Geochemical heterogeneity within mid-ocean ridge lava £ows ...

Earth and Planetary Science Letters 188 (2001) 349^367

www.elsevier.com/locate/epsl

Geochemical heterogeneity within mid-ocean ridge lava £ows:
insights into eruption, emplacement and global variations in

magma generation

K.H. Rubin aY*, M.C. Smith a, E.C. Bergmanis a, M.R. Per¢t b, J.M. Sinton a,
R. Batiza aYc

a Department of Geology and Geophysics, University of Hawaii, Honolulu, HI 96822, USA
b Department of Geological Sciences, P.O. Box 112120, Williamson Hall, University of Florida, Gainesville, FL 32611, USA

c MGG Division of Ocean Sciences, National Science Foundation, 4201 Wilson Blvd., Arlington, VA 22230, USA

Received 5 September 2000; accepted 28 March 2001

Abstract

Compositional heterogeneity in mid-ocean ridge (MOR) lava flows is a powerful yet presently under-utilized
volcanological and petrological tracer. Here, it is demonstrated that variations in pre- and syn-eruptive magmatic
conditions throughout the global ridge system can be constrained with intra-flow compositional heterogeneity among
10 discrete MOR flows. Geographical distribution of chemical heterogeneity within flows is also used along with
mapped physical features to help decipher the range of conditions that apply to seafloor eruptions (i.e. inferred vent
locations and whether there were single or multiple eruptive episodes). Although low-pressure equilibrium fractional
crystallization can account for much of the observed intra-flow compositional heterogeneity, some cases require
multiple parent magmas and/or more complex crystallization conditions. Globally, the extent of within-flow
compositional heterogeneity is well correlated (positively) with estimated erupted volume for flows from the northern
East Pacific Rise (EPR), and the Mid Atlantic, Juan de Fuca and Gorda Ridges; however, some lavas from the
superfast spreading southern EPR fall below this trend. Compositional heterogeneity is also inversely correlated with
spreading rate. The more homogeneous compositions of lavas from faster spreading ridges likely reflect the relative
thermal stability and longevity of sub-ridge crustal magma bodies, and possibly higher eruption frequencies. By
contrast, greater compositional heterogeneity in lavas at slower spreading rates probably results from low thermal
stability of the crust (due to diminished magma supply and greater hydrothermal cooling). Finally, the within-flow
compositional variations observed here imply that caution must be exercised when interpreting MOR basalt data on
samples where individual flows have not been mapped because chemical variations between lava samples may not
necessarily record the history of spatially and temporally distinct eruptions. ß 2001 Elsevier Science B.V. All rights
reserved.

Keywords: eruptions; lava £ows; igneous rocks; geochemistry; magmas; mid-ocean ridge basalts; petrology; submarine volcanoes;
volcanism

* Corresponding author. Tel.: +1-808-956-8973; Fax: +1-808-956-5512; E-mail: [email protected]

0012-821X / 01 / $ ^ see front matter ß 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 0 1 2 - 8 2 1 X ( 0 1 ) 0 0 3 3 9 - 9

350 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

1. Introduction and background a fraction of those have the geologic information
necessary to place them within the context of a
Mid-ocean ridges (MORs) are dynamic vol- speci¢c eruption.
canic environments. Although remote and di¤cult
to observe, MOR volcanism produces lavas that Since the earliest studies of oceanic volcanism,
cover roughly 2/3 of Earth's surface, and MOR MORBs have been known to be less composition-
crust is an important chemo-physical interface be- ally variable than lavas from other Earthly vol-
tween the mantle, ocean, and marine biosphere canic terrains (including subaerial sites of primar-
(e.g. [2^4]). ily tholeiitic extrusion). The lack of broad
compositional variability coupled with the di¤-
To completely understand the physical and culty in obtaining samples with a clearly de¢ned
chemical history of MORs, it is important to context has resulted in spatial variation in com-
study them over a range of temporal and geo- positions of individual MORB samples commonly
graphic length scales, including those related to being used literally to ascribe spatial and/or tem-
the emplacement of individual eruptive units. poral variability to petrogenetic processes and/or
MOR lava £ows, like those in other volcanic ter- mantle compositions. There has been little at-
rains, have complex emplacement histories and tempt to distinguish petrogenetic heterogeneity
preserve information about processes from initial between £ows from heterogeneity within a £ow.
melt generation to post-emplacement cooling. Yet, the signi¢cance of broader scale spatial and
Through comparisons of individual £ows from temporal variations in lava composition cannot be
di¡erent MORs, intra-£ow compositional hetero- completely assessed without understanding both
geneity is used here to constrain variations in the e¡ects of the eruptive process itself (in terms
magmatic/petrologic conditions through the glob- of creating or destroying magmatic compositional
al ridge system. Furthermore, the spatial distribu- variability) and the extent to which the composi-
tion of chemical heterogeneity within a £ow is tional variability in MORB is accounted for by
used along with mapped physical features to pre- and post-eruptive petrologic histories of sin-
help decipher the cryptic history and range of gle eruptive units.
conditions that apply to eruption and emplace-
ment of speci¢c submarine lava £ows. An active MOR eruption has yet to be wit-
nessed. Still, the range of geomorphic features
Numerous studies of MOR basalt (MORB) that constitute an individual submarine lava
compositions over the past four decades have pro- £ow have recently begun to be recognized
vided insights into the formation of the oceanic through a combination of detailed near-bottom
lithosphere and, because MORB is largely unaf- surveying, active monitoring and fortuitous arriv-
fected by crustal contamination, how mantle com- al at the scene of recent volcanism (e.g. [8^12]).
positions/heterogeneities are sampled by volcan- Such features are easiest to identify in very recent
ism (e.g. [5^7] and references therein). However, lavas because weathering and sedimentation can
MOR data have mostly been collected at coarse/ rapidly obscure £ow contacts between older lavas.
incompletely resolved length and spreading time As a result, the database of mapped discrete sub-
scales (10 s of km and kyr), without the bene¢t of marine lava £ows (or £ow sequences) is still small
detailed geological maps needed to place volcanic and limited primarily to young £ows (e.g. [7,13]).
and other structures in the context of individual We know little of the composition, mode of em-
eruptions. MOR volcanoes are relatively inacces- placement or aerial extent of s 99% of the lava
sible and are primarily sampled from surface ships £ows within even the most active/youngest region
(e.g. wax coring or rock dredging), which is of the global MOR system. In most cases, there
broadly equivalent to sampling Hawaiian volca- are inadequate vertical cross-sectional exposures
noes from a helicopter £ying at 2 km altitude. with which to identify physical variations arising
Only a small percentage of MORB samples in from multiple emplacement lobes [14] and in£a-
the literature have been collected using in situ tion/cooling episodes [15]. Instead, MOR £ows
observation (by submersible or ROV), and only are typically observed and sampled from the

K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367 351

upper crust or in collapse features. Nevertheless, implies no compositional heterogeneity outside of
detailed study of the MOR lava £ow database analytical uncertainty. Higher values indicate an-
provides useful, albeit preliminary, insights into alytically signi¢cant compositional variation. HI
MOR processes. is referred to here as the `heterogeneity index',
since this name seems more appropriate for an
2. Compositional heterogeneity of subaerial and index for which larger values correspond to great-
MORB lava £ows er compositional variance.

