ArticlePDF Available

The crustal magma chamber of the Katla volcano in south Iceland revealed by 2-D undershooting

Authors:

Abstract and Figures

Results of a 2-D, seismic undershooting experiment on the Katla central volcano in south Iceland are reported. Large localized traveltime anomalies (0.4s) are observed on an array within the Katla caldera. the traveltimes are forward modelled using a wavefront tracker developed in Appendix A. Thus, non-linear effects encountered in traveltime tomography are avoided as well as common problems with ray tracing in the presence of strong lateral heterogeneity. an extreme variation in compressional velocity is required to extend over a significant volume in order to model the data. the resulting model is not unique, but constraints on the allowable range of velocities (2.5-6.0 km/s) render the basic features well constrained. A clear S-wave shadow is closely associated with delays in traveltime due to a shallow slow anomaly. Low-amplitude P waves go hand in hand with early arrivals due to thin structural features flanking the slow anomaly. the model is interpreted in terms of a magma chamber containing exte
Content may be subject to copyright.
Geophys.
J.
Int.
(1994)
119,
277-296
The crustal magma chamber
of
the Katla volcano in south Iceland
revealed
by
2-D
seismic undershooting
0.
Gudmundsson,' B. Brandsdottir,2
W.
Menke3 and
G.
E.
Sigvaldason4
'
Research School
of
Earth Sciences, Australian National University, Canberra,
ACT
0200,
Australia
'
Larnont-Doherty Earth Observatory and Dept
of
Geological Sciences, Columbia University, Palisades, New York,
USA
Science Institute
of
the University
of
Iceland, Reykjavik, Iceland
Nordic Volcanological Institute, University
of
Iceland, Reykjavik, Iceland
Accepted
1994
March
25.
Received
1994
March
25;
in
original
form
1993
July
21
SUMMARY
Results
of
a 2-D, seismic undershooting experiment on the Katla central volcano in
south Iceland are reported. Large localized traveltime anomalies (0.4
s)
are observed
on an array within the Katla caldera. The traveltimes are forward modelled
using a wavefront tracker developed in Appendix A. Thus, non-linear effects en-
countered in traveltime tomography are avoided as well as common problems
with ray tracing in the presence
of
strong lateral heterogeneity.
An
extreme
variation in compressional velocity is required to extend over a significant volume
in
order to model the data. The resulting model is not unique, but constraints
on
the
allowable range
of
velocities (2.5-6.0 km
s-')
render the basic features well
constrained.
A
clear S-wave shadow is closely associated with delays in traveltime
due to a shallow slow anomaly. Low-amplitude
P
waves go hand in hand with early
arrivals due to thin structural features flanking the slow anomaly. The model
is
interpreted in terms
of
a magma chamber containing extensively molten rock. The
magma chamber is shallow, with a bottom at a depth
of
about 1.5 km below
sea-level
(3.0
km below surface), and measures about
5
km across. The depth
of
the
chamber is roughly at the level
of
buoyant equilibrium for basaltic melt in the crust.
Owing
to
poor vertical resolution at shallow depths in the undershooting geometry
the top
of
this shallow magma chamber is not well resolved. On the other hand, the
bottom
of
the chamber
is
well resolved. The chamber is underlain by rocks
of
average
or
high velocity for that depth. The magma chamber is a persistent feature,
big enough
(10
km3) to supply magma for large eruptions and to supply heat to
permit remelting
of
hydrated basaltic crust to produce silicic magmas at shallow
levels. The chamber is fed by magma fracturing from below. The model agrees with
phenomenological models
of
magma chambers in Iceland based
on
geological
observations and provides a quantification
of
those models in terms
of
depth and
size. On the other hand, it
is
fundamentally different from recent models
of
magma
chambers at mid-ocean ridges which may be more akin to the pervasive region
of
partial melt at depth beneath Iceland. This underlines the important effect
of
the
Icelandic hotspot on tectonics and volcanism in Iceland and implies a substantially
different crustal and thermal structure in Iceland from that
of
'normal' mid-ocean
ridges.
Key
words:
magma chamber, thermal structure, velocity structure, wavefront
tracking.
settings. In some cases considerable detail in the structure
of
magma chambers and their surroundings
has
been mapped.
1
INTRODUCTION
Seismology has, in recent years, been successful in detecting The principal observations have been traveltimes
in
the presence
of
magma chambers in various volcanic refraction experiments and the timing and amplitudes
of
277
278
0.
Gudmundsson
et
al.
reflected waves. These experiments have yielded models
of
highly variable compressional velocity, where the low
velocities are explained in terms
of
near-solidus tempera-
tures
or
the presence
of
melt and the high velocities are
explained in terms
of
crystalline (aporous) intrusives.
Reflection techniques have been most successfully applied
in the ocean. Recent experiments on the East Pacific Rise
have revealed a bright reflection
off
a negative impedance
contrast (Detrick
et
al.
1987; Harding
et
al.
1989; Vera
et
al.
1990). Efforts to recover sufficiently deep reflections over
land volcanoes have on the other hand been less successful
(Black, Deemer
&
Smithson 1991) and are often
complicated by the rough terrain in volcanic regions.
Early refraction studies utilized available local data from
permanent monitoring networks (e.g. the work of Thurber
(1984) on Hawaii and Benz
&
Smith (1984) on
Yellowstone). Such studies often suffer from a poor
distribution
of
seismicity and poorly known ray geometry
and source locations. Active surveying methods, although
costly, are therefore preferable. Nercessian, Hirn
&
Tarantola (1984) applied a technique often referred to as
undershooting on Mont Dore. Refractions from the Moho
discontinuity were used to illuminate a crustal target from
below using explosives at distance. This method has the
advantage
of
good lateral resolution and stable ray paths,
but suffers from poor vertical resolution. Achauer, Evans
&
Stauber (1988) and Evans
&
Zucca (1988) applied a similar
understooting technique without a discrete refractor at
depth. On the other hand Toomey
et
al.
(1990) had both
artificial sources and instruments contained within their
study area on the East Pacific Rise. Thus resolution was
more isotropic, but ray paths are potentially unstable in such
a geometry.
For
a thorough review refer to Thurber
&
Aki
(1987).
In order to study the 3-D structure
of
a volcano,
refraction experiments on volcanoes are generally conducted
with an areal array
of
a
large number of instruments, use
large amounts
of
explosives, and may require drilling.
Consequently costs run high. Here, we present the results
of
a linear deployment
of
a moderate number
of
instruments
across the Katla central volcano in south Iceland. the
resulting 2-D interpretation has obvious limitations. We do,
however, hope to convince the reader that valuable
information can be obtained from such a low-cost
experiment.
2
MAGMA CHAMBERS IN
THE
ICELANDIC CRUST
The structure
of
the divergent Mid-Atlantic plate boundary
across Iceland is characterized by volcanic systems, which
are arranged en-echelon along the neovolcanic zones. Each
volcanic system consists
of
a central volcano, where the
productivity is highest, and a fissure swarm cutting across it
roughly parallel to the volcanic zone.
The
fissure swarms are
characterized by normal faulting during subsurface lateral
magma excursions from the central volcano and fissure
eruptions when those magma excursions reach the surface
(Bjornsson
et
al.
1977). A number
of
volcanic systems in
Iceland have been studied by various geophysical methods,
which yield information about their magma reservoirs. The
best studied volcano is Krafla in the northern part
of
the
eastern volcanic zone
of
the country. Einarsson (1978) used
local recording
of
seismicity to delineate a body causing a
shadow zone for
S
waves, presumably a magma chamber, at
depths between 3 and up to 7 km within the Krafla caldera.
Uncertainties in event locations and ray geometry render
the shape
of
this body uncertain. Tryggvason (1986) used tilt
and geodetic measurements to infer a possibly fourfold level
of
magma reservoirs under Krafla, a small chamber within
the crust at a depth
of
2.5 km, a second small chamber at
the base
of
the crust (depth >10km) and two additional
larger reservoirs in the mantle. Arnott (1990) used local
earthquakes and artificial sources recorded at
a
local array
of
seismic stations to map the shallow crustal structure in
and around the Krafla caldera. Limited depth penetration
of
rays and deteriorating resolution with depth precluded the
detection
of
a magma chamber, but fast bodies were
detected around the rim
of
the Krafla caldera at shallow
depths (depth
<3
km), interpreted as crystalline intrusives.
Brandsdottir
&
Menke (1992) synthesized multiple arrivals
in local three-component recordings to construct a l-D
model
of
compressional velocity. Their model contains a
1
km thick low-velocity layer at a depth of about
3
km in the
northern part
of
the Krafla caldera, which they interpret as a
magma chamber. Results from other volcanoes are more
sketchy. Toomey
&
Foulger (1989) mapped a weak
low-velocity anomaly at a depth
of
3
km under the Hengill
volcanic complex in south-west Iceland, which they
attributed to partial melt. Kjartansson
&
Gronvold (1983)
and Sigmundsson, Einarsson
&
Bilham (1992) derived a
poorly constrained depth
of
a magma reservoir under the
Hekla volcano in south Iceland at about 9 km from geodetic
measurements. Gudmundsson (1987) and others have
argued on the basis
of
geologic observations
of
extinct,
eroded central volcanoes in Iceland that largely molten,
shallow, crustal magma chambers are common. From the
above it is clear that little is known about their shape, size,
location,
or
stability.
In the above we have summarized what is known about
shallow, crustal magma chambers in Iceland.
