Crustal deformation associated with the 1996 Gjálp subglacial eruption, Iceland: InSAR studies in affected areas adjacent to the Vatnajökull ice cap

Abstract

Crustal deformation signals associated with the September 30–October 13, 1996 Gjálp subglacial eruption, Vatnajökull ice cap, Iceland, have been identified using interferometric analysis of Synthetic Aperture Radar images (InSAR) in areas outside the ice cap. On September 29, 1996 an M w 5.6 earthquake occurred at the nearby Bárdarbunga volcano and on September 30 seismicity propagated 20 km southwards where the Gjálp eruption occurred. Analysis of interferograms spanning different times from 1992 to 2000 allows us to separate two distinct co-eruptive deformation periods in areas outside the ice cap. Diking at the Bárdarbunga caldera rim appears to be responsible for deformation during the first week of eruption while significant deflation occurred at Bárdarbunga only after October 6 when most of the magma had already been erupted at Gjálp. A pressure connection between the Bárdarbunga volcano and the Gjálp eruptive fissure is inferred. Fault slip in three areas up to 30 km from the center of the Bárdarbunga volcano was triggered by the deflation. Local deformation signals there are consistent with small fault movements.

Full text

Page 1

Crustal deformation associated with the 1996 Gjálp subglacial
eruption, Iceland: InSAR studies in affected areas adjacent
to the Vatnajökull ice cap
Carolina Paglia,⁎,1, Freysteinn Sigmundssona, Rikke Pedersena, Páll Einarssonb,
Thóra Árnadóttira, Kurt L. Feiglc,2
aNordic Volcanological Center, Institute of Earth Sciences, University of Iceland, Reykjavík, Iceland
bInstitute of Earth Sciences, University of Iceland, Reykjavík, Iceland
cCentre National de la Recherche Scientifique, Toulouse, France
Received 20 July 2006; received in revised form 4 April 2007; accepted 5 April 2007
Available online 19 April 2007
Editor: C.P. Jaupart
Abstract
Crustal deformation signals associated with the September 30–October 13, 1996 Gjálp subglacial eruption, Vatnajökull ice cap,
Iceland, have been identified using interferometric analysis of Synthetic Aperture Radar images (InSAR) in areas outside the ice
cap. On September 29, 1996 an Mw5.6 earthquake occurred at the nearby Bárdarbunga volcano and on September 30 seismicity
propagated 20 km southwards where the Gjálp eruption occurred. Analysis of interferograms spanning different times from 1992 to
2000 allows us to separate two distinct co-eruptive deformation periods in areas outside the ice cap. Diking at the Bárdarbunga
caldera rim appears to be responsible for deformation during the first week of eruption while significant deflation occurred at
Bárdarbunga only after October 6 when most of the magma had already been erupted at Gjálp. A pressure connection between the
Bárdarbunga volcano and the Gjálp eruptive fissure is inferred. Fault slip in three areas up to 30 km from the center of the
Bárdarbunga volcano was triggered by the deflation. Local deformation signals there are consistent with small fault movements.
© 2007 Elsevier B.V. All rights reserved.
Keywords: magma intrusion; Gjálp eruption; multiple magma chambers; fault slip triggering
1. Introduction
A subglacial eruption beneath the Vatnajökull ice cap
in 1996, at the Gjálp eruptive site, was the largest in
Iceland in terms of volume (0.45 km3of erupted
magma) since 1967 (Gudmundsson et al., 2004). The
eruption occurred on a fissure midway between two of
Iceland's most active volcanoes, Grímsvötn and Bár-
darbunga (Fig. 1). The process leading to the eruption as
well as links between the eruption and the magma
systems of nearby volcanoes is enigmatic. Seismic
activity has been interpreted in terms of magma
movement from Bárdarbunga (Einarsson et al., 1997)
and as ring faulting on the caldera (Nettles and Ekström,
1998), whereas a different seismic interpretation
(Vogfjörd et al., 1999) and a geochemical study
Earth and Planetary Science Letters 259 (2007) 24–33
www.elsevier.com/locate/epsl
⁎Correspondingauthor.Tel.:+3524666446273;fax:+3524666446567.
E-mail address: carolina.pagli@uni.lu (C. Pagli).
1Present address: Faculty of Sciences, Technology and Commu-
nications, University of Luxembourg, Luxembourg.
2Present address: Department Geology & Geophysics, University of
Wisconsin, Madison, United States.
0012-821X/$ - see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsl.2007.04.019