Physically, most lava £ows are complex, with The original HI study included 10 major ele-
few, if any, known examples of a simple basaltic ments (Si, Ti, Al, Fe, Mn, Mg, Ca, Na, K, P)
£ow (one that is not divisible into £ow units hav- and seven trace elements (Nb, Zr, Sr, Zn, V, Ni,
ing distinguishable extrusion and cooling histories Cr) analyzed by XRF on 80^100 g rock powders
[14]). Not surprisingly, their compound character of 5^12 samples per Mauna Loa £ow [1]. HI
leads to small-scale internal compositional varia- ranged from 1.1 to 13.9, except for picrite erup-
tions. Early works on subaerial basalt £ows de- tions in 1852 (HI = 30.3) and 1868 (HI = 34.0) that
¢ned heterogeneity by variations in phenocryst had copiously accumulated olivine. Composition-
populations (e.g. [14,16]) and liquid chemistry al variability was not a simple function of em-
(e.g. [16^18]), and led some authors to caution placement rate or time, but larger area £ows
against characterizing lava £ow composition were typically more heterogeneous [1]. Roughly
with only one or a few samples. Such heterogene- 50% and V85% of the compositional heterogene-
ity was attributed to a combination of emplace- ity in the non-picrites and the picrites could
ment mechanism, assimilation, £ow di¡erentia- be mathematically `eliminated' by correcting to
tion, in situ fractionation, localized oxidation, constant MgO for low-pressure crystal frac-
zoned magma chambers and random mixing. tionation or accumulation (with only olivine on
These early studies also noted that although the liquidus) [1]. The remaining heterogeneity
only loosely quanti¢ed, analytical errors contrib- (HI = 1^5) was presumed to re£ect other pro-
ute to apparent heterogeneity between multiple cesses. Some Kilauea lavas with smaller £ow areas
samples of a £ow. were also noted to be more heterogeneous than
those at Mauna Loa, suggesting that these other
The ¢rst systematic attempt to quantify `extra- processes were related to the scale and com-
analytical' chemical heterogeneity in lava £ows plexity of crustal magma reservoirs at both volca-
and to evaluate the source(s) of this composition- noes [1].
al variability examined 16 historical basaltic erup-
tions from 1843 to 1975 at Mauna Loa, Hawaii, As on land, compositional variability in a
USA, using a measure known as the `homogeneity MORB £ow can re£ect spatial/temporal variabil-
index' (HI) [1]. The HI uses multiple chemical ity in source composition, melting, crystallization/
constituents in di¡erent lava £ow samples, nor- assimilation histories, and evolution during em-
malized by the precision to which they can be placement. Previous studies of individual MOR
measured. It is de¢ned as: lava £ows [7,10,13,19^27] suggest that composi-
tional ranges within a single £ow can be substan-
HI ˆ 4…Si †a…Pi † tial relative to the range of compositions along its
n respective ridge segment (i.e. Fig. 1), and element^
element variations within MOR £ows do not al-
where Si and Pi are the standard deviation (2c) ways follow expected crystallization trends (e.g.
about the mean and analytical precision for ele- smooth variations with MgO).
ment i, and n is the number of elements used.
Precision is quanti¢ed by replicate analyses on a Intra-£ow compositional variability in MORB
single sample or standard that is similar in com- was discussed qualitatively using the variance of
position to the unknowns. An index value of 9 1 individual chemical elements in Icelandic [28] and
submarine [7,24] lava £ows. Later, HI was used to
make the ¢rst quantitative comparison of the

352 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

Fig. 1. Major element oxide compositions plotted versus MgO content for three individual MOR lava £ows from the NE Paci¢c
Ocean. Individual £ow data are shown as ¢lled circles (1996 N. Gorda Ridge and uncertain-aged N. Cleft Sheet £ow) and open
triangles (mid 1980s Cleft Pillow Mounds). Typical 2c errors are depicted by the cross symbol in each panel. Filled ¢elds repre-
sent `whole ridge segment' compositional data from the vicinity of these lava £ows for comparison (references given in the ¢g-
ure). Details about these and other MOR £ows discussed in this paper are given in the EPSL Online Background Dataset1.

compositional variability within a newly erupted strong linear correlation of HI and £ow volume
(1996) lava £ow on the North Gorda Ridge and (R2 s 0.95), despite potential errors in estimating
¢ve other discrete MORB lava £ows from the both parameters (discussed below). The spatial
literature [26]. HI values were based on 8^24 sam- scales and compositional attributes of the intra-
ples per £ow, using electron microprobe major £ow heterogeneity were not discussed. The ¢elds
element glass analyses (trace element data were for MORB and Mauna Loa overlap on a HI ver-
not used because they were not available for all sus estimated £ow volume plot [26], although the
£ows). The pre-eruptive histories of those MOR Mauna Loa ¢eld extends to greater values of each
magmas were found to be generally more variable (Mauna Loa £ow volumes from [29] and HI from
as erupted volume increased, as indicated by a [1,30]).

Table 1
Physical and compositional data for some MOR lava £ows

Flow name Location Reference Volume Number of SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2O K2O P2O5 HI

samples
(106 m3)

1996 N. Gorda Gorda 42³40PN a, 1 18 12 mean wt% 51.04 1.26 15.81 8.57 0.13 8.37 12.26 2.85 0.10 0.11 2.06
5 21 0.56 0.09 0.54 0.41 0.02 0.35 0.34 0.15 0.01 0.02
1991-2 BBQ EPR 9³50PN b, 2 0.03 8 Si (2c) 0.27 0.03 0.18 0.18 0.02 0.15 0.23 0.07 0.01 0.01 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367
Pi (2c)
ODP EPR 9³30PN c, 3 10.5 12 mean wt% 49.97 1.27 15.51 9.32 0.18 8.54 12.20 2.57 0.10 0.14 1.55
8.7 24 0.30 0.06 0.25 0.23 0.02 0.32 0.11 0.13 0.03 0.06
N. Cleft `Young JdFR 44³56PN d, 4 Si (2c) 0.38 0.06 0.13 0.18 0.02 0.14 0.13 0.12 0.01 0.02
Sheet Flow' 60 22 Pi (2c)
200 53 mean wt% 50.39 1.66 14.39 10.58 0.20 7.08 11.90 2.60 0.14 0.13 1.26
1993 Coaxial JdFR 46³31PN e, 5 140 51 0.21 0.03 0.14 0.10 0.02 0.08 0.03 0.06 0.005 0.01
26 11 Si (2c) 0.31 0.05 0.14 0.20 0.02 0.13 0.16 0.09 0.02 0.02
Serocki Volcano MAR 22³55PN f, 6 55 12 Pi (2c)
mean wt% 50.28 1.57 14.49 10.80 n.a. 7.38 11.95 2.47 0.14 0.15 2.04
Animal Farm EPR 18³37PS g, 7 0.75 0.07 0.18 0.37 n.a. 0.26 0.30 0.08 0.03 0.02
Si (2c) 0.25 0.02 0.25 0.22 n.a. 0.13 0.07 0.04 0.02 0.04
Aldo-Kihi EPR 17³30PS h, 8 Pi (2c)
mean wt% 50.48 1.68 13.55 12.77 0.24 6.73 11.32 2.67 0.12 0.15 1.64
Moai EPR 18³11PS i, 9 0.23 0.05 0.19 0.21 0.02 0.11 0.11 0.06 0.01 0.02
Si (2c) 0.25 0.02 0.10 0.09 0.01 0.08 0.17 0.08 0.01 0.01
N. Cleft Pillow JdFR 45³05PN j, 10 Pi (2c)
Mounds mean wt% 50.04 1.68 15.84 9.98 n.a. 7.50 11.20 2.94 0.14 0.17 3.27
0.61 0.24 0.70 0.82 n.a. 0.62 0.29 0.14 0.03 0.05
Si (2c) 0.37 0.07 0.08 0.22 n.a. 0.11 0.16 0.09 0.03 0.03
Pi (2c)
mean wt% 50.57 1.66 14.45 10.57 0.20 7.27 11.72 2.96 0.09 0.14 1.20
0.29 0.05 0.19 0.20 0.02 0.13 0.14 0.08 0.01 0.01
Si (2c) 0.29 0.06 0.18 0.16 0.02 0.15 0.14 0.11 0.00 0.01
Pi (2c)
mean wt% 50.25 1.53 14.77 9.93 0.18 7.88 12.20 2.53 0.11 0.13 2.00
0.45 0.13 0.42 0.42 0.02 0.40 0.27 0.17 0.02 0.02
Si (2c) 0.28 0.05 0.12 0.24 0.03 0.14 0.14 0.12 0.01 0.01
Pi (2c)
mean wt% 49.78 1.35 15.36 9.62 0.18 8.12 12.52 2.62 0.08 0.11 1.62
0.27 0.10 0.25 0.32 0.02 0.27 0.18 0.17 0.01 0.02
Si (2c) 0.29 0.05 0.12 0.16 0.02 0.10 0.19 0.14 0.01 0.01
Pi (2c)
mean wt% 50.64 1.54 14.45 10.54 0.20 7.69 11.73 2.52 0.18 0.14 3.05
0.24 0.26 0.55 0.90 0.02 0.69 0.33 0.14 0.02 0.04
Si (2c) 0.43 0.03 0.15 0.13 0.01 0.19 0.17 0.11 0.02 0.02
Pi (2c)

All data collected by electron microprobe analysis of glass chips (¢ve spots per sample except 1996 N. Gorda samples, which are three spots per sample). Loca-
tion/volume references (letters): a. [27]; b. [26]; c. (see EPSL Online Background Dataset1); d. [22]; e. [27]; f. [23]; g^i. [13]; j. [27]. Composition references (num-
bers): 1. [28]; 2. [7] and Per¢t, unpublished; 3. (see EPSL Online Background Dataset1); 4. [29]: 5. [29]; 6. [21]; 7^9. [13] and Bergmanis et al., unpublished data;
10. [42] and Smith et al., unpublished data. Additional metadata can be found in the EPSL Online Background Dataset1.