A
separate
feature
of
Icelandic volcanism is what appears to be a
pervasive region
of
partial melt under the country. Beblo
&
Bjornsson (1980) and Eysteinsson
&
Hermance (1985)
mapped layers
of
high conductivity under profiles crossing
e
volcanic zones
of
Iceland using magnetotelluric
measurements. This layer extends beyond the flanks
of
the
volcanic zones. It is shallowest under the volcanic zones,
mapped there at a depth
of
approximately 10 km, but the
depth is rather uncertain. Furthermore, a strong tradeoff
persists between this layer’s conductivity and thickness.
Nevertheless, a high-conductance layer is required at depth
to explain the magnetotelluric data, a layer with properties
consistent with a degree
of
partial melting (Schmeling 1985).
A fairly clear picture
of
crustal magma chambers has
emerged from mid-ocean ridges in recent years. As
summarized by Sinton
&
Detrick (1992), fast, and
intermediate-spreading ridges appear to have thin molten
lenses at the top
of
broad mush zones
of
slight partial melt.
Slow ‘ridges, such as the Mid-Atlantic Ridge around Iceland,
on the other hand, have not been observed to have a molten
lens on top
of
the overall slow structures
of
the crust at the
ridge crest. Sinton
&
Detrick (1992) take this to suggest that
at slow-spreading ridges accumulation
of
magma in the crust
Katla volcano's crustal magma chamber
279
would have
to
be a transient affair. This contradicts the
geological evidence from Iceland, but more quantitative
geophysical studies are needed.
3
KATLA-TECTONIC
SETTING
The Katla central volcano is situated beneath the
central-southeastern part
of
Myrdalsjokull (a glacier), which
is located in the southern part
of
the Eastern Volcanic Zone
(EVZ), close to the south coast
of
Iceland (Fig. 1). The
EVZ is propagating southwestward, away from the centre
of
the hotspot in east-central Iceland into old oceanic crust at a
rate
of
3.5-5
cm yr-' (Einarsson 1991; Oskarsson, Steint-
harsson
&
Sigvaldason 1985). Several volcanic systems are
situated near the tip
of
this propagating volcanic zone. The
Vestmannaeyjar volcanic system (Heimaey and Surtsey),
off
the Iceland coast,
is
largely submarine. Two subglacial
volcanic systems are aligned along the south coast. They are
tectonically interconnected by east-west-striking faults and
eruptive fissures (see Fig. 1). The Eyjafjallajokull volcano,
further to the west, is seismically quiet, whereas the
volcano(es) beneath Myrdalsjokull is (are) persistently
active. The bimodal pattern of seismicity evident in Fig.
1
and the eruptive history suggest that two volcanoes coexist
under Myrdalsjokull (Thorarinsson 1975; Brandsdottir
&
Einarsson 1992), although they may represent two venting
areas
of
the same central volcano. Katla is the larger and
furthest to the east
of
the two.
The volcanic edifice of Myrdalsjokull is circular in outline,
with a diameter
of
about
30
km, and reaches a height of
about 1470 m above sea-level. The subglacial caldera has
been mapped by radio-echo sounding (Bjornsson, Palsson
&
Gudmundsson 1994) and is approximately 13 km in diameter
and up
to
600m deep. The Katla central volcano is one
of
the most active volcanoes of the EVZ with 17 documented
eruptions in the past
11
centuries. The last large eruption
occurred in 1918, and a small eruption may have occurred in
1955 without braking through the ice. Geothermal activity
persists to the present day.
The portion
of
the EVZ south
of
about 64"N is postulated
to represent a propagating volcanic zone. The reduced
significance
of
extensional features at the surface relative to
the stable rift zone to the north implies that this propagating
zone is not rifting as vigorously as the rift zone as
a
whole
(Einarsson 1991). The transitional character
of
the petrology
in this part
of
the volcanic zone
of
Iceland from olivine
fissure
swarm
-
Figure
1.
Location
of
and seismicity in the Myrdalsjokull area. Myrdalsjokull and Eyjafjallajokull are shown in black at the southern end
of
the eastern volcanic zone
(EVZ)
on
the insert map
of
Iceland in the upper left hand corner
(WVZ
on insert represents the western strand
of
the neovolcanic zone in southern Iceland). Two glaciers are found in the area, outlined here with
a
solid line and labelled by name
(Eyjafjallajokull to the west, Myrdalsjokull
to
the east). The caldera under Myrdalsjokull, the summit crater
of
Eyjafjallajokull, and some
normal faults in between the two glaciers are outlined with nicked lines. Eruptive fissures are indicated with fat solid lines. The south coast
of
Iceland appears as a stippled line at the bottom
of
the figure. Solid dots are all earthquakes located by the University
of
Iceland Seismic
network
during a seven year period
(1979-85).
The network operated four stations in the near vicinity
of
Myrdalsjokull (marked as triangles)
as part
of
a more extensive regional network
in
southern Iceland during this period. The straight line across the figure shows the location of the
profile measured in this study. The name Katla is used here to refer to the caldera under Myrdalsjokull and the area
of
high seismic activity
furthest to the east beneath the measured profile.
280
0.
Gudmundsson
et
al.
tholeiites of the rift zone through transalkalic basalt of the
propagating zone and alkali basalts at its southern tip
(Vestmannaeyjar) supports the notion that rifting tapers
off
to the south along the propagating zone (Oskarsson et
al.
1985; Steinthorsson, Oskarsson
&
Sigvaldason
1985).
Einarsson (1991) refers to Katla as an intraplate volcano.
The tectonics and petrology
of
the propagating volcanic
zone clearly distinguish it from the rift zone to the north,
but the two nevertheless form a continuous band of
volcanism.
4
THE EXPERIMENT
Our seismic experiment was designed to detect a possible
shallow magma chamber beneath the Katla caldera and
outline its structure by seismic undershooting. Explosions
were detonated to the north and south
of
the volcano and
2
E
25
N
I
seismic traveltimes measured on a linear profile crossing it
(Fig.
2).
The seismic waves dive into the Earth, turn at
depths
of
up to
7
km, and rise obliquely to the surface to
illuminate the magma chamber from below. Where paths
from different shot points cross, the data provide
2-D
resolution.
The profile consisted of approximately 40 seismic stations
crossing the Katla central volcano at an azimuth of N20W,
extending from the coast to about lOkm beyond the
northern edge
of
the glacier. Instrumentation consisted of
15
Scintrex PRS-4 digital recorders (100 samples
s-')
with
Rugby radio timing and Mark Products
L22D
three-
component
2
Hz
geophones. Repeated shots, consisting
of
15-100
kg
of
dynamite, were detonated in water at four sites
on the profile. One shot site was north
of
the glacier, the
other three to the south (see Fig.
2).
Two sections
of
the profile, at the northern and southern
C
Gaesavatn
b
Heidarvatn
Figure
2.
Layout
of
refraction profile and topography under the profile. (a) The Myrdalsjokull glacier and south coast
of
Iceland are drawn
for
reference. The broken solid line outlines the caldera under Myrdalsjokull. Dots represent cauldrons as observed in summer
of 1991,
at the time
of the experiment. Crosses represent instrument sites, solid stars explosive shot points in the ocean at
A
and in the lakes Heidarvatn,
Gaesavatn, and Alftavatn. (b) Topography
of
glacier surface and underlying bedrock under profile.
Katla volcano's crustal magma chamber
281
rims of the glacier, were inaccessible due to fracturing in the
ice. One shot
in
Alftavatn to the north failed due to failure
of
radio communication. Otherwise, data recovery was
about
90
per cent.
Instrument clocks were set by Rugby before each
deployment period and then compared with Rugby again as
the instruments were pulled out
of
the field. Thus, a clock
drift measurement was obtained over a time interval
of
the
order
of
one day. Time was then corrected in the data by
assuming linear clock drift within each recording interval.
We estimate that clock uncertainty is
of
the order
of
10-20 ms.
Shots were fired using a Nanometrics clock, which outputs
a pulse every minute. The clock was calibrated with the
Rugby time signal in the same manner as the recorder
clocks. Similar drifts and uncertainties are involved. The
dynamite was lowered to the bottom (50m in ocean,
20-30m in lakes). Only minor fountaining occurred, and
was limited to the larger shots.
Off the glacier, hard rock sites were sought out to deploy
the instruments. In all but a few cases
a
suitable outcrop was
found. On the glacier a hole about 2 feet deep was dug in
the surface firn, the instrument was then placed in the hole,
levelled, and the hole closed. Only in a few cases was
levelling significantly off when instruments were collected.
Overall response was good. The two sites at 10.3 and
11.6 km in Fig. 3 clearly produce anomalously low
amplitudes, however. They are both on a significant slope
on a length scale
of
tens and hundreds
of
metres.
5
THE DATA
Figures
3
and 4 show examples
of
the data collected on the
profile in record section. Fig. 3 is a record section
of
the
radial component from the two southernmost shot points.
Each section is globally normalized after scaling with
distance squared and inverse charge size. Distance is
measured from the southernmost shot point in both
sections.
P
and
S
waves are clearly identified at short
distances. At a distance
of
30-35 km the traveltime curve
for the
P
waves develops
a
sigmoidal shape and has clearly
late arrivals. Furthermore, the amplitudes of the
S
waves
are suddenly and drastically diminished in the same distance
range (30-35 km) and then reappear at distances beyond
40 km. This is already an indication that a highly attenuating
and seismically slow region exists beneath the array at this
location. A slight shift in this pattern between the two
sections (slightly further along profile (further from coast)
for the closer in shot at Heidarvatn, where incidence angles
are larger) implies that the source
of
these effects is at some
depth under the array. A similar shift in this pattern is
observed for the third of the southern shots and a reverse
shift for the shot from the north.
In Fig. 4 we show vertical and transverse components
of
the record section for the shot point in the ocean. The
arrival times vary smoothly and coherently with distance.