Page 2

(Sigmarsson et al., 2000) favor a link to the Grímsvötn
magma system.
Bárdarbunga has been one of the most seismically
active volcanoes in Iceland. No earthquakes reaching
magnitude 5 had been reported in the area since
instrumental measurements began in the 1920's until
1974 (Einarsson, 1991). Then, activity increased
abruptlyinJune1974, when anearthquake ofmagnitude
5 occurred (Einarsson, 1991). Since then, ten Mw
5.0–5.6 earthquakes have been recorded at Bárdar-
bunga. The earthquake epicenters cluster slightly NE of
the caldera and fault plane solutions indicate mostly
reverse faulting. The seismic activity has been inter-
preted as a response to decreasing pressure in a magma
chamber, occurring since 1974 (Einarsson, 1991).
Eleven eruptions have been recognized at Bárdarbunga
since 1200 AD based on ash layers (Larsen et al., 1998),
with the most recent rifting episode in 1862–1864.
Grímsvötn has the highest eruption frequency of all
volcanoes in Iceland, with at least thirty eruptions since
1200 AD (Larsen et al., 1998). The most recent
eruptions occurred in 1983, 1998 (Sturkell et al.,
2003a) and 2004 (Vogfjörd et al., 2005). On the other
hand, large earthquakes are not known in the vicinity of
Grímsvötn. Seismicity there is most pronounced
preceding and during eruptions (Einarsson and Brands-
dóttir, 1984; Vogfjörd et al., 2005). Grímsvötn showed
increased seismic activity in the 3–5 months before its
1983 eruption, probably due to inflation of a magma
chamber beneath the southeastern rim of its caldera
(Einarsson, 1991). The seismicity stopped during the
eruption but then increased again after the eruption. The
activity ceased in 1984 with a burst of volcanic tremor
(Björnsson and Einarsson, 1990).
TheGjálpsubglacialeruptionbeneaththeVatnajökull
ice cap lasted from September 30 to October 13, 1996
(Einarsson et al., 1997; Gudmundsson et al., 1997). An
Mw5.6 earthquake occurred at 3.6 km depth near the
northern caldera rim of the Bárdarbunga volcano on
September29at10:48:19GMT(Fig.1)(Einarssonetal.,
1997; Nettles and Ekström, 1998; Vogfjörd et al., 1999).
Unlike previous large events at Bárdarbunga, this
earthquake was immediately followed by intense
swarm activity on the western rim of the caldera. The
swarm propagated 5 km south within few hours,
suddenly stopping at the SW edge of the Bárdarbunga
caldera around 14:00:00 GMT (Vogfjörd et al., 1999).
Earthquake activity increased again on September 30,
now south of Bárdarbunga, and reached a maximum
before decreasing around 22:00:00 GMT, when a N–S
trending 7 km long eruptive fissure opened midway
between Bárdarbunga and Grímsvötn volcanoes at a
distance of 20 km from the Mw5.6 earthquake (Fig. 1).
During the first two days of the eruption some
earthquakes occurred around the eruptive fissure and at
the SE and NE edges of the Bárdarbunga caldera. The
eruption produced 0.8 km3of basaltic glass, correspond-
ing to a total magma volume of 0.45 km3assuming a
magma density of 2600 kg/m3(Gudmundsson et al.,
2004). The eruption rate was estimated to be
3.5×106kg/s during the first three days of eruption but
it was considerably lower during the second week. Melt
Fig. 1. SAR amplitude image of NW Vatnajökull from 06/10/96 to 13/
07/97. Overlaid are central volcanoes, calderas (dotted lines) and
outlines of fissure swarms (black lines) from Einarsson and
Saemudsson (1987): B—Bárdarbunga, G—Grímsvötn and T—
Tungnafellsjökull, (Gj) Gjálp eruptive fissure. Epicenter location of
the Mw 5.6 event on September 29, 1996 is shown with a star.
Earthquakes from September 29 to October 13, 1996 are shown with
white asterisks. Note the lack of earthquakes at the Grímsvötn caldera.
The rectangular box gives the location of the interferograms in Fig. 2.
Inset shows the plate boundary in Iceland with the study area indicated
by a box.
25C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33