353

354 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

2.1. The MORB HI dataset

A small, 10 lava £ow dataset forms the basis of Fig. 2. (a) HI (de¢ned in the text) versus estimated £ow vol-
the present study. Four mapped lava £ows have ume for 10 individual MOR lava £ows. Note the high degree
been added since our initial investigation [26]: the of correlation for all seven £ows (¢lled diamonds) from lo-
mid 1980s Cleft Pillow Mounds eruption on the cales other than the superfast spreading S-EPR (open
Juan de Fuca Ridge (JdFR) [8] and three repre- squares). Two of three £ows from the latter area are signi¢-
sentatives from a group of recently mapped £ows cantly less compositionally heterogeneous. (b) The range of
on the superfast spreading southern East Paci¢c MORB values relative to subaerial tholeiitic rift zone lavas.
Rise (S-EPR) [13]. HI, location and volume data For comparison, 14 historical lavas from Mauna Loa volca-
for each £ow are in Table 1; physical, composi- no plus the ongoing Puu Oo eruption of Kilauea volcano
tional and age metadata are in table 1 in the (both in Hawaii) are plotted as data points, and our estimate
EPSL Online Background Dataset1. Flow volumes of the range of tephras from the 1783^1784 Laki eruption
are very di¤cult to accurately estimate [7]. Vol- (in Iceland) is plotted as a vertical bar (data sources de-
umes (from the literature) were derived from ei- scribed in the text). Note that despite the considerable scat-
ther (£ow area)U(estimated thickness), depth-dif- ter, the broad relationship between estimated £ow volume
ference anomalies before and after eruption, or and compositional heterogeneity is similar for the Hawaiian,
both. Generally, £ow area can be constrained to Icelandic and MOR datasets.
within 10^20% error by mapped boundaries (W.
Chadwick, personal communication). There can 3. Quantifying compositional heterogeneity
be more error in volume from thickness estimates
(potentially 50^100%) and unaccounted for inter- Before interpreting the MORB HI results, it is
nal porosities (which locally can be up to 70%, important to address the appropriateness and ac-
largely hidden beneath the upper crusts of some curacy of the index for measuring true composi-
£ows from lava drain-out during emplacement). tional variability within a £ow. The logic of using
Propagated, these individual (maximum) errors multiple chemical elements is that the HI then
result in potential volume uncertainties of V85^ integrates over a range of processes that might
125%. a¡ect individual elemental abundances in varying

Adding the Cleft Pillow Mounds to the pre-
vious HI database improves the correlation with
estimated £ow volume (R2 = 0.964 ; Fig. 2a). Two
of three S-EPR £ows have signi¢cantly lower
compositional heterogeneity for their volume;
however, they do lie within the broad range of
historical Mauna Loa lavas (Fig. 2b). The largest
S-EPR £ow (`Animal Farm') is very composition-
ally homogeneous (HI = 1.20) and would fall o¡
the Fig. 2a trend even if the volume were errone-
ously large by 300% (which is outside of the un-
certainty discussed above). This result is inter-
preted in a later section.

1 http://www.elsevier.nl/locate/epsl; mirror site: http://
www.elsevier.com/locate/epsl

K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367 355

proportion. Mapping lava £ow compositional het- term `np-HI') is not well correlated with HI
erogeneity onto a single variable is useful for ex- (R2 = 0.38 ; Fig. 3a), although this is dominated
amining temporal^spatial evolution of an erup- by non-systematic variations in three typically
tion, but study of individual element variations low-precision elements (Mn, P and K). Removing
is also necessary to determine the source(s) of
the heterogeneity. Fig. 3. Various other means of estimating chemical heteroge-
neity plotted versus the HI (symbols as in Fig. 2). a: Np-HI
True analytical precision is di¤cult to con- (index of non-precision-normalized standard deviations, Si,
strain, yet consideration of such uncertainty in a for all 10 basaltic major elements). b: Np-HI* (similar index
dataset is preferable to using `raw' data to exam- without the three typically lowest precision elements Mn,
ine `true' compositional signals. A rigorous meth- P and K). c: HI data that have been corrected using the Stu-
od of accounting for analytical uncertainty is par- dent's t distribution based on the number of samples and
ticularly important where compositional variance standards data. The Serocki precision dataset was based on
between and within £ows is small relative to an- only four analyses (each the mean of ¢ve spots), which is
alytical error (i.e. most MORB) and when com- why it appears somewhat anomalous in this panel (open dia-
paring data produced by di¡erent laboratories ex- mond).
hibiting a range of instrumental and analytical
qualities (i.e. the case here). Reproducibility anal-
yses of standard materials that are similar in com-
position to the unknowns are probably the best
uncertainty measure. Instrument-related variabil-
ity can a¡ect precision even when a single oper-
ator makes analyses on a sample suite over only a
few months, so `Pi' values are best constrained
simultaneous to unknowns analysis using at least
as many standard measurements as unknown
measurements. `Historical' laboratory precision
values (accumulated over a longer time period)
are not as desirable because instrument drift
that does not apply to a particular set of un-
knowns might occur (note slight variations in
standards reproducibility between groups of anal-
yses made over 4 months at the University of
Hawaii in the Animal Farm, Aldo-Kihi and
Moai data, Table 1).

It is important to investigate if analytical error
correction using Pi is introducing uncertainty or
biases into the data. In this MORB dataset, pre-
cision-normalization had minimal e¡ect on the
relative apparent compositional heterogeneity of
individual elements (2c): non-precision-normal-
ized (i.e. Si) and precision-normalized (i.e. Si/Pi)
data are well to moderately well correlated for
most elements in Table 1 (Si vs. Si/Pi of Mg, Ti,
Fe, Al, Si have R2 = 0.7^0.85 and Ca, P have
R2 = 0.5). However, a few elements are only
poorly correlated (Na, K, Mn have R2 6 0.2).
An index of combined non-precision-normalized
elemental heterogeneity (e.g. 4Si/n, which we

356 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

the latter from the index (termed `np-HI*') im- changes by 25%, primarily because Pi is based on
proves the correlation with HI signi¢cantly only six, seven and four standards, respectively.
(R2 = 0.82; Fig. 3b). Apparently, the subset of el- HI and T-factor-adjusted HI are well correlated
ements in np-HI* preserves aspects of the ¢rst- (R2 = 0.84, improving to 0.96 without the Serocki
order compositional variability, however, remov- datum; Fig. 3c). The basic correlation of esti-
ing incompatible major elements from the index mated £ow volume and compositional heteroge-
might sacri¢ce important information. Np-HI and neity is unchanged by sample size correction, so
estimated £ow volume are not well correlated small (nV10) sample sets may be adequate
(R2 = 0.25), although np-HI* and volume are (although not ideal) to de¢ne compositional het-
(R2 = 0.86), the scatter in the latter being in the erogeneity in typical cases.
most homogeneous £ows. However, precision-
normalized HI and estimated £ow volumes are A third issue is whether parametric standard
even more strongly correlated (R2 = 0.964), which deviations [1] are the best mathematical descrip-
is not a mathematical artifact of the precision- tion of compositional (Si) and analytical (Pi) var-
normalization because 4Pi/n of each £ow and vol- iability among multiple samples of a single lava
ume are not correlated (R2 = 0.0034 and 0.069, £ow. This method presumes that chemical varia-
with and without Mn, P and K, respectively). bility is normally distributed, which may not be
To summarize, there is a heterogeneity signal in entirely true for variations amongst a group of
the raw compositional data (uncorrected for ana- elements that are at least partially correlated.
lytical precision), but precision-normalization us- With extremely small sample datasets displaying
ing replicate analyses helps to adjust for variable limited compositional diversity, it is also possible
data quality at the elemental and inter-laboratory that some Si ranges are Poissonian (e.g. dominat-
scales. ed by rare `events'). It is particularly important to
test for such behavior because the di¤culty with
A second issue is how many samples are re- which submarine lava £ows are mapped and
quired to con¢dently determine the compositional sampled might lead to accidental inclusion of
variability within a single lava £ow. A su¤cient samples that are actually not from that £ow. At
number of sample and standard measurements to the same time, local variations in cooling history
encompass the actual data distribution is required may result in real Poissonian distributions if an
to derive a statistically signi¢cant measure of the anomalous sample representing a volumetrically
true chemical variation. The appropriate number insigni¢cant part of a £ow is included in a small
of samples for Si is also likely de¢ned in part by dataset.
£ow size and physical morphology, and samples
must be spatially well distributed to be represen- The importance of `rare events' in the data was
tative of the variability length scales within a £ow. investigated by using ranges (as opposed to stan-
Many of the existing MOR (n = 8^60) and Mauna dard deviations) for either or both terms in the
Loa (n = 3^12, except n = 67 for the 1984 lava) HI. In each case (HI based on range/range, sdev/
individual £ow datasets are not ideal for this pur- range or range/sdev) these heterogeneity indices
pose because they are small, as are many of the are only moderately correlated with the normal
associated standard replicates (n = 4^25 for the (sdev/sdev) HI, suggesting that the compositional
MOR unknowns). True variability can theoreti- extremes within the sample and standards data do
cally be gleaned from small sample sets using not dominate the variability signal : range/range
the Student's t distribution, which provides con- slope = 0.90, R2 = 0.66, range/sdev slope = 0.93,
¢dence levels from low sample populations by R2 = 0.76, sdev/range slope = 0.61, R2 = 0.36.
broadening the 2c width in inverse proportion Each of these indices also resulted in a poorer
to the number of samples. Such correction to correlation with estimated £ow volume. Thus,
both Si and Pi has little e¡ect on most HI values parametric standard deviations appear to ad-
(1^6%); but, HI of the North Cleft Sheet and equately measure Si and Pi. In undertaking the
1993 Coaxial £ows change by V12%, and Serocki range analysis we identi¢ed and then rejected
one spurious lava in the Serocki dataset, changing