The sinusoidal variation of traveltime has a wavelength
of
about 10 km, which is much greater than the wavelength
of
the dominantly 5-10 Hz waves recorded (<1 km). The
smoothness of the variation in traveltime is an indication
that clock integrity in the experiment was good. The
signal-to-noise ratio is generally larger than 10 on the
vertical components. We can, therefore, pick the onset
of
the first wave with an accuracy
of
the order of 10-20ms.
The combined uncertainty (clocks and picks) of traveltime
measurements is thus about 25 ms. The traveltime anomaly
in Fig. 4 has an amplitude about 0.4s
(400ms),
more than
an order
of
magnitude larger than the uncertainty. The
traveltime differences between adjacent seismograms, less a
broader trend, are
of
the order
of
a few tens
of
ms.
One site
(at
31
km) was revisited with a second shot and has a
traveltime discrepancy
of
only 10 ms.
Our profile traverses a structure that is expected to be
strongly heterogeneous in three dimensions. Unfortunately,
the geometry
of
our
experiment precludes
3-D
inter-
pretation, and we are forced to interpret
our
observations in
terms
of
a cross-section. This assumption is justified by the
observation that the first 0.1-0.2
s
after the first arrival are
virtually devoid
of
energy on the transverse component (Fig.
4). Consequently, the first arrival
is
a wave that does not
travel significantly out
of
the plane
of
the profile.
The two sections in Fig. 4 are mutually normalized after a
simple scaling
of
the traces by the distance squared to
account for geometrical spreading effects. This is clearly a
very crude and artificial way
of
accounting for geometrical
spreading, but has been applied before (e.g. by Palmason
1971). Note the similarity
of
amplitude on the vertical
component in the two traces recorded at the same site with
shots
of
different size and only 300 m separation
of
the shot
points (at distance
of
31 km in Fig.
4,
traces drawn as bold
lines). This argues that the inverse scaling
of
traces with
charge size is appropriate. The highest amplitudes are
observed at the shortest range. They decay with distance
throughout the fast traveltime anomaly at distances between
25 and 31 km, particularly beyond the fastest arrivals at
28 km. Amplitudes increase again within the centre
of
the
slow anomaly at 33 km, but are very low in the fast anomaly
at 37-40 km.
A
similar pattern is observed for other shot
points. Two focusing effects may be identified in this
pattern. The slow anomaly at 33 km has a clear, but spatially
confined focusing effect, such as one might expect. The fast
anomaly, which causes the early arrivals at 37-40 km, has a
pronounced defocusing effect, indicating that this may be an
elongated structure along the ray paths. Note, that the
record section in Fig. 4 does not give the impression that
traveltimes can be determined accurately within this
low-amplitude anomaly (37-40 km). This is due to the
global normalization
of
the section. Despite a low-amplitude
signal, the signal-to-noise ratio remains high on the vertical
component, as is brought out, e.g. in a trace-normalized
section, thus allowing for accurate picking.
6
THE REGIONAL STRUCTURE
While the primary goal
of
this study is to observe and
describe the magma chamber under Katla, we chose to
extend observations as far as possible along the entire length
of
the profile with the purpose
of
building a crude regional
model. This regional model then serves the purpose of
defining ray paths or wave propagation outside the central
volcano, as well as an absolute reference for traveltime and
velocity variations within the volcano.
Before interpreting the traveltime measurements along
282
0.
Gudmundsson
et
a].
20
15
10
5
0
t
He
n
-
radial comDonent
!I
0-
10
20
30
1-
10
adial comDonent
distance (km)
40
Figure
3.
Examples of record sections compiled from the data. The radial components
for
two shots from the south are shown, with distance
measured along profile from the southernmost shot in the ocean. Note the correlation
of
shear-wave shadows with late arrivals
of
P
waves in
both sections. Each section is globally normalized after scaling with propagation distance and charge size as described in text.
the profile a number of corrections must be made:
(1)
for the significant topography that occurs along the
(2)
for the effect
of
propagation through glacial ice; and
(3)
for propagation through the sandy sedimentary
bottoms
of
the lakes in which shots were fired.
Fortunately, we have observations in the close vicinity
of
all the shot points except the one in the ocean. We cannot
measure the velocity
of
the surface sediment directly,
profile;
because
of
sometimes large differences
in
elevation between
shot and adjacent observations, and because observations
are ordinarily on hard rock (hyaloclastite). We can,
however, determine the thickness and velocity
of
these
sediments crudely from the nearby measurements of
traveltime, taking into account the topography and our own
observations
of
rock types at each site. We take the velocity
of
bedrock to be
3
km
s-'.
This is near the average
of
surface layer
0
velocities from Palmason
(1971).
This
is
also
in the range
of
velocities interpreted by Haraldsson
&
Palm
Katla volcano's crustal magma chamber
283
0.5
0.4
0.3
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
Ocean
-
vertical component
0.5
-
0.4
-
0.3
-
0.2
-
0.1
-
0.0
-
-0.1
-
-0.2
-
-0.3
-
-0.4
-
Ocean
-
transverse component
35 40 45
distance (km)
Figure
4.
Reduced-time record sections showing the vertical and transverse components
of
motion
for
shots in the ocean. Times
of
first-arrival
picks are marked with small solid bars. Note the smoothness
of
the traveltime anomaly and the good agreement between adjacent picks.
Also,
note the lack
of
energy on the transverse component during the first
0.1-0.2s
after the pick. Both sections are mutually and globally
normalized after scaling with propagation distance and charge size as described in text. Variable dashing helps tracing
of
individual
seismograms.
A
reduction velocity
of
5.79
km
s-'
was used.
(1980) as Quaternary eruptives, primarily composed
of
hyaloclastic tuffs and breccias, a description which fits well
the exposed rocks around much
of
Myrdalsjokull. We
determine velocities between
1.3
and 1.4 km s-l for the lake
sediments and thicknesses
of
100-150 m for the three
lakes. These velocities are lower than the speed
of
sound in
water (1.5 kms-'). They are, however, reasonable in light
of
the results
of
Morgan (1969) and Haraldsson
&
Palm
(1980). The resulting source corrections are about
50
ms far
away from the shot points, but are more significant at short
distance. We cannot claim these estimates to be more
accurate than within
20
ms. On the other hand they do serve
well as first-order estimates
of
the effect of source structure.
Much seismic work was done in the 1960s
on
glacial ice
caps and core samples
of
ice. Roethlisberger (1972) presents
a synthesis
of
much
of
this work. Fully compacted
(density
=
0.9 g cm-') cold ice (temperature
<
-10
"C)
has
a compressional velocity around
3.85
kms-' and a shear
velocity
of
1.95 kms-', with very limited scatter in
observations. Full degree
of
compaction is reached at
a
284
0.
Gudmundsson
et
al.
8-
n
E
is,
N
-
(4
I
I
I
I
I
//
V
0.4
I
1
I
I
I
-@'
-
~
'"0
10
20
30
40
50
60
I
I
I
I I I
I
0
10
20
30
40
50
60
distance
(km)
Figure
5.
Regional analysis
of
traveltimes. (a) A simple,
2-D,
regional velocity model constructed by perturbing earlier refraction results in a
forward modelling procedure.
(b)
Raypaths in the model in (a). (c) Measured traveltimes (symbols) corrected for clock drift, source effects and
topography, and the traveltimes predicted by the model in (a) and paths in
(b)
(thin curves). (d) The differences between the data and
theoretical traveltime curves in
(c)
define the residuals shown here. In (c) and (d) solid circles represent data from shot point in ocean, open
circles represent data from shot point in Heidarvatn, solid squares represent data from shots in Gaesavatn and open squares data from shots in
Alftavatn.
Katla volcano's crustal magma chamber
285
depth
of
a few tens
of
metres. Glaciers in Iceland are
temperate, and are at their freezing point throughout.
Owing to a marked temperature effect near the melting
point, temperate ice is somewhat softer than polar ice, and a
value
of
about 3.6kms-' is more appropriate for the
compressional velocity. This agrees with the velocities
modelled by Gudmundsson (1989) in the ice cap over the
Grimsvotn caldera in Vatnajokull, central Iceland. Shear
velocities fall in the range 1.7-1.8 km
s-'.
A
limited number
of observations presented by Roethlisberger indicate a
quality factor
of
the order
of
Q
=
100
for compressional
waves in temperate ice. No data on the quality factor for
shear waves in ice are known to us. The transition from
bedrock
(V,
=
3.0 km
s-',
p
=
2.5 to ice
(V,
=
3.6 km
s-l,
p
=
0.9 g cm-') represents only a minor velocity
contrast but a significant impedance contrast (with a
reflection coefficient of 0.3 at vertical incidence).
Reduced traveltime data and the building
of
the regional
reference model are summarized in Fig.
5.
Traveltimes and
residuals are presented after correction for source structure
(lake sediments) and topography. The topographic correc-
tion is done by full ray-tracing computations, using the
topography and ice thickness in Fig. 2(b) (Bjornsson,
Palsson
&
Gudmundsson 1994), projecting the endpoints
of
each ray on to the common level
of
500m above sea-level.
This correction,
or
equalization
or
reference, is done
assuming a velocity
of
2.5 km
s-'
above the datum
(0.5
km
above sea-level) and a velocity model below that level which
fits the average behaviour
of
our traveltime measurements
with offset (the best-fitting 1-D model). The topography and
ice thickness are well known (to within 10m, Bjornsson,
Palsson
&
Gudmundsson 1994). The velocity
of
ice is also
well known (to within
0.1
km
s-I,
Roethlisberger 1972).
