repeating coupled earthquakes at shishaldin volcano,...
TRANSCRIPT
www.elsevier.com/locate/jvolgeores
Journal of Volcanology and Geotherm
Repeating coupled earthquakes at Shishaldin Volcano, Alaska
Jacqueline Caplan-AuerbachT, Tanja Petersen
Alaska Volcano Observatory, Geophysical Institute, University of Alaska Fairbanks, 903 Koyukuk Dr.,
Fairbanks, AK 99775-7320, United States
Received 13 November 2003; accepted 20 January 2005
Abstract
Since it last erupted in 1999, Shishaldin Volcano, Aleutian Islands, Alaska, has produced hundreds to thousands of long-
period (1–2 Hz; LP) earthquakes every day with no other sign of volcanic unrest. In 2002, the earthquakes also exhibited a
short-period (4–7 Hz; SP) signal occurring between 3 and 15 s before the LP phase. Although the SP phase contains higher
frequencies than the LP phase, its spectral content is still well below that expected of brittle failure events. The SP phase was
never observed without the LP phase, although LP events continued to occur in the absence of the precursory signal. The two-
phased events are termed bcoupled eventsQ, reflecting a triggered relationship between two discrete event types. Both phases are
highly repetitive in time series, suggestive of stable, non-destructive sources. Waveform cross-correlation and spectral
coherence are used to extract waveforms from the continuous record and determine precise P-wave arrivals for the SP phase.
Although depths are poorly constrained, the SP phase is believed to lie at shallow (b4 km) depths just west of Shishaldin’s
summit. The variable timing between the SP and LP arrivals indicates that the trigger mechanism between the phases itself
moves at variable speeds. A model is proposed in which the SP phase results from fluid moving within the conduit, possibly
around an obstruction and the LP phase results from the coalescence of a shallow gas bubble. The variable timing is attributed to
changes in gas content within the conduit. The destruction of the conduit obstacle on November 21, 2002 resulted in the abrupt
disappearance of the SP phase.
Published by Elsevier B.V.
Keywords: Shishaldin Volcano; volcano seismology; long-period earthquake
1. Introduction
Since late 1999, shortly after it produced a
Subplinian eruption, Shishaldin Volcano, Unimak
0377-0273/$ - see front matter. Published by Elsevier B.V.
doi:10.1016/j.jvolgeores.2005.01.011
T Corresponding author. Now at the Alaska Volcano Observatory,
U.S. Geological Survey, 4200 University Dr., Anchorage, AK
99508, United States. Tel.: +1 907 786 7460.
E-mail address: [email protected] (J. Caplan-Auerbach).
Island, Alaska (Fig. 1), has exhibited up to thousands
of daily small long-period (LP) earthquakes. This
level of seismicity is in stark contrast to the other 24
volcanoes monitored by the Alaska Volcano Observ-
atory (AVO), which generally experience a few
earthquakes per week (Dixon et al., 2003). While
swarms of earthquakes are typical on volcanoes, the
continuous nature of Shishaldin seismicity makes it
unique among Alaskan volcanoes and unusual among
al Research 145 (2005) 151–172
-165o 00'
-165 00'
-164o 40'
-164 40'
-164o 20' -164o 00' -163o 40' -163o 20' -163o 00' -162o 40' -162o 20'
54o 20'
54o 40'
55o 00'
55 20'
-165 00' -164 40'
55 20'
54o 20'
54o 40'
55o 00'
54o 20'
-164o 20' -164o 00' -163o 40' -163o 20' -163o 00' -162 o40' -162o 20'
SSLN
SSLS
SSLW
BRPK
ISNN
ISTK
False Pass
Cold Bay
Shishaldin
Isanotski
Fisher
50˚
180
180
-170
-170
-160
-160
-150
-150
50 50
60 60
70 70
0 5 10
km
Westdahl
Fig. 1. Map of Unimak Island, Aleutian Arc, and the Shishaldin seismic network. Shishaldin Volcano is marked with an arrow, as are Westdahl,
Fisher and Isanotski volcanoes. The solid circles represent the locations of short-period seismic stations operated by the Alaska Volcano
Observatory. The nearest population centers, the villages of False Pass and Cold Bay, are marked by triangles.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172152
monitored volcanoes in general. Shishaldin’s behavior
is also perplexing in that the elevated seismicity is
apparently unassociated with other evidence of
volcanic unrest such as thermal anomalies. Finally,
this activity is notable for the extreme similarity
evidenced among events over long time scales
(Petersen et al., 2002).
Volcanic seismicity commonly heralds eruptive
activity. While there is a great deal of variety in the
type and duration of pre-eruptive seismicity, it is
largely accepted that a high degree of seismicity on a
volcano reflects the activation of internal eruption
processes such as the rise of magma from depth, dike
propagation, or pressure increase related to changes in
gas flux (e.g. Klein et al., 1987; Lahr et al., 1994;
McNutt, 2000; Chouet, 2003). LP earthquakes in
particular are commonly seen as indicators of volcanic
unrest, as they are believed to relate to the presence of
fluids such as magma (Fischer et al., 1994; Lahr et al.,
1994; Chouet, 1996). Thus, a continuously high level
of LP activity is troubling, especially for an organ-
ization such as AVO which is responsible for eruption
forecasting and hazard assessment.
In early 2002, the appearance of Shishaldin seis-
micity changed, with the addition of a precursory short-
period (SP) phase to the LP events. We refer to this pair
of earthquake phases as a bcoupledQ event and show
that it is unlike other multi-phase earthquakes com-
monly presented in the volcano seismology literature.
In this article we first give an overview of Shishaldin
seismicity, and then explain how the coupled earth-
quakes differ from more common earthquake types
such as LP or hybrid events. We describe the coupled
events in detail, discussing event similarity, earthquake
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 153
location and the relationship between the signal’s two
phases. Finally we propose a mechanism by which
coupled events may be generated and discuss the
implications for hazards monitoring at Shishaldin.
2. Geologic setting
Shishaldin Volcano is among the most active
volcanoes in Alaska, with 29 confirmed eruptions in
historic time (Miller et al., 1998; Nye et al., 2002). Its
eruptions are composed primarily of basalt and
basaltic andesite, although rare lavas with up to 64
wt.% silica have been identified (Fournelle and
Marsh, 1991; Miller et al., 1998). The tallest volcano
in the Aleutians, Shishaldin’s summit reaches 2857 m
above sea level and glaciers and snowfields blanket its
surface year-round. A puffing gas plume nearly
always emanates from the volcano’s summit.
Although the composition of the plume is unknown,
observations have been made of sulfur deposits within
the mouth of the conduit and a strong sulfur smell has
been associated with gas puffs (P. Stelling, personal
communication, 2003). This suggests that some
portion of the plume results from magmatic gases.
Although most of Shishaldin’s eruptions have
consisted of small ash and steam plumes, at least 11
Subplinian deposits have been identified in Holocene
stratigraphy (Beget et al., 1998). The most recent
eruption, in April–May 1999, consisted of both
Strombolian bubble and ash bursts and a 16-km
Subplinian eruption column. Thus, while the majority
of its eruptive activity is small and local to the region,
the fact that Shishaldin is capable of generating high
and voluminous ash plumes confirms that it must be
recognized as a hazardous volcano.
Unimak Island, the easternmost island in the
Aleutian chain (Fig. 1) comprises 5 volcanoes,
including Shishaldin. The two nearest population
centers are False Pass, located ~60 km from Shish-
aldin on the island’s eastern coast, and Cold Bay,
located on the Alaska Peninsula ~92 km from
Shishaldin. Unfortunately, Shishaldin cannot be seen
from False Pass because the nearer volcanoes Round
Top and Isanotski block the view. AVO is therefore
dependent on local pilots or observers in Cold Bay for
visual descriptions of the volcano and its activity.
Overall, however, visual observations are minimal and
AVO relies heavily on telemetered seismic data and
satellite imagery for volcanic monitoring. This lack of
visual observation makes it difficult to examine
possible correlations between Shishaldin seismicity
and factors such as precipitation, or the strength of the
ubiquitous gas plume at the volcano’s summit.