K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367 357

HI from 3.53 to 3.27 (sample 1690-5, K/ most undi¡erentiated (MgO-rich) sample in each
Ti s 200% of the mean of the other 21 samples). £ow as parent compositions. Two stage liquid
lines of descent (LLDs) best corresponding to ob-
Finally, there is the issue of how many and served data trends of each £ow were used to
which elements to include in the HI, which in `back-out' crystallization e¡ects on major element
the present case was limited by the available data- compositions of each sample to the parent MgO.
sets. More elements can be added but one should The most evolved stage (a¡ecting one or two sam-
consider petrogenetic causes for their variability ples per £ow) was a plagioclase^olivine^clinopyr-
(see below) because they could weight HI toward oxene fractionation trend; the second portion (af-
measuring a particular cause of heterogeneity. To fecting most samples) was a plagioclase^olivine
examine the potential di¡erence between HI based fractionation trend.
on only major elements (i.e. [26]) versus that in-
cluding trace elements (i.e. [1]), we have calculated LLDs were calculated for total pressures rang-
a second HI for the BBQ £ow (erupted in 1991 ing from 0.2 to 2.5 kbar, and fO2 of 0.1 times the
and 1992 near 9³50PN EPR) using six XRF trace QFM bu¡er. Slightly di¡erent crystallization con-
elements (Nb, Zr, Sr, V, Y, Rb) as well as the ditions yielded the `best' LLD for each £ow. Pres-
microprobe glass major element data. In this sures of 0.5, 0.5 and 1.5 kbar best ¢t the Cleft
case, the HI values are negligibly di¡erent (1.543 Pillow Mounds, the 1996 N. Gorda and the Se-
and 1.551 with and without trace elements, re- rocki data, respectively. Water content is another
spectively). important variable, although data exist only for
the Cleft Pillow Mounds (0.15 wt% H2O in the
It is extremely di¤cult to place an error limit parent; Dixon et al., unpublished data). Higher
on the calculated HI values in Table 1, given the Al2O3 in the N. Gorda and Serocki £ows requires
di¡erences in size and quality of the various data- slightly higher parent water contents (0.2 wt% and
sets. Qualitatively, HI values within V10% or so 0.3 wt%, respectively) to prevent plagioclase from
are probably not distinguishable. being a sole liquidus phase (which would not ¢t
the data trends).
4. Petrological sources of observed compositional
heterogeneity Qualitatively, the Cleft Pillow Mounds samples
adhere to the simulated LLD with little scatter,
Pre- and post-eruptive low-pressure fractional implying that much of the observed heterogeneity
crystallization during cooling imparts an easily may have arisen from low-pressure equilibrium
distinguished chemical signature to basaltic lavas fractional crystallization from a single parent
(e.g. strong trends on MgO variation diagrams). melt. This can be demonstrated quantitatively by
Data from some £ows (e.g. 1993 Coaxial and a factor of two reduction in the HI of the lava
Cleft Pillow Mounds £ows, Fig. 1) display such before and after fractionation correction (in both
trends on MgO variation diagrams whereas data cases HI is calculated without MgO since it is
from others do not (e.g. 1996 N. Gorda £ow, Fig. ¢xed in the fractionation correction, see Table 2).
1, and Serocki [19]). Compositions of three of the
most heterogeneous £ows (N. Gorda £ow, Cleft Both the N. Gorda and Serocki data are more
Pillow Mounds, and Serocki £ows) were modeled scattered about the constant-pressure LLDs, par-
to determine the potential e¡ects of low-pressure ticularly the Al2O3^MgO and CaO^MgO trends,
fractional crystallization, allowing approximation suggesting that they did not crystallize from a
of their relative amounts of crystallization- and single liquid composition. Instead, their arrays
non-crystallization-related compositional variabil- may have been formed through mixing of di¡er-
ity. Early crystallizing assemblages in MORB are ent parental melts and/or more complicated crys-
typically multi-phase. Simulated constant-pressure tallization histories (also noted by [19] for Se-
liquid evolution paths were generated using the rocki). This can again be demonstrated
`Melts' computer program [31] starting with the quantitatively with the HI of LLD-corrected
data, which changed by only 8% (Serocki) and
1% (N. Gorda) in these £ows (Table 2).

358 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

Thus, although some of the observed intra-£ow and using the np-HI* vs. HI correlation (Fig. 3b).
heterogeneity in some of the highest HI £ows can This calculation results in tephra HI of 8.1^14.9,
be explained by crystallization, this mechanism using the two estimate methods, respectively (Fig.
cannot be the only control. Instead, other process- 2b). The upper HI estimate is close to the value
es (e.g. assimilation, magma recharge, parental predicted by the HI vs. volume correlation of 14.7
magma mixing) likely acted on the pre-erupted at 0.4 km3. HI of the lava £ow can be likewise
magmas. estimated to be 7.6^14.7 (very similar to that of
the tephra), but there are limited lava data
5. Discussion of global trends (n = 13) and sample coverage of the £ow ¢eld
was very restricted (to primarily near the vent
As mentioned previously, within-£ow composi- ¢ssures [33]). Although much more data on
tional heterogeneity and estimated £ow volume more geographically dispersed samples are neces-
are strongly correlated for seven of the 10 MOR sary to determine how heterogeneous the entire
lavas in Table 1 (R2 = 0.964). One important ques- £ow volume is, the well sampled portions of the
tion is over what volume range does this correla- eruption products have an HI that is 50^100% of
tion hold? No submarine MOR lava £ow larger that predicted by the HI vs. volume correlation.
than Serocki has been mapped and sampled to the
degree that compositional heterogeneity can be Considering the potential sources of error in
calculated reliably. However, there is indirect evi- calculating HI and volumes of MOR £ows, it is
dence that the relationship might £atten out perhaps remarkable that any global HI^volume
above some critical volume: HI [26] of lavas correlation exists. The relationship indicates that
erupted from 1983 to 1992 in the ongoing Puu the pre-solidi¢cation history of an erupted batch
Oo eruption of Kilauea ([32] and references there- of lava is generally more complex at larger vol-
in) is lower than the correlation would predict. ume and suggests that many of the same processes
Data from the 1783^1784 Laki eruption (Iceland), a¡ect both the volume and compositional hetero-
which produced 14.7 km3 of lava and 0.4 km3 geneity of an MOR eruption. Yet, the three newly
(dense rock equivalent) of tephra ([33] and refer- analyzed S-EPR lavas do not adhere to this trend
ences therein), can also be used to indirectly ad- within the error limits (Fig. 2). Because HI varia-
dress this question. We calculated an HI of the tions at given volume are most apparent in lavas
basaltic tephra (n = 51) using glass compositions erupted at ridges with superfast spreading rates, it
analyzed with the University of Hawaii micro- suggests that volume is not the only scaling pa-
probe [33]. No precision data were reported but rameter for compositional heterogeneity.
HI can be estimated from both historical Pi values
HI in the 10 MOR lava £ows also varies in-
versely with spreading rate. Power law regressions
to the data yield excellent correlations (Fig. 4).

Table 2
Raw and fractionation-corrected HI data for selected £ows

Flow HI Number of samples Highest MgO HI (no MgO) HI (LLD-corrected) HITiFex

LLD correction 8.65 2.01 2.03 1.92
8.16 2.97 2.72 2.26
1996 N. Gorda 2.51 12 8.09 2.99 1.48 2.34

Serocki 3.27 21 1.38 1.33
0.24 0.28
Cleft Pillow Mounds 3.05 12 1.50 0.95

LLD correction in £ow subdomains

Cleft Mound 1 1.37 3

Cleft Mound 5^6 0.23 2

Cleft Mound 8 1.53 6

Fractionation correction along model-estimated LLDs is to highest MgO wt% in each £ow (MgO = `x'). HI `no MgO' and HI
`LLD-corrected' do not include MgO content as this parameter is ¢xed in the fractionation correction. Calculations and TiFex
are explained in the text; HITiFex is Si of TiFex at x = highest MgO, divided by Pi of TiFe in the precision standards.