Page 3

water flowed from the eruption site accumulating in the
Grímsvötn caldera, causing a catastrophic glacial out-
burst flood on November 4. Isotopic composition of the
magmaeruptedatGjálpsuggeststhatitoriginatedfroma
subsidiary magma chamber of the Grímsvötn magma
system located beneath Gjálp (Sigmarsson et al., 2000).
2. Radar interferometric data analysis
An interferogram can be described as a map of one
component of the displacement field, where one fringe
(one full cycle of colors) represents 28 mm of motion
along the line of sight from the satellite to the ground for
the ERS images used here. Using images acquired by
the ERS satellites between 1992 and 2002, we have
calculated over 200 interferograms with the standard
two-pass method (Massonet and Feigl, 1998).
Specifically, we applied version 3 of the PRISME/
DIAPASON software developed at the French Space
Agency (CNES) to frame 2295 of descending tracks
009, 238 and 281. The combination of precise orbital
trajectories from the Department of Earth Observation
and Space Systems (DEOS) at Delft University of
Technology (Scharroo and Visser, 1998) and a digital
elevation model (from the National Land Survey of
Iceland) with 10-meter accuracy implies that the
presence of topographic fringes is not significant in
our interferograms. Uncertainties of 10 mm for ERS
interferograms have been previously used in Iceland
(Pedersen et al., 2003; Pagli et al., 2006), when using a
digital elevation model with 30-meter accuracy and
orbital information from the German Aerospace Center
(DLR). However, we are now using an improved digital
elevation model and precise Delft orbital information.
Fig. 2. Interferograms covering the co-eruptive period (a) from 03/06/95 to 06/10/96, (b) from 31/05/95 to 03/10/96, (c) from 06/10/96 to 13/07/97,
and (d) from 03/10/96 to 23/09/99. The boxes in panel (c) and (d) give the location of the unfiltered interferograms in panel (e) and (f), respectively.
The arrows mark local deformation signal north of Tungnafellsjökull. The ERS tracks and the orbit numbers are indicated. The altitude of ambiguity
(meter) is in the lower right corner, it corresponds to the difference in topographic elevation that produces one (artefactual) fringe in an interferogram.
Color scale bar in panel (a) applies to all interferograms.
26C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33

Page 4

Therefore the uncertainties on our interferograms may
be lower than 10 mm. In subsequent modeling, we tried
uncertainties between 5 and 10 mm and we calculated
the resulting reduced χ2for all models. This parameter
equals rT∑−1r/(n−m), where r is the residual, ∑ is the
data covariance matrix, n is the number of data and m is
the number of model parameters. The residual is the
difference between the observed values and the model
predictions (evaluated at the center point of each pixel).
The reduced χ2is expected to be 1 if optimal solutions
are found and our assessments of the data errors are
correct. We found that uncertainties of 7 mm gave
reasonable reduced χ2, between 0.5–3 for all models.
Therefore, for the modeling we used uncertainties of
7 mm for all the interferograms.
Irregular spacing of usable ERS images for inter-
ferometry in the study area limits our capability to
perform time-series analysis of deformation in the area.
Only images acquired with snow-free conditions outside
the icecap(mostlysummertime)canbesuccessfullyused
to construct interferograms. The snow, covering the area
for most of the year, causes complete incoherence.
However, our InSAR dataset allows us to distinguish
between at least two different deformation episodes
duringthe course of the 1996 Gjálp eruption. Co-eruptive
interferograms, spanning only the first few days of the
eruption and including the Mw5.6 earthquake, show a
modest but consistent range decrease over a small area at
the western flank of Bárdarbunga at the NW edge of
Vatnajökull (Fig. 2a and 2b). Co-eruptive interferograms,
spanning the last week of the eruption and the following
nine months, show an inversion of the deformation from
range decrease to range increase. A more widespread
deflation signal of about two fringes at the flank of
Bárdarbunga (Fig. 2c and 2d) is observed. These co-
eruptive interferograms also show three other distinct
local deformation signals. The strongest of these is
located north of the Tungnafellsjökullvolcano. It consists
of about half a fringe arranged in one lobe (Fig. 2c and
2d). A second smaller signal is northwest of the volcano
and a third one north of Bárdarbunga. In Fig. 2e and 2f
the two unfiltered interferogram of Fig. 2c and 2d,
respectively, are shown to illustrate that sharp deforma-
tion signals northeast of Tungnafellsjökull can be
identified also in the unfiltered interferograms. All other
interferograms have been smoothed using a power
spectrum filter based on the approach described by
Goldstein and Werner (1998). We used their method with
a filtering factor of 0.9, where a factor of 0 represents no
filtering and a factor of 1 represents strong filtering. The
three signals appear in the same locations in many
independent co-eruptive interferograms, but not in our
Fig. 3. Interferograms covering: pre-eruptive period (a) from 08/07/95 to 01/09/96, and (b) from 04/09/93 to 13/08/96, post-eruptive period (c) from
13/07/97 to 16/09/98, and (d) from 02/08/98 to 13/06/99. The ERS tracks and the orbit numbers are indicated. The altitude of ambiguity (meter) is in
the lower right corner. Color scale bar in panel (a) applies to all interferograms.
27C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33