3. Seismic network and history
In 1997, AVO installed a 6-station seismic network
near Shishaldin, and in 1998 a network was installed
on Westdahl volcano west of Shishaldin (Fig. 1). Five
of the Shishaldin stations have short period Mark
Products L-4 vertical seismometers with 1 Hz natural
frequency. Station SSLS has a 3-component Mark
Products L-22 seismometer with 2 Hz natural
frequency. A Setra 239 pressure sensor was co-located
with station SSLN on the volcano’s north flank, but
stopped functioning in late 1999, and was replaced in
2003 by a Chaparral Model 2 microphone. Data from
all instruments are telemetered to the nearby towns of
Cold Bay and King Cove and then to AVO over
analog telephone lines. Once in Fairbanks the data are
digitized at 100 Hz with 12-bit resolution.
While the 1999 eruption proved that the network is
effective for volcanic monitoring (Moran et al., 2002;
Nye et al., 2002; Thompson et al., 2002; Caplan-
Auerbach and McNutt, 2003), the large distance
between the outer stations and Shishaldin’s summit
(14–19 km; Fig. 1) makes location of small events
extremely difficult. Furthermore, station ISNN suffers
from a poor telemetry link, resulting in large data gaps
from that station. Finally, LP and coupled earthquakes
do not exhibit clear S-waves. Thus for earthquake
locating we can typically use a maximum of 5 P-
waves, with 4 phases the more common situation.
Consequently, conventional locations of Shishaldin
earthquakes are not always possible, although pre-
liminary locations place them beneath the summit at
shallow (0–3 km) depths (Dixon et al., 2003).
The seismicity commonly observed in volcanic
regions can be described using an event classification
scheme such as those developed by Minakami (1960)
and Lahr et al. (1994) based on the present under-
standing of source mechanisms. The volcanic event
types most commonly observed include volcano-
tectonic (VT) earthquakes, LP events and hybrids.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172154
Volcano tectonic (VT) events are high-frequency
earthquakes, occurring within or beneath the volcanic
edifice. They represent the brittle failure of rocks due
to stress changes within the volcano. Although the
source of the stress may be volcanic in nature, such as
magmatic activity or transport of volatiles to the
surface (Lahr et al., 1994; Moran, 2003), VT earth-
quakes are otherwise indistinguishable from tectonic
earthquakes. VT events show characteristics similar to
common double-couple tectonic earthquakes, e.g.
they have clear P- and S-phases with significant
energy at frequencies N5 Hz and a variety of first-
motions (Lahr et al., 1994). In March 1999, a shallow
ML 5.2 earthquake occurred 16 km to the west of
Shishaldin, followed by a several-month long after-
shock sequence of VT events (Moran et al., 2002).
Since then, only a few VT earthquakes have been
located at Shishaldin, almost exclusively restricted to
the area 10–15 km west of the summit.
LP events are low-frequency (1–2 Hz) earthquakes
commonly associated with fluids moving in volcanic
conduits or cracks (Chouet, 1988, 1996; Julian, 1994;
Neuberg et al., 1998). They are characterized by
emergent onsets due to a gradual increase of ground
shaking caused by fluid-driven oscillation (Lahr et al.,
1994). The S-waves of LP events are not distinct and
their codas are extended. At Shishaldin, the vast
majority of earthquakes are LP events. The omnipre-
sent steam plume and steam puffs suggest that these
events might be related to degassing processes,
however, the underlying source mechanisms are still
poorly understood.
A third type of earthquake, termed the hybrid, is
related to both VT and LP events. Hybrid earthquakes
typically begin with an impulsive, short-period onset
followed by an LP-like coda. Initial descriptions of
hybrid events noted that, like VT events they
commonly exhibit a variety of first motions and clear
S-phases (Lahr et al., 1994; White et al., 1998).
However, other researchers have discarded this
requirement and note that hybrids are part of a
continuum of events that include LPs (Neuberg et
al., 2000). Lahr et al. (1994) suggest that hybrids
result from brittle failure on a plane intersecting a
fluid body, thereby combining the source mechanism
for both VT and LP events. Earthquakes of this sort
have been observed at a number of volcanoes,
including Deception Island (Ibanez et al., 2003),
Galeras (Gil Cruz and Chouet, 1997), Montserrat
(Neuberg et al., 1998; White et al., 1998), and
Redoubt (Lahr et al., 1994).
Prior to March 2002, Shishaldin activity was
composed almost exclusively of LP events. On March
13, however, a new type of earthquake was identified,
and became the predominant event type until Novem-
ber. These earthquakes begin with a short-period (SP)
phase, followed by what appears to be a normal LP
event. Subsequent investigation of earlier records
indicated that these events were present prior to March
13 but in small numbers and at weak amplitudes. We
refer to these events as coupled because, like hybrid
earthquakes, they involve two discrete phases. How-
ever, for a number of reasons, we believe that the source
process commonly invoked for hybrid events, brittle
failure near a fluid filled crack, is unlikely for
Shishaldin events. First of all, the SP phase of
Shishaldin coupled earthquakes is strongly bandlimited
between 4 and 7 Hz, which is quite low for local brittle
failure. There are also few identifiable S-phases for
either the SP or LP phases. As we will show, the SP
portion of Shishaldin events is highly repetitive for
prolonged time periods (days to weeks). It is unlikely
that repeating brittle failure event could retain such
similarity in waveform over such a long time, since
with time the source process would alter conditions at
the source region. Because the SP phases are generally
small, and somewhat emergent, first motions are
difficult to constrain. Unlike traditional hybrid events,
the timing between SP and LP phases in Shishaldin
coupled earthquakes is highly variable, as will be
described later in this article. Finally, the LP phase
bears significant similarity to bnormalQ LP events,
suggesting that it is triggered by the SP phase and is not
simply a later phase of the same earthquake. Con-
sequently we call these events bcoupled earthquakesQand propose a different mechanism for their formation.
4. Coupled events observed elsewhere
Because some Shishaldin coupled events bear
similarity to hybrid earthquakes recorded elsewhere,
it is non-trivial to determine how common such events
are at other volcanoes. However, events that appear
similar in time series to Shishaldin coupled events
have been noted on a handful of other volcanoes
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 155
(Galeras, Gil Cruz and Chouet, 1997; Stromboli, J.
Lees, personal communication, 2002; Deception
Island, Ibanez et al., 2003; Koryakski, Gordeev and
Senyukov, 2003; Popocatepetl, R. White, personal
communication, 2003). Signals called bsecondaryeventsQ were identified as precursory to LP events at
Redoubt volcano, but were only visible on a single
station and thus could not be thoroughly examined
(Stephens and Chouet, 2001). Precursory signals
identified at Koryakski volcano, Kamchatka were
interpreted by Gordeev and Senyukov (2003) to be
related to magma injection into cracks. Gil Cruz and
Chouet (1997) note their occurrence on Galeras
volcano in 1991 and suggest a source mechanism
involving two discrete cracks and a rising gas bubble.
Ibanez et al. (2003) were able to determine that the
short and long-period phases of these earthquakes have
different locations and attribute their source to brittle
failure on faults lubricated by fluids. Thus, although
they are not common features, events similar to
Shishaldin coupled events have been identified at a
number of volcanoes. However, Shishaldin events
appear to be unique in the variable timing observed
between the SP and LP phases, and the spectral content
of the SP phase is generally lower than that observed
50 100 150 200 250 3TIM
180 185 190 195 200 205TIM
-400
-200
0
200
400
-200
0
200
400
Fig. 2. Cross-correlation technique used to extract repeating events from th
correlated with segments of the continuous record and aligned at the point o
the frequency band of the reference event (4–7 Hz for SP events) and eve
continuous data. Note that the first event shown in the top panel was not s
analysis.
elsewhere. Consequently, we propose a slightly
modified mechanism for the events described here.