K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367 359

Fig. 4. HI versus spreading rate for 10 individual MOR lava gree of fracturing), and the e¤ciency of magma
£ows (squares). The means for the various spreading rates cooling/solidi¢cation. All depend primarily upon
are also shown (open triangles). Notice the inverse relation- thermal conditions within the upper mantle and
ship between spreading rate and compositional heterogeneity. MOR crust, the latter of which are controlled by
Two £ows are anomalously heterogeneous for their given the interplay of the rates of magma supply and
spreading rate (open squares) relative to the others (¢lled loss of heat. To a ¢rst approximation, spreading
squares). The two di¡erent power law regressions are de- rate can be considered a proxy for magma supply
scribed in the text. rate (at ridge segments away from the excess melt
production of a hot-spot).
Two regressions were calculated in somewhat dif-
ferent manners to help identify outliers: the ¢rst Magma enters the crust (continuously or in
(solid line, R2 = 0.92) uses the mean HI at each pulses) with spatial and temporal variations in
spreading rate for which we have data and the composition acquired in the source region. In an
second (dashed line, R2 = 0.82) uses all the indi- unevenly cooled magma chamber some or all of
vidual £ow data except for two anomalously het- this compositional variance can be homogenized
erogeneous £ows (Cleft Pillow Mounds and Aldo- and/or overprinted by convection and crystalliza-
Kihi). Although the dataset is small, both regres- tion. Some subaerial volcanoes approach steady-
sion curves are functionally very similar, suggest- state wherein erupted volumes are positively cor-
ing that the two anomalous £ows are true outliers related with the duration of the preceding repose
relative to the mean HI at their spreading rate. [34], from which one can generalize that large-
Both anomalous £ows are volumetrically large volume eruptions will typically require longer
relative to others at their spreading rate, indicat- melt accumulation periods than small-volume
ing that very large £ows can exceed the typical eruptions. In the absence of e¤cient convective
compositional heterogeneity that is characteristic homogenization, longer melt accumulation time
of a speci¢c spreading rate. should add to the potential for compositional het-
erogeneity.
The inter-relationship of HI, volume and
spreading rate provides fundamental constraints In a generic sense, how much magma accumu-
on MOR magmatic processes. Overall, the devel- lates at a particular volcano before an eruption
opment of compositional heterogeneity within an takes place re£ects a balance of forces driving
eruptable magma volume and the lava £ow it and preventing eruption, such that eruptions oc-
produces re£ect variations in how magma accu- cur when hydrostatic pressure in the reservoir ex-
mulates within and is tapped from crustal magma ceeds a critical lithostatic con¢nement force. Con-
bodies. The volume, shape and compositional sequently, for magmas of similar volatile content,
range of crustal magma bodies re£ect the coupling larger pre-eruptive volumes will accumulate in
of magma input and eruption rates, repose period, stronger crust. This probably explains the inverse
chamber geometry, crustal strength (thickness, de- correlation of MOR lava £ow volume and spread-
ing rate [7]. A recent seismic re£ection experiment
[35] provides additional constraint on maximum
volumes of eruptable magma within axial magma
chambers along the superfast spreading S-EPR,
where magma supply rate is high. Namely, the
commonly referred to `melt lens' contains segre-
gated discontinuous pods of low crystallinity (5^
10%) eruptable magma volumes (V100^200U106
m3) separated by 15^20 km domains of mostly
high crystallinity (40^60%) uneruptable mush
[35]. Signi¢cantly larger erupted volumes at that
spreading rate would likely require recharge dur-
ing eruption, enhancing compositional heteroge-

360 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

neity in the resulting lava £ow. Similar constraints erupted from geographically distinct vents spread
have not been established for slower spreading over 26 km on the Southeast Rift Zone [40]
ridges, yet it is likely that the largest-volume erup- and required a strati¢ed magma body or mul-
tions would also require recharge. tiple, relatively homogeneous magma bodies [1].
Although similar scenarios are di¤cult to com-
Thus the global variations in HI with volume pletely resolve for a submarine MOR eruption
and spreading rate lead to the general conclusion without ¢ne scale stratigraphic control from direct
that where the MOR crust is hot and weak, erup- observation/monitoring or high-resolution age
tions are likely frequent, smaller and composition- dating, some eruption constraints are provided
ally more homogeneous; where the crust is cold by geographic distribution of compositional het-
and strong, eruptions are likely to be less frequent erogeneity within known lava £ows; examples are
and pre-eruptive magma batches should have given below.
more opportunities to develop compositional var-
iability. Furthermore, a steady-state eruptable 6.1. Tubeworm BBQ £ow (EPR)
magma lens is more likely to occur beneath a
ridge segment as thermal stability, magma supply, High-resolution mapping, sampling and 210Po^
and spreading rate increase [36], which should 210Pb dating of this compound lava £ow near
promote compositional homogeneity in erupted 9³50PN indicate that lavas were emplaced over
lavas. In contrast eruption of more heterogeneous V1 yr between Spring 1991 and January 1992
and poorly mixed magma batches should be fa- [9,11] from a nearly continuous 8.5 km long ¢s-
vored at slower spreading rates. sure [9], resulting in a relatively homogeneously
composed £ow [24]. The ¢rst eruptive phase was
6. Compositional heterogeneity and £ow geography in February^April 1991 [11] (three dated and 15
undated samples were collected by Alvin in April^
Local di¡erences in £ow morphology can be May 1991). The £ow ¢eld was reactivated in late
related to changes in emplacement rate as erup- December 1991 [11] and produced indistinguish-
tions progress (e.g. [37,38] and references therein). ably composed lavas (one dated and two undated
The geographic distribution of compositional var- samples collected in Spring 1992). HI of lavas
iability within a £ow can also be used to help from both episodes is 1.55 (see Fig. 5); removing
understand emplacement history. Such geograph- the three 1992-collected samples changes the HI
ical variability might arise from simultaneous by 6 5%. Two lava pillars from a collapsed
eruption of chemically di¡erent lavas from di¡er- part of this £ow have a similar HI of 1.57 (based
ent vents, temporal variability in compositions on V500 elemental analyses at 100 Wm spacing
erupted from a single vent, and/or syn/post-erup- on seven transects through glassy pillar crusts
tive di¡erentiation. HI in two Mauna Loa £ows [41]).
(1942 and 1950) helps to illustrate the range of
conditions that might also apply to MOR erup- Lavas were also erupted on the adjacent ridge
tions. Both formed in short-duration (2 weeks) segment to the north during the second phase
eruptions along nearly continuous single eruptive of activity: one sample collected from the axis
¢ssures. The compositionally homogeneous V5 km north of the main mapped £ow (from a
(HI = 1.6), moderate-volume 1942 lava formed feature not present in 1991) erupted in early Jan-
two sequential £ow ¢elds, ¢rst erupting from a uary 1992 (210Po^210Pb date) and is signi¢cantly
vent near the summit and then from a V1 km more compositionally evolved. Adding this sam-
long ¢ssure some 10 km distant [39]. In contrast, ple to the 9³50PN £ow ¢eld increases the HI by
the very compositionally variable (HI = 13.9) V20%, indicating that a larger area of the EPR
high-volume 1950 lava formed in seven major rift was activated during this second phase and
lobes from primitive, intermediate and evolved multiple magma compositions were erupted, in a
magmas (11^12, 9^10 and s 7 wt% MgO) geometrically similar fashion to the 1942 and 1950
Mauna Loa eruptions.