Page 5

pre-eruptiveorpost-eruptiveinterferograms(Fig.3a–3d).
Using pair-wise logic, we conclude that these three
signals are related to the co-eruptive seismo-volcanic
activity, rather than some other geophysical phenomena
orprocessingartifact. In the following,we concentrate on
modeling the strongest signal north of Tungnafellsjökull.
3. Modeling
We model the deformation signals assuming a
uniform elastic isotropic half-space with a Poisson's
ratio, ν=0.25 and a rigidity, μ=30 GPa. The data were
first unwrapped and quadtree partitioned (Fig. 4). The
unwrapping transforms the interferometric deformation
signal, expressed as a series of fringes, into a continuous
scale, and the quadtree partitioning reduces the data size
[e.g. Jónsson et al., 2002]. To describe the source of the
deformation, we estimate dislocation parameters using a
simulated annealing algorithm followed by a derivative
based method, as described by Cervelli et al. (2001). All
models include an additional offset parameter to account
for uncertainty in identifying the fringe corresponding to
zero displacement.
3.1. Co-eruptive deformation prior to October 6, 1996
The interferogram in Fig. 2a was selected for the
modeling as it has the best spatial resolution over the
area, and with comparison to other interferograms it
appears to be free of atmospheric effects. The Gjálp
eruption took place on a ∼7 km long eruptive fissure,
starting on September 30, 1996, with the highest
eruption rate occurring during the first three days of
the eruption. The propagating earthquakes leading to the
eruption can be used to argue that magma moved from
the Bárdarbunga volcano. Thus, we first considered a
model with diking at the Gjálp site, and extending
towards Bárdarbunga in order to test if such processes
could contribute to the observed deformation signal
outside the ice cap. We used two vertical dikes, each
8 km long and striking north. One dike is located at the
Gjálp eruptive site and a second one is just north of it,
and extending towards Bárdarbunga (Fig. 5). The
location and length of the two dikes are based on the
seismicity, but as the eruption took place under the ice
cap, no direct observation of the dike opening is
available. For the dike closer to Bárdarbunga we
assumed 0.5 m of opening, extending from 5 to 2 km
depth below the surface, while the other dike reached
the surface and had 1 m of opening. These values were
selected to broadly conform to experience of recorded
diking in Iceland [e.g. Sigmundsson, 2006]. This model
does not cause any significant deformation in areas
outside the ice cap edge, indicating that diking processes
occurring southeast of Bárdarbunga have little influence
on our results. We cannot constrain deformation sources
located in this area.
An Mw 5.6 earthquake and intense seismicity
occurred at the Bárdarbunga caldera during the first
week of eruption. One interpretation suggests ring
faulting on the caldera (Nettles and Ekström, 1998).
Therefore, we attempted to model the fringe pattern at
Bárdarbunga with a rectangular dislocation with uni-
form slip. Fixing the fault plane location to correspond
to the Mw5.6 earthquake, assuming zero fault opening,
we find two solutions on perpendicular planes with
either ∼3 m of left-lateral or right-lateral strike slip and
∼1.5 m of dip slip (Fig. 5). Both these models have a
seismic moment Mo=6×1018N m corresponding to an
earthquake with a moment magnitude Mw of 6.5,
assuming Mw¼2
vary the location of the fault plane, we find no model
with an Mwlower than about 6.2. Such large magnitudes
are inconsistent with the observed seismicity. The
earthquakes activity prior to the eruption may also
reflect dike injection at Bárdarbunga. Thus, we tried a
third model with dike opening at the Bárdarbunga
caldera rim at a distance of 20 km from the Gjálp
eruptive site. The optimal model is a 6 km long vertical
dike, striking N31°W, extending from 6 km depth to
2.5 km below the surface with 4.7 m of opening
corresponding to a volume increase of 0.16 km3(Figs.
4c, 4d and 5). This model gives an RMS misfit of
3.2 mm (a null model has an RMS of 19 mm), and a
reduced χ2of 0.5. Our results indicate that diking at the
Bárdarbunga caldera rim can explain most of the
deformation signal observed prior to October 6.
3logðMoÞ ? 6:03. Although we can
3.2. Deformation after October 6, 1996
Modeling of the deflation at Bárdarbunga after
October 6 has been attempted with several different
sources. We use a Mogi source, an ellipsoid, and a
rectangular dislocation with uniform opening under the
Bárdarbunga volcano (Fig. 5). The deformation signal
north of Tungnafellsjökull was modeled with a
rectangular dislocation simulating either a slipping
fault or an opening dike. One interferogram was selected
for the modeling, showing high coherence and spanning
the shortest time interval including the last week of the
eruption (Fig. 2c). Our initial model includes a Mogi
source at Bárdarbunga and a uniform slip dislocation
north of Tungnafellsjökull. We find a best-fitting Mogi
source located beneath the northern rim of the
28C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33