5. Identification and extraction of repeating
earthquakes
A preliminary examination of Shishaldin seismic-
ity shows that many of its earthquakes, both LP and
coupled, appear similar in time series (Petersen et al.,
2002). Because these events are small, only a low
percentage trigger the data acquisition system using
the EARTHWORM triggering algorithms Carlstatrig
and Carsubtrig (Dixon et al., 2003). Consequently,
most events are not automatically segmented for
further analysis. To examine a larger subset of coupled
events we took advantage of the fact that the earth-
quakes are repetitive and developed a method by
which similar events could be extracted from the
continuous record (Fig. 2). In this algorithm, a
breference eventQ is selected against which other
earthquakes will be compared. The reference event
is selected somewhat arbitrarily, with the only criteria
being a good signal-to-noise ratio and no overlap with
other events. The continuous record is divided into
00 350 400 450 500 550E (s)
210 215 220 225 230E (s)
e continuous record. An SP reference event, shown in red, is cross-
f maximum correlation. The spectral coherence is then calculated in
nts with coherence estimates exceeding 0.9 are segmented from the
ufficiently similar to the reference event to be considered for further
Table 1
Time frames and number of events for the eight time periods
selected in this study
Event group Start date End date Number of
events extracted
by SP phase
1 March 3 April 27 1442
2 April 27 June 19 1258
3 June 14 July 25 1773
4 July 14 July 21 426
5 July 22 August 19 380
6 August 24 September 16 1151
7 September 7 October 9 202
8 September 28 November 21 943
Each period reflects a time in which a single reference event was
dominant. Time periods were considered to end if two days passed
in which no events similar to the reference event were found. Note
that there is some overlap in the time periods, reflecting times in
which some variation in event appearance was observed.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172156
segments of length equal to that of the reference
event, and each segment is cross-correlated with the
reference event. The maximum cross-correlation value
yields the lag position at which the two signals
correlate most strongly, and a data segment beginning
at that time is selected from the continuous record for
coherence testing. The spectral coherence between
this data segment and the reference event is calculated
and the mean coherence in a specified frequency band
is examined. All data segments with mean coherence
exceeding a threshold value (typically 0.9 for the SP
phase of coupled events) are selected as part of an
event group. The initial selection of events was made
using station SSLN, the station on which Shishaldin
events typically exhibit the highest signal-to-noise
ratio, although we were successful in performing the
same task using other stations. Once the repeating
events are selected, a 30-s window around each event
is extracted for all stations in the network, and the data
are thereby segmented. This method allows us to
select a sequence of repeating events from the
continuous record that are too small to trigger the
data acquisition system. This method is similar to the
one used by Stephens and Chouet (2001) in their
analysis of LP seismicity at Redoubt volcano. As with
all other methods used to count events at Shishaldin
(Petersen et al., 2002), this method only selects a
small subset of earthquakes. However, this subset was
sufficient to examine the coupled events, their
locations and their temporal behavior. Petersen et al.
(2002) performed a similar analysis for the LP events.
Each reference event was used for a time period
ranging from days to weeks, until no events were
found that fit the coherence criteria for selection. When
the number of daily extracted events dropped to zero
for more than 2 days we selected another reference
event and tested an overlapping time period. In total,
we were able to extract N7000 coupled events between
March 13 and November 21, 2002, using the SP phase
and eight reference events (Table 1). We note that
although we have used eight reference events, this
does not reflect eight distinct event families, but rather
a continuously evolving source process.
It is important to note that the extraction algorithm
could be simplified by selecting events with high
correlation coefficients rather than spectral coheren-
cies, since coherence requires conversion to the
frequency domain. We chose to use the spectral
coherence for two reasons. First of all, the SP phase
of the coupled events typically has much smaller
amplitude than the LP phase. Consequently, we often
find the highest correlation coefficients when earth-
quakes are aligned with their LP rather than SP
portions in phase. Secondly, because the signals are
fairly monochromatic, they may be misaligned by a
unit number of periods, yet still have a high
correlation coefficient.
6. How similar are coupled events?
The coupled events discussed in this work were
selected because their SP phases were determined to
be similar to a reference event. Thus, each SP event is
known to have a coherence estimate of 0.9 relative to
one specific SP earthquake. This does not provide
information about how similar all SP phases in the
cluster are to one another, and how they compare to
events extracted using other reference events. Nor
does it provide information on the repeatability of the
LP phases of coupled events. Thus, we sought a
means by which the entire data set, or a generalized
subset thereof, could be examined.
As previously noted, Shishaldin coupled events do
not typically trigger the data acquisition system. Thus,
in order to investigate all of the coupled events at
Shishaldin, not just those that meet specific similarity
criteria, we needed to segment the data in a manner
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 157
that would identify all earthquakes rather than those
that meet similarity criteria. To this end we first
filtered the data between 3 and 8 Hz, to enhance the
SP phase of coupled events. A new short-term-
average/long-term-average (STA/LTA) ratio (STA
period=0.2 s, LTA period=5 s, ratio=4) was used on
the filtered data to select all events in that frequency
band occurring on a given day. With this algorithm we
were able to identify hundreds of SP events per day,
regardless of their similarity. Because the sheer
quantity of events made it impossible to perform this
function on all of the data, we selected a subset of
days for which signal-to-noise was strong and
identified all events for those days. We chose 1 day
for each time period covered by a reference event,
with the exception of the first time period, for which
we examined 2 days, and period 4 which overlaps
both periods 3 and 5 (Table 2). This gave us a subset
of data with which we could examine the similarity
between events from different time periods as well as
events from within the same time period.
Because spurious signals may be selected using
the STA/LTA algorithm, we examined all of the
segmented events and manually deleted any events
that were obviously not SP phases of coupled events.
Events that were identified by the algorithm included
calibration pulses, electronic spikes due to telemetry
problems and regional earthquakes, all of which
were removed from the data subset. Additionally, we
removed earthquakes that overlapped the coda of
previous events. Of 2847 events of all types initially
extracted, 1400 coupled events were selected for
further analysis.
Table 2
Number of events extracted on selected days using eight different referen
Date Time
period
Number of coupled
events (STA/LTA)
Ref
event 1
Ref
even
March 26 1 234 59 10
March 30 1 166 55 5
May 11 2 142 0 21
July 10 3 265 0 0
July 25 5 169 0 0
August 29 6 311 0 0
September 23 7 134 0 0
October 24 8 244 0 0
In general, events were extracted using only the reference event associated
given day were found to be similar to multiple reference events. Thi
simultaneously, or that events were similar to several reference events, re
One of the goals of this analysis was to evaluate our
decision to use separate SP reference events for each of
eight time periods (Table 1). To that end we calculated
the spectral coherence between each reference event
and all 1400 events in the data subset. In general, only
days in the time period originally assigned to a
reference event contained earthquakes selected by that
event. In other words, events on May 11, a day initially
assigned to time period 2, were selected only when
reference event 2 was used. No events occurring on
May 11 were selected using any other reference event.
Events in time periods 6 and 7 were found to be mostly
interchangeable, suggesting that the two time periods
comprise similar event types. The only other interesting
exception is that reference event 2 was found to be
highly similar (coherenceN0.9) to 15 events that
occurred during first time period. In contrast, no
earthquakes on the day selected from time period 2
were extracted using reference event 1. This suggests
that there may have been two distinct SP sources in the
initial time period, but that the second type became
dominant by late April. Overall, these results (Table 2)
confirm that the decision to use different reference
events in each time period was robust.
Next we examined all earthquakes in the data
subset to investigate their similarity. To that end we
calculated the coherency between the SP phases of all
1400 events in the data subset. To ensure that we are
looking at the similarity of the SP phase only, we
calculated the mean coherence in the frequency range
of the SP phase, between 4 and 7 Hz. The total results
and mean coherence values for each event are shown
in Fig. 3a. The matrix diagonal (running from the
ce events
t 2
Ref
event 3
Ref
event 4
Ref
event 5
Ref
event 6
Ref
event 7
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
56 11 0 0 0
0 17 0 0 0
0 0 75 93 0
0 0 8 18 2
0 0 0 0 14
with that day’s event group. In some instances, however, events on a
s suggests either that there were multiple event types occurring
flecting a gradually evolving source.
