Oxygen isotope constraints on the sources of Central
American arc lavas
John M. Eiler
Division of Geological and Planetary Sciences, California Institute of Technology, MC 100-23, Pasadena, California
91125, USA (eiler@gps.caltech.edu)
Michael J. Carr
Department of Geological Sciences, Rutgers-State University of New Jersey, Piscataway, New Jersey 08854, USA
Mark Reagan
Department of Geology, University of Iowa, Iowa City, Iowa 52242, USA
Edward Stolper
Division of Geological and Planetary Sciences, California Institute of Technology, MC 170-25, Pasadena, California
91125, USA
[
1
]
Oxygen-isotope ratios of olivine and plagioclase phenocrysts in basalts and basaltic andesites from
the Central American arc vary systematically with location, from a minimum
d
18
O
olivine
value of 4.6
(below the range typical of terrestrial basalts) in Nicaragua near the center of the arc to a maximum
d
18
O
olivine
value of 5.7 (above the typical range) in Guatemala near the northwest end of the arc. These
oxygen-isotope variations correlate with major and trace element abundances and with Sr and Nd
isotope compositions of host lavas, defining trends that suggest variations in
d
18
O reflect slab
contributions to the mantle sources of these lavas. These trends can be explained by a model in which
both a low-
d
18
O, water-rich component and a high-
d
18
O, water-poor component are extracted from the
subducting Cocos slab and flux melting in the overlying mantle wedge. The first of these components
dominates slab fluxes beneath the center of the arc and is the principal control on the extent of
melting of the mantle wedge (which is highest in the center of the arc); the second component
dominates slab fluxes beneath the northwestern margin of the arc. Fluxes of both components are
small or negligible beneath the southeastern margin of the arc. We suggest that the low-
d
18
O
component is a solute-rich aqueous fluid produced by dehydration of hydrothermally altered rocks
deep within the Cocos slab (perhaps serpentinites produced in deep normal faults offshore of
Nicaragua) and that the high-
d
18
O component is a partial melt of subducted sediment on top of the
plate.
Components:
17,022 words, 10 figures, 3 tables
.
Keywords:
arc; basalt; Central America; oxygen isotope.
Index Terms:
1065 Geochemistry: Major and trace element geochemistry; 1040 Geochemistry: Radiogenic isotope
geochemistry; 3640 Mineralogy and Petrology: Igneous petrology.
Received
16 July 2004;
Revised
28 February 2005;
Accepted
14 March 2005;
Published
19 July 2005.
Eiler, J. M., M. J. Carr, M. Reagan, and E. Stolper (2005), Oxygen isotope constraints on the sources of Central American arc
lavas,
Geochem. Geophys. Geosyst.
,
6
, Q07007, doi:10.1029/2004GC000804.
G
3
G
3
Geochemistry
Geophysics
Geosystems
Published by AGU and the Geochemical Society
AN ELECTRONIC JOURNAL OF THE EARTH SCIENCES
Geochemistry
Geophysics
Geosystems
Article
Volume 6
,Number7
19 July 2005
Q07007, doi:10.1029/2004GC000804
ISSN: 1525-2027
Copyright 2005 by the American Geophysical Union
1 of 28
1. Introduction
[
2
] The geochemistry of mafic lavas from the
Central American arc (Figures 1a, 1b, and 1c)
varies systematically with distance along the arc
and covaries in general both with the dip of the
subducting Cocos slab (Figure 1d) and the crustal
thickness of the overriding Caribbean and (in the
northwestern end of the arc) North American plates
(Figure 1e; geophysical parameters are based on
estimates from
Carr
[1984],
Carr et al.
[1990], and
Carr et al.
[2004] and are likely to be revised soon
on the basis of recent and planned geophysical
experiments in Central America). For example,
strong ‘‘slab’’ trace element signatures (e.g., high
U/Th and Ba/La) and low Na contents are typical
of lavas in the center of the arc, where the slab dip
is high and crustal thickness low; whereas weaker
‘‘slab’’ trace element signatures and higher Na
contents are typical of lavas from the margins of
the arc, where slab dip is low and crustal thickness
high. In addition, there are subtle geochemical
differences between the two ends of the arc, and
these are not simply correlated with slab dip or
crustal thickness (which are not significantly dif-
ferent at the northern and southern ends of the arc).
For example, whereas Costa Rican lavas (the
southeast end) have U/Th and Ba/La ratios
approaching those typical of mid-ocean-ridge and
ocean-island basalts (referred to as MORB and
OIB), Guatemalan lavas (the northwestern end)
are similarly low in U/Th but higher in Ba/La.
[
3
] The covariations of lava chemistry and slab dip
shown in Figure 1 have been interpreted as evi-
dence that subduction-zone geometry controls
fluxes of slab-derived fluids through the overlying
mantle wedge. For example,
Carr et al.
[1990]
hypothesized that fluids released from a steeply
dipping slab are focused over a narrow interval of
the mantle wedge (measured perpendicular to the
trench) whereas a flatter slab distributes the fluids
over a wider interval; this behavior is plausible if
fluids are released from the slab over a constant
range of depths. Highly focused fluxes of slab-
derived fluids would lead to relatively high inte-
grated extents of melting in the infiltrated parts of
the mantle wedge, producing lavas with low abun-
dances of Na and trace-element ratios dominated
by subducted components (i.e., similar to Nicara-
guan lavas). In contrast, less focused fluxes of the
same total amounts of fluid would result in more
widely distributed, smaller degrees of melting,
producing more Na-rich magmas having trace
element ratios more influenced by the mantle
wedge (i.e., similar to Guatemalan and Costa Rican
lavas, although, as mentioned above, these two
suites differ from one another in some aspects of
their geochemistry). In addition, recent geophysical
evidence shows that the Cocos slab beneath
Nicaragua has an exceptionally high degree of
hydration [
Abers et al.
, 2003], providing a second
reason why fluxes of slab-derived fluid might be
unusually high beneath the center of the arc. Never-
theless, although it can account for key aspects of
the geochemistry of Central American lavas, with-
out modification, this hypothesis does not account
for the contrasting behaviors of the U/Th ratio
(which varies symmetrically about the center of the
arc; Figure 1b) and the Ba/La ratio and perhaps
Na
6.0
(which differ between the northwest and
southeast ends of the arc; Figures 1a and 1c).
[
4
] An alternative hypothesis that might explain
some of the trends in Figure 1 is that the degree of
melting of the mantle wedge is influenced by the
distance over which it upwells (i.e., as for decom-
pression melting beneath mid-ocean ridges) rather
than being dominantly set by the availability of
slab-derived fluids, and that this distance is con-
trolled by the thickness of the overriding plate
(which might equal or correlate with, crustal thick-
ness [
Plank and Langmuir
, 1988]). In this context,
high-degree melts with low Na contents would be
produced by the longer and on average lower
pressure columns of upwelling mantle wedge un-
derlying thinner crust (i.e., as in Nicaraguan),
whereas low-degree melts with high Na contents
would be produced by shorter and on average
higher pressure upwelling columns underlying
thicker crust (i.e., as in Guatemala and Costa Rica).
Variations in the vertical distance over which
melting takes place could also plausibly depend
on slab dip if the depth of the base of the melting
column is related to the depth of the slab. This
hypothesis, at least as detailed by
Plank and
Langmuir
[1988], does not make specific predic-
tions about the distribution of slab trace element
signatures along the length of the arc. However,
one can easily imagine variants of this hypothesis
that combine processes of fluxed and decompres-
sion melting and that might also explain these
geographic variations in trace elements.
[
5
] The fluxed-melting hypothesis of
Carr et al.
[1990] predicts a strong relationship between the
degree of partial melting in the mantle wedge and
the amount of slab-derived fluid or hydrous melt
added to the wedge, whereas the decompression-
melting hypothesis does not require such a rela-
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
2of28
Figure 1.
Geographic distribution of compositions of Central American arc lavas (Figures 1a–1c and 1f ) and
geophysical properties of the subducting slab and overriding plate (Figures 1d and 1e). Note that strong trace element
signatures of subducted components (high U/Th and Ba/La ratios), low Na contents, and low
d
18
O values are
concentrated in the center of the arc, where slab dip is steep and the Caribbean plate is thin. Each data point in
Figure 1a is the average Na
2
O content, in wt.%, of a suite of lavas from a single volcanic center, normalized to
6 wt.% MgO (the average MgO content of the lavas we analyzed). These Na
6.0
values were calculated on the basis of
linear regressions of measurements of wt.% Na
2
O versus wt% MgO for several lavas from each volcano, based on
data from a database maintained by M. J. Carr (http://www-rci.rutgers.edu/
carr/index.html). U/Th and Ba/La ratios
(Figures 1b and 1c) are for individual samples of basalt and basaltic andesite from the M. J. Carr database. Slab dip
and crustal thickness (Figures 1d and 1e) are estimates taken from
Carr and Stoiber
[1990] and
Carr et al.
[2004].