These uncertainties translate into an insignificant uncer-
tainty in this rereferencing (less than 10 m
s,
combined).
Even if the reference velocity was incorrect by
0.5
km
s-l
over a depth interval
of
0.5
km, the error in the topographic
correction would be only on the order
of
25ms, which is
comparable with the combined picking and clock uncer-
tainty
of
individual measurements. The topographic
correction is thus robust and cannot account for the
magnitude
of
time residuals evident in Fig. 4.
The parameterization
of
the regional model is defined as
follows. First,
1-D,
piece-wise-linear velocity models are
defined for each
of
the three regions: south
of
the volcano
(S),
within the volcano (V), and north
of
the volcano (N).
The domain
of
each
of
these regions can be clearly seen in
Fig. 5(a). Then, surfaces
of
equal velocity are linearly
interpolated through
5
km wide transition regions. The
model in Fig. 5(a) was built by forward modelling. The fits
produced to the reduced traveltimes are shown in Fig. 5(c).
Obviously, we neither try to fit the traveltime variations
within the volcano (23-42 km) at this stage, nor can we by
using the present parameterization. It is in fact difficult to
improve this fit to the data as a whole with the present
parameterization. The ray paths in this model are shown in
Fig. 5(b), and traveltime residuals, as defined by the
difference between corrected, observed, traveltimes and
those predicted by the regional reference model, are shown
in Fig. 5(d).
Clearly, we do not resolve the regional structure within all
of
the cross-section shown in Fig. 5(a). Within the volcano
and to its north the deepest rays reach a depth
of
6-7 km.
To
its south the deepest rays turn at about
5
km depth.
Nevertheless, significant differences are resolved between
the region to the north
of
the volcano and to its south. The
top few km must be significantly faster in the south in order
to fit the observations near each shot point (see Fig. 5c).
Below a depth
of
about 2 km the structure to the south
requires
a
low velocity gradient, the velocity increasing by
about
0.1
km
s-'
per km from
a
velocity
of
5.0
km
s-'.
This
is in agreement with Palmason's results (1971) for profile 34,
which was recorded to the west from a shot point in
Heidarvatn. His interpretation has a thick Layer
2
(more
than 4.5 km thick) with a velocity
of
5.2 km
s-'
extending
from a depth
of
2.7 km. On the other hand, the modelled
velocity to the north continues to increase with depth, much
like models proposed by Flovenz (1980) and Flovenz
&
Gunnarsson (1991) for the uneroded, upper, Icelandic crust,
i.e. models appropriate for the neovolcanic zone. The
region at the northern end
of
the profile is within the EVS.
This regional velocity model should be considered crude
for the simple reason that we allow ourselves limited
freedom in its modelling by a rigid choice of parameteriza-
tion. It should, however, serve our purpose
of
describing the
absolute timing and ray geometry as the rays enter the
volcano. Having this regional reference model we proceed
to interpret the residuals in Fig. 5(d) in terms
of
the
structure within the volcano.
7
THE
LOCAL STRUCTURE
The traveltime residuals in Fig. 5(d), which fall within the
volcano, are reproduced in Fig. 6(a). Before modelling the
residuals it is worthwhile bringing out graphically their main
features. They all display a similar pattern. In Fig. 6(b) we
have taken the residuals from each
of
the shot points at
Heidarvatn, Gaesavatn, and Alftavatn and have shifted
them both vertically and laterally as a whole in order to
optimize their correlation with observations from the shot
point in the ocean.
A
simple pattern appears with limited
scatter. This demonstrates that the data from all shot points
can be thought
of
as containing the same pattern. The shifts
in time (vertical) are smaller than 0.1
s,
but are not
systematic among the three data curves. Neither are they
very significant. Slight modifications
of
the regional
reference model, not accommodated by the parameteriza-
tion employed in its construction, can easily generate such
shifts. Those shifts marginally exceed uncertainties in the
source corrections. The spatial shifts (horizontal) are, on the
other hand, both systematic and significant. If we define
these shifts such that the shots at Heidarvatn from the south
and Alftavatn from the north, which are at comparable
distance, are evenly distributed around zero, we get: +1.6,
-1.3, -1.6 and -2.1 km for Alftavan ocean, Heidarvatn,
and Gaesavatn, respectively. This is a regular pattern with
the angle
of
incidence at the array. The shots from the south
have projected an anomaly at depth northward by a
progressively greater distance as the angle of incidence is
increased and need to be pulled back southward to optimize
correlation. Conversely, the shot from the north has
projected the same anomaly southward, and the pattern
needs to be shifted back northward to optimize correlation.
The implication is clearly that a pattern
of
high, low, and
286
0.
Gudmundsson
et
al.
0.4
-(a)
I
I
I
I
20 25 30 35
40
45
I
I
I I
-0.41
,
I I
4
20
25
30
35
40 45
I
I
I
I
20
25
30
35
40
45
distance
(km)
Figure
6.
The traveltime residuals measured over the Katla
volcano. In (a) we show a portion
of
Fig. 5(d) enlarged. In (b) we
have applied a common shift in time and distance
to
all data from
each shot point, in order
to
optimize the coherency
of
pattern in the
residuals. For description and justification refer
to
text. In (c) we
show the residual data after modelling. Symbols are the same as in
Fig.
5.
high velocities exists at depth under the centre
of
the Katla
caldera. The sizes
of
the above shifts can be crudely taken to
imply a depth
of
a few km to this pattern.
The variation in traveltime residuals in Fig.
6
is
of
the
order of
0.4s,
which is roughly equivalent to the total,
vertical transit time of
P
waves through the top 1.5 km
of
the crust.
A
shallow source
of
such a large signal implies
extreme variations in velocity. This, in turn, causes
problems for standard techniques
of
ray tracing (Thurber
1987). As pointed out by Vidale (1988), multipathing and
artificial, numerical, ray-tracing shadows become prominent
in strongly heterogeneous media. He suggested an algorithm
for tracking the initial wavefront as an efficient method
of
predicting first arrival times in heterogeneous models. We
adopt his philosophy and develop a simple, efficient
algorithm in Appendix A.
Our
method is designed for a
smooth velocity structure and essentially applies Huygens'
principle at an arbitrary, densely sampled wavefront in
order to advance it by a fixed time-step. One drawback
of
such an approach lies in the fact that ray paths are not
generated, although they can in principle be constructed as
normals to the wavefronts. Thus, an inverse problem is not
readily set up. The simplicity
of
the geometry
of
the present
experiment, and the simplicity
of
the patterns
of
traveltime
residuals in its results (see Fig. 6b) do not, however, call for
an inversion procedure for interpretation. The task
of
forward modelling the data is simple enough. We note that
the wavefront tracker developed in Appendix A only tracks
the first arrival, which may be
of
very low amplitude. We
argue, however, that the high signal-to-noise ratio on the
recorded seismograms (see e.g. Fig.
4)
justifies such an
approach in this case.
We apply three basic constraints in performing the
forward modelling. The most basic assumption is that the
anomalies in traveltime observed from shots from the south
are caused by the same structural anomalies as traveltime
anomalies seen from the north. Strictly speaking this need
not be the case, because significant portions
of
the
propagation distance
of
the observed waves towards the
ends
of
the profile have
no
2-D
coverage. Consequently, we
are to some degree free to determine structure there entirely
independent
of
other parts
of
the model. We do not allow
ourselves this freedom for two reasons. First, structure is
more likely on geological grounds to exist under the volcano
than elsewhere. Velocity variations are most likely caused
by thermal anomalies and/or porosity anomalies. Thermal
anomalies are most likely associated with the plumbing
system
of
the Katla volcano, porosity anomalies are most
likely associated with the volcano's extrusion and intrusion
processes. Since the observed traveltime signals on the four
profiles have a common explanation in terms
of
structure
within the volcano, that is its most likely explanation.
Second, large perturbations, which evolve on the wavefront
near a shot point, due to heterogeneity there, will only
partially survive propagation to the array of instruments
over the volcano. Advanced portions
of
the wavefront
would expand and thereby heal
or
mask its delayed portions
(an effect called wavefront healing).
As
a consequence, we
actually do have limited freedom to explain the data with
structure towards the
ends
of
the profile. The wavefront
healing process is essentially a geometrical diffraction effect
and the scale over which it is effective is the first Fresnel
zone
of
constructive interference. At 5Hz frequency and
30 km source-receiver distance (characteristic
of
traveltime
measurements within the caldera) the width
of
the Fresnel
zone exceeds
5
km outside the caldera. The characteristic
length scale
of
the traveltime variation in Fig.
4
is
5
km.
Therefore, the traveltime signal seen in the data must be
explained by structure within the confines
of
the Katla
volcano. Additional constraints are that velocity not exceed
a value
of
6.0 km
s-l
in the top few km and that it not be
less than
2.5
km
s-'.
The upper bound velocity is near the
velocity
of
crystalline rocks
of
basaltic composition at
0
"C.
The lower bound velocity is near the acoustic velocity
of
basalts at
or
above their liquidus, according to Murase
&
McBirney (1973). It is also near the lower end
of
the
distribution
of
surface velocities observed by Palmason
Katla volcano's crustal magma chamber
287
Figure
7.
The results
of
forward modelling the observed traveltimes over the Katla volcano by wavefront tracking. The larger frame at the top
shows
our
model under the entire array defined by a grey shade ranging in velocity from
3.0
to
6.5
km
SKI,
with wavefronts at
0.1
s
intervals
for
a shot in Heidarvatn (explosion) superimposed. The smaller frame at the bottom shows the velocity perturbations
to
a 1-D model using the
same grey shade as in the upper frame, but representing velocity perturbations in the range
-2
to
2
km
s-'.