Fig. 3. Estimated spectral coherencies for (a) the SP phase and (b) the LP phase of 1400 coupled earthquakes occurring over the span of the time period of this study (March–November
2002). For the SP phase, the coherence was calculated for each event pair in the frequency band between 4 and 7 Hz for the 5 s around the P-wave. A frequency band of 0.5–3 Hz was
used to estimate the spectral coherence for the first 5 s of the LP phase. The scale for spectral coherence is shown to the right. Values lying on the diagonals represent the auto-coherence
between an event and itself and are everywhere equal to 1. Points close to the diagonal represent events that occur closely spaced in time while those farther from the diagonal are
separated by longer time periods. The days from which data were extracted are shown on the top of the figure. While each event exhibits a range of coherence values, the overall values
are high, typically exceeding 0.7 for SP phases and 0.6 for LP phases, indicating a high degree of similarity over 9 months. Also evident is the fact that for the SP phase, certain time
periods exhibit long-term similarity while others change more rapidly. Events occurring between March and May are highly similar to one another, as are events occurring between
August and October. This suggests a minor but long-term change in the SP source process. In contrast, the LP phase shows little temporal evolution.
J.Caplan-Auerb
ach,T.Petersen
/JournalofVolca
nologyandGeotherm
alResea
rch145(2005)151–172
158
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 159
lower left to upper right in Fig. 3a) shows the auto-
coherence for an event (the spectral coherence with
that event and itself) and is equal to 1. Values close to
the diagonal represent events that are closely spaced
in time, while values plotted farther from the diagonal
represent the coherence between events spanning
longer time scales. In all cases, coherence values are
relatively high (0.7–0.9; Fig. 3a), with highest values
clustering near the diagonal. This indicates that while
all SP events are fairly similar, events are more likely
to be similar if they occur over a short time scale when
source conditions have not changed significantly.
Furthermore, events that occurred between March
and May have high coherence values relative to one
another, as do events occurring between August and
October. There is a strong decrease in coherence when
events in the former group are compared with events
in the latter (Fig. 3a). In contrast, events extracted in
July bear strong similarity only to each other. This
indicates that there was a slight but significant change
in the SP phase between May and August 2002.
A similar test was performed for the LP phase of
the 1400 coupled events included in the SP analysis.
As with the SP phase, the LP portions of coupled
events were cross-correlated with one another and the
coherencies calculated for a 5-s window (Fig. 3b).
This analysis shows that overall, LP phases have
slightly lower correlation coefficients than the SP
phases for the same events, averaging 0.6–0.8.
Furthermore, LP phases do not show the same
temporal trends evident in the SP phase analysis.
This suggests that LP phases do not change signifi-
cantly over long time scales and may have a more
stable source mechanism.
7. Coupled event delay times
A benefit of identifying groups of events by
extracting similar events is that the events are
segmented such that the earthquakes are aligned in
time series. This method of viewing the activity
promotes identification and timing of different phases.
Examples of such a plot for a single day of coupled
earthquakes are shown in Fig. 4. This perspective
clearly shows that when the SP phases are aligned in
time series, the arrival times of the LP phases occur
anywhere from 3 to 15 s after the SP phase (Fig. 4). In
this article, the time period between the SP and LP
phases will be called the bSP–LP delay timeQ, or
simply the bdelay timeQ. At times the delay time is
found to vary dramatically over short time scales (Fig.
4a and b) while at other times the delay time is
relatively consistent (Fig. 4c and d). As previously
noted, when delay times change rapidly, there does
not appear to be a systematic timing change. In some
instances a change in delay time is followed by a long
sequence of events while in other cases, only a few
events occur before delay times change yet again. The
observed changes in delay time are visible at all
stations, suggesting that the delay results from a
source effect rather than one associated with site or
path. Finally, more distant stations exhibit slightly
longer delay times (Fig. 4). This timing difference
indicates that the two phases have different source
locations or different phase velocities.
To estimate delay times for all events we need to
identify the start of both phases. The SP phase onset is
easily determined as it is identified during extraction of
similar events. To determine the beginning of the LP
phase, we filtered the data between 0.5 and 4 Hz and
used an STA/LTA algorithm. Because the LP phase
often begins in the coda of the precursory SP phase, and
because the onset of LP activity is emergent, we
estimate a 2-s uncertainty in the actual delay time.
Although the actual delay times are approximate, the
general trend, that of wide variability at some times and
consistent delay times at others, is robust.
When delay time is plotted for the time period of the
study (March–November 2002; Fig. 5) we see that
there are time periods (e.g. late March) with highly
variable delay times, while at other times (e.g. late July,
November) the two phases occur at predictable
intervals. Furthermore, there are general trends in the
data: delay times are short in April and May but longer
between June and September. This suggests that the
mechanism acting between the two phases not only
changes on short time scales but occasionally stabilizes
at the maximum or minimum observed values.
To further investigate SP–LP delay times, we
plotted delay time against several parameters for
Shishaldin coupled events (Fig. 6). We found no
correlation between delay time and interevent time
(Fig. 6a). There is only a weak correlation between the
maximum amplitude of the SP phase and the delay
time (Fig. 6b); high amplitude SP phases correlate
0 5 10 15 20 25 30TIME (s)
ALL EVENTS BELOW, STACKED
ISTK:SHZ, 3/29/02
0 5 10 15 20 25 30TIME (s)
ALL EVENTS BELOW, STACKED
SSLN:SHZ, 3/29/02
0 5 10 15 20 25 30TIME (s)
ALL EVENTS BELOW, STACKED
ISTK:SHZ, 5/1/02
0 5 10 15 20 25 30TIME (s)
ALL EVENTS BELOW, STACKED
SSLN:SHZ, 5/1/02
a) b)
c) d)
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172160
04/01 05/01 06/01 07/01 08/01 09/01 10/01 11/010
5
10
15
20
SP
-LP
DE
LAY
(s)
DATE, 2002
Fig. 5. SP–LP delay time plotted against date for station SSLN. Some time periods, such as mid-March and mid-July exhibit wide variation in
delay time, while other time periods, such as early August and mid-November, show consistent delay times. In most cases the variability in
delay time is no more than 5–7 s.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 161
with short delay times. However, at small delay times
the apparent maximum amplitude for SP events may
be influenced by overlapping signal from the larger
LP phase resulting in larger apparent amplitudes.
Although there appears to be a slight decrease in event
count with the time of day (Fig. 6c), this may be due
to an increase in noise resulting from afternoon winds
(occurring toward the end of the UTC day). Spectral
analysis of event occurrence times shows no obvious
peaks, thereby ruling out tidal or diurnal effects.
Coupled events occurred throughout the time period
from March to November 2002, suggesting that there
is no correlation between seismic activity and
precipitation. However, since there are no meteoro-
logical stations on Unimak Island, this lack of
correlation cannot be positively confirmed.
8. Relationship between LP phases and LP events
The LP phase of the coupled event bears strong
similarity to the LP events that occurred prior to March
13 and after November 21. The possibility therefore
Fig. 4. (a) Shishaldin coupled events occurring on March 29, 2002, record
from the continuous record using a common 5-s reference event taken from
extracted events, plotted according to their time of occurrence during the d
panels show stacks of all of the events plotted below. Although the SP phas
~15 seconds after the SP phase. (b) Shishaldin coupled events occurring on
variable timing of SP and LP phases is evident on both stations. (c) Shishal
represents a day in which the time between the SP and LP phases is relativ
Again, the timing of the two phases is consistent at all stations.
exists that the LP phase visible within coupled events
is the same as other LP events, with the only difference
being the SP trigger. Thus, to better understand the
coupled events, we investigate the relationship
between Shishaldin LPs and the LP phase of coupled
earthquakes. We can examine the LP signals by
comparing the time series of a btypicalQ LP event (that
is to say, unassociated with an SP precursor) and the
LP phase of a coupled event (Fig. 7a). In this format it
is evident that the two events are highly similar from
the onset time at 20 s until approximately 37 s into the
file. This similarity is visible at all stations, again
suggesting that the similarity is due to source rather
than path or site effects. When the data are bandpass-
filtered to enhance SP energy (Fig. 7b) it is clear that
only one of the LP events has an associated SP
precursor. These data suggest that the LP phase of
Shishaldin coupled events shares a common source
mechanism and location with other Shishaldin LPs; the
only difference between the two events is the presence
of the precursory SP phase for the coupled events.