Values of
d
18
O
olivine
(Figure 1f ) are from this study and are compiled in Table 1. Two lavas from behind the volcanic
front in Honduras are distinguished by open symbols in Figure 1f. The geographic distribution of
d
18
O
plagioclase
values (Table 1) resembles that for
d
18
O
olivine
in Figure 1f. The inset is a map of the Central American arc; the heavy
dashed line was used to measure distances from the Mexican-Guatemalan border.
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
eiler et al.: central american arc lavas
10.1029/2004GC000804
3of28
tionship. Therefore determination of the amounts
of slab-derived components sampled by lavas pro-
duced by known degrees of partial melting of
peridotites in the mantle wedge would provide a
test of the Carr et al. hypothesis and valuable
constraints on any model to explain the connec-
tions between geochemical and geophysical fea-
tures of the Central American arc.
[
6
] Oxygen-isotope ratios of arc lavas provide a
potentially powerful means of characterizing and,
under certain circumstances, quantifying the input
of slab-derived components into their sources
[
Eiler et al.
, 2000a; see also
Eiler
, 2001, and
references therein]. Oxygen-isotope ratios, espe-
cially in conjunction with other geochemical indi-
cators that constrain source composition and
degree of melting, are particularly useful for two
reasons:
[
7
] 1. All common rocks and geological fluids
contain similar concentrations of oxygen. Conse-
quently, mixing trajectories are roughly linear
functions of oxygen-isotope ratios, making it rela-
tively easy to relate variations in
d
18
O of lavas to
amounts of isotopically exotic fluids or melts
added to their sources. This is not usually the case
for radiogenic-isotope ratios or abundance ratios of
incompatible elements because their concentrations
can differ significantly in various slab-derived and
mantle components [
Elliott
, 2004].
[
8
] 2. The oceanic lithosphere is heterogeneous in
d
18
O, so the
d
18
O values of slab-derived compo-
nents could provide information on the part(s) of
the slab from which they come. Although
d
18
O
SMOW
values of most mantle peridotites span
a narrow range (
5.5 ± 0.2
%
[
Mattey et al.
,
1994]; see
Eiler et al.
[1996a] and
Ducea et al.
[2002] for exceptions), many of the candidates
for the sources of slab-derived components are
distinct from this range. For example, marine
carbonates (
d
18
O
SMOW
= 25–32
%
), siliceous
oozes (
d
18
O
SMOW
=35–42
%
), pelagic clays
(
d
18
O
SMOW
= 15–25
%
)[
Kolodny and Epstein
,
1976;
Arthur et al.
, 1983], and weathered and
hydrothermally altered upper oceanic crust
(
d
18
O
SMOW
= 7–15
%
)[
Gregory and Taylor
,
1981;
Alt et al.
, 1986;
Staudigel et al.
, 1995] are
all richer in
18
O than typical upper mantle perido-
tites, whereas hydrothermally altered lower oceanic
crust (
d
18
O
SMOW
=0–6
%
)[
Gregory and Taylor
,
1981;
Altetal.
,1986;
Staudigel et al.
, 1995],
hydrothermally altered ultramafic rocks (e.g., ser-
pentinites) from the ocean lithosphere (
d
18
O
SMOW
=
0–6
%
, although these are poorly known [
Magaritz
and Taylor
, 1974;
Cocker et al.
, 1982]), and pore
waters in marine sediments (
d
18
O=0to
3
%
;
perhaps as low as
15
%
[
Schrag et al.
, 1992] are
all poorer in
18
O than typical upper mantle perido-
tites. Thus oxygen-isotope ratios can help to distin-
guish among various choices of possible sources of
slab-derived components, especially in the context
of other geochemical discriminators among these
components.
[
9
] In this paper, we report measurements of oxy-
gen-isotope ratios in phenocrysts from 51 Central
American arc basalts and basaltic andesites from
Costa Rica to Guatemala. We focus on phenocrysts
because they are less susceptible to posteruptive
alteration than groundmass or glass [
Eiler
, 2001].
These samples span most of the geographic range of
the arc and have been characterized geochemically
in previous studies. These measurements allow us to
revisit existing hypotheses for the origins of geo-
graphically correlated geochemical variability of
Central American lavas and have the potential to
constrain the amounts of slab-derived components
in the mantle sources of these lavas, the properties of
slab-derived fluids and/or melts, and the part(s) of
the slab from which they were derived.
2. Samples and Methods
[
10
] We analyzed oxygen-isotope ratios of olivine
and/or plagioclase phenocrysts from 51 lavas from
the volcanic front in Guatemala, El Salvador,
Nicaragua and Costa-Rica, and from behind the
volcanic front (‘‘bvf’’) in Honduras (Table 1). All
of the analyzed samples are relatively mafic mem-
bers (average MgO = 5.8 wt.%; average Mg# =
0.54) of a suite that has been previously character-
ized for major and trace element and radiogenic-
isotope geochemistry by M. J. Carr and colleagues.
A database containing these previous measure-
ments is available on request from M. J. Carr
(http://www-rci.rutgers.edu/
carr/index.html).
[
11
] Oxygen-isotope ratios were measured on
hand-picked olivine and plagioclase mineral sepa-
rates of
200–500
m
m grains. Minerals containing
visible inclusions of fluid, glass, or other minerals
were avoided when recognized. Adhered dust was
removed from the mineral separates by blowing
filtered, compressed air over them. The samples
were then analyzed by laser fluorination using a
50-watt CO
2
laser, BrF
5
reagent, and apparatus for
gas purification and conversion of O
2
to CO
2
based
on designs by
Sharp
[1990] and
Valley et al.
[1995]. Oxygen yields were 96 ± 3% (1
s
), based
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
4of28
on the typical major element compositions of
phenocrysts in rocks of similar composition. Mea-
surements of
d
18
O were made on twelve different
days. Two to five analyses of the SCO-1 olivine
standard [
Eiler et al.
, 1995] and two to seven
analyses of the UWG-2 garnet standard [
Valley et
al.
, 1995] were made each day; reproducibility for
repeat measurements of each standard on a given
day averaged ±0.05
%
(1
s
). The mean
d
18
O values
for standards were generally within 0.1
%
of their
nominal values based on previous analyses [
Eiler
et al.
, 1995, 1996a, 1996b, 1996c, 2000a, 2000b,
Table 1.
Oxygen Isotope Compositions of Phenocrysts From Central American Arc Lavas
Country
Volcano
Sample
Mg#
d
18
O
olivine
±1
sd
18
O
plagioclase
±1
s
Guatemala
Tecuamburro
Qam12
0.48
—
—
6.55
0.00
Guatemala
Tecuamburro
Qam13
0.47
—
—
6.36
0.00
Guatemala
Tecuamburro
Qam14
0.48
5.67
0.09
6.41
0.00
Guatemala
Tecuamburro
Qam15
0.53
5.57
0.02
6.34
0.02
Guatemala
Tecuamburro
T301
0.54
5.59
0.00
6.29
0.00
Guatemala
Tecuamburro
T302
0.59
5.51
0.04
6.21
0.09
Guatemala
Pacaya
GUP701
0.44
5.43
0.03
6.29
0.05
Guatemala
Pacaya
Pacaya 1985
0.47
5.34
0.08
6.20
0.03
El Salvador
Cerro Verde
CV1
0.52
5.60
0.04
6.40
0.00
El Salvador
Izalco
IZ101
0.52
5.43
0.01
6.34
0.04
El Salvador
Izalco
IZ103
0.51
5.49
0.01
6.21
0.07
El Salvador
Izalco
IZ104
0.50
5.44
0.05
6.31
0.02
El Salvador
Izalco
IZ108
0.51
5.45
0.02
6.25
0.03
El Salvador
Izalco
IZ 109
0.46
—
—
6.45
0.01
El Salvador
Izalco
IZ 112
0.52
5.41
0.04
6.15
0.03
El Salvador
Izalco
IZ 114
0.46
5.42
0.01
6.44
0.04
El Salvador
Izalco
IZ119
0.53
5.44
0.03
6.31
0.03
El Salvador
Izalco
IZ122
0.46
5.45
0.00
6.23
0.01
El Salvador
Apastapeque
AP-3
0.53
5.10
0.03
5.81
0.09
Honduras (bhf)
Yohoa
YO3
0.61
5.33
0.01
—
—
Honduras (bhf)
Yohoa
YO10
0.51
5.64
0.04
6.44
0.17
Nicaragua
Telica
TE3
0.54
—
—
5.51
—
Nicaragua
Telica
TE6
0.56
5.02
0.03
5.80
0.03
Nicaragua
Telica
TE112
0.42
5.07
0.01
5.63
0.07
Nicaragua
Telica
TE114
0.45
4.92
0.03
5.72
0.03
Nicaragua
Telica
TE119
0.47
4.82
0.01
5.65
0.04
Nicaragua
Telica
TE123
0.53
4.94
0.03
5.80
0.09
Nicaragua
Telica
TE124
0.55
4.84
0.06
5.75
0.03
Nicaragua
Cerro Negro
CN1
0.49
4.83
0.00
5.54
—
Nicaragua
Cerro Negro
CN2
0.47
4.90
0.01
5.59
0.01
Nicaragua
Cerro Negro
CN3
0.60
4.79
0.02
5.69
0.05
Nicaragua
Cerro Negro
CN5
0.64
4.88
0.01
5.57
0.03
Nicaragua
Cerro Negro
CN10
0.53
4.88
0.05
5.65
0.08
Nicaragua
Cerro Negro
CN11
0.51
4.78
0.00
5.55
0.02
Nicaragua
Cerro Negro
CN12
0.56
4.80
0.01
5.68
0.02
Nicaragua
Cerro Negro
1995 bomb
—
4.94
0.01
5.47
0.09
Nicaragua
San Cristobal
SC2
0.53
4.78
0.06
5.69
0.07
Nicaragua
Las Pilas
LP2
0.57
4.62
0.04
5.39
0.01
Nicaragua
Nejapa
NE3
0.54
4.92
0.01
5.55
0.11
Nicaragua
Nejapa
NE5
0.62
4.98
0.04
5.62
0.07
Nicaragua
Nejapa
NE6
0.63
5.23
0.01
5.79
0.14
Nicaragua
Nejapa
NE203
0.65
5.05
0.17
5.63
0.19
Nicaragua
Granada
GR5
0.60
5.07
0.08
5.59
0.01
Nicaragua
Granada
GR101
0.63
5.01
0.09
—
—
Nicaragua
Grenada
GM 92-3
0.59
4.97
0.00
5.75
0.07
Costa Rica
Tenorio
TE9
0.52
5.18
0.02
6.05
0.06
Costa Rica
Platanar
PP7
0.76
5.09
0.02
—
—
Costa Rica
Barba
B7
0.60
5.13
0.01
6.08
0.02
Costa Rica
Irazu
IZ-63A
0.62
5.28
0.02
5.98
0.02
Costa Rica
Turrialba
T24-2
0.62
5.10
0.02
—
—
Costa Rica
Turrialba
T91
0.62
5.04
0.01
—
—
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
5of28
2000c;
Valley et al.