Distance is measured northward
along fhe profile (Fig.
1)
away from shot point in the ocean.
(1971) in the Icelandic crust. We perform the forward
modelling in terms of absolute time, not residuals, and solve
the full, non-linear problem.
Results
of
the modelling are shown in Fig. 7. The model is
shown in terms
of
both absolute velocity and velocity
perturbation. Also shown is an example
of
wavefronts in the
model
for
a shot in Heidarvatn. The main feature
of
the
model is a strong velocity depression at a depth of
2-3
km
beneath the central part
of
the glacier, flanked by thin,
sharp, oblique, fast anomalies. This is the source
of
the
fast/slow/fast sequence
of
anomalies, which dominates the
data. In that sense, this model is qualitatively inevitable.
There are critical questions. At what depths do the
structures belong? What are their shapes? What are their
geometrical sizes? What are their amplitudes? In a forward
modelling procedure these questions can be addressed by
performing experiments in which the model is perturbed.
Thus,
we estimate the depth
of
the bottom
of
the main slow
anomaly as 1.5
f
0.5
km below sea-level, and its width as
5
f
1
km. The model in Fig.
7
has velocity around
3
km
s-'
to a depth
of
0.5
km below sea-level, below which it falls to
2.5
kms-' within the slow anomaly. This small range
of
velocities cannot be varied much, given the lower bound we
place on the model velocity
(2.5
km
s-').
The fast flanks
of
the slow anomaly have a maximum velocity
of
6.0
km
s-'
to
the south and
5.5
km
s-'
to the north.
On
both sides they
are modelled as thin features. This is necessitated by the
need to advance very significantly the waves that travel
along trajectories which roughly parallel those structures,
while not advancing much the waves passing through them
at a right angle. The amplitude relations, e.g. evident in Fig.
4,
qualitatively support this model feature. The early
arrivals at
34-40
km in Fig.
4
are very low in amplitude and
have clearly been strongly defocused as
if
they have
travelled along a reverse waveguide. With the constraint of
the upper bound on velocity we are forced to infer high
velocities beneath the slow anomaly.
Owing to a small number
of
late arrival times near the
northern end
of
the glacier
we
infer a relatively weak
(1.0
kms-'), slow anomaly at shallow depths (at and below
sea-level) and at a distance
of
40
km from the southern end
of
the profile. This
is
based on consistent features on the
traveltime curves from all shot points, i.e. two adjacent
traveltime picks which are delayed by about
50ms.
The
observations, however, come from only four instruments.
Given that
50
ms only marginally exceeds estimated
uncertainties
of
25ms, and that
so
few instruments are
involved, which conceivably could have coherent clock
288
0.
Gudmundsson
et
al.
errors of this order, this feature must be interpreted
tentatively.
The root-mean-square (rms) value
of
the
95
data points in
Fig. 6(a) which were used in the modelling
of
the local
structure is
0.124
s.
The residual rms value is
0.023
s
for the
model in Fig.
7.
Thus, about
97
per cent
of
the variance is
explained, and the
level
of
residual data is comparable with
the error estimates discussed earlier. The residual data are
shown in Fig. 6(c).
In summary we can conclude that velocities as low as
2.5
km
s-'
exist to a depth of about
1.5
km below sea-level
under the centre
of
the Katla caldera. This low-velocity
anomaly is flanked by fast, thin, tilted structures of
velocities about 5.5-6.0 km
s-'
and underlain by velocities,
which are moderately high for that depth (5.5-6.0 km
s-I).
These structures are set against a background regional
velocity structure of velocity
4.5-5.5
km
s-'
at that depth.
We cannot determine the shapes
of
those main anomalies in
any detail. The model in Fig.
7
contains the detail needed to
explain the traveltimes. Adding further detail is neither
warranted by the uncertainty
of
the data, nor does the finite
wavelength
of
5Hz waves allow for more detail to be
interpreted.
The wavefronts in Fig.
7
show some interesting features
near the surface under the array. At a distance
of
about
25
km, near the southern fast anomaly, the angle
of
incidence varies quickly from steep to very oblique. It then
becomes progressively more oblique until at a distance
of
about
34km,
where it again changes quickly to become
essentially vertical. Where the wavefronts are convex,
focusing occurs. Where the wavefronts are horizontal,
compressional energy is all on the vertical component,
whereas for oblique angles of incidence the radial
component should carry a significant portion
of
the energy.
The shape
of
the wavefronts thus predicts a pattern in the
variation of amplitude along the profile. This pattern is seen
in
the data. The amplitudes of the radial component shown
in
Fig.
3
increase between distances of
25
and
32
km, where
the waves become progressively more oblique. Further on,
they are drastically reduced as the incoming waves become
vertical. The amplitude on the vertical component in Fig.
4,
on the other hand, gradually decreases between
25
and
32km,
increases again at
34
km due to focusing and a
change to a steep angle
of
incidence. Between
34
and
36
km
the amplitudes remain moderately high, until the angle of
incidence changes again to a more oblique one and the
strong defocusing effect
of
the northern fast anomaly is
encountered. Thus, amplitude relations in the compressional
waves support the modelling of traveltimes, at least
qualitatively.
8
DISCUSSION AND CONCLUSIONS
We interpret the low-velocity anomaly under the Katla
caldera as a magma chamber. Velocities as low as
2.5
km
s-l
are required to extend to a depth
of
2-3
km
within this anomaly, but could be lower yet from the
modelling point
of
view. That is, however, not feasible.
Extensive sedimentation would not occur under the flowing
ice cap
of
Myrdalsjokull and the Katla caldera is sufficiently
eroded
so
that water does not accumulate under the ice
(Bjornsson, Palsson
&
Gudmundsson
1994).
A velocity
of
2.5
kms-' is a feasible velocity
if
we interpret the
low-velocity anomaly as due to the presence of melt. That is
the speed
of
sound measured in basaltic melts (Murase
&
McBirney
1973).
Our interpretation is further supported by
the abrupt termination
of
shear-wave phases on the
Heidarvatn and ocean profiles from the south (and on the
northern profile from Alftavatn to a lesser degree), just as
the shear waves cross the proposed magma chamber (Fig.
3).
The third profile from the south (Gaesavatn) shows no
clear shear waves at all (apparently they were not strongly
excited by the explosive source). This shear-wave shadow is
an indication that the seismic waves have traversed a region
that contains fluids. Again, water in the ice
or
under the ice
cannot explain the shear-wave shadows. The locations
of
the
shadows correlate neither with ice thickness nor depressions
in the bedrock topography.
A possible, albeit unlikely, alternative interpretation
of
the low-velocity anomaly is in terms
of
caldera
fill.
Highly
porous rocks would have a low velocity and high
attenuation. Fresh Holocene lava fields in Iceland are found
to have an average surface velocity
of
about
2.5
km
s-',
sometimes as low as
2.1
km
s-'
(Palmason
1963, 1971),
and
porosity as high
as
20
per cent. On the other hand, recent
blocks of hyaloclastic material are found to have an average
surface velocity
of
2.9
km
s-l
(based on data
of
Palmason,
1963, 1971)
and much lower porosity. Given crude estimates
of
the volcanic productivity
of
Katla (Larsen
1994)
we can
estimate that the bottom
of
a caldera
fill
at
2.5
km depth
would have been erupted a few thousand centuries ago.
During most
of
that time the Katla volcano would have been
subglacial and the caldera
fill
therefore primarily composed
of
hyaloclastites. In order to model the observed
traveltimes, a velocity comparable with,
or
lower than, the
surface velocity
of
hyaloclastites is required to extend to a
depth
of
2-3
km. We therefore discard this intepretation.
We interpret the fast structures on both sides
of
the
magma chamber as crystalline intrusives. Again, higher
velocities are allowed from the modelling point
of
view, but
would be unrealistic. Velocities
of
5.5-6.0 km
s-'
are
feasible if we interpret these structures as volumes
possessing a high fraction
of
crystalline (aporous) rocks.
This is the compressional velocity
of
crystalline basaltic
rocks or gabbros as measured by Murase
&
McBirney
(1973).
These fast structures are thin and could be fans
of
conical dykes above the magma chamber. The fast velocities
could represent the chilled margins
of
an evolving magma
chamber where they occur at the sides and bottom
of
the
low-velocity anomaly. Note the asymmetry
of
these fast
structures. Velocities are significantly higher to the south
than to the north. This could to some degree be explained
by higher temperatures to the north, i.e. towards the EVZ.
Another possible explanation is that the volume fraction
of
intrusions is less to the north than south. This may be
related to Katla's position near the tip
of
a propagating
volcanic zone. A fissure swarm can be traced to the north
of
the caldera, but not to the south (Fig.
1).
Perhaps cooling
(and freezing) is more rapid to the south while magma has
more freedom to move laterally within the fissure swarm to
the north and thus forms conical intrusions less frequently.
The fast structures to the north of the magma chamber
lead up to the northern caldera rim. On the southern side,
on the other hand, they do not lead up to the caldera rim,
Katla volcano’s crustal magma chamber
289
but rather a hill
or
small mountain some distance away from
it, within the caldera. Comparison with Fig.
1
reveals that
the region of dense seismicity within the caldera is
terminated near this small mountain. It is likely that the
seismicity in Fig.
1
is confined to the region above the
magma chamber. The geometrical association
of
these fast
structures and the region of dense seismicity further
supports our interpretation
of
the modelled structures under
Katla as a magma chamber and associated structures.