As a final check as to the relationship between LP
and coupled events, we scanned through several days
ed at station SSLN, 6.5 km from the vent. All events were extracted
the SP phase of a coupled earthquake. The lower panels show the
ay (morning events plotted toward the top of the panel). The upper
es are aligned in time series, the LP phases occur anywhere from 3 to
March 29 2002, recorded at station ISTK, 17 km from the vent. The
din coupled events on April 30 2002, recorded at station SSLN. This
ely constant, at ~3 s. (d) The same events as (c), recorded at ISTK.
0
5
10
15
20
25
0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
INTEREVENT TIME (s)
SP
-LP
DE
LAY
TIM
E (
s)
0 5 10 15 200
50
100
150
200
250
300
350
HOUR OF DAY
NU
MB
ER
OF
EV
EN
TS
SP
-LP
DE
LAY
TIM
E (
s)
SP PHASE MAX AMPLITUDE (cts)0 50 100 150 200 250 300
0
5
10
15
20
25
a)
b)
c)
Fig. 6. Representations of the SP–LP delay against various parameters associated with the coupled events. (a) Interevent time for repeating
coupled events plotted against delay time. (b) Maximum amplitude of the SP phase versus delay time. (c) Number of events that occurred during
each hour of the day. No strong correlation exists between delay time and interevent time. There are slightly fewer events towards the end of the
UTC day, corresponding to late afternoon Alaska time. This probably reflects an increase in background noise due to wind rather than an actual
decrease in event numbers. There is a very weak correlation between the amplitude of the SP phase and the delay time; high amplitude SP
phases occur for short delay times only.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172162
0 5 10 15 20 25 30 35TIME (s)
0 5 10 15 20 25 30 35TIME (s)
-200
0
200
a)
b)
-50
0
50
Fig. 7. (a) Waveforms for an LP event at 10:31 UTC with no apparent SP precursor (blue), and the LP phase of a coupled event at 11:36 UTC
(red), both on November 21. The strong similarity between events suggests that they share a common source. b) The same waveforms shown
above, bandpass filtered between 4 and 8 Hz to enhance any SP precursory signal. The blue seismogram shows no noticeable increase in SP
energy prior to the LP phase, while the red event shows a clear SP signal between 5 and 10 s.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 163
of data to estimate the ratio of bnormalQ LP events to
coupled events. Identifying LP events that are not
associated with SP precursors is difficult because SP
phases are small and may easily be masked by noise,
giving the event the appearance of a standard LP.
However, our estimated counts suggest that during the
time frame of this study (March–November 2002),
there were roughly twice as many coupled events as
LPs. In contrast, on January 3 2002, 183 LP events
were counted, compared to only 38 coupled events.
This activity occurred before the coupled events were
identified as a distinct event type. Intriguingly, we
note that when coupled events were dominant, the LP
phase of the coupled events were significantly larger
than plain LPs. Similarly, on January 3, when LP
events were among the largest noted at Shishaldin, the
amplitude of the SP phase of coupled events was so
small that they were not identified except in retro-
spect. Thus there appears to be a trade-off in energy
content; the bulk of the energy released by Shishal-
din’s seismicity resides in one event type at a time.
9. Earthquake locations
Shishaldin earthquakes are extremely difficult to
locate due to their small magnitude and the station
geometry described above. Initial locations of the
largest coupled and LP events place their hypocenters
at 0–3 km directly beneath Shishaldin’s summit
(Dixon et al., 2003), but hypocenters for the majority
of events cannot be calculated by traditional means.
We made use of the similarity of Shishaldin activity to
estimate locations of the smaller earthquakes.
Locating the SP phase of the coupled events first
required a means to enhance the first arrivals. To
that end, we began by stacking all of the events
selected with a single reference event. In some cases
this consisted of events spanning several weeks
while for other reference events we found similar
events occurring for only a few days (Table 1).
Following the removal of noisy traces, events for
each group were aligned by cross-correlation and
stacked at each station. P-wave arrival times were
picked for the stacks, and the events were located
using HYPOELLIPSE (Lahr, 1999). The velocity
model used for Shishaldin is composed of seven
constant-velocity layers and was originally derived
for Pavlof volcano (McNutt and Jacob, 1986). We
were unable to identify P-wave arrivals at station
BRPK (Fig. 1), even after stacking several hundred
earthquakes, so that station was not used in the
locations. As previously mentioned, station ISNN
suffers from a poor telemetry link, so for event
group 5 (Table 1) we found no usable traces at that
station.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172164
Hypocentral locations for the stacked events are
shown in Fig. 8. Although the events are clearest at
station SSLN they occur just west of the summit,
midway between stations SSLN and SSLS. Depths are
very poorly constrained due to (a) the lack of reliable
S-phases and (b) the low number of stations recording
the activity, but the best fit locations are at shallow
(b4 km below the summit) depths. Thus, although we
do not know exact hypocentral depths, we take 4 km
to be the maximum likely depth for the source of the
SP phase. Note that the apparent bairquakesQ visible inFig. 8 (earthquakes whose depths are above regional
topography) are a consequence of the location
algorithm’s assumption of flat topography. These
events are obviously poorly located but indicative of
shallow depths.
Although the small amplitude of Shishaldin events
makes it impossible to pick reliable P-waves for
individual events, we were able to estimate arrival
times by cross-correlating each waveform with the
stacked events. The number of lags at maximum
correlation yields the number of samples by which
each event is offset relative to the stack and allows us
to pick an arrival time to within 0.01 s. Picks were
weighted according to the correlation coefficient, such
164o 10'
164o 10'
164o 00'
164o 00'
163o 50
163o 50
54o 40'
54o 50'
SSLS
SSLW
SSLN
a)
Fig. 8. (a) Approximate hypocentral locations for stacks of events extracte
hypocenters for 1370 coupled events, with P-wave arrivals selected by cros
with white triangles. (b) North–south cross section for the hypocenters sho
shallow portion of the edifice. Hypocenters that plot above ground are
topographic relief.
that events with correlations exceeding 0.8 were given
full weight, and those with correlations of b0.4 were
not used in the location. Of the ~7000 earthquakes
selected for this study, 1400 were located with
HYPOELLIPSE quality factors of A or B, indicating
68% confidence levels for horizontal and vertical
locations of b2.67 km (Lahr, 1999). Not surprisingly,
those locations cluster tightly in the same region as the
stacked events, on the west flank of Shishaldin, near
the summit. As with the stacked events, depths are
poorly constrained between 0 and 4 km. In all cases,
however, the events appear to be shallow and within
the Shishaldin edifice. While these hypocenters may
be considered suspect due to the low number of
stations, we note that in the vast majority of cases, the
best correlations for all stations were found with
offsets of less than 2 samples (0.02 s), requiring that
all of the events have their source within ~100 m of
each other. No temporal trends in earthquake locations
are evident.
We attempted the same stacking analysis on the LP
phase of the coupled earthquakes but were unsuc-
cessful in determining reliable hypocenters. Because
the LP phase occurs after the SP phase, first arrivals in
LP data were often difficult to identify. Furthermore,
'
'
ISNN
-3
-2
-1
0
1
2
3
54.7o 54.8o
54.7o 54.8o
-3
-2
-1
0
1
2
3
ELE
VA
TIO
N (km
)
b)
d with a common reference event (black stars) and the most reliable
s-correlation (gray circles). Short period seismic stations are marked
wn in panel (a). Depths are very poorly constrained but cluster in the
due to the fact that the location algorithm does not account for
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 165
even in stacked data, the onset of the event was too
emergent to accurately time first arrivals. Because of
the low-frequency nature of the event, a half-wave-
length uncertainty in first arrivals corresponds to a
long time interval (~0.5 s) and consequently a large
degree of location uncertainty. However, there is
reason to believe that, like the SP phase, the LP phase
of the coupled event has its source within Shishaldin’s
edifice. Firstly, delay times are slightly offset between
stations; an event with a 5-s delay time at station
SSLN (6.3 km from the vent) exhibits just over 5–6 s
of delay at SSLW (10 km from the vent) and 7–8-s of
delay at ISTK (17.1 km from the vent). This allows us
to conclude that the SP and LP sources have discrete
locations, since delay times would be consistent
across the network for co-located phases. The
relatively small move-out further suggests that the
two phases have sources near one another, or at least
within the immediate vicinity of Shishaldin. If one
source region lay significantly farther we would
expect to see larger move-out at the more distant
stations. Finally, conventional location of pure LP
events places them at shallow depths beneath the
summit (Dixon et al., 2003) and stacking analysis of
pure LP events yield similar hypocenters (Petersen et
al., 2004). If LP phases share a common source with
LP events, then they would also be located in the
shallow edifice.