K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367 361

6.2. North Cleft Pillow Mounds (JdFR) other mounds and imply that magmas feeding
them either cooled more slowly on the surface
This compound lava £ow has a strong geo- (syn/post-emplacement crystallization) or resided
graphic distribution of compositions along its 17 longer in crust before eruption (magma chamber
km length (Fig. 5). All eight mounds (numbered crystallization). Because each of the pillow
sequentially as M1^M8 from S to N) probably mounds is morphologically similar and thus their
erupted between bathymetric surveys in 1981 extrusion (and cooling) rates should likewise have
and 1987 and although they were not necessarily been similar, we favor a scenario where the M5/
erupted simultaneously, they are clearly related to M6 lavas acquired their evolved composition pri-
a single 9 6 yr long volcanic event (possibly 9 2 yr or to eruption. Thus, compositional variations
[8]). About half of the compositional heterogene- suggest the £ow was emplaced from a relatively
ity within this £ow is due to low-pressure crystal long ¢ssure by sampling of a horizontally strati-
fractionation (discussed earlier). There is a regular ¢ed sub-ridge magma lens or lenses, possibly over
geographical pattern of compositional heterogene- a few years. Multiple extrusion events can also be
ity amongst the mounds, suggesting that each be- inferred from hydrothermal `mega-plumes' in the
haved as separate cooling units. MgO content de- water column above the ridge observed in 1986
creases towards the middle (latitudinally) of the and 1987 [43].
£ow (Fig. 5). Even with the limited number of
samples, it is apparent that the HI values of in- 6.3. North Cleft sheet £ow (JdFR)
dividual mounds are lower than the HI of the
entire £ow: HI of M1 and M8 are V50% of This sheet £ow was emplaced just to the south
the overall HI (Table 2, Fig. 5). The most com- of the Cleft Pillow Mounds. It has been mapped
positionally evolved mound (M2) is essentially ho- as a single contiguous £ow with a narrow con-
mogeneous (HI 6 1) within the resolution of the striction at 44³58.1PN dividing it into a small
limited (n = 2) dataset (e.g. analyses are less vari- northern lobe and broader southern lobe [22,38].
able than predicted by the Pi values). After cor- The £ow is moderately heterogeneous and lacks a
recting for low-pressure fractionation, the M1 HI strong geographic variation in composition,
changes by only 4%, whereas the M8 HI changes although samples have only been recovered from
by 63% (to V1, Table 2, Fig. 5). Thus, all of the the northern 2/3 of the £ow and sample density is
compositional heterogeneity in the most heteroge- lower in the southern portion than in the north
neous and volumetrically largest mound (M8) can (Fig. 5). Magmatic heterogeneities were fairly
be ascribed to low-pressure crystal fractionation, equally dispersed throughout the £ow during
but e¡ectively none of the heterogeneity of M1 is eruption, consistent with this being a high em-
related to this process. Instead, M1 appears to be placement rate sheet £ow [22]. Splitting the sam-
a composite structure with chemical compositions ple set at 44³58.1PN yields HI values 18% greater
re£ecting magma mixing (also inferred independ- (northern domain) and 27% lower (southern do-
ently from trace element data [7,42]). main) than the £ow overall. In addition, the most
primitive and most evolved compositions occur in
Overall the geographic distribution of composi- the north. Geological mapping suggests the erup-
tional heterogeneity in the mounds is inconsistent tive ¢ssure was con¢ned to the northern half of
with a simple one-time diking and extrusion the sheet £ow, that the eruption was longest lived
event. M1 was either built in multiple episodes in this area, and that much of the £ow south of
by lavas of somewhat di¡erent composition or the constriction was fed from the north [22]. The
erupted from a chemically heterogeneous sub- compositional data also suggest that the northern
ridge magma reservoir. The large, LLD-controlled portion of the £ow was active for a longer period
compositional range of M8 is also easiest to ra- of time and the vent(s) sequentially erupted a
tionalize by sequential extrusion of di¡erent com- range of compositions that did not all spread to
positions. Glasses from central mounds M5 and the south. However, it is also possible that the
M6 are compositionally more evolved than the

362 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367 363

Fig. 5. HI histograms (HI values given above the columns) and MgO content versus location (¢lled diamonds) within ¢ve MOR
lava £ows. Geographic variations of compositional heterogeneity in these £ows are interpreted in the text. Typical 2c error is de-
picted by the vertical bar in each lower panel. A: The Tubeworm BBQ £ow data are from samples collected in 1991 and 1992, a
subset of which were dated as having erupted in 1991 and 1992. Notice that the lone sample from the next ridge segment to the
north (also erupted in 1992) is compositionally distinct. B: The Cleft Pillow Mounds display a regular variation in MgO content
on a S^N trajectory through the £ow, which consists of eight mounds spread over 17 km of ridge axis. Mean MgO contents for
individual mounds are also shown (open squares). Corresponding HI values for the entire £ow and for mound 1, mounds 5+6
and mound 8 are shown in the upper panel. This panel includes HI data after fractionation correction to constant MgO, which
is why each mound and the overall £ow are shown as triplicates of columns. In each case the left-most column represents raw
HI, the middle column represents raw HI without MgO, and the right-most column represents LLD-corrected HI values. C: The
North Cleft Sheet £ow was geographically subdivided based upon £ow morphology and there is no obvious geographical varia-
tion in composition within it. In contrast, the 1993 Coaxial (D) and Aldo-Kihi (E) £ows both show strong geographic variations
in composition. The Coaxial £ow was subdivided based on chemical groupings. The Aldo-Kihi £ow can be geographically subdi-
vided on the basis of £ow morphology (`physical division') or £ow composition (`chemical division'), with only minor di¡erences
in HI (both of which are shown by the doublet columns in the histogram panel for this £ow).
6

compositionally heterogeneous magma was sup- parts of the £ow, slowing e¡usion rate (inhibiting
plied all at once to the northern vent(s) and these magma mixing during emplacement) and greater
were mixed during emplacement as the £ow thickness all suggest that vents in the south^cen-
spread out from this location. tral portion of the £ow were active for a longer
duration.
6.4. 1993 Coaxial £ow (JdFR)
6.5. Aldo-Kihi £ow (S-EPR)
This £ow has a strong N^S variation in com-
position and heterogeneity between £ow subdo- This £ow at 17³30PS has a strong geographic
mains (more ma¢c/more heterogeneous in the distribution of compositions and lava morpholo-
south and less ma¢c/less heterogeneous in the gies along its 8.5 km axial length [13,45,46]. In
north; Fig. 5) despite its relatively low HI overall. plan view the £ow ¢eld is separated into a wide
Seismicity propagated northward during this northern portion with en echelon axial collapse
eruption [44], suggesting that the eruption started troughs and a narrow southern portion lacking
in the south. It may have ¢rst tapped a more a collapse trough. In the north it grades semi-con-
ma¢c and later a more evolved liquid from a com- centrically from high e¡usion rate jumbled sheets
positionally zoned magma body; alternatively, the on-axis through lower e¡usion rate lobate lavas to
signature may have arisen during dike propaga- pillows on the o¡-axis margins, whereas in the
tion, either by fractionation within the dike or by southern portion it is all lobate £ows and pillow
mixing of young ma¢c and older evolved magma constructions. Lava compositions are more ma¢c
[27]. Similar magma mixing occurred in the early and less compositionally variable south of 17³30PS
phases of the Puu Oo eruption [32] and accounts (HI is V35% 6 HI of the £ow overall) and more
for roughly half of the HI value of that £ow. The evolved and more compositionally variable north
Coaxial £ow is a series of coalesced mounds that of 17³30PS (HI is only V20% 6 HI of the £ow
initially formed from a long eruptive ¢ssure that overall; Fig. 5). Splitting the £ow morphologi-
then localized to a series of vents [23]. The great- cally at 17³30PS and splitting the £ow on geo-
est bathymetric anomaly (thickness) is in the graphic^chemical variations gives only slightly
south^central portion. Wax modeling of £ow di¡erent HI values (Fig. 5). Shibata et al. [46]
morphology suggests that s 50% of the £ow vol- argued that the compositional pattern formed by
ume erupted in 2 h from a 2.5 km long ¢ssure, northward and southward fractional crystalliza-
followed by slower extrusion for 10 days from tion at or near the surface from a ma¢c eruptive
localized vents [37]. Together, the greater compo- center near 17³30PS, noting also that this is not
sitional heterogeneity of the southern and middle the shallowest or broadest point on the £ow. An-

364 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

other possibility is that the compositional varia- been low-pressure fractionation-corrected to con-
tions resulted from vertical diking from an along stant MgO (7.3 wt% [47] and 8.0 wt% [48] in those
axis compositionally zoned magma body [45]. cases). Variations in Mg, Ti and Fe abundances in
Neither scenario considers that there is a more both N-MORB datasets are mostly small, but a
chemically heterogeneous subdomain between handful of geochemically distinct transitional or
17³26PS and 17³27PS characterized by high e¡u- enriched MORBs also occur in both locales. An-
sion rate lavas and the most evolved compositions alytical precision is only given for TiFex in those
(Fig. 5), the presence of which suggests a third works (2c = 0.4), which can be used to show that
scenario: an initially slow e¡usion, long ¢ssure TiFex variability in those entire N-MORB data-
eruption expelled ma¢c lava in the south and un- sets (2c = 1.1 and 0.74) is only Vthree times [47]
known composition in the north, followed by vent and two times [48] the precision. TiFex varies over
focusing to a higher e¡usion rate, primary erup- a similar range in the three single £ows fraction-
tion center between 17³26PS and 17³27PS that ex- ation-corrected here (2c = 0.34, 0.47 and 0.57 in
pelled a more evolved composition near the end the Cleft Pillow Mounds, 1996 N. Gorda and Se-
of the eruption. The last phase may have largely rocki £ows, respectively). Converting the latter to
covered earlier lavas erupted in the north, making HI values (Si/Pi) shows that TiFex variability in
it di¤cult to distinguish between along axis com- these three £ows is Vtwo times the analytical
positional zonation and a temporal control to the precision (see HITiFex in Table 2). Qualitatively,
compositional patterns in the £ow with existing some of the compositional variability in both
samples. EPR sample grids might be due to spatial and
temporal variability in magmatic conditions.
7. Implications of within-£ow compositional Yet, because TiFex variability is equal to or
heterogeneity for other MOR petrologic studies only slightly greater in the sample grids than in
many individual £ows, inter-£ow and intra-£ow
On average, the 10 MOR lava £ows in Table 1 compositional variability cannot be con¢dently
are twice as compositionally heterogeneous as the resolved, making this form of `mapping' ambigu-
mean precision for major element analysis (mean ous in these cases. TiFex and other more subtle
HI is 2.0). Precision-normalized variability of in- chemical variations in these 12³N samples were
dividual elements can be even greater (up to subsequently used to identify [49] and model [50]
eight). This range of compositions in individual a new MORB composition known as `o¡-axis'
MOR lava £ows indicates that care must be MORB, yet such an exercise seems premature
used when interpreting subtle chemical variations based on the preceding discussion.
between MORB samples with an unknown rela-
tionship to individual £ow boundaries as being 8. Summary and future directions
due to spatially, temporally and compositionally
distinct eruptions. There are many examples of interesting per-
spectives about magmatic conditions gained by
Likewise, caution should be used in experi- studying individual subaerial lava £ows, but logis-
ments designed to `map' magmatic variation tical di¤culties have in the past made this signi¢-
through time using MORB compositions from cantly more challenging at submarine volcanoes.
sample grids along and across axis. Two examples Results of this study demonstrate that a great deal
of such mapping at 12³N [47] and 9.5^10³N [48] can also be learned by studying within-£ow com-
on the EPR used major element oxide composi- positional heterogeneity in individual MOR lava
tions (e.g. MgO) and TiFex. The latter parameter £ows.
was proposed to sensitively distinguish subtle dif-
ferences in magma chemistry [47] and is de¢ned as 1. The strong relationships within the dataset pre-
the sum of the di¡erences of 1.5UTi and Fe con- sented here between HI, estimated lava £ow
centrations from their means after Ti and Fe have