The size
of
the Katla magma chamber is
of
interest. The
nearly constant velocity from the top
of
bedrock to the
bottom of the chamber does not clearly define its top. We
can, however, crudely use the velocity perturbation to
estimate it. We take a -1kms-’ velocity perturbation to
define the vertical extent
of
the magma chamber. Thus, we
estimate the thickness as
1
km and the width as
5
km as
before. We can use the relationship with the seismicity in
Fig.
1
to estimate the magma chamber’s width in the
direction transverse to the profile. The seismicity is
distributed over an oval area
of
width 11.5 km along the
profile and 9 km transverse to it. Assuming that the magma
chamber has the same aspect ratio in the horizontal plane,
its width is 4 km in the transverse direction. We thus
estimate a volume of about 10 kmP3, assuming an ellipsoidal
shape with the above dimensions. At a velocity of
2.5
kms-’ most
of
this volume is in a molten state. Were
the velocity
to
be somewhat higher, say 3kms-’, which
does not contradict our observations, one still estimates a
melt fraction of at least
50
per cent using the lower
Hashin-Shtrikman bound and reasonable values for the
elastic properties of melt and crystalline basalt. We,
therefore, estimate that the Katla magma chamber contains
at least
5
km3
of
melt. Tephra production in the 1918
eruption in Katla is estimated to have been 0.7km3
(Thorarinsson 1975), which corresponds to 0.2-0.4 km3
of
melt. Hence, even a large eruption, such as the one in 1918,
removes less than
8
per cent of the available melt. Katla has
erupted twice per century in the last 1000 years. Thus, a
production rate
of
less than 0.8 km3 century-’ (in terms
of
melt) can be inferred. That, in turn, implies magma
residence times in the Katla chamber
of
the order of at least
1000 years. The thermal (diffusive) lifetime
of
a magma
chamber
of
this size is on the other hand of the order
of
10
000
years.
The above interpretation
of
our results is in general
agreement with results from a dense gravity survey recently
conducted on Myrdalsjokull (Gudmundsson 1994). A broad
gravity high over the volcano is consistent with dense
material and high velocities at depth, which may correspond
to the fast region beneath the magma chamber. A
horseshoe-shaped gravity high within and around the
caldera rim, which is open to the north, suggests shallow
intrusions which correspond to the fast bodies that we model
on both sides on the chamber. The fact that this
horseshoe-shaped anomaly is open to the north agrees with
our result that the fast anomaly to the south is more
pronounced than to the north. The gravity map suggests that
the Katla magma chamber is indeed roughly circular in
outline and that it coincides with the eastern, circular cluster
of
seismicity in Fig.
1.
The volcanic production of Katla is primarily transalkalic
FeTi basalts. Fused silicic zenoliths are frequently found in
tephra emitted by the volcano. Phenocrysts in chilled glasses
are primarily plagioclase and augite (Steinthorsson
et
al.
1985, and references therein). As discussed by Meyer,
Sigurdsson
&
Schilling (1985) the morphology of phenocry-
sts from Katla may be explained in terms
of
mixing of
magmas
of
different compositions, such as evolved FeTi
basalt in a magma chamber an primitive olivine tholeiite
injected from below. With reference to physical modelling
of
convection and mixing in magma chambers (Huppert
&
Sparks 1980; Huppert, Turner
&
Sparks 1982; see Turner
&
Campbell (1986) for a comprehensive review), Meyer and
coworkers conclude that a magma chamber beneath Katla
should be characterized by long intervals between
replenishment, long residence times, and consequently a
relatively great degree
of
fractionation. This can be taken as
supporting evidence for our interpretation of the structure
under Katla, at least in terms
of
the existence
of
a chamber
and residence times, although an alternative interpretation
of
the petrology and geochemistry to that
of
Meyer
et
al.
(1985) may be equally viable (Oskarsson
et
al.
1985;
Steinthorsson
et
al.
1985). The replenishment interval is
uncertain.
The silicic rocks (xenoliths)
of
Katla are likely formed by
remelting
of
hydrated crust around the magma chamber as
this appears to be the case for Icelandic rhyolites on the
basis
of
light oxygen content, explained by incorporation
of
meteoric water (Muehlenbachs 1973), and other petro-
chemical and isotope data (O’Nions
&
Gronvold 1973;
Sigmarsson
et
al.
1991). As discussed by Turner
&
Campbell
(1986) this can occur at the magma chamber’s walls and
roof, and will inevitably occur to some degree since the
solidus of mantle-derived melt is much higher than the wet
solidus
of
significantly hydrated basalt. The light rhyolitic
melt would accumulate at the top
of
the chamber. Marsh
et
al.
(1991) discuss mechanisms by which this process can
become efficient, and by which the silicic melt can be
mobilized, i.e. by the initial melting
of
granophyric veins in
the chamber’s host rock, which results in spalling of the wall
rock, and subsidence
of
the chamber’s roof, which results in
extensive reworking
of
its rocks. The silicic melt may rise
above the magma chamber proper to form sills above the
chamber, such as Marsh
et
al.
(1991) suggest and which are
found, e.g. in the extensively drilled caldera
fill
of
the Krafla
central volcano in northern Iceland (Armannsson, Gud-
mundsson
&
Steingrimsson 1987). Thus, rhyolitic sills may
contribute to the overall slow velocities in the top
1.0-1.5 km above the Katla magma chamber.
High velocities underneath the Katla magma chamber are
indicative
of
the absence
of
melt and a reversed temperature
gradient. Therefore, the chamber is not fed by ascending
blobs
of
melt through a viscous rock matrix, but by episodic
injections through dykes. The most likely mechanism for
such injections is magma fracturing (e.g. Turcotte 1982).
For a discussion
of
the initiation
of
magma fracture see
Sleep (1988). Lister
&
Kerr (1991) analyse the mechanics
of
magma-fracture propagation after initiation. They show that
for reasonable values
of
magma flux, magma viscosity,
buoyancy, and depth, the width
of
the feeder dyke would be
of
the order of
1
m
or
less,
i.e.
comparable with dyke widths
found in Iceland (Gudmundsson 1983) and elsewhere.
Consequently, the feeder dyke is orders of magnitude too
thln to be seismically detectable and has a thermal life time
290
0.
Gudmundsson
et
al.
much shorter than for example the average 60yr intervals
between eruptions in Katla. Note that the eruption interval
need not necessarily be the same as the replenishment
interval. Eruptions from the shallow chamber may be
triggered by an extensional stress field near the surface
(Gudmundsson 1990) or the enhancement
of
buoyancy due
to fractionation within the chamber (Turner
&
Campbell
1986).
Lister
&
Kerr (1991) conclude that the dominant forces
governing magma fracture are a buoyancy driving force and
viscous dissipation. Owing to the local nature
of
this force
balance, magma fracturing ceases near the level
of
neutral
buoyancy. If a magma chamber is formed, and it is fed by a
feeder dyke, it should form at this level or overshoot it
slightly. The density
of
basaltic melt is 2.6-2.7g~m-~
(Murase
&
McBirney 1973). This corresponds to a seismic
velocity for oceanic layer 2 of the order of 5.0kms-'
(Christensen
&
Salisbury 1975). This velocity is found at a
depth
of
about
1.5
km below
sea-level
in the area around
Katla (see Fig.
5).
Hence, the Katla magma chamber resides
roughly at the level of buoyant equilibrium.
Katla is located near the tip of a volcanic zone which is
propagating into old oceanic crust towards the south. This
portion of the rift zones in Iceland is not believed to be
taking a full part in the rifting (Oskarsson
et
al.
1985;
Einarsson 1991). However, the depth to and size of the Katla
magma chamber is comparable with those
of
KraAa and
Askja, which are central volcanoes within the actively rifting
plate boundary (Brandsdottir, Menke
&
Gudmundsson
1992). These magma chambers in Iceland are fundamentally
different in form from the one on the East Pacific rise (EPR)
(Toomey
et
al.
1990), although both exhibit a shallow
emplacement in the crust in an extensional environment and
both may be steady state. The EPR magma chamber is
elongated along the ridge, while Katla and the other central
volcanoes in Iceland are well separated and circular in
outline, with roughly circular calderas, and presumably are
Elated to localized magma reservoirs. The molten zone at
the EPR seems to be very thin (<loom) and underlain by
seismically slow material, while the chamber under Katla
appears to be thick (about
1
km) and underlain by
seismically fast material. On the other hand, the volume
above the EPR chamber has relatively high velocities, while
the Katla chamber is overlain by seismically slow rocks, and
a clear velocity (impedance) contrast may not exist at its
roof. The depth
of
the EPR chamber also seems to be
considerably greater than would be predicted by buoyancy
(Hooft
&
Detrick 1993). The EPR magma chamber may be
more analogous to the partial-melt zone under Iceland,
which has been inferred from seismic and magnetotelluric
measurements (Schmeling 1985) and presumably forms
under a freezing horizon.
We
place these differences in the context
of
a thermal
model developed in Appendix B in order to emphasize the
difference between Iceland on one hand and mid-ocean
ridges on the other. The model is a simple, kinematic
convection model, which involves the rising
of
hot mantle
through a spreading zone
of
finite width, and thus finite
strain rate, and a continuous temperature distribution.
Cooling is by conduction
only.
Admittedly, other processes
play a role, such as crustal accretion and hydrothermal
cooling. The model does underscore the effect
of
the width
of
the volcanic zone, in which lies a major morphological
difference between Iceland and mid-ocean ridges proper.
Because
of
the slow spreading rate
of
the Mid-Atlantic
Ridge near Iceland (2 cm y-') the rate
of
upwelling is low.
Owing to the large width
of
the volcanic zone in Iceland
(50
km) as compared with normal ridges
(1
km) (Macdonald
1982), over which the upwelling is distributed, the rate
of
upwelling is further retarded in Iceland relative to the ridge
system as a whole.