Page 6

Fig. 4. (a) Unwrapped interferogram from 03/06/95 to 06/10/96 (Fig. 2a), (b) quadtree interferogram, (c) best-fit dike model, the pink thick line marks
the location of the dike (d) residual between Figs. 2a and 4c, the pink thick line marks the location of the dike, epicenter location of the Mw5.6 event
on September 29, 1996 is shown with a star and earthquakes September 29–October 06, 1996 are shown with white asterisks, the color scale bar in
panelb appliestopanelsa andb,andthe colorscale barasin Fig.2aappliesto panelsc andd.(e)unwrappedinterferogram from06/10/96to 13/07/97
(Fig. 2c), (f) quadtree interferogram, (g) best-fit model, the black thick line marks the location of the fault and the pink triangle gives the location of
the Mogi source (h) residual between Figs. 2c and 4g, the black thick line marks the location of the fault and the pink triangle gives the location of the
Mogi source, earthquakes October 06–13, 1996 are shown with white asterisks, the color scale bar in panel f applies to panels e and f, and the color
scale bar as in Fig. 2a applies to panels g and h.
29 C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33

Page 7

Bárdarbunga caldera (64.67°N, 342.57°E) at 10 km
depth, with a volume decrease of 0.07 km3(Fig. 4g and
4h). The best-fit fault solution is vertical and 7 km long,
striking N37°E located at 64.88°N, 342.33°E, extending
from the surface to 6 km depth. It has 5 cm of left-lateral
strike–slip and 3 cm of dip slip (Fig. 4g and 4h). This
fault model combined with the Mogi source gives an
RMS misfit of 7.9 mm, significantly better than the
18 mm value for a null model, and a reduced χ2of 2.4.
An alternative dislocation solution consists of a 6 km
long vertical dike, striking N37°E, extending from the
surface to 2 km depth with 6 cm of uniform opening.
This dike model combined with the Mogi source gives
an RMS misfit of 7.6 mm and a reduced χ2of 1.9.
For both of the above models, some residual
deflation can be identified southwest and northeast of
the Mogi source (Fig. 3h). To improve the fit, we test a
model with an ellipsoidal source (Tiampo et al., 2000)
instead of the Mogi source. We fix its major axis to be
horizontal and estimate the remaining seven model
parameters. The best-fit ellipsoidal source is located at
64.69°N, 342.60°E (Fig. 5) at 8.4 km depth, has a
pressure change of −37 MPa and a volume decrease of
0.05 km3. The solutions for the deformation signal north
of Tungnafellsjökull are similar to our previous results.
The ellipsoidal source does not significantly improve
the fit found with a Mogi source. The slipping fault
solution combined with the ellipsoidal source gives an
RMS of 7.4 mm and a reduced χ2of 2.3, while the
opening dike solution gives an RMS of 8.3 mm and a
reduced χ2of 2.9. Therefore we prefer the Mogi source
model, as it gives a comparable fit to the data using
fewer model parameters.
Alternatively, we can model the deflation at Bárdar-
bunga with a slipping fault near the caldera rim (Fig. 5).
The inversion finds a 16 km long vertical fault, located
southeast of the caldera, extending from the surface to
10 km depth with 2.9 m of left-lateral strike–slip and
1.3 m of dip slip. The model finds a fit comparable as
that of the pressure sources, but cannot improve upon it.
The model corresponds, however, to an earthquake with
magnitude Mw6.8. Since, no such large earthquake
occurred in the area, we reject this possibility. Accord-
ingly, we conclude that the source under Bárdarbunga
decreased in volume, compatible with extraction of
magma from a chamber.
The ice cover limits our ability to detect deformation
sources. Magma could be extracted from several sources
without any deformation signals outside the ice cap. A
Mogi source at the center of the Bárdarbunga caldera, at
3 km depth, could contract 0.03 km3without producing
a signal larger than ∼1 cm outside the Vatnajökull ice
cap detectable with radar interferometry. Similarly, a
source farther from the well-correlated fringes in the
interferograms could undergo a larger volume change
without producing a measurable signal. For example, a
Mogi source at the Gjálp eruptive site at 6 km depth,
could have decreased in volume up to 0.25 km3without
detection by InSAR.
The deformation north of Tungnafellsjökull is con-
sistent with either a slipping fault or a dike injection. In
eithercasethedisplacementscausedbythesesourcesare
lessthan8cm.DuringtheGjálperuptionseismicactivity
occurred there on September 30 and October 1, earth-
quakes started migrating west and north of the volcano
on October 4 and seismic activity was recorded in the
area until October 13. Seismic data alone may suggest
that a dike was injected northwest of the Tungnafellsjö-
kull volcano, beginning on October 4. However, the
Fig. 5. SAR amplitude image, volcanic systems and earthquake
epicenters as in Fig. 1. Overlaid are the different deformation sources
used in the modeling. Pink color marks deformation sources used to
model deformation before October 6, 1996, while blue color marks
deformation sources used to model deformation after October 6, 1996.
The two pink thick lines, striking north, represent two dikes, the pink
box at the Bárdarbungacaldera rim gives the location of one of the two
perpendicular best-fit faults and the pink thick line at the northeastern
Bárdarbunga caldera rim marks the best-fit dike model (see text for
discussion). The blue triangle and star mark the center of the best-fit
Mogi and ellipsoidal source, respectively, coupled with the best-fit
fault model north of Tungnafellsjökull (the blue thick line), and the
blue thick line south of the Bárdarbunga caldera gives the location of
the best-fit fault–slip model (see text for discussion). The area marked
with a squared pattern and a question mark shows that deformation
sources located there cannot be constrained using our InSAR dataset.
30 C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33