K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367 365

volume and spreading rate suggest that compo- intra-£ow compositional variability of other
sitional heterogeneity within £ows can be used geochemical tracers (e.g. trace element concen-
to help decipher global variations in magma trations and ratios, radiogenic isotope ratios)
supply and thermal conditions at the MOR was investigated in greater detail because it
system. Where the MOR crust is hot and would help to re¢ne the resolution of inter-
weak, eruptions are likely frequent, smaller £ow compositional di¡erences. Such data
and more compositionally homogeneous than might also allow for a series of HIs applicable
where the crust is cold and strong. Addition of to di¡erent aspects of MORB petrogenesis (e.g.
more £ows to the database and more geochem- for mantle heterogeneity, melting processes
ical tracers would further elucidate the range of and low-pressure crystallization).
MOR magmatic conditions and global crustal
accretion variability. Acknowledgements
2. Compositional heterogeneity within individual
MOR lava £ows can be used to study the erup- We would like to thank D. Fornari and T.
tions that produced them (vent locations, con- Gregg for helpful reviews, B. Chadwick for com-
ditions of emplacement). In the optimum, lava ments on an earlier version of the manuscript, M.
sampling and chemical analysis for individual Rhodes for discussions, W.I. Ridley for micro-
£ow studies should be conducted at the highest probe analyses (Cleft Mounds and 1993 Coaxial
possible sampling density and spatial coverage. £ows), the many scientists whose papers could not
Uncertainty in the results will be minimized be cited here but were nonetheless important to
with large (n v 30^50) sample and standard our study, and the scientists and crews of the
replication datasets. With additional composi- various mapping and sample collection cruises
tional data and detailed physical mapping, it for their hard work at sea. Special thanks to
should be possible to further decipher eruptive B. Embley and the NOAA PMEL group for
processes resulting in morphological and com- their seminal studies of MOR eruptions and
positional domains within a £ow. These data, lava £ows in the NE Paci¢c. This study was
along with hydrothermal vent and faunal dis- supported by the NSF (Grants OCE-9905463
tribution maps, should also help to unravel the and OCE-9633268 (K.H.R.), OCE-9986874 and
physiochemical and ecological evolution of OCE-9530299 (M.R.P.), and OCE-9633398 and
newly erupted £ows (e.g. cooling history, the 9415989 (J.M.S.)). This is SOEST contribution
development of geothermal circulation, and the # 5408.[EB]
colonization of the £ow by organisms).
3. Individual £ows can be as compositionally var- References
iable as regional MOR sample sets. It is there-
fore important to simultaneously conduct high- [1] J.M. Rhodes, Homogeneity of lava £ows: chemical data
resolution geologic mapping and sampling to for historic Mauna Loan eruptions, J. Geophys. Res. 88A
know the volcanologic context of the samples. (1983) 869^879.
With additional compositional data on individ-
ual £ows from more spreading environments, [2] H. Elder¢eld, A. Schulz, Mid-ocean ridge hydrothermal
petrologists will be able to better determine £uxes and the chemical composition of the ocean, Ann.
how mantle compositions and melting condi- Rev. Earth Planet. Sci. 24 (1996) 191^224.
tions are mapped into along and across ridge
variations in MORB compositions found in [3] K.H. Rubin, Degassing of metals and metalloids from
samples from geologically un-mapped terrains erupting seamount and mid-ocean ridge volcanoes: obser-
(e.g. most MORB studies). It will be important vations and predictions, Geochim. Cosmochim. Acta 61
for future studies of MORB compositions to (1997) 3525^3542.
consider the range of intra-£ow variations de-
scribed here. It would also be advantageous if [4] J.R. Delaney, D.S. Kelley, M.D. Lilley, D.A. Butter¢eld,
J.A. Baross, W.S.D. Wilcock, R.W. Embley, M. Summit,
The quantum event of oceanic crustal accretion: impacts
of diking at mid-ocean ridges, Science 281 (1998) 222^230.