0 5 10 15 20
TIME (
3/26/02, 03:07 UTC
3/26/02, 14:30 UTC
Fig. 9. Shishaldin coupled earthquakes recorded on March 26, 2002. Bot
correlation, although the SP–LP delay time varies by ~3 s. The SP phase
coherence between the LP phases at 0.5–3 Hz is 0.76.
10. A proposed model for Shishaldin coupled
earthquakes
Several parameters must be considered when
proposing a model for Shishaldin coupled earth-
quakes. The first is the coupled nature of the event;
each SP phase is followed by an LP phase, and no SP
events have been identified without a succeeding LP
phase. In contrast, LP events occur without the SP
precursor, and waveforms for some standard LPs are
virtually identical to the LP phase of Shishaldin
coupled events. Secondly, the repetitive nature of both
the SP and LP phases suggests that each represents a
stationary, non-destructive source process, although
there is more variability in the appearance of the LP
phases (Fig. 3b; Petersen et al., 2002). Nonetheless,
events with variable SP–LP delay times have been
identified for which both the SP and LP phases are
highly similar (Fig. 9). Thus, if both source regions
are fixed, then the variable timing between the two
phases must be caused by variations in the mechanism
that relates the two, i.e. the trigger mechanism. Next,
the prolonged nature of the SP–LP delays, between 3
and 15 s, requires a trigger mechanism that can
change speed by a factor of 5. And finally, since both
source regions are believed to lie within the Shishal-
din edifice, the trigger mechanism cannot travel more
than a few kilometers over the delay time, and
25 30 35 40
s)
h the SP and LP phases of these events exhibit a strong degree of
s have a spectral coherence between 4 and 7 Hz of 0.79 while the
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172166
therefore must move at slow speeds (b1 km/s). In fact,
if we take the maximum depth of the SP phase to be 4
km below the summit (the maximum depth calculated
for hypocentral locations), and assume that the LP
phase is shallower, we can place constraints on the
velocity of the triggering mechanism. The approx-
imate range in delay time of 3–15 s for 4 km spacing
yields a velocity range of approximately 270–1333 m/
s. If the SP phase has a shallower source, or the LP
source is located at greater depth, then these values
would be considerably lower.
We believe that both the SP and LP phases of
Shishaldin coupled events result from the presence of
fluids within the volcano. This is consistent with the
low frequency of these signals as well as the
repeatability of the source. Furthermore, the plume
at Shishaldin’s summit confirms that gas is constantly
escaping the volcano, although the relative contribu-
tions of magmatic and meteoric gas are unknown.
While no correlation has been confirmed between
seismicity and the puffing steam plume, the observed
timing between successive gas puffs (1–3 min) is
similar to the timing between LP events recorded by
the seismic network. Furthermore, the fact that the
plume comes in bpuffsQ indicates that gas is releasedin discrete bursts. Finally, the installation of a more
sensitive pressure sensor in 2003 allowed us to
correlate LP events at Shishaldin with infrasonic
signals (Petersen et al., 2004). Although a detailed
investigation of this relationship is beyond the scope
of this article, this correlation encourages us to assume
a shallow, gas-related source for Shishaldin LPs and
the LP phase of coupled events. Consequently, all the
models we consider are based upon a volcano with
two fluid-based sources.
There are many means by which fluids, and gas in
particular, may generate seismic signals. One involves
the sudden coalescence of foam into a single gas slug.
This mechanism has been observed in laboratory
experiments (Jaupart and Vergniolle, 1988, 1989) and
has been invoked as a source of LP seismicity (Ripepe
and Gordeev, 1999). When the slug reaches the top of
the fluid it oscillates prior to bursting, inducing
seismic and infrasonic signals (Vergniolle and Bran-
deis, 1994, 1996; Gil Cruz and Chouet, 1997;
Vergniolle et al., 2004). Alternatively, LP events
could be due to explosions resulting from magma–
water, or fuel–coolant, interactions (Wohletz, 1986,
2002; Buttner and Zimanowski, 1998). An additional
means by which fluid may generate seismic signals is
by promoting oscillation of a crack or reservoir
(Chouet, 1985, 1988; Julian, 1994; Garces, 1997).
Conduit vibration may also be induced by choked
flow at obstructions (Julian, 1994; Morrissey and
Chouet, 1997) or by turbulent flow (Hellweg, 2000).
Finally, fluid transfer through volcanic conduits has
been invoked to explain LP events and tremor at a
number of volcanoes, including Kilauea (Aki et al.,
1977; Almendros et al., 2001), Lascar (Hellweg,
2000) and Sakurajima (Honda and Yomogida, 1993).
We first considered a model in which gas bubbles
coalesce into a slug at a deep boundary and rise under
their own buoyancy through magma or hydrothermal
fluid. The SP phase of the coupled event is produced
when the bubbles coalesce at depth. At some point as
it rises through the conduit, the slug enters a second
fluid body, perhaps a magma reservoir or crack,
initiating the LP signal. This signal may result from
the vibration of the reservoir walls or from the
oscillation of the bubble itself. Thus the trigger event
between the two phases of the coupled earthquake is a
mechanical one, the rising of the slug itself. In this
model we hypothesize that variations in SP–LP delay
time result from changes in the rise speed of slugs at
depth due to slug radius or fluid density and viscosity.
There are several reasons to immediately discount
such a model. If the SP phase is both the trigger and
the source for the LP phase, then the LP events cannot
occur without the precursor. Yet the LP signals are
observed to occur before, coincident with and after the
time period in which coupled events were dominant.
Thus it is more likely that the two phases have
independent sources, but that the LP phase can be
triggered into activity by the SP phase. Secondly, even
for large slugs in dense fluid, the rise velocity is only a
few m/s. For coupled events in which the LP phase
follows immediately after the SP phase, this would
require the SP and LP reservoirs to be within a few
tens of meters of each other. It seems to be highly
unlikely that two discrete reservoirs could coexist at
such proximity for months without merging. Sec-
ondly, bubbles coalescing in a reservoir tend to form a
slug which fills the conduit through which it rises;
variable radii are not observed in laboratory models
(Jaupart and Vergniolle, 1989; Vergniolle et al., 1996).
Finally, we have shown that velocities for the
SP SIGNAL
LP SIGNAL
Fig. 10. A proposed model for Shishaldin coupled earthquakes
Fluid moving within the Shishaldin conduit encounters an obstacle
on the conduit wall. The restricted flow promotes oscillations of the
conduit walls, producing the SP phase. Seismic energy from the SP
signal propagates through the fluid at a rate dependent on the void
fraction in the fluid. The wave encounters a region with high void
fraction near the top of the conduit and induces bubble coalescence
and the LP seismic signal.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 167
triggering mechanism are likely to be as high as
several hundred m/s, a speed that is unattainable by a
rising gas slug in magma.
In our preferred model, the trigger mechanism is
seismic energy generated by the precursory SP event,
rather than mechanical transport. In this model, the
velocity of the trigger mechanism depends on the
acoustic velocity of the material through which the
wave propagates; in this case, the conduit fluid. We
will show that small changes in gas content are
sufficient to explain the observed variability in delay
times.