As
a consequence, the crust and mantle
immediately underneath the volcanic zone in Iceland are
much cooler than beneath the volcanic zones
of
the ridges.
In particular, the depth
of
the 1200°C isotherm (mantle
solidus, liquidus
of
basaltic crust) is an order
of
magnitude
greater in Iceland than at an intermediate-spreading ridge.
Taking the deep mantle temperature to be 1350°C under
ridges and
1550
"C under Iceland, the width
of
the volcanic
zone at a ridge to be
1
km and 40 km in Iceland, and taking
the spreading rate to be
8
cm yr-' at the ridge and 2 cm yr-I
in Iceland, our thermal model predicts a depth
of
about
1
km to the 1200 "C isotherm at the ridge and about 10 km
under the Icelandic rift zone. While this difference cannot
be taken literally, due to the simplicity
of
our thermal
model, it is a clear indication
of
a dramatic difference in the
thermal regime and a semi-quantitative estimate thereof.
For Katla one expects this difference in thermal structure to
be even greater, because Katla is placed outside the fully
active rift zones
of
Iceland. Also, one expects the normal
spreading ridge to be cooler than is predicted by our thermal
model due to unmodelled hydrothermal cooling. However,
velocity models of the EPR chamber are consistent with a
positive thermal gradient with depth under the chamber and
a mush zone
of
near-solidus rock (lOOO°C) under the
chamber (Macdonald 1989; Sinton
&
Detrick 1992). In
other words, the 1000°C isotherm reaches a depth
of
only
1-2 km under the volcanic zone
of
a fast-spreading ridge on
a regional scale. This suggests that the hydrothermal cooling
effect is confined to the top
1
km
of
the crust, and that it
represents only a relatively minor perturbation to our
thermal model (at least in terms
of
the deeper thermal
structure). On the other hand, the localized Katla chamber
is clearly underlain by a reverse thermal gradient.
On the ridge system, which is characterized by passive
upwelling, temperatures remain near the melting curves
of
crustal and mantle rocks up
to
a
freezing horizon at 1-2 km
depth and magma ascends smoothly up to this level, where
it may accumulate. In Iceland, on the other hand, the crust
in the volcanic zone is cool due to slow upwelling under a
broad volcanic zone. The regional freezing horizon
of
melt
is far below the level
of
neutral buoyancy (LNB). Ascending
magma encounters a barrier to its quest to reach the LNB,
magma accumulates underneath this barrier, and magmatic
overpressure builds up until the magma breaks its way
through the overyling lithosphere to the LNB or all the way
to the surface. On the ridge system the LNB is shallow
(<400m depth) (Hooft
&
Detrick 1993) and the
overshooting
of
the LNB by magma fractures, predicted on
theoretical grounds by Lister
&
Kerr (1991) and observed in
tank experiments (Turner
&
Campbell 1986), reduces the
likelihood that magma accumulates at the LNB at
a
normal
spreading centre.
Melt accumulation in
a
region
of
high melt fraction under
a freezing horizon is likely to cause some thermal erosion
of
Katla volcano's crustal magma chamber
291
the lithosphere above if melt ascent below it is spatially
uneven. Thus, localized magma wells can form as Meyer
et
af.
(1985)
envisaged (their fig.
17)
over which a central
volcano could evolve. It is interesting to note that this
appears to occur in the deep, broad, and possibly thick
magma reservoir under Iceland, but not in the shallow,
narrow, and relatively thin magma reservoirs under the
ridges.
ACKNOWLEDGMENTS
We thank
H.
Brynjolfsson,
H.
Olafsson,
J.
Holmjarn,
K.
Ingoifsson, K. Palsson, R. Ragnarsson,
D.
Reynisson, G.
Skagfjord,
E.
Sturkell, M.
T.
Gudmundsson, and the
helicopter crew
of
the Icelandic Coast Guard for their
valuable assistance in the field. We are grateful
to
T.
Palsson and the family at Litla Heidi and the family at
Kirkjuvegur in Vik for their help and hospitality. Amy
Clement helped prepare and calibrate the instrumentation.
The National Life Saving Association
of
Iceland and the
National Energy Authority supplied boats and offroad
vehicles. We are indebted to H. Bjornsson and
F.
Palsson
and
M.
T.
Gudmundsson for making their ice thickness and
gravity maps available to us prior to publication. The
manuscript was significantly improved by the critical reviews
of
Pall Einarsson, Colin Thomson, Malcolm Sambridge,
Brian Kennett and
an
anonymous reviewer. This work was
supported by the Icelandic Science Foundation, the
Icelandic Road Authority, the Icelandic Coast Guard, the
US
National Science Foundation, the Australian National
University, and the Fulbright Foundation.
REFERENCES
Achauer,
U.,
Evans, J.R.
&
Stauber, D.A., 1988. High resolution
seismic tomography of compressional wave velocity structure at
Newberry volcano, Oregon Cascade Range,
J.
geophys. Res.,
Armannsson, H., Gudmundsson,
A.
&
Steingrimsson, B., 1987.
Exploration and development of the Krafla geothermal area,
Arnott, S.K., 1990. A seismic study of the Krafla volcanic system,
Iceland,
PhD thesis,
University of Durham.
Beblo, M.
&
Bjornsson, A., 1980. A model of electrical resistivity
beneath NE-Iceland, correlation with temperature,
J.
Geophys.,
47, 184-190.
Benz, H.M.
&
Smith, R.B., 1984. Simultaneous inversion for
lateral velocity variations and hypocenters in the Yellowstone
region using earthquake and refraction data,
J.
geophys. Rex,
Bjornsson, A., Saemundsson, K., Einarsson,
P.,
Tryggvason, E.
&
Gronvold, K., 1991. Current rifting episode in north Iceland,
Nature,
266,
318-323.
Bjornsson,
H.,
Palsson, F.
&
Gudmundsson, M.T., 1994. The
topography of the Katla caldera beneath the ice cap
Myrdalsjokull, south Iceland,
EOS Trans. Am. geophys. Un.,
75, 321.
Black, R.A., Deemer,
S.J.
&
Smithson, S.B., 1991. Seismic
reflection studies in Long Valley caldera, California,
J.
geophys. Rex,
%,
4389-4300.
Brandsdottir, B.
&
Einarsson, P., 1992. Volcanic tremor and
low-frequency earthquakes in Iceland, in
Volcanic Seismology,
eds Gasparini
et
al.,
IA
VCEI Proc. Volcan.,
3,
212-222.
Brandsdottir, B.
&
Menke,
W.,
1992. Thin low-velocity zone within
93, 10 135-10 148.
Jokull,
37, 13-29.
89, 1208-1220.
the Krafla caldera, attributed to a small magma chamber,
Geophys. Res. Lett.,
19, 2381-2384.
Brandsdottir, B., Menke,
W.H.
&
Gudmundsson,
O.,
1992. The
role of shallow crustal magma chambers in accommodating
rising magma and their influence on remelting and recycling of
the crust: Case histories from Askja, Krafla, and Katla central
volcanos, Iceland,
EOS Trans. Am. geophys. Un.,
73,
647.
Christensen, N.I.
&
Salisbury, M.H., 1975. Structure and
constitution of the lower oceanic crust,
Rev. Geophys. Space
Claerbout, J.F., 1985.
Imaging the Earth's Interior,
Blackwell
Scientific Publications, Oxford.
Detrick,
R.S.,
Buhl, P., Vera, E., Mutter, J., Orcutt, J., Madsen,
J. and Brocher, T., 1987. Multichannel seismic imaging of a
crustal magma chamber along the East Pacific Rise,
Nature,
Einarsson, P., 1978. S-wave shadows in the Krafla Caldera
in
NE-Iceland, evidence for a magma chamber in the crust,
Bull.
Volcanol.,
41, 187-195.
Einarsson,
P.,
1991. Earthquakes and present-day tectonism in
Iceland,
Tectonophysics,
189, 261-279.
Evans, J.R.
&
Zucca, J.J., 1988. Active high resolution seismic
tomography
of
compressional wave velocity and attenuation
structure at Medicine Lake volcano, northern California
Cascade Range,
J.
geophys. Res.,
93,
15 016-15 036.
Eysteinsson, H.
&
Hermance, J.F., 1985. Magnetotelluric
measurements across the eastern neovolcanic zone in south
Iceland,
J.
geophys. Res.,
90,
10093-10 103.
Flovenz, O.G., 1980. Seismic structure of the Icelandic crust above
layer three and the relation between body wave velocity and
the alteration
of
the basaltic crust,
J.
Geophys.,
47, 21 1-220.
Flovenz,
O.G.
&
Gunnarsson, K., 1991. Seismic crustal structure in
Iceland and surrounding area,
Tectonophysics,
189, 1- 17.
Flovenz, O.G.
&
Samundsson, K., 1993. Heat flow and geothermal
processes in Iceland,
Tectonophysics,
225,
123-138.
Gudmundsson, A., 1983. Form and dimensions of dikes in eastern
Iceland,
Tectonophysics,
95, 295-307.
Gudmundsson,
A.,
1987. Formation and mechanics of magma
reservoirs in Iceland,
Geophys.
J.
R. astr. SOC.,
91,27-41.
Gudmundsson, A., 1990. Emplacement
of
dikes, sills and crustal
magma chambers at divergent plate boundaries,
Tectonophysics,
176, 257-275.
Gudmundsson, M.T., 1989. The Grimsvotn caldera, Vatnajokull:
Subglacial topography and structure of the caldera infill,
Gudmundsson, M.T., 1994. The structure
of
Katla, a central
volcano in a propagating rift zone, south Iceland, from gravity
data,
EOS Trans
Am.
geophys. Un.,
75, 335.