, 1995;
Cooper et al.
, 2004].
Differences in measurements from day to day were
normalized by correcting all data for unknowns by a
constant value equal to the average difference
between measured and accepted values for all the
measurements of standards on that day (the absolute
value of this correction averaged 0.07
%
). All
unknown samples were analyzed two to nine (typ-
ically three) times, often distributed over several
days, with an average reproducibility for a given
sample of ±0.04
%
(1
s
) after correction for day-to-
day variations in measured values for standards.
3. Results
3.1. Isotopic Distributions Among
Coexisting Minerals
[
12
] Phenocryst populations in arc lavas are often
complex mixtures of crystals precipitated from
multiple liquids [e.g.,
Cooper and Reid
,2003]
and could include xenocrysts (i.e., crystals precip-
itated from liquids other than their host magma)
with oxygen-isotope compositions that do not
reflect those of their host magmas. We measured
oxygen-isotope ratios of both olivine and plagio-
clase in most samples, with the expectation that
rocks containing few or no xenocrysts would exhibit
near-equilibrium plagioclase-olivine fractionations,
whereas those containing abundant xenocrysts
might exhibit disequilibrium plagioclase-olivine
fractionations. Furthermore, because we analyzed
most samples multiple times, we anticipated that
rocks containing isotopically exotic xenocrysts
might be revealed by poor reproducibility. Although
it is possible that all measured grains of both
olivine and plagioclase in any given sample
could be xenocrysts that are in oxygen-isotope
exchange equilibrium with one another but not
with their host magma, we assume on the basis
of previous studies of the oxygen-isotope geo-
chemistry of phenocrysts and xenocrysts in
basalts [
Garcia et al.
, 1998;
Maclennan et al.
,
2003;
Wang et al.
, 2003] that this would be an
unusual circumstance.
[
13
] We observe intrasample reproducibility in
most samples similar to our analytical precision
(Table 1). Although the difference is small, the
average reproducibility for olivine phenocrysts
(±0.03
%
,1
s
) is somewhat better than for plagio-
clase (±0.05
%
,1
s
), perhaps indicating that plagio-
clase xenocrysts are more common than olivine
xenocrysts in these lavas. Nevertheless, all but a
handful of samples are homogeneous in
d
18
O
given the precision of our measurements. Measured
plagioclase-olivine fractionations average 0.76 ±
0.11
%
(Table 1; Figures 2a and 2b), consistent
with the range of
0.6to1.1
%
expected for
equilibrium between forsterite and An
50–100
pla-
gioclase at temperatures of 1150–1350
C[
Chiba et
al.
, 1989;
Eiler et al.
, 1995]. Moreover, plagioclase-
olivine fractionations are correlated weakly with
the Mg# of the host lava (Figure 2b), as expected
on the basis of the more albitic plagioclases and
lower temperatures characteristic of more differen-
tiated magmas. These results suggest to us that
isotopically exotic crystals, though possibly present
Figure 2.
(a) Comparison of
d
18
O values of coexisting
plagioclase and olivine phenocrysts from lavas examined
in this study. The diagonal gray lines mark the range of
fractionations consistent with equilibrium at magmatic
temperatures [
Chiba et al.
, 1989]. (b) Comparison of the
plagioclase-olivine fractionation to the Mg# of the host
lava. Values of Mg# were calculated as (mole fraction
MgO)/(mole fraction MgO + 0.9x (mole fraction total
iron as FeO) and are based on data for the same samples
analyzed for
d
18
O
olivine
, taken from the M. J. Carr
database. Although the relationship between these two
variables is scattered, most lavas having relatively low
D
plagioclase-olivine
values (0.7 or less) are relatively
primitive (Mg# greater than
0.58). This relationship
is consistent with the decrease in plagioclase-olivine
fractionation expected for the higher liquidus tempera-
tures and more Ca-rich plagioclase compositions typical
of high-Mg# lavas.
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
6of28
in some rocks, do not control the oxygen-isotope
systematics of our sample suite.
3.2. Range and Geographic Distribution
of Oxygen-Isotope Compositions
[
14
] Oxygen-isotope compositions of phenocrysts
from Central American arc lavas vary systemati-
cally with distance along the strike of the volcanic
front, with a minimum in Nicaraguan lavas
(Figure 1f ). Broadly similar-looking minima or
maxima are also observed for other geochemical
variables (Figures 1a–1c), suggesting to us that a
common set of petrogenetic processes could be
responsible for all of these trends. Note, however,
that oxygen-isotope ratios of phenocrysts from
lavas from the northwest end of the arc (Guatemala
and El Salvador) are systematically higher than
those from the southeast end of the arc (central
Costa Rica); thus the geographic trend in
d
18
Ois
asymmetric, like those for Ba/La ratios and perhaps
also Na
6.0
values (Figures 1c and 1a), rather than
symmetric, like that for U/Th ratios (Figure 1b).
[
15
] Values of
d
18
O
olivine
of 4.6–4.9
%
from the
center of the arc (Nicaragua) are lower than values
typical for arc-related basalts (
d
18
O
olivine
= 5.2 ±
0.2 [
Eiler et al.
, 2000a;
Eiler
, 2001]), most mantle
peridotites (
d
18
O
olivine
= 5.2 ± 0.2
%
[
Mattey et al.
,
1994;
Eiler
, 2001]), and normal mid-ocean-ridge
basalts (NMORB;
d
18
O
olivine
5.0–5.2
%
, based
on measurements of glass [
Eiler et al.
, 2000b;
Cooper et al.
, 2004]). Basaltic lavas with
d
18
O
oli-
olivine
values as low as the Nicaraguan phenocrysts
are only known to be abundant in Iceland [
Gee et
al.
, 1998;
Eiler et al.
, 2000c] and Hawaii [
Eiler
et al.
, 1996b, 1996c;
Wang et al.
, 2003]; they
also occur, although less commonly, in other
ocean-island volcanic centers (e.g., St. Helena
and the Canary Islands [
Eiler et al.
, 1996a;
Thirlwall et al.
, 1997]). In contrast, phenocrysts
from lavas from the northwestern margin of the
arc (Guatemala and El Salvador;
d
18
O
olivine
=
5.10–5.67; 5.46
%
average) and behind the vol-
canic front (Honduras;
d
18
O
olivine
=5.33and
5.64) are higher in
d
18
O than typical mantle
peridotite and most basaltic lavas. These relatively
high values are within the range common for
‘‘EM-2’’ type ocean island basalts (including those
from French Polynesia and Samoa;
d
18
O
olivine
=
5.4–6.1 [
Eiler et al.
, 1996a]), Hawaiian lavas from
Koolau and Lanai having relatively radiogenic Sr
isotope ratios (
d
18
O
olivine
= 5.6–6.0 [
Eiler et al.
,
1996c]), and some subduction-related basalts
from Vanuatu and the Taba-Lihir-Tanga-Feni arc
(
d
18
O
olivine
= 5.5–5.8 for boninites and low-Ti
basalts and basaltic andesites [
Eiler et al.
, 2000a]).