Page 8

InSAR data shows no detectable deformation prior to
October 6 in this area. Thereafter magma may have
migrated to shallow depths and a dike may have formed
in the uppermost 2 km of the crust without reaching the
surface. The opening dislocation may also represent any
combination of dike and tension fractures. Alternatively
the deformation may be due to fault slip after October 6.
In any case, the deformation in the area north of
Tungnafellsjökull took place between October 6, 1996
and July 13, 1997, as did the observed deflation at the
nearby Bárdarbunga volcano. Since no deformation was
detected at the Tungnafellsjökull volcano, we consider a
dike originating there an unlikely source of our signal.
We, therefore, prefer the fault slip solution.
4. Discussion
The fringe pattern at Bárdarbunga prior to October 6,
1996 is consistent either with a slipping fault or a dike
injection near the caldera rim. The fault slip solution
gives at least a Mw=6.2, while the largest earthquake in
the area had Mw=5.6, thus implying a large component
of aseismic slip. We favor the dike injection solution
because it correlates with the areas of earthquake
swarms during the first week of the eruption. Diking
nearby the eruption site is a likely event and simpler
explanation than aseismic faulting, although we cannot
exclude that possibility. Subsidence at Bárdarbunga is
not observed before October 6, suggesting that the dike
originated either from a subsidiary magma chamber
located southeast of Bárdarbunga or directly under
Gjálp where deformation signals cannot be detected by
InSAR due to ice cover.
InSAR analysis indicates that significant deflation
occurred at Bárdarbunga but only after October 6, when
most of the magma had already been erupted at the
Gjálp site. Modeling results indicate that a 10 km deep
deflating Mogi source can explain most of the
deformation with a volume decrease of the source of
0.07 km3. Using the formula by Johnson et al. (2000)
and considering also the effect of volumetric decom-
pression of residing magma, we argue that a larger
volume of magma could have flowed out of a magma
chamber under Bárdarbunga between October 6, 1996
and July 13, 1997. Assuming a bulk modulus of 17 GPa
we find a magma volume of 0.23 km3. However, the
bulk modulus of the residing magma is uncertain as this
parameter depends strongly on e.g. the amount of
volatiles dissolved or exsolved in the magma, which are
difficult to determine. One GPS site (GJAL) located
north of the Bárdarbunga volcano moved horizontally
approximately 8 cm in the SSE direction between 1993–
1997 (Hreinsdóttir et al., 1998) in broad agreement with
our model of a deflating Mogi source. We interpret the
deflation of Bárdarbunga as indicative of a pressure
connection between the volcano and Gjálp. We suggest
that magma flowed out of Bárdarbunga to refill a
subsidiary magma chamber located southeast of the
volcano and connected to Gjálp, or directly towards the
Gjálp eruptive site. Part of the magma may also have
erupted from October 6 to 13, 2006. During this period
the eruption rate was low (about 0.25×106kg/s, total
magma volume of 0.06 km3) and any eruptive products
were presumably completely subglacial. Our dataset
cannot be directly used to determine if magma flowed
from the center of the Grímsvötn magmatic system
(under the Grímsvötn caldera) during the Gjálp eruption
as the ice cover excludes the use of InSAR for crustal
deformation studies there. However, campaign GPS
measurements at a single GPS site (on a nunatak
sticking out of the ice) at the Grímsvötn caldera 1992–
1997 (Sturkell et al., 2003b) show uplift and outward
movement from the caldera. This pattern is consistent
with continuous recharging of the Grímsvötn volcano in
the years preceding an eruption there in 1998 (two years
after the Gjálp eruption), and no outflow of magma from
a source beneath the Grímsvötn caldera during the Gjálp
eruption. However,the resolution of the GPS time-series
is limited, with observations closest in time to the Gjálp
eruption being collected 4 years prior to the eruption (in
1992), and 10 months after the eruption (in August
1997). Another argument against magma movements
from a source below the Grímsvötn caldera during the
Gjálp eruption is the lack of any unusual seismicity at
Grímsvötn during or preceding that eruption.
The deflation of Bárdarbunga after October 6 caused
a surface radial strain of 1.4 μstrain at the distance of the
deformation signal north of Tungnafellsjökull. This
value corresponds to about seven years of strain
accumulation due to plate motion in the area, assuming
a spreading rate of 2 cm/yr over a 100 km wide zone. We
suggest that the Bárdarbunga deflation imparted some
additional stress to the area north of Tungnafellsjökull
promoting earthquakes and aseismic fault movement.
Such processes have already been observed in Iceland
(Pagli et al., 2003; Árnadóttir et al., 2000). A Mw6.5
earthquake in the South Iceland Seismic Zone in June
2000 triggered widespread seismicity in south and
southwest Iceland. As a part of that activity, an aseismic
fault slip occurred at a distance of 85 km from the initial
main event. Furthermore, the deformation signal north
of Tungnafellsjökull is situated in the middle of a fissure
swarm, an area already intensely faulted and fissured.
Therefore the slip may have occurred on a preexisting
31 C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33