366 K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367

[5] A. Prinzhofer, E. Lewin, C.J. Alle©gre, Stochastic melting neities in a single Icelandic tholeiite £ow, Geochim. Cos-
of marble cake mantle: evidence from local study of the mochim. Acta 45 (1981) 15^31.
East Paci¢c Rise at 12³50PN, Earth Planet. Sci. Lett. 92 [19] S.E. Humphris, W.B. Bryan, G. Thompson, L.K. Autio,
(1989) 189^206. Morphology, geochemistry, and evolution of Serocki vol-
cano, Ocean Drilling Program, 106/109, College Station,
[6] Y. Niu, D.G. Waggoner, J.M. Sinton, J.J. Mahoney, TX, 1990, pp. 67^84.
Mantle source heterogeneity and melting processes be- [20] R.W. Embley, W.W. Chadwick Jr., M.R. Per¢t, E.T.
neath sea£oor spreading centers: The East Paci¢c Rise, Baker, Geology of the northern Cleft segment, Juan de
18³^19³S, J. Geophys. Res. 101 (1996) 27711^27733. Fuca Ridge: Recent lava £ows, sea-£oor spreading, and
the formation of megaplumes, Geology 19 (1991) 771^
[7] M.R. Per¢t, W.W. Chadwick, Jr., Magmatism at mid- 775.
ocean ridges: Constraints from volcanological and geo- [21] W.B. Bryan, S.E. Humphris, G. Thompson, J.F. Casey,
chemical investigations, in: W.R. Buck, P. Delaney, J.A. Comparative volcanology of small axial eruptive centers
Karson, Y. Lababrille (Eds.), Faulting and Magmatism at in the MARK area, J. Geophys. Res. 99 (1994) 2973^
Mid-Ocean Ridges, Geophysical Monograph 106, Am. 2984.
Geophys. Union, Washington, DC, 1998, pp. 59^115. [22] R.W. Embley, W.W. Chadwick Jr., Volcanic and hydro-
thermal processes associated with a recent phase of sea-
[8] W.W. Chadwick Jr., R.W. Embley, C.G. Fox, Evidence £oor spreading at the northern Cleft segment: Juan de
for volcanic eruption on the southern Juan de Fuca Ridge Fuca Ridge, J. Geophys. Res. 99 (1994) 4741^4760.
between 1981 and 1987, Nature 350 (1991) 416^418. [23] W.W. Chadwick Jr., R.W. Embley, C.G. Fox, SeaBeam
depth changes associated with recent lava £ows, CoAxial
[9] R.M. Haymon, D.J. Fornari, K.L. Von Damm, M.D. segment, Juan de Fuca Ridge: Evidence for multiple erup-
Lilley, M.R. Per¢t, J.M. Edmond, W.C.I. Shanks, R.A. tions between 1981^1993, Geophys. Res. Lett. 22 (1995)
Lutz, J.M. Grebmeier, S. Carbotte, D. Wright, E. 167^170.
McLaughlin, M. Smith, N. Beedle, E. Olson, Volcanic [24] T.K.P. Gregg, D.J. Fornari, M.R. Per¢t, R.M. Haymon,
eruption of the mid-ocean ridge along the East Paci¢c J.H. Fink, Rapid emplacement of a mid-ocean ridge lava
Rise crest at 9 degrees 45^52PN; direct submersible obser- £ow on the East Paci¢c Rise at 9³ 46P^51PN, Earth Planet.
vations of sea£oor phenomena associated with an erup- Sci. Lett. 144 (1996) E1^E7.
tion event in April, 1991, Earth Planet. Sci. Lett. 119 [25] W.W. Chadwick Jr., R.W. Embley, T.M. Shank, The
(1993) 85^101. 1996 Gorda Ridge eruption: Geologic mapping, sidescan
sonar, and SeaBeam comparison results, Deep-Sea Res. II
[10] W.W. Chadwick Jr., R.W. Embley, Lava £ows from a 45 (1998) 2547^2569.
mid-1980s submarine eruption on the Cleft Segment, [26] K.H. Rubin, M.C. Smith, M.R. Per¢t, D.M. Christie,
Juan de Fuca Ridge, J. Geophys. Res. 99 (1994) 4761^ L.F. Sacks, Geochronology and petrology of lavas from
4776. the 1996 North Gorda Ridge eruption, Deep-Sea Res. II
45 (1998) 2571^2597.
[11] K.H. Rubin, J.D. Macdougall, M.R. Per¢t, 210Po^210Pb [27] M.C. Smith, Geochemistry of Eastern Paci¢c MORB:
dating of recent volcanic eruptions on the sea £oor, Na- Implications for MORB Petrogenesis and the Nature of
ture 368 (1994) 841^844. Crustal Accretion Within the Neovolcanic Zones of Two
Recently Active Ridge Segments, Ph.D. thesis, University
[12] C.G. Fox, W.E. Radford, R.P. Dziak, T.K. Lau, H. Mat- of Florida, FL, 1999.
sumoto, A.E. Schreiner, Acoustic detection of a sea£oor [28] J.M. Sinton, Eruptive history of the Hengill volcanic sys-
spreading episode on the Juan de Fuca Ridge using mili- tem: mid-atlantic ridge in the Western Rift Zone of Ice-
tary hydrophone arrays, Geophys. Res. Lett. 22 (1995) land, EOS Trans. Am. Geophys. Union 78 (1997) F645.
131^134. [29] J.P. Lockwood, P.W. Lipman, Holocene eruptive history
of Mauna Loa volcano, US Geol. Surv. Prof. Pap. 1350
[13] J. Sinton, E. Bergmanis, K.H. Rubin, R. Batiza, T.K.P. (1987) 509^536.
Gregg, K. Gro«nvold, S. White, K. Macdonald, Volcanic [30] J.M. Rhodes, Geochemistry of the 1984 Mauna Loa erup-
eruptions at superfast spreading: the East Paci¢c Rise, tion: Implications for magma storage and supply, J. Geo-
17³^19³S, J. Geophys. Res. (2001), submitted. phys. Res. 93 (1988) 4453^4466.
[31] M.S. Ghiorso, R.O. Sack, Chemical mass transfer in mag-
[14] G.P.L. Walker, Compound and simple lava £ows and matic systems IV. A revised and internally consistent ther-
£ood basalts, Bull. Volcanol. 35 (1971) 579^590. modynamic model for the interpolation and extrapolation
of liquid^solid equilibria in magmatic systems at elevated
[15] S. Self, T. Thordarson, L. Keszthelyi, Emplacement of temperatures and pressures, Contrib. Mineral. Petrol. 119
continental £ood basalt lava £ows, in: J.J. Mahoney, (1995) 197^212.
M.F. Co¤n (Eds.), Large Igneous Provinces: Continen- [32] M. Garcia, J.M. Rhodes, F.A. Trusdell, A. Pietruszka,
tal, Oceanic and Planetary Flood Volcanism, Geophysical
Monograph 100, Am. Geophys. Union, Washington, DC,
1997, pp. 381^410.

[16] J.R. Carden, A.W. Laughlin, Petrochemical variations
within the McCartys basalt £ow, Valencia County, New
Mexico, GSA Bull. 85 (1974) 1479^1484.

[17] M.M. Lindstrom, L.A. Haskin, Causes of compositional
variations within mare basalt suites, Proc. 9th Lunar
Planet. Sci Conf., 1978, pp. 466^486.

[18] M.M. Lindstrom, L.A. Haskin, Compositional inhomoge-

K.H. Rubin et al. / Earth and Planetary Science Letters 188 (2001) 349^367 367

Petrology of lavas from the Puu Oo eruption of Kilauea mid-ocean ridge basalt, J. Geophys. Res. 99 (1994) 4787^
Volcano: III. The Pond stage (1986^1992), Bull. Volca- 4812.
nol. 58 (1996) 359^379. [43] E.T. Baker, J.W. Lavelle, R.A. Feely, G.J. Massoth, S.L.
[33] Th. Thordarsson, S. Self, N. Oè skarsson, T. Hulsebosch, Walker, J.E. Lupton, Episodic venting of hydrothermal
Sulfur, chlorine, and £uorine degassing and atmospheric £uids from the Juan de Fuca Ridge, J. Geophys. Res.
loading by the 1783^1884 AD Laki Skafta¨r Fires eruption 94 (1989) 9237^9250.
in Iceland, Bull. Volcanol. 58 (1996) 205^225. [44] R.P. Dziak, C.G. Fox, A.E. Schreiner, The June^July
[34] G. Wadge, Steady state volcanism: evidence from erup- 1993 seismo-acoustic event at CoAxial segment, Juan de
tion histories of polygenetic volcanoes, J. Geophys. Res. Fuca Ridge: Evidence for a lateral dike injection, Geo-
87 (1982) 4035^4049. phys. Res. Lett. 22 (1995) 135^138.
[35] S.C. Singh, G.M. Kent, J.S. Collier, A.J. Harding, J.A. [45] E.C. Bergmanis, J.M. Sinton, S. White, K.C. Macdonald,
Orcutt, Melt to mush variations in crustal magma proper- R. Batiza, K. Rubin, T. Gregg, C. Van Dover, K. Gro« n-
ties along the ridge crest at the southern East Paci¢c Rise, vold, Anatomy of a mid-ocean ridge volcanic eruption:
Nature 394 (1998) 874^878. the Aldo-Kihi £ow between 17³24P and 17³34PS, East Pa-
[36] J.M. Sinton, R.S. Detrick, Mid-ocean ridge magma cham- ci¢c Rise, Trans. Am. Geophys. Union 80 (1999) F1075.
bers, J. Geophys. Res. 97 (1992) 197^216. [46] T. Shibata, O. Okano, R.W. Embley, K. Nogami, K.
[37] T.K.P. Gregg, J.H. Fink, Quanti¢cation of submarine Fujioka, K. Sayanagi, M. Kinoshita, Along-axis fraction-
lava-£ow morphology through analog experiments, Geol- ation of basalts: samples from recent lava £ows at 17³24^
ogy 23 (1995) 73^76. 35PS, East Paci¢c Rise, Trans. Am. Geophys. Union 80
[38] W.W. Chadwick Jr., T.K.P. Gregg, R.W. Embley, Sub- (1999) F1075.
marine lineated sheet £ows: a unique lava morphology [47] J.R. Reynolds, C.H. Langmuir, J.F. Bender, K.A. Kas-
formed on subsiding lava ponds, Bull. Volcanol. 61 tens, W.B.F. Ryan, Spatial and temporal variability in the
(1999) 194^206. geochemistry of basalts from the East Paci¢c Rise, Nature
[39] G.A. Macdonald, The 1942 eruption of Mauna Loa, Ha- 359 (1992) 493^499.
waii, Am. J. Sci. 241 (1943) 241^256. [48] M.R. Per¢t, D.J. Fornari, M.C. Smith, J.F. Bender, C.H.
[40] G.A. Macdonald, Activity of Hawaiian volcanoes during Langmuir, R.M. Haymon, Small-scale spatial and tempo-
the years 1940^50, Bull. Volcanol. 15 (1954) 119^179. ral variations in mid-ocean ridge crest magmatic process-
[41] T.K.P. Gregg, D.J. Fornari, M.R. Per¢t, W.I. Ridley, es, Geology 22 (1994) 375^379.
M.D. Kurz, Using submarine lava pillars to record mid- [49] J.R. Reynolds, C.H. Langmuir, Identi¢cation and impli-
ocean ridge eruption dynamics, Earth Planet. Sci. Lett. cations of o¡-axis lava £ows around the East Paci¢c Rise,
178 (2000) 195^214. Geochem. Geophys. Geosyst. G3 1 (2000).
[42] M.C. Smith, M.R. Per¢t, I.R. Jonasson, Petrology and [50] M. Spiegelman, J.R. Reynolds, Combined dynamic and
geochemistry of basalts from the southern Juan de Fuca geochemical evidence for convergent melt £ow beneath
Ridge: Controls on the spatial and temporal evolution of the East Paci¢c Rise, Nature 402 (1999) 282^285.