3.3. Comparisons of Oxygen Isotopes
With Other Geochemical Properties
[
16
] Figure 3 compares
d
18
O
olivine
values with other
geochemical variables previously measured for
whole rock samples of the same specimens. Note
that Na
6.0
values are based on regressions of com-
positions of multiple lavas (see Figure 1 caption), so
each plotted point in Figure 3a is the average for all
analyzed samples from a single volcano, whereas
each point represents a single hand specimen in
Figures 3b–3g. Variations in
d
18
O
olivine
correlate
with most of the plotted geochemical parameters:
Na
6.0
values correlate positively with
d
18
O
olivine
,
whereas Cu concentrations, Sr and Nd isotope
ratios and U/Th and Ba/La abundance ratios corre-
late negatively with
d
18
O
olivine
. These correlations
further support our conclusion that
d
18
O values
of olivines generally reflect those of their host
magmas; i.e., the concentrations and ratios
plotted against
d
18
O
olivine
in Figure 3 are based
on elements that are strongly concentrated in the
groundmass, and thus they would not in general
be expected to correlate with
d
18
O
olivine
unless
the olivines were cogenetic with the ground-
mass. Note that the La/Sm ratio is uncorrelated
with
d
18
O
olivine
; it is shown to illustrate the
unusual characteristics of some of the Costa Rican
lavas (discussed below; see also
Herrstrom et al.
[1995]).
[
17
] Note that high
d
18
O
olivine
is associated with a
relatively ‘‘depleted’’ (unradiogenic) Sr isotope
composition on the one hand (Figure 3f ) and a
relatively ‘‘enriched’’ (also unradiogenic) Nd iso-
tope composition on the other (Figure 3g). This
reflects the unusual positive correlation between Sr
and Nd isotope ratios exhibited by Central Amer-
ican lavas [
Feigenson and Carr
, 1986], which
contrasts with the negatively sloped array typical
of MORBs and most ocean-island basalts (Figure 4)
[see also
Hofmann
, 1997]. This peculiarity has
been previously interpreted to reflect mixing
between an enriched mantle wedge having Sr and
Nd isotope compositions plotting on the ‘‘mantle
array’’ in Figure 4 (perhaps resembling enriched,
so-called ‘‘E’’ MORBs) and a slab-derived com-
ponent with higher
87
Sr/
86
Sr and
143
Nd/
144
Nd
plotting above and to the right of the mantle array
on Figure 4 [
Feigenson and Carr
, 1986;
Carr et
al.
, 1990, 2004]. Carr and his coworkers have
hypothesized that this slab-derived component has
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
7of28
relatively ‘‘depleted’’ (i.e., radiogenic) Nd isotope
ratios because of the depleted nature of the Pacific
mantle sources from which the Cocos plate was
derived, but ‘‘enriched’’ Sr isotope ratios due to
the addition of seawater Sr to the plate when it
was hydrothermally altered before subduction.
[
18
] The trends in Figures 3a–3d, 3f, and 3g
suggest that variations in
d
18
O might be controlled
by the same petrogenetic processes controlling
other geochemical indices previously used as the
basis for models of Central American magmatism:
Low-
d
18
O values are associated with high inferred
integrated degrees of mantle melting (i.e., Na-poor
lavas), high abundances of some fluid-soluble
elements (e.g., Ba and Cu; see
Stolper and
Newman
[1994] and
Noll et al.
[1996] for previous
studies of this often-overlooked element in arc
lavas), high ratios of fluid-soluble to fluid-
insoluble elements (i.e., Ba/La and U/Th), and Sr
Figure 3.
Correlations of
d
18
O
olivine
values measured in this study for Central American lavas (Table 1) with other
geochemical parameters previously measured for these same samples and available through the M. J. Carr database.
Figure 3a compares Na
6.0
values, calculated as described in the caption to Figure 1, to the average
d
18
O
olivine
value of
all analyzed lavas from the same volcanic center.
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
8of28
and Nd isotope compositions thought to be charac-
teristic of weathered and/or hydrothermally altered
rocks in the Cocos slab. In contrast, high
d
18
O
values are associated with lower inferred integrated
degrees of melting and weaker slab signatures.
4. Discussion
4.1. Two-Component Models
[
19
] The geographic trend in
d
18
O
olivine
(Figure 1f )
and correlations between
d
18
O
olivine
and other geo-
chemical variables (Figure 3) suggest that oxygen-
isotope data can be used to test or extend previous
models for the petrogenesis of Central American
arc lavas. The best-defined trends in Figure 3 are
roughly linear and could be explained by petroge-
netic models involving only two components, one
relatively low in
d
18
O and the other relatively high
in
d
18
O. We consider two models that might
explain these trends in terms of only two compo-
sitionally distinct end-members:
[
20
] 1. Higher
d
18
O values occur on the northern
and southern extremities of the arc where the crust
is thick and lower
d
18
O values occur near the
middle of the arc where the crust is thin. The first
possibility that we consider is that all mantle-
derived magmas delivered to the arc are initially
low in
d
18
O but that contamination by crustal rocks
in the overriding plates (North American crust in
the extreme northwest end of the arc; Caribbean
crust elsewhere) is responsible for higher
d
18
O
values, and that crustal contamination is more
extensive where the crust is thicker. This hypoth-
esis would imply that much of the geochemical
variability previously interpreted in terms of mantle
processes [e.g.,
Carr et al.
, 1990;
Patino et al.
,
2000;
Plank et al.
, 2002] actually reflects system-
atic variations along the length of the arc in the
degrees of crustal contamination.
[
21
] 2. Because the low-
d
18
O extreme is associated
with ‘‘slab-like’’ geochemical signatures and high
inferred degrees of melting (and the high-
d
18
O
extreme with the opposite), the data might be
explained by a two-component model where the
two components are a high-
d
18
O mantle wedge
beneath Central America and a low-
d
18
Oslab-
derived, water-rich component that induces melting
of the mantle wedge. This is similar to the petro-
genetic models of
Carr
[1984],
Carr et al.
[1990],
Patino et al.
[2000], and
Plank et al.
[2002]. If
such a model can successfully account for the full
array of geochemical data, the constraints placed
on the oxygen-isotope ratios of the mantle and
slab-derived components would allow evaluation
of the plausibility of the model.
[
22
] The following paragraphs examine the consis-
tency of these two models with the geochemical
data.
4.1.1. Contamination by the Upper Plate?
[
23
] The simplest hypothesis of this kind would
interpret lower
d
18
O values in Nicaraguan lavas as
Figure 4.
Plot of Sr versus Nd isotope compositions of the lavas we studied and comparison with the compositions
of Atlantic and Pacific mid-ocean ridge basalts and basement gneisses from southeast Guatemala (based on data from
the PetDB database maintained by the Lamont-Doherty Oceanographic Observatory, http://petdb.ldeo.columbia.edu/
petdb/, and data reported by
Walker et al.
[1995]).
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
9of28
relatively uncontaminated and higher
d
18
O values
in Guatemalan, El Salvadoran, Honduran, and
Costa Rican lavas as more strongly contaminated.
This hypothesis is difficult to reconcile with sev-
eral geochemical observations:
[
24
]1.
Walker et al.
[1995] and
Patino et al.
[2000]
presented evidence that relatively radiogenic Sr
and nonradiogenic Nd isotope ratios in lavas erup-
ted behind the volcanic front (‘‘bvf’’) in Guatemala
reflect contamination by crystalline basement rocks
of the continental North American plate. However,
they did not find evidence for contamination in
high-Mg# lavas in the volcanic front or in lavas
erupted further to the southeast through the Carib-
bean plate (i.e., in samples comparable to those
investigated here). Moreover, the quartzo-feld-
spathic gneisses identified by
Walker et al.
[1995]
as the assimilants in Guatemalan bvf lavas are
higher in
87
Sr/
86
Sr and lower in
143
Nd/
144
Nd than
any Central American arc lavas and could not
represent an end-member in the positively sloped
Central American Sr-Nd isotope array (Figure 4).
Oxygen isotopes compound this problem since,
although the
d
18
O values of these basement rocks
are unknown, they consist of lithologies that are
typically higher in
d
18
O than the mantle (i.e.,
typically
8–14
%
[
Eiler
, 2001;
Simon and
Lecuyer
, 2005]), and therefore crustal contamina-
tion, at least in Guatemala, would be expected to
produce a positive trend in Figure 3f rather than the
negative trend we observe.
[
25
] 2. There is no correlation between
d
18
O and
Mg# in the lavas we studied (Table 1 and
Figure 5a). Crustal contamination in variably dif-
ferentiated magmatic suites generally increases
with the extent of crystallization differentiation
[e.g.,
Taylor
, 1980;
DePaolo
, 1981;
Nicholson et
al.
, 1991;
Eiler et al.
, 2000c], so low-Mg# lavas are
expected to be systematically higher in
d
18
O than
high-Mg# lavas if crustal contamination strongly
influences their oxygen-isotope ratios, but no such
relationship is observed. Similarly, suites of lavas
from the same volcanic edifice spanning ranges in
Mg# are nearly uniform in
d
18
O, regardless of the
average
d
18
O value of that suite or its geographic
location along the length of the arc (Figure 5b). The
increase in
d
18
O of Tecuamburro lavas with de-
creasing Mg# is a possible exception, but even if so,
this shift is small relative to the 1
%
range observed
across our sample suite (note that this trend at
Tecuamburron is also consistent with the effects
of crystal fractionation alone [
Eiler
, 2001]).