Page 9

fault plane. Our best-fit fault plane solution in the area
north of Tungnafellsjökull gives a seismic moment Mo
of 6×1016Nm while the total seismic moment release in
the area from October 4 to 13 was 6×1013Nm. Low
level seismicity continued throughout November, 2006.
5. Conclusions
Analysis of seismic and InSAR data spanning the
Gjálp eruption allows us to identify deformation events
that were previously unknown and helps to constrain the
sequence of events in an otherwise enigmatic eruption.
We suggest that the sequence began with the Mw5.6
earthquake on the Bárdarbunga caldera. It caused
fissuring that in turn ruptured a subsidiary magma
chamber located southeast of Bárdarbunga, possibly
under Gjálp. There an eruption started on September 30,
1996. Our observations indicate that diking occurred as
well near the Bárdarbunga caldera rim during the first
week of the eruption. The Gjálp eruption was associated
with or followed by flow of magma from a source under
Bárdarbunga after October 6, 1996. Outflow of magma
from a source at Bárdarbunga in response to pressure
decrease in the source that fed the Gjálp eruption is
inferred, suggesting a link between the sources. Timing
of the earthquakes north of Tungnafellsjökull suggests
that stress changes due to the Bárdarbunga deflation
triggered faulting in these areas.
Acknowledgments
We thank Kristín Vogfjörd at the Icelandic Meteor-
ological Office for providing the seismic data and
Eyjólfur Magnússon for the outline of Vatnajökull.
Comments from Pete LaFemina and an anonymous
reviewer helped us improve the manuscript. We grate-
fully acknowledge the European Space Agency (ESA)
for providing their copyrighted ERS and ENVISAT data
under the terms and conditions of several AO and Cat-I
projects as well as partial support from the Icelandic
Research Council (RANNÍS), the National Power
Company of Iceland (Landsvirkjun), the University of
Iceland Research Fund, and from a grant by the
European Union to the FORESIGHT project (SSPI-
CT-2003-511139).
References
Árnadóttir, T., Geirsson, H., Einarsson, P., 2000. Coseismic stress
changesandcrustal deformation onthe ReykjanesPeninsuladueto
triggered earthquakes on 17 June. J. Geophys. Res. 109, B09307,
doi:10.1029/2004JB003130.
Björnsson, H., Einarsson, P., 1990. Volcanoes beneath Vatnajökull,
Iceland: evidence from radio echo-sounding, earthquakes and
jökulhaups. Jökull 40, 147–168.
Cervelli, P., Murray, M.H., Segall, P., Aoki, Y., Kato, T., 2001.
Estimating source parameters from deformation data, with an
applicationto the March 1997 earthquake swarm off Izu Peninsula,
Japan. J. Geophys. Res. 106, 11217–11237.
Einarsson, P., 1991. Earthquakes and present day tectonism in Iceland.
Tectonophysics 189, 261–279.
Einarsson, P., Brandsdóttir, B., 1984. Seismic activity preceding and
during the 1983 volcanic eruption in Grímsvötn, Iceland. Jökull
34, 13–23.
Einarsson, P., Saemundsson, K., 1987. Earthquake epicenters
1982–1985 and volcanic systems in Iceland. In: Sigfusson,
T.I. (Ed.), Í hlutarins edli. Menningarsjódur, Reykjavik.
Einarsson, P., Brandsdóttir, B., Gudmundsson, M.T., Björnsson, H.,
Grönvold, K., Sigmundsson, F., 1997. Center of the Iceland
hotspot experiences volcanic unrest. Eos 78, 369–375.
Goldstein, R.M., Werner, C.L., 1998. Radar interferograms
filtering for geophysical applications. Geophys. Res. Lett. 25,
4035–4038.
Gudmundsson, M.T., Sigmundsson, F., Björnsson, H., 1997. Ice-
volcano interaction of the 1996 Gjálp subglacial eruption,
Vatnajökull, Iceland. Nature 389, 954–957.
Gudmundsson, M.T., Sigmundsson, F., Björnsson, H., Högnadóttir, T.,
2004. The 1996 eruption at Gjálp, Vatnajökull ice cap, Iceland:
efficiency of heat transfer, ice deformation and subglacial water
pressure. Bull. Volcanol. 66, 46–65, doi:10.1007/s00445-003-
0295-9.
Hreinsdóttir, S., Einarsson, P., Jónsson, S., Sigmundsson, F., 1998.
GPS Geodetic Measurements Around Bárdarbunga Volcano,
Central Iceland, in 1997, RH-13-98. Science Inst., Univ. Iceland,
Reykiavík, p. 42.
Johnson, D.J., Sigmundsson, F., Delaney, P.T., 2000. Comment on
“Volume of magma and accumulation or withdrawal estimated
from surface uplift or subsidence, with application to the 1960
collapse of Kilauea volcano” by P.T. Delaney and D.F. McTigue.
Bull. Volcanol. 61, 491–493.
Jónsson, S., Zebker, H., Segall, P., Amelung, F., 2002. Fault slip
distribution of the 1999 Mw 7.2 Hector mine Earthquake,
California, estimated from satellite radar and GPS measurements.
Bull. Seismol. Soc. Am. 92, 1377–1389.
Larsen, G., Gudmundsson, M.T., Björnsson, H., 1998. Eight centuries
of periodic volcanism at the center of the Iceland hotspot revealed
by glacier tephrostratigraphy. Geology 26 (10), 943–946.
Massonet, D., Feigl, K.L., 1998. Radar interferometry and its
application to changes in the earth's surface. Rev. Geophys. 36,
441–500.
Nettles, M., Ekström, G., 1998. Faulting mechanism of anomalous
earthquakes near Bárdarbunga volcano, Iceland. J. Geophys. Res.
103, 17973–17983.
Pagli, C., Pedersen, R., Sigmundsson, F., Feigl, K.L., 2003. Triggered
fault slip on June 17, 2000 on the Reykjanes Peninsula, SW-
Iceland captured by radar interferometry. Geophys. Res. Lett. 30,
1273, doi:10.1029/2002GL015310.
Pagli, C., Sigmundsson, F., Áranadóttir, T., Einarsson, P., Sturkell, E.,
2006. J. Volcanol. Geotherm. Res. 152, 97–108, doi:10.1016/j.
volgeores.2005.09.014.
Pedersen, R., Jónsson, S., Árnadóttir, T., Sigmundsson, F., Feigl, K.L.,
2003. Fault slip distribution of two June 2000 Mw 6.5 earthquakes
in South Iceland estimated from joint inversion of InSAR and GPS
measurements. EPSL 213, 487–502.
32C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33