We propose that coupled events initiate when a
pressure transient within the conduit fluid creates the
SP signal. Julian (1994) has shown that a small
perturbation to steady-state flow within a conduit can
result in transient seismic events. While many of the
source mechanisms described above can also explain
a signal such as the SP phase, a key parameter
constraining the SP source is the fact that the
similarity evident in SP waveforms is independent
of the observed delay time. This rules out any source
mechanism associated with conduit resonance, as
crack wave velocities are dependent on the acoustic
wave speed of the fluid (Chouet, 1988, 1996; Gil Cruz
and Chouet, 1997). Because many fundamental
parameters of the SP phase, such as source depth
and regional material properties, remain unknown, an
investigation of the SP power spectrum is beyond the
scope of this study.
The same seismic energy recorded by the seis-
mometers as the SP phase also travels within the
volcano itself to a shallow portion of the conduit or
reservoir where it triggers the LP source. There are
two mechanisms that we considered for the LP source:
magma–water interactions (Wohletz, 1986; Buttner
and Zimanowski, 1998) and bubble coalescence
(Jaupart and Vergniolle, 1988, 1989). In the former
case, seismic waves impinging on a magma–water
boundary collapse an insulating film layer at the
boundary, causing superheated water to flash to steam.
In the latter model, SP phase energy passes through a
shallow layer of bubbles or foam (Fig. 10), causing
the bubbles to decompress and rapidly coalesce into a
bubble slug. The slug coalescence generates an LP
signal and releases a pulse of gas. Both models can
generate LP events with or without a triggering event:
Superheated water can flash to steam without a trigger
.
if it continues to heat past a critical temperature
(Wohletz, 2002) and a foam layer that grows beyond a
critical thickness will likewise collapse on its own
(Jaupart and Vergniolle, 1988, 1989).
While both models are reasonable for Shishaldin,
we prefer the bubble coalescence model. Magma–
water interactions can generate instabilities at the
magma surface, promoting magma break-up and an
ensuing explosive eruption (Wohletz, 1986, 2002),
which was not observed at Shishaldin in 2002. The
ability of the volcano to continually supply water that
is rapidly superheated is also questionable. Finally,
there is no evidence of shallow magma in the
Shishaldin conduit. In the case of the bubble
coalescence model, there is a large supply of gas
available in the form of volcanic and hydrothermal
gases, and shallow magma is not required. In the
models presented here, we assume the presence of
magma but discuss the consequences for the case of
other fluids.
To explain the variable SP–LP delay timing in this
model, we seek a means by which the velocity of
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172168
seismic energy can be dramatically changed over
short time periods. We invoke changes in fluid void
fraction as a means of accelerating or decelerating the
trigger mechanism. After Gil Cruz and Chouet (1997)
we calculate am, the velocity of a pressure wave in a
fluid according to
1
am¼ 1
alþ
ffiffiffiffiffiffiffiffiffiffiqmegcmP
rð1Þ
(Hsieh and Plesset, 1961; Gil Cruz and Chouet, 1997).
The fluid velocity is given by al, P represents the
pressure in the reservoir, and eg is the void fraction of
the fluid. qm, the density of the gas–liquid mixture,
depends on the void fraction eg and liquid and gas
densities ql and qg according to,
qm ¼ egqg þ 1� eg� �
ql ð2Þ
Because these values have not been measured
precisely at Shishaldin, we use commonly accepted
values for gas and basalt and take ql to be 2700 kg/m3
(Williams and McBirney, 1979) and qg to be 1 kg/m3.
The seismic velocity of the liquid magma is estimated
0 0.1 0.20
500
1000
1500
VOID FR
P-W
AV
E V
ELO
CIT
Y (
m/s
)
εg= 0
εg= 0.0004
εg= 0.037
εg= 0.042
INC
REA
SIN
G P
RES
SUR
E
0
Fig. 11. Velocity of seismic waves through gas-rich basaltic magma. Veloc
pressure within the fluid. Thin solid curves represent values calculated at
solid line represents a pressure of 10 MPa for a magma-filled conduit. Th
water to a depth of 4 km, resulting in static pressure of ~4 MPa. White ci
maximum and minimum velocities estimated for the coupled event trigge
as 2500 m/s. The isentropic coefficient of the mixture
cm depends on the specific heats of the gas and liquid
as well as the mass ratio of gas to liquid. Because
these values are not known specifically for Shishaldin
we follow Gil Cruz and Chouet (1997) and take the
value of cm to be 1.0066.
Seismic velocity in low-viscosity fluid is strongly
dependent on void fraction qg, with only a small
amount of gas dropping the velocity far below that of
either pure liquid or pure gas. In Fig. 11 we show
seismic velocity for 2-phase fluids with a range of
void fractions and reservoir pressures. Because the
composition of the fluid within the Shishaldin conduit
is unknown, we cannot precisely determine the
pressure within the conduit. We can, however,
estimate the regional pressure for the endmember
cases of a conduit filled with magma (q=2700 kg/m3)
and one filled with water (q=1000 kg/m3). We
estimate the pressure as P=qgh where q is fluid
density and is related to void fraction according to Eq.
(2) above, g is the acceleration due to gravity and h is
the depth within the conduit, between 0 and 4 km
below the summit, according to the hypocentral
0.4 0.5 0.6
ACTION εg
P = 1 MPa
P = 100 MPa
.3
ity is dependent on the fluid void fraction qg as well as the ambient
pressures between 1 and 100 MPa, at 10 MPa intervals. The heavy
e heavy dashed line represents the same conduit filled with gas-rich
rcles mark the gas velocities of 270 and 1300 m/s, the approximate
r mechanism.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 169
depths of The SP phase. These values give us an
approximate pressure range of 5–100 MPa. All of
these values decrease, of course, if we consider the
presence of gas within the fluid. While we cannot
determine the precise pressure within the conduit,
these values allow us to investigate the effects of on
seismic velocity at pressures that are reasonable for
the conditions (Fig. 11).
The values shown in Fig. 11 confirm that only a
small change in qg is required to explain the
observed variability in delay time. For example, a
magma at 10 MPa (heavy solid line in Fig. 11)
experiences a 5-fold decrease in seismic velocity for
a change in eg of 0.0004–0.042. More importantly,
the predicted seismic velocities for this range of egare between 270 and 1300 m/s, the precise range of
velocities required to explain the observed delay
times for an SP–LP source separation of 4 km. The
results are similar for an identical water filled
conduit, with slightly lower values for eg. At 4 km
depth, pressure in a water-filled conduit is ~0.39
MPa (heavy dashed line in Fig. 11) and velocity is
found to range between 270 and 1300 m/s for egbetween 0 and 0.037 (Fig. 11).
The relationship described above, in which
increased gas content results in increased SP–LP
delay time, causes us to wonder if there might be a
relationship between the interval between successive
seismic events and the delay time. In the bubble
coalescence model, long delay times indicate that
more gas is available to generate bubble layers or
foam, which would presumably result in more LP
events. No correlation between interevent time and
delay time was initially visible in our data (Fig. 8a).
However, the interevent times presented in Fig. 8a are
for SP phases only. To compare delay times to the
total number of events, we selected 2 days and
counted all earthquakes regardless of event type. On
May 11, a day when delay times were low (3–5 s),
indicating low gas content, we counted a total of 368
events. In contrast, we counted 957 earthquakes on
July 25, when high delay times (10–12 s) suggest high
gas content. This fact supports the bubble coalescence
model for the LP phase.
As previously noted, when coupled earthquakes
are the predominant event type, pure LP events appear
to be of low amplitude. This model may indicate that
the SP phase forces overpressurized bubbles to
coalesce. Ripepe and Gordeev (1999) have shown
that bubbles layers that coalesce freely do so when the
hydrostatic pressure has diminished beyond a critical
value. Bubbles that are forced to coalesce, however,
exhibit larger overpressure and generate stronger
seismic waves.
11. Demise of the coupled events
The last day on which the coupled earthquakes
were dominant at Shishaldin was November 21,
2002. We find coupled events occurring until
approximately 09:15 UTC, after which the events
appear to be normal LP earthquakes. No other
unusual features, such as VT earthquakes or thermal
anomalies, were observed at the time, and the overall
occurrence rate of earthquakes appears unchanged.