[
26
] 3. In order to explain the low
d
18
O values of
the Nicaraguan samples, this hypothesis requires
that the mantle wedge beneath the Central Amer-
ican arc is lower in
d
18
O than the sources of normal
MORBs. This is possible, but the scarcity of low
d
18
O basalts elsewhere suggests this is unlikely.
[
27
] Although we cannot absolutely rule it out, we
conclude that crustal contamination is an implau-
sible explanation of the oxygen-isotope variations
of relatively high-Mg# volcanic-front lavas in the
Central American arc.
4.1.2. Melting Fluxed by a Single,
Low-
D
18
O Slab-Derived Component?
[
28
] As noted above, the geochemistry of Central
American arc lavas has been previously interpreted
in terms of a mantle wedge consisting of fertile
peridotite that melts in response to influx of slab-
derived aqueous fluids that are more abundant
beneath the center of the arc than beneath its
northern and southern ends [
Carr et al.
, 1990].
Following this reasoning, we could interpret our
data as evidence that the mantle wedge is relatively
Figure 5.
Comparison of
d
18
O
olivine
values for lavas examined in this study (Table 1) to the Mg# values of host
lavas. Symbols are as in Figure 3.
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
10 of 28
high in
d
18
O (like lavas from the northern end of
the arc) and the slab-derived metasomatic agent is
relatively low in
d
18
O (like lavas from the center of
the arc).
[
29
] Figure 6 compares the correlation we observe
between
d
18
O and Na
6.0
with that predicted for a
representative fluxed melting model in which a
high-
d
18
O peridotite (
d
18
O
olivine
=5.6
%
) melts in
response to infiltration by low-
d
18
O aqueous fluid
(
d
18
O=0
%
). The solid black curve and the solid
gray curve in Figure 6 represent the compositions
of magmas produced by batch and fractional melt-
ing, respectively, assuming the low-
d
18
Ofluid
contains 90 wt.% H
2
O and 2 wt.% Na
2
O, and a
productivity of fluxed melting as defined by
Eiler
et al.
[2000a] (see Appendix A for further details).
For comparison, we also plot as a thin, long-and-
short dashed black curve the compositions of
magmas generated by batch melting, assuming a
composition for the low-
d
18
O fluid (44 wt% H
2
O
and 42.6 wt.% Na
2
O) and a productivity of fluxed
melting from
Stolper and Newman
[1994]. These
three models all predict that only small amounts
of fluid (<wt. 1%) are required to produce the
observed range of lava Na contents, but such
small fluid fluxes can produce only
0.1 to 0.2
%
differences between the
d
18
O values of Na-rich and
Na-poor lavas. This is inconsistent with our obser-
vations and suggests that Central American arc
lavas cannot be the products of melting a high-
d
18
O, enriched mantle wedge in response to fluxing
by low-
d
18
O aqueous fluids, at least as envisioned
by these relatively simple models.
[
30
] An alternative is that the slab-derived metaso-
matic agent driving melting in the mantle wedge is
a hydrous melt rather than an aqueous fluid. In this
case, large amounts of that slab melt might be
required to drive high degrees of melting in the
mantle wedge because the amount of melt gener-
ated by addition of hydrous components to peri-
dotite is roughly proportional to the amount of
water added, whatever its form [
Hirschmann et al.
,
1999;
Eiler et al.
, 2000a;
Gaetani and Grove
,
2004]. Previous studies [
Carr et al.
, 1990;
Patino
Figure 6.
Correlation of average
d
18
O
olivine
values versus Na
6.0
values for suites of Central American arc lavas; each
data point is the average for a single volcanic center (reproduced and transposed from Figure 3a). The curves are
trends predicted by models of fluxed melting of fertile peridotite by low-
d
18
O aqueous fluid (solid or thin, long-and-
short dashed curves) or hydrous melt (dashed curves). See text and Appendix A for further details. Symbols are as in
Figure 3.
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
11 of 28
et al.
, 2000;
Leeman et al.
, 1994] have shown that
Central American arc lavas having unusually
strong ‘‘slab’’ signatures (i.e., those from central
Nicaragua) also have dominantly ‘‘fluid-like’’ trace
element signatures (e.g., high Ba/La and U/Th),
and lack clear evidence of ‘‘slab melt’’ signatures
(e.g., high Th and La/Sm [
Elliott
, 2004, and
references therein]), arguing against this hypothe-
sis. These arguments are mitigated in the case of
Nicaragua because sediments on the downgoing
Cocos plate have unusually high Ba/La and U/Th
ratios and unusually low Th concentrations [
Plank
et al.
, 2002], and so partial melts of those sedi-
ments might have trace element compositions
comparable to those usually associated with slab-
derived fluids. Given this ambiguity in the trace
element evidence, we have examined whether
fluxing by slab-derived melts can explain the
correlation of
d
18
O with Na
6.0
.
[
31
] The dashed curves in Figure 6 show the
compositions predicted for magmas produced by
batch melting of peridotite in response to addition
of a hydrous melt. Two models of this kind are
shown (see Appendix A for further details): (1) The
black dashed curve illustrates a model assuming
the productivity for fluxed melting given by
Eiler
et al.
[2000a] (appropriate for a mantle wedge
colder than its dry solidus), a mantle wedge
d
18
O
olivine
of 5.9
%
, and fluxing by slab-derived
melt containing 5 wt.% H
2
O and 1.1 wt.% Na
2
O
and having a
d
18
O value of 0
%
. Other models with
slightly different parameters are similar, but all that
fit the overall trend of our data require large
amounts (
10 wt.%) of a water- and sodium-poor
(
5–10% and
1–2%, respectively) slab melt.
(2) The gray dashed curve illustrates an alternative
model that assumes the wedge is hotter than its dry
solidus (so the degree of melting for a given amount
of water addition is higher), has a
d
18
O
olivine
of
5.6
%
, and is infiltrated by slab melt having 15%
H
2
O, 5.58
%
Na
2
O and a
d
18
Oof0
%
. Both of these
slab-melt-fluxed models capture the key feature
of the Na
6.0
versus
d
18
O
olivine
trend defined by
Central American arc lavas (the significant decrease
in both parameters from the edges to the center of
the arc), whereas models in which the flux is
envisioned to be a hydrous fluid do not do so.
[
32
] Although the models outlined above can fit
the observed relationship between
d
18
Oand
Na
6.0
, we nevertheless regard them as implausible
because the characteristics of the required end-
members are inconsistent with known or expected
properties of the mantle wedge and/or slab-derived
metasomatic agents.
[
33
] 1. The lithologies at the top of the Cocos slab
that one might expect to melt on subduction (i.e.,
metamorphosed hemipelagic sediments at the top
of the plate [
Plank et al.
, 2002]) are expected to
have bulk
d
18
O values of 20–30
%
[
Kolodny and
Epstein
, 1976;
Arthur et al.
, 1983]. Thus such a
component would lead to
increases
rather than
observed
decreases
in
d
18
O with increasing slab
inputs. Low-
d
18
O slab melt could come from
partial fusion of hydrothermally altered rocks
within the slab interior [
Gregory and Taylor
,
1981;
Muehlenbachs
, 1986], but thermal models
of subduction suggest deep interiors of slabs
generally heat enough to de-water but not enough
to melt [
Peacock
, 2004;
Kelemen et al.
, 2004a].
[
34
] 2. The unmodified mantle wedge required by
this model must have a
d
18
O
olivine
comparable to
the highest known for olivines from mantle peri-
dotites and basalts [
Eiler
, 2001]. This is possible,
but known high-
d
18
O mantle-derived materials
[
Eiler et al.
, 1996a, 1996b;
Eiler
, 2001] have excep-
tional properties (such as
87
Sr/
86
Sr ratios
0.705)
that would be inconsistent with the mantle wedge
component inferred by the simple two-component
model shown in Figure 6.
[
35
] 3. According to this model, a large amount
(nearly 20 wt.%) of the low-
d
18
O slab melt would
have to have been added to the sources of Nicar-
aguan lavas, suggesting that they might have
geochemical characteristics resembling other arc
lavas that have been proposed to be derived largely
or entirely from slab melts (e.g., ‘‘adakites’’ and
high-Mg andesites [
Defant and Drummond
, 1990;
Kelemen et al.
, 2004b]). However, Nicaraguan
lavas, in addition to having unusually low concen-
trations of most incompatible trace elements (e.g.,
50 ppm Zr and 5 ppm La) and trace element
ratios usually regarded as indicative of fractiona-
tions between fluid and silicate minerals (e.g.,
U/Th ratios near 1 [
Johnson and Plank
, 1999;
Elliott
, 2004]), lack trace element characteristics
previously suggested to be indicative of direct
partial melts of the subducted slab. For example,
their Th/La ratios are
0.05–0.15, in contrast to
expected partial melts of subducted sediment
(
0.5–1.0 [
Johnson and Plank
, 1999;
Elliott
,
2004]), and their Sr/Y ratios are
15–25, in con-
trast to expected partial melts of eclogitic sources
(
500–1000 [
Kelemen et al.