Page 10

Scharroo, R., Visser, P.N.A.M., 1998. Precise orbit determination and
gravity field improvement for the ERS satellites. J. Geophys. Res.
103, 8113–8127.
Sigmarsson, O., Karlsson, H.R., Larsen, G., 2000. The 1996 and 1998
subglacial eruptions beneath the Vatnajökull ice sheet in Iceland:
contrasting geochemical and geophysical inferences on magma
migration. Bull. Volcanol. 61, 468–476.
Sigmundsson, F., 2006. Iceland Geodynamics, Crustal Deformation
and Divergent Plate Tectonics. Springer-Praxis, Chichester, UK.
Sturkell, E., Einarsson, P., Sigmundsson, F., Hreinsdóttir, S., Geirsson,
H., 2003a. Deformation of Grímsvötn volcano, Iceland: 1998
eruption and subsequent inflation. Geophys. Res. Lett. 30, 1182,
doi:10.1029/2002GL016460.
Sturkell, E., Einarsson, P., Sigmundsson, F., Hreinsdóttir, S., Geirsson,
H., 2003b. Deformation of Grímsvötn volcano, Iceland: 1998
eruption and subsequent inflation. Geophys. Res. Lett. 30, 1182,
doi:10.1029/2002GL016460.
Tiampo, K.F., Rundle, J.B., Fernandez, J., Langbein, J.O., 2000.
Spherical and ellipsoidal volcanic sources at Long Valley caldera,
California, using a genetic algorithm inversion technique. J.
Volcanol. Geotherm. Res. 102, 189–206.
Vogfjörd, K.S., Rögnvaldsson, S.T., Slunga, R., et al., 1999.
Propagation of seismicity during the Sept/Oct, 1996 subglacial
eruption episode near Bardarbunga volcano, Iceland. Seismol.
Res. Lett. 2.
Vogfjörd, K.S., Jakobsdóttir, S.S., Gudmundsson, G.B., et al., 2005.
Forecasting and monitoring a subglacial eruption in Iceland. Eos,
Trans. - Am. Geophys. Union 86, 245–248.
33 C. Pagli et al. / Earth and Planetary Science Letters 259 (2007) 24–33