Haraldsson,
H.
&
Palm, H., 1980.
A
Seismic Investigation in the
Marka@jot Sandur Area, Southern Iceland,
Societas Upsalien-
sis pro Geologia Quaternaria, Upsala.
Harding, A.J., Orcutt, J.A., Kappus, M.E., Vera, E.E., Mutter,
J.C., Buhl, P., Detrick, R.S.
&
Brocher, T.M., 1989. Structure
of young Oceanic crust at
13"N
on the East Pacific Rise from
expanded spread profiles,
J.
geophys. Rex,
94,
12 163-12 196.
Hooft, E.E.
&
Detrick, R.S., 1993. The role of density in the
accumulation of basaltic melts at mid-ocean ridges,
Geophys.
Res. Lett.,
26,
423-426.
Huppert, H.
&
Sparks, R.S.J., 1980. The fluid dynamics
of
a
basaltic magma chamber replenished by influx
of
hot, dense
ultrabasic magma,
Contrib. Mineral. Petrol.,
75, 279-289.
Huppert, H., Turner, J.S.
&
Sparks, R.S.J., 1982. Replenished
magma chambers: Effects of compositional zonation and input
rates,
Earth planet. Res. Lett.,
57, 345-357.
Kjartansson, E.
&
Gronvold, K., 1983. Location of a magma
reservoir beneath Hekla volcano, Iceland,
Nature,
301,
Larsen, G., 1994.
Sur
og basisk gjoskulog fra Kotlu-er gossaga
Phys.,
13,
57-86.
326,
35-41.
Jokull,
39, 1-19.
139- 141.
292
0.
Gudmundsson
et
al.
sidustu loo0 ara daemigerd fyrir eldstodina? (Rhyolitic and
basaltic tephra from Katla-are the past
lo00
years
characteristic
of
the volcano’s history, in Icelandic),
Kotlustefna, Proc.
geol.
SOC. Iceland,
4-5, Reykjavik, Iceland..
Lin, J.
&
Parmentier, E.M., 1989. Mechanisms
of
lithosphere
extension at mid-ocean ridges,
Geophys.
J.
lnt.,
%,
1-22.
Lister, J.R.
&
Kerr, R.C., 1991. Fluid-mechanical models
of
crack
propagation and their application to magma transport in dikes.
J.
geophys. Res.,
%,
10049-10077.
Macdonald, K.C., 1982. Mid-ocean ridges: fine scale tectonic,
volcanic and hydrothermal processes within the plate boundary
zone,
Ann. Rev. Earth planet. Sci.,
10,
155-190.
Macdonald, K.C., 1989. Anatomy
of
the magma reservoir,
Nature,
Marsh, B.D., Gunnarsson,
B.,
Congdon, R.
&
Carmody, R. 1991.
Hawaiian basalt and Icelandic rhyolite: indicators
of
differentiation and partial melting,
Geol. Rundschau,
80,
McKenzie,
D.,
1984. The generation and compaction
of
partially
molten rock,
J.
Petrol.,
25,
713-765.
Meyer, P.S., Sigurdsson,
H.
&
Schilling,
J.G.,
1985. Petrological
and geochemical variations along Iceland‘s neovolcanic zones,
J.
geophys. Res.,
90,
10 043-10 072.
Morgan, N.A., 1969. Physical properties
of
marine sediments as
related to seismic velocities,
Geophysics,
34,
529-545.
Muehlenbachs,
K.,
1973. The oxygen isotope geochemistry
of
siliceous volcanic rocks from Iceland,
Carnegie Institution
of
Washington Year
Book,
72, 593-597.
Murase,
T.
&
McBirney, A.R., 1973. Properties
of
some common
igneous rocks and their melts at high temparatures,
Geol. SOC.
Am.
Bull.,
84,3563-3592.
Nercessian, Al., Hirn, Al.
&
Tarantola, Al., 1984. Three-
dimensional seismic transmission prospecting
of
the Mont Dore
volcano, France,
Geophys.
J.
R. astr. SOC.,
76,307-315.
O’Nions, R.K.
&
Gronvold, K., 1973. Petrogenetic relationship
of
silicic and basic rocks in Iceland:
Sr
isotopes and rare earth
elements in late and post glacial volcanics,
Earth planet. Sci.
Lett.,
19,
397-409.
Oskarsson, N., Steinhorsson,
S.
&
Sigvaldason, G.E., 1985. Iceland
geochemical anomaly: origin, volcanotectonics, chemical
fractionation and isotope evolution
of
the crust,
J.
geophys.
Res.,
90,
10011-10025.
Palmason, G., 1963. Seismic refraction investigation
of
the basalt
lavas in northern and eastern Iceland,
Jokull,
13,
40-60.
Palmason, G., 1971.
Crustal Structure
of
lceland from Explosion
Seismology,
Societas Scientarum Islandica, Reykjavik.
Palmason, G., 1986. Model
of
crustal formation in Iceland, and
application to submarine mid-ocean ridges, in
The Geology
of
North America,
Vol. M,
The Western North Atlantic Region,
pp. 87-97, eds Vogt, P.R.
&
Tuchilke, B.E.
Phipps-Morgan, J.
&
Chen, Y.J., 1993. The genesis
of
oceanic
crust: Magma injection, hydrothermal circulation, and crustal
flow,
J.
geophys. Res.,
98,
6283-6297.
Phipps-Morgan, J., Parmentier, E.M.,
&
Lin,
J.,
1987. Mechanisms
for the origin
of
mid-ocean ridge axial topography: Implications
for
the thermal and mechanical structure at accreting plate
boundaries,
J.
geophys. Res.,
92,
12 823-12 836.
Podvin, P.
&
Lecomte, I., 1991. Finite difference computation of
travel times in very constrasted velocity models: a massively
parallel approach and its associated tools,
Geophys.
J.
Int.,
Roethlisberger, H., 1972.
Seismic Exploration in Cold Regions,
Monograph
11-A2,
Cold Regions Research and Engineering
Laboratory, Hanover,
NH.
Schmeling, H., 1985. Partial melt below Iceland: a combined
interpretation
of
seismic and conductivity data,
J.
geophys.
Res.,
90,
10 105-10 116.
Sigmundsson,
F.,
Einarsson, P.
&
Bilham, R., 1992. Magma
339,
178-179.
481-510.
I
105,
271-284.
chamber deflation recorded
by
the global positioning system:
The Hekla 1991 eruption,
Geophys. Res. Lett.,
19,
1483-1486.
Sigmarsson,
0..
Hemond,
C.,
Condomines, M., Fourcade,
S.
&
Oskarsson,
N.,
1991. Origin
of
silicic magma in Iceland
revealed by Th isotopes,
Geology,
19,
621-624.
Sinton, J.M.
&
Detrick, R.S., 1992. Mid-ocean ridge magma
chambers,
J.
geophys.
Res.,
97,
197-216.
Sleep, N.H., 1988. Tapping
of
melt by veins and dikes,
1.
geophys.
Res.,
93,
10 255-10 272.
Snieder, R.
&
Sambridge, M., 1992. Ray perturbation theory for
travel times and ray paths in
3D
heterogeneous media,
Geophys.
J.
lnt.,
109,
294-322.
Steinthorsson,
S.,
Oskarsson,
N.
&
Sigvaldason, G.E., 1985. Origin
of
alkali basalts in Iceland: a plate tectonic model,
J.
geophys.
Res.,
90,
10
027-10 042.
Thorarinsson,
S.,
1975. Katla og annall Kotlugosa (Katla and its
eruption history), in
lceland Travel Society Yearbook,
pp.
Thurber, C.H., 1984. Seismic detection
of
the summit magma
complex
of
Kilauea volcano, Hawaii,
Science,
223,
165-167.
Thurber, C.H., 1987. Analysis methods for kinematic data from
local earthquakes,
Rev.
Geophys.,
24,
793-805.
Thurber, C.H.
&
Aki, K., 1987. Three-dimensional seismic
imaging,
Ann. Rev. Earth planet. Sci.,
15,
115-139.
Thurber, C.H.
&
Ellsworth, W.L., 1987. Rapid solution
of
ray
tracing problems in heterogeneous media,
Bull. seism.
SOC.
Am.,
70,
1137-1148.
Toomey, D.R.
&
Foulger, G.R., 1989. Tomographic inversion
of
local earthquake data from the Hengill-Grensdalur central
volcano complex, Iceland,
1.
geophys.
Rex,
94,
17 497-17
510.
Toomey, D.R., Purdy, G.M., Solomon,
S.
&
Wilcox, W., 1990.
The three dimensional seismic velocity structure
of
the East
Pacific Rise near latitude 9”30‘N,
Nature,
347,
639-644.
Tryggvason,
E.,
1986. Multiple magma reservoirs in a rift zone
volcano; ground deformation and magma transport during the
September 1984 eruption
of
Krafla, Iceland,
J.
Volcan.
Geotherm. Res.,
28,
1-44.
Turcotte, D.L., 1982. Magma migration,
Ann.
Rev.
Earth planet.
Sci.,
10,
397-408.
Turcotte, D.L.
&
Schubert, G., 1982.
Geodynamics: Applications
of
Continuum Physics to Geological Problems,
John Wiley
&
Sons, New York.
Turner, J.S.
&
Campbell, I.H., 1986. Convection and mixing in
magma chambers,
Earth Sci. Rev.,
23,
255-352.
Vera,
E.E.,
Mutter, J.C., Buhl, P., Orcutt,
J.A.,
Harding,
A.J.,
Kappus, M.E., Detrick, R.S.
&
Brocher, T.M., 1990. The </