, 2004b]).
[
36
] We conclude that while it is possible to
construct simple melt-fluxed models that can ex-
plain our observations, the sets of geochemical
characteristics of the end-members of successful
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
12 of 28
models of this kind are implausible. Given the
underlying strength of evidence that oxygen-
isotope variations of Central American basalts are
related to their primary petrogenesis (Figures 1–3),
we proceed by developing a model that is more
complex (i.e., has more than two end-members),
but also capable of explaining all the geochemical
constraints satisfactorily.
4.2. Melting Fluxed by Mixtures of Two
Slab Components
[
37
] To address the difficulties of the two-compo-
nent mixing-melting models presented in Figure 6,
we consider in this section a model involving the
peridotitic mantle wedge and two slab-derived
components, one higher and one lower in
d
18
O
than common mantle peridotites. This model has
much in common with previous hypotheses [
Plank
and Langmuir
, 1993;
Elliott et al.
, 1997;
Elliott
,
2004] that invoke both a sediment-derived melt
(which we expect to be anomalously high in
d
18
O)
and an aqueous fluid derived from dehydration
metamorphism of rocks in the slab interior (some
of which are expected to be anomalously low in
d
18
O). As we demonstrate below, this model can
explain the geochemistry of Central American arc
lavas because all such lavas with strong trace
element signatures of slab-derived components
(e.g., Ba/La
40) are
either
anomalously
18
O-rich
or
anomalously
18
O-poor. Conversely, the
only major group of Central American lavas that are
consistently ‘‘normal’’ in
d
18
O (those from Costa
Rica), consistently have trace element patterns
indicative of small or negligible contributions from
slab-derived components. These observations could
be at least qualitatively explained if the mantle
wedge under the Central American arc were normal
in
d
18
O (i.e., like the sources of most basalts) and
variable proportions of two different slab-derived
components (one high in
d
18
O and the other low in
d
18
O) were added to it. Such a model is also
consistent with the recent suggestion of
Cameron
et al.
[2003] that volcanic-front lavas from Guate-
mala are derived from sources metasomatized by
both sediment melt and fluid evolved from the slab
interior. In this section, we examine this class of
model quantitatively in light of our new oxygen-
isotope data and previously published trace element
and radiogenic-isotope data.
[
38
] As we did for the simpler two-component
models, we examine this hypothesis using an
approach similar to that presented by
Eiler et al.
[2000a]. They assumed a mineralogy and chemical
composition for the mantle wedge,
d
18
O values for
both the mantle wedge and a slab-derived compo-
nent, and a water content for the slab-derived
component. Given these assumptions, and con-
straints on the melting properties of hydrous peri-
dotite from experiments and the MELTS model,
they then estimated the amount of slab-derived
phase required to have been added to the sources
of lavas from the Vanuatu arc in order to explain
simultaneously their
d
18
O values, TiO
2
contents,
and Yb/Sc ratios and the Cr/(Cr + Al) ratios of their
spinel phenocrysts. They then used these results,
combined with the known trace element chemistry
of the lavas and the assumed trace element
chemistry of their mantle wedge sources prior to
metasomatism to solve for the trace element
composition of the slab-derived component added
to the source of each lava. We follow a similar
approach here, but assume two slab-derived com-
ponents and modify some of the assumed prop-
erties of the mantle wedge and slab-derived
components to match better the particulars of
the Central American arc; we also use a different
algorithm to estimate the trace element composi-
tions of slab-derived components based on those
of the lavas.
4.2.1. Model Assumptions
[
39
] 1. The unmetasomatized mantle wedge be-
neath the Central American arc at the depth of
melt segregation is assumed to be a spinel lherzo-
lite (51.6 wt.% olivine; 28.7. wt.% orthopyroxene;
16.0 wt.% clinopyroxene; 3.7 wt.% spinel, as
assumed in the model of
Eiler et al.
[2000a]).
Our default assumption is that the minor and trace
element composition of the mantle wedge is the
same as the mantle wedge source given by
Eiler et
al.
[2000a]; i.e., it has a
d
18
O
olivine
value of 5.0
%
,
and its Sr and Nd isotope ratios are within the
range of normal Pacific and Atlantic MORBs [
Ito
et al.
, 1987]. However, we also include in our
modeling a second mantle wedge component in
order to account for compositions of lavas from
southeast of central Costa Rica, which are geo-
chemically distinct from lavas elsewhere in the arc
in ways that are best attributed to variability in the
mantle wedge (e.g., as shown in Figure 3f they are
higher in La/Sm; they also have distinctive Pb
isotope compositions [
Herrstrom et al.
, 1995;
Carr
et al.
, 2004]).
Herrstrom et al.
[1995] suggested
that Costa Rican lavas are derived from a part of
the mantle wedge resembling the sources of some
ocean-island basalts (OIB). We thus include a
mantle wedge component with such characteristics.
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
13 of 28
The properties of the ‘‘OIB-like’’ mantle wedge
component are based on the assumption that sam-
ple T24-2 from Turrialba (the best-characterized,
high-La/Sm Costa Rican lava examined in this
study) is a 5% batch partial melt of only that
source (we further assume that this melting oc-
curred in the spinel peridotite stability field, and we
normalize both the model and lava compositions to
6% MgO by olivine addition or subtraction to
account for differentiation of T24-2 before erup-
tion). This lava has a
d
18
O
olivine
value of 5.1
%
, and
we assume this value is representative of this OIB-
like source. The compositions of both the assumed
MORB-like and OIB-like mantle wedge compo-
nents are summarized in Table 2.
[
40
] 2. As described above, the key feature of these
models is that the mantle wedge is assumed to be
metasomatized by variable amounts of two slab-
derived components: The first is meant to approx-
imate a partial melt of sediments from the top of
the slab; it has a
d
18
O of +25
%
(similar to the
expected weighted average for pelagic sediments
[
Kolodny and Epstein
, 1976;
Arthur et al.
, 1983;
Plank et al.
, 2002]) and a water content of 10 wt.%
(similar to inferred volatile contents of high-
pressure partial melts of hemipelagic sediments
[
Johnson and Plank
, 1999]). The second is
meant to approximate an aqueous fluid derived
from dehydration metamorphism of hydrothermally
altered gabbros and/or serpentinites in the slab
interior; it is assumed to have a
d
18
O value of
0
%
. This
d
18
O value is at the lower end of the
range for altered layer-3 gabbros and serpentinites
in ophiolites [
Muehlenbachs
, 1986;
Shanks
, 2001]
and similar to values expected for water in equilib-
rium at 400 to 500
C with silicate minerals having
d
18
O values of +3
%
[
Chacko et al.
, 2001] (3
%
is
similar to the average value for ocean lithosphere
hydrothermally altered at high temperatures). We
further assume this component has a water content
of 50 wt.%, similar to estimates for slab-derived
fluids by
Stolper and Newman
[1994] and
Grove et
al.
[2002] and consistent with the high-pressure
fluid-melt solvus of
Shen and Keppler
[1997].
[
41
] 3. Implementation of this model requires
knowledge of the degree of melting of the
mantle wedge as a function of its water content,
including both the trace of water intrinsic to that
mantle and water added to it as a result of
metasomatism by slab-derived components. This
relationship depends on P, T, and solid and fluid
compositions, but we simply fit the average of
the two functions presented by
Eiler et al.
[2000a] for partial melting of peridotite mixed
with ‘‘hydrous melt’’ on one hand and ‘‘aqueous
fluid’’ on the other. Although clearly an over-
simplification, this allows the use of a single
relationship to approximate the melting properties
of peridotites mixed with different proportions of
hydrous melt and solute-rich fluid (and is justi-
fied by the fact that the two functions presented
by
Eiler et al.
[2000a] are so similar). The
Table 2.
Compositions of Model Components
‘‘MORB-Like’’ Mantle Wedge ‘‘OIB-Like’’ Mantle Wedge Low-
d
18
O Slab Phase High-
d
18
O Slab Phase
Ba, ppm
0.317
33.93
2573
614
U, ppm
0.0042
0.09
2.60
0.28
Th, ppm
0.011
0.25
2.75
1.25
K
2
O, wt.%
0.030
0.08
2.10
0.3
Nb, ppm
0.137
1.09
14.7
0.5
La, ppm
0.138
1.89
25.8
9.3
Pb, ppm
0.031
0.22
17.3
7.9
Ce, ppm
0.440
4.23
59.4
17.5
H
2
O, wt.%
0.026
0.25
50.0
10.0
Nd, ppm
0.472
2.29
40.5
11.5
Sr, ppm
5.87
58.61
3071
921
Zr, ppm
5.45
13.44
215
25
Sm, ppm
0.20
0.49
9.7
1.6
Cu, ppm
4.80
5.67
1150
32
Na
2
O, wt.%
0.20
0.24
8.0
2.1
TiO
2
, wt.%
0.17
0.18
0.75
0.1
Yb, ppm
0.48
0.43
0.5
0.5
d
18
O
olivine
5.0
5.1
0.0
25
87
Sr/
86
Sr
0.70250
0.70368
0.7041
0.7040
143
Nd/
144
Nd
0.51315
0.51294
0.5131
0.5128
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
14 of 28
average relationship was fit to the following
polynomial: Melt fraction (F) = 20.3[H
2
O]
350[H
2
O]
2
, where both melt fraction and water
concentration are in weight fraction units (i.e., a
melt fraction of 0.1 indicates 10% partial melt-
ing). Because this function dictates that any
amount of water in the source leads to partial
melting, our ‘‘MORB-like’’ mantle component
will undergo 0.5% partial melting even in the
absence of a slab-derived component, as would
be expected if the mantle wedge is a few tens of
degrees below its dry solidus. This function
generates less melt for a given amount of water
than calculated on the basis of data for Mariana
trough lavas (F = 0.05 + 61[H
2
O];
Stolper and
Newman
, 1994), but more closely resembles
hydrous peridotite melting experiments conducted
near the dry solidus [
Gaetani and Grove
, 2004].
