-Zr basalts, the content of residual clinopyroxene in the source of the low-Zr primary magmas should be much smaller than that in the source of the primary high-Zr magmas. Our results suggest that the residual mantle should contain only 46 wt % of clinopyroxene after production of the low-Zr primary magmas but should contain
10% clinopyroxene after production of the high-Zr magmas.
xsx
View this table:
[in this window]
[in a new window]
|
Table 5: Ratios of trace element concentrations in two lavas (low-Zr and high-Zr) and degrees of partial melting of the source (F) of the primary magmas estimated using these element pair ratios
|
|
 |
ISOTOPES
|
|---|
Sr, Nd and Pb isotope data for the Sumisu basalts are listed
in
Table 2. Samples 575R4 and 575R5 and 575R7 and SI14 are representative
low-Zr basalts and high-Zr basalts, respectively. Importantly,
the low-Zr and high-Zr basalts overlap in isotopic composition.
These new Sr, Nd and Pb isotopic data for basalts from Sumisu
caldera volcano are integrated with previously published data
for Quaternary frontal volcanoes in the IzuBonin arc
(Taylor & Nesbitt, 1998

; Ishizuka
et al., 2003

) and the
Sumisu Rift (Hochstaedter
et al., 1990

).
Figure 15 shows along-arc
SrNdPb isotopic and Ba/Zr variations of lavas
from frontal volcanoes and the Sumisu Rift. Each frontal volcano
has a limited range of isotope compositions compared with those
from the Sumisu Rift. Similar systematic along-arc variations
for the lavas of the IzuBonin arc have already been reported
by Taylor & Nesbitt (1998)
![Go]()