[
42
] If the OIB-like mantle wedge component
has a H
2
O/Ce ratio within the range of mid-
ocean-ridge and back-arc basin basalts (
200–
600, on a weight fraction basis [
Stolper and
Newman
, 1994;
Michael
, 1995], then the water
content of that source would yield
2–5% melting
even without addition of slab-derived components
based on the hydrous peridotite melting function
adopted here. We assume this source contains
0.25 wt.% H
2
O and so undergoes 5% partial melt-
ing, as was assumed when calculating the minor
and trace element composition of that source.
[
43
] 4. The melting reaction was assumed to
have the same stoichiometry as that used by
Eiler et al.
[2000a] for hydrous melting fluxed
by aqueous fluid: Fluid
1
+ Clinopyroxene
9.8
+
Orthopyroxene.
9.9
+ Spinel
1.5
= Melt
14.6
+ Olivine
7.6
(on a mass basis). All melting was modeled as a
batch process, using the equations of
Shaw
[1970] and the distribution coefficients listed in
Appendix A.
4.2.2. Compositions of the Slab-Derived
Components
[
44
] We calculated the minor and trace element
compositions of the two slab-derived components
by assuming that only the high-
d
18
O component
metasomatized the sources of the five highest-
d
18
O volcanic-front lavas from Tecuamburro and
Cerro Verde, and only the low-
d
18
O component
metasomatized the sources of the five lowest
d
18
O lavas from San Cristobal, Las Pilas, and
Cerro Negro. This approach results in the small-
est possible difference in composition between
the high-
d
18
Oandlow-
d
18
O components (i.e.,
their compositions might be more different from
one another than our approach suggests if, in
reality, both are present in the sources of all
studied lavas).
[
45
] For each sample in these two groups, we
first normalized the concentrations of all incom-
patible minor and trace elements to account for
differences in their extents of crystallization dif-
ferentiation by adding or subtracting small incre-
ments of equilibrium olivine until the lava
contained 6 wt.% MgO. These corrections were
generally small (less than 10%, relative). We then
averaged for each group the concentrations of
each of the minor elements Na, K, and Ti, and
the trace elements Ba, Sr, Pb, U, Th, Nb, Zr, La,
Ce, Sm, Nd, Yb, and Cu, and the isotopic
indices
d
18
O
olivine
,
87
Sr/
86
Sr, and
143
Nd/
144
Nd.
We refer to these fractionation-corrected, group
averages as the ‘‘model high-
d
18
O end-member
lava’’ and the ‘‘model low-
d
18
Oend-member
lava’’. Because none of the highest-
d
18
O lavas
were analyzed for Cu, we assume the model
high-
d
18
O lava contains 70 ppm Cu (a round
number slightly lower than concentrations in
other Guatemalan and Salvadoran lavas having
d
18
O
olivine
values of
5.4).
[
46
] We then calculated by mass balance the
amount of high-
d
18
O, slab-derived component that
must be added to the peridotitic mantle wedge to
match the
d
18
O value of the model high-
d
18
O
end-member lava (2.5 wt.%), and similarly the
amount of low-
d
18
O, slab-derived component
required to explain the
d
18
O value of the model
low-
d
18
O end-member lava (3.0 wt.%). On the
basis of the assumed water content of each
component and our melting function for hydrous
peridotite (above), we then calculated the degree
of batch melting required to generate each of the
two model lavas. On the basis of these degrees
of melting, the amount of low- or high-
d
18
O slab-
derived component in the sources of these end-
member lavas, and the assumed composition of the
MORB-like mantle wedge component, we then
solved for the minor and trace element concentra-
tions in the low- and high-
d
18
O slab-derived com-
ponents. These calculations assumed batch melting,
the nonmodal peridotite melting reaction given
above, and the mineral-melt distribution coeffi-
cients given in Appendix A. Finally, we solved
for the Sr and Nd isotope composition of each slab-
derived component required to match the isotopic
compositions of the respective model end-member
lavas to which they contributed. The compositions
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
15 of 28
of the high- and low-
d
18
O slab derived components
are listed in Table 2.
4.2.3. Comparison of Model End-Members
to Materials and Model Components From
Previous Studies
[
47
] The compositions of all mantle-wedge and
slab-derived components used in this model are
plotted in Figure 7, along with previous estimates
of the compositions of slab-derived metasomatic
agents [
Stolper and Newman
, 1994;
Eiler et al.
,
2000a;
Grove et al.
, 2002] and Cocos plate sedi-
ments [
Plank et al.
, 2002].
[
48
] The composition of the MORB-like mantle
wedge (a horizontal line in this figure since it was
used to normalize all other patterns) is similar in
most respects to mantle wedge compositions as-
sumed or calculated in models from
McCulloch
and Gamble
[1991],
Stolper and Newman
[1994],
and the ‘‘depleted source’’ of
Grove et al.
[2002].
The OIB-like mantle wedge is relatively incompat-
ible element enriched, but has a pronounced neg-
ative K anomaly. A negative K anomaly is
characteristic of the ‘‘HIMU’’ type of OIB [
Sun
and McDonough
, 1989], and so our trace element
model is consistent with the
Herrstrom et al.
[1995] suggestion that the enriched component
beneath Costa Rica is a HIMU-like source.
[
49
] The Na
2
O content of the model high-
d
18
O slab-
derived component (2.1 wt.%) is consistent with it
being a high-degree partial melt of hemipelagic
sediments, based on the compositions of Cocos plate
sediments and previous sediment melting experi-
ments [
Johnson and Plank
, 1999;
Plank et al.
,
2002]. Note that although we assumed this compo-
nent was a hydrous melt and set its H
2
O content
(10 wt.%) accordingly, the Na
2
O content was solved
for, so this correspondence with a partial melt of
sediment supports the internal consistency in the
model. The Na
2
O content of the model water-rich,
low-
d
18
O slab-derived component (8.0 wt.%),
though much higher than that of the high-
d
18
O
component, is lower than the calculated water-rich
components of
Stolper and Newman
[1994]
(42.6 wt.%) and
Grove et al.
[2002] (26–32 wt.%),
and higher than the water-rich component assumed
by
Eiler et al.
[2000a] (2 wt.%).
[
50
] The model minor and trace element composi-
tions of the high- and low-
d
18
O slab-derived com-
ponents have many similarities with each other. At
the resolution of the logarithmic scale in Figure 7,
the low-
d
18
O component (i.e., the component as-
sumed to be a fluid rather than a melt) is similar to
model slab-derived components inferred for the
sources of lavas in the Mariana trough [
Stolper
and Newman
, 1994], the Vanuatu arc [
Eiler et al.
,
2000a], and the Cascade arc [
Grove et al.
, 2002]:
All are characterized by significant enrichments in
Ba, Sr, H
2
O, and Pb and depletions in high-field-
strength elements such as Nb, Ti, and perhaps Zr
relative to elements of similar compatibility during
peridotite partial melting. Note that it is possible
that these depletions in high-field-strength ele-
Figure 7.
Abundances of selected minor and trace
elements in model MORB-like and OIB-like mantle
wedge components (heavy solid and dashed gray lines,
respectively), the model high-
d
18
O, water-poor compo-
nent (heavy solid black line), the model low-
d
18
O,
water-rich component (heavy dashed black line), 2.5 Ma
hemipelagic sediment on the Cocos plate (light solid
black line; data from
Plank et al.
[2002]), the model
slab-derived aqueous fluid from
Eiler et al.
[2000a]
(thin solid gray line), model slab-derived component
from
Stolper and Newman
[1994] (thin short-dashed
gray line), and a representative model slab-derived
component from
Grove et al.
[2002] (thin long-dotted
gray line). All concentrations are normalized to those in
our model MORB-like mantle wedge component
(Table 2). Elements are plotted in their approximate
order of increasing compatibility in residual solids
during peridotite partial melting (see Appendix A),
adjusted to be consistent with common convention (Nb
was placed between K and La). For reference, labels for
the lanthanide elements are on the bottom horizontal
axis, and those for all other elements are on the top
horizontal axis.
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
16 of 28
ments in fact reflect depletion of those elements in
the mantle wedge prior to its melting beneath the
arc [e.g.,
Woodhead et al.
, 1993;
Stolper and
Newman
, 1994], rather than properties imposed
on undepleted sources by addition of the slab-
derived component.
[
51
] Figure 7 shows that despite some similarities
(e.g., the elevations in Ba, Pb and Sr relative to
neighboring elements), the model high-
d
18
O com-
ponent is readily distinguishable in its trace and
minor element abundance pattern from the model
low-
d
18
O component. For example, the low-
d
18
O
component is approximately six-times higher in
Cu/Sm ratio, four-times higher in U/Th and K/Th
ratios, and two-times higher in Ba/Th ratio than the
high-
d
18
O component (Table 2). These differences
are broadly consistent with the low-
d
18
O component
being an aqueous fluid and the high-
d
18
O component
being a hydrous silicate melt [
Brenan et al.
, 1995;
Elliott et al.
, 1997;
Johnson and Plank
, 1999;
Elliott
,
2004], as we assumed when choosing their water
concentrations. The trace element abundances of the
model high-
d
18
O component are particularly close to
those of sediments being subducted on the Cocos
plate [
Plank et al.
, 2002] (concentrations of all
plotted elements are within 25% of each other, on
average), strongly supporting our inference that this
component represents a high-degree partial melt of
such sediments. Moreover, the pattern of the model
low-
d
18
O component is also broadly similar to these
sediments, consistent with their being the sources of
much of the trace element inventory of this slab-
derived component as well. The high concentrations
of
10
Be (which must come from recently subducted
sediment) in many Central American lavas [
Morris
et al.
, 1990] support our suggestion that sediment is
the source of much of the trace element inventory of
slab-derived components, as does recent evidence
that spatial and temporal variations in trace element
compositions of Central American arc lavas mimic
those in Cocos plate sediments [
Patino et al.
, 2000;
Plank et al.
, 2002]. It is, however, not so easy to
reconcile the radiogenic-isotope composition of our
slab-derived components with these sediments (see
section 4.2.5).
4.2.4. Model Algorithm and Application
to the Central American Arc
[
52
] To this point, although we have described how
we used the compositions of extremes in the
Central American sample suite to define chemical
and isotopic characteristics of model slab-derived
components in this system, we have not actually
demonstrated that our model can successfully ac-
count for the compositional variations of other
lavas from the Central American arc. In this
section, we examine the degree to which our model
can explain the geochemical properties of all
Central American arc lavas examined in this study,
and go on to examine the geographic distribution
of model components implied by spatial trends in
the compositions of those lavas (Figure 1).
[
53
] We first calculated an average, fractionation-
normalized composition for each of the 16 volcanic
centers in the volcanic front examined in this study.
For relatively well-studied suites, this was done
empirically by regressing the concentrations of all
elements of interest as a function of wt.% MgO
and calculating the value of that regression line at
6 wt.% MgO. Some suites of samples were not
sufficiently well characterized to calculate mean-
ingful regressions, so in those cases we corrected the
compositions of each lava from that suite to 6 wt.%
MgO by addition or subtraction of small increments
of equilibrium olivine and then averaged the oliv-
ine-adjusted concentrations. Although this exercise
results in internal consistency and was thus in our
view worth doing, in fact these normalizations
generally involved small changes and have little
impact on our model fits.
[
54
] We then modeled each of the fractionation-
normalized, suite-average compositions as a partial
melt of a mixture of high- and low-
d
18
O slab-
derived components and a mantle source containing
variable amounts of the MORB-like or OIB-like
mantle components (Table 2). The best-fit amounts
of these four components in the source of each suite-
average were determined by initially defining there
to be 0.1 wt.% of each slab-derived component in the
source, and a ratio of MORB-like to OIB-like mantle
components in that source needed to match the
La/Sm ratio of the suite in question. We then
iterated the abundances of each slab component
and the OIB-like component (subject to the
constraint that the amounts of the four compo-
nents had to sum to 100%) to minimize the
misfit between model and data. In each iteration,
the degree of melting of the mixed source was
set by its bulk water content (as described in
section 4.2.1) and all calculations assumed batch
melting, the melting-reaction stoichiometry given
above, and the mineral-melt distribution coeffi-
cients given in Appendix A. A successful fit was
taken as one that produced a model partial melt
that matched the fractionation-normalized suite
average simultaneously to within the following
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
17 of 28
tolerances: <0.1
%
in
d
18
O; <10
4
in
87
Sr/
86
Sr and
143
Nd/
144
Nd; <10% in both the Na
2
O concentra-
tion and La/Sm ratio; and the concentrations of
other minor and trace elements shown in Figure 7
within <20%, on average. It was often possible to
find fits to the suite averages that were factors of two
or more closer matches in all these criteria, but there
were also cases where we failed to find solutions
within one or more of these limits. These exceptions
are discussed below. Figures 8 and 9 compare the
best-fit model compositions to the fractionation-
normalized, suite-average compositions for all of
the 16 volcanic-front suites examined in this study,
including the calculated versus observed trends in
plots of
d
18
O
olivine
versus Na
6.0
(Figure 8a) and
87
Sr/
86
Sr versus
143
Nd/
144
Nd (Figure 8b), and the
calculated versus observed trace element abundance
patters of each suite (Figures 9a–9p). Figure 10
shows the best-fit proportions of the source
components and the degree of partial melting
for the 16 suits as a function of distance along
the arc.
[
55
] The model we present succeeds at simulta-
neously explaining all the noteworthy features of
geochemical variability of Central American arc
lavas, including the minor and trace element abun-
dances and the O, Sr, and Nd isotope compositions
of basalts and basaltic andesites distributed along
the full length of the arc. Note that the model is fit
by varying only three independent variables (i.e.,
the abundances of three of the four source compo-
nents, since their sum is fixed at 100%), yet
succeeds in describing variations in many elemen-
tal abundances and isotope ratios among 16 differ-
ent sample suites. Although this success is not
surprising for the five volcanic centers that contain
one or more the eleven samples used to constrain
the properties of the high and low-
d
18
Oslab
components and the OIB-like mantle source, there
are 40 samples that were not included in these
constraints, and 11 volcanic centers that contained
none of the samples used to calculate model end-
member lavas.
[
56
] Several aspects of the comparison between the
model and suite average compositions are of par-
ticular interest:
[
57
] 1. Most Nicaraguan lavas are best fit by fluxed
melting of a mantle wedge dominated by (>90%)
the MORB-like peridotite source, driven by addi-
tions of
1–4 wt.% of the water-rich, low-
d
18
O
slab-derived component and smaller amounts
(
0–0.5 wt.%) of the water-poor, high-
d
18
O slab-
derived component.
[
58
] 2. Most Guatemalan and El Salvadoran lavas
are best fit by fluxing a mantle wedge also dom-
inated by the MORB-like component, but contain-
ing, on average, slightly more enriched (OIB-like)
component than that beneath Nicaragua, driven by
addition of
1.5–2.4 wt.% of the water-poor,
high-
d
18
O slab-derived component and traces (0–
0.1 wt.%) of the water-rich, low-
d
18
O slab-derived
component. Samples from Apastapeque (Figure 9l)
have
d
18
O
olivine
values near 5.1
%
and require a
similar mantle wedge but fluxed by somewhat
more of the low-
d
18
O component and significantly
less of the high-
d
18
O component.
[
59
] 3. Costa Rican lavas from Platanar and volca-
nic centers southeast of it require partial melting of
a mantle wedge dominated by (>98%) the OIB-like
source, to which traces (
0–0.3 wt.%) of high-
and/or low-
d
18
O slab derived component were
added.
[
60
] 4. Tenorio, in northwest Costa Rica, requires
partial melting of a source made up mostly of
MORB-like peridotite, but with a substantial
(20%) contribution of OIB-like peridotite and sig-
nificant amounts of the high-
d
18
O (0.9 wt.%) and
low-
d
18
O (0.4 wt.%) slab-derived components.
This volcanic center thus appears to require a
Figure 8.
Comparison of the fractionation-corrected,
suite-average Sr, Nd, and O isotope compositions and
Na
6.0
values of Central American arc lavas (solid
circles) with those fit by the model described in the
text (open circles).
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
18 of 28
source transitional between the dominantly OIB-
like mantle beneath Costa Rica and the dominantly
MORB-like mantle beneath the rest of the arc, and
between the abundant slab-derived components
seen in the middle of the arc and the minimal
slab-derived components characteristic of its south-
east end.
[
61
] Our model implies that the amount of the low-
d
18
O, water-rich, slab-derived component is rela-
Figure 9.
Comparison of the fractionation-corrected, suite-average trace element concentrations of Central
American arc lavas (black lines) with those fit by the model described in the text (gray lines). All patterns are
normalized to the MORB-mantle composition used in the modeling. Lines connect data and model-fit compositions
for each suite. These fits are the same as those used to generate model points in Figure 8. Each panel also lists the
volcanic edifice and nation where that suite comes from, and its distance along the arc from the Mexico-Guatemala
border (to aid comparisons with Figures 1 and 10).
Geochemistry
Geophysics
Geosystems
G
3
G
3
eiler et al.: central american arc lavas
10.1029/2004GC000804
19 of 28