This suggests that the source of the SP phase was
destroyed rapidly, and that the LP phase was
unaffected by the change. The most likely explan-
ation is that the conduit obstruction that caused the
SP phase was destroyed on November 21, although
no associated seismic signal was observed. LP
activity continued without trigger activity after the
coupled events ended. The total number of events at
Shishaldin remained similar following the cessation
of the SP phase.
12. Discussion
The model proposed here indicates variations in
SP–LP delay time are likely due to changes in gas
content within the conduit. Events such as these can
therefore provide insight into the amount of gas within
the system and hence the state of unrest of the
volcano. Because Shishaldin is not visible from any
populated area, and because little is known about gas
flux at the volcano, events such as the coupled
earthquakes could provide a rare window into changes
in gas content. At present, the conduit system at
Shishaldin appears to be open, allowing constant
degassing, so increases in gas flux may not indicate a
hazardous change in the volcano’s plumbing system.
In 1931, however, Hubbard (1935) reported that
Shishaldin’s summit crater was filled with a solidified
lava lake, indicating that it has the capacity of shutting
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172170
and blocking gas release. At a volcano with a closed
system, events such as the coupled earthquakes could
be used as an indicator of gas content for use in hazard
evaluations.
As previously noted, there are times when SP–LP
delay times vary significantly over short time scales
(Figs. 4 and 5). More commonly, however, the LP
phase seems to follow the SP phase within a few
seconds of a predictable time (Figs. 4 and 5). This
suggests that gas flux and gas content at Shishaldin
usually exhibit relatively steady state behaviour, but
occasionally fluctuate rapidly. There appear to be
times (e.g. near June 1; Fig. 5), in which delay
times suddenly increase markedly. This reflects the
addition of new gas into the system. The sharp
change from short delay times in April–June to
consistently long delay times evident between June
and October may indicate that a new fracture
opened through which gas could enter the system.
Increased gas content may be expected to correlate
with increased gas flux at the vent, but no
observations exist to evaluate this relationship.
Periodic sealing of the crack could regulate delay
times over short periods.
The presence of the SP phase is the key feature
that defines a coupled earthquake. If this phase does
indeed result from an obstruction in the conduit or
other such feature, then it may be a simple case of
good fortune that these signals occurred such that
they could be identified and studied. However,
earthquakes similar to Shishaldin coupled events
have been noted on a handful of other volcanoes
(Galeras, Gil Cruz and Chouet, 1997; Popocatepetl,
R. White, personal communication, 2003; Stromboli,
J. Lees, personal communication, 2002; Deception
Island, Ibanez et al., 2003; Koryaksi, Gordeev and
Senyukov, 2003), so they may be more common than
has been previously noted. In other instances, the
association between the two phases may not be
noted, possibly due to the small amplitude of the SP
phase. The increased popularity of cross-correlation
techniques in the study of volcanic seismicity (e.g.
Gillard et al., 1996; Lees, 1998; Aster and Rowe,
2000; Wolfe et al., 2003; Rowe et al., 2003; Battaglia
et al., 2004) may promote the identification of
similar types of events. We are therefore optimistic
that this type of locally triggered seismicity may be
found in other seismically active regions. Events
such as these could provide a telling view on the
internal structure of active volcanoes and a new
means by which gas content can be remotely and
safely monitored.
Acknowledgments
We are grateful to our colleagues at the Alaska
Volcano Observatory, notably S.R. McNutt, for
helpful discussions regarding Shishaldin seismicity.
The work was greatly improved following helpful
reviews by Timothy Horscroft, Eisuke Fujita and an
anonymous reviewer. The manuscript further bene-
fited from discussion with R. White and S. Prejean
of the U.S.G.S. Several figures were produced
using the Generic Mapping Tools software by
Wessel and Smith (1998).
References
Aki, K., Fehler, M., Das, S., 1977. Source mechanism of
volcanic tremor: fluid-driven crack models and their applica-
tion to the 1963 Kilauea eruption. J. Volcanol. Geotherm.
Res. 86, 7095–7110.
Almendros, J., Chouet, B.A., Dawson, P.B., 2001. Spatial extent of
a hydrothermal system at Kilauea Volcano, Hawaii, determined
from array analyses of shallow long-period seismicity: 2.
Results. J. Geophys. Res. 106, 13581–13597.
Aster, R., Rowe, C., 2000. Automatic phase pick refinement and
similar event association in large seismic datasets. In:
Thurber, C.H., Rabinowitz, N. (Eds.), Advances in Seismic
Event Location. Kluwer Academic Publishers, the Nether-
lands, pp. 231–263.
Battaglia, J., Thurber, C.H., Got, J.-L., Rowe, C.A., White, R.A.,
2004. Precise relocation of earthquakes following the 15 June
1991 eruption of Mount Pinatubo (Philippines). J. Geophys.
Res. 109, B07302.
Beget, J., Nye, C., Stelling, P., 1998. Postglacial collapse and
regrowth of Shishaldin Volcano, Alaska, requires high eruption
rates. EOS Trans. 70, S359.
Buttner, R., Zimanowski, B., 1998. Physics of thermohydraulic
explosions. Phys. Rev., E 57, 5726–5729.
Caplan-Auerbach, J., McNutt, S.R., 2003. New insights into the
1999 eruption of Shishaldin Volcano, Alaska, based on acoustic
data. Bull. Volcanol. 65 (6), 405–417.
Chouet, B.A., 1985. Excitation of a buried magmatic pipe: a
seismic source model for volcanic tremor. J. Geophys. Res.
90, 1881–1893.
Chouet, B.A., 1988. Resonance of a fluid-driven crack: radiation
properties and implications for the source of long-period events
and harmonic tremor. J. Geophys. Res. 93, 4373–4400.
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172 171
Chouet, B.A., 1996. Long-period volcano seismicity: its source and
use in eruption forecasting. Nature 380, 309–316.
Chouet, B.A., 2003. Volcano seismology. Pure Appl. Geophys. 160,
739–788.
Dixon, J.P., Stihler, S.D., Power, J.A., Tytgat, G., Estes, S., Moran,
S.C., Paskievitch, J., McNutt, S.R., 2003. Catalog of earthquake
hypocenters at Alaskan volcanoes: January 1, 2002 through
December 31, 2002: U.S. Geological Survey Open-File Report
OF 03_267, 58 pp.
Fischer, T.P., Morrisey, M.M., Calvache, M.L., Gomez, D., Torres,
R., Stix, J., Williams, S.N., 1994. Correlations between SO2 flux
and long-period seismicity at Galeras volcano. Nature 368,
135–137.
Fournelle, J., Marsh, B., 1991. Shishaldin Volcano; Aleutian high-
alumina basalts and the question of plagioclase accumulation.
Geology 19, 234–237.
Garces, M., 1997. On the volcanic waveguide. J. Geophys. Res. 102
(10), 22547–22564.
Gil Cruz, F., Chouet, B.A., 1997. Long-period events, the most
characteristic seismicity accompanying the emplacement and
extrusion of a lava dome in Galeras Volcano, Colombia, in
1991. J. Volcanol. Geotherm. Res. 77, 121–158.
Gillard, D., Rubin, A.M., Okubo, P., 1996. Highly concentrated
seismicity caused by deformation of Kilauea’s deep magma
system. Nature 384, 343–346.
Gordeev, E.I., Senyukov, S.L., 2003. Seismic activity at Koryaksi
volcano in 1994: hybrid seismic events and their implications
for forecasting volcanic activity. J. Volcanol. Geotherm. Res.
128, 225–232.
Hellweg, M., 2000. Physical models for the source of Lascar’s
harmonic tremor. J. Volcanol. Geotherm. Res. 101, 183–198.
Honda, S., Yomogida, K., 1993. Periodic magma movement in the
conduit with a barrier: a model for the volcano tremor. Geophys.
Res. Lett. 20 (2), 229–232.
Hsieh, D.Y., Plesset, M.S., 1961. Theory of rectified diffusion of
mass into gas bubbles. J. Acoust. Soc. Am. 33, 206–215.
Hubbard, B.R., 1935. Cradle of the Storms. Dodd, Mead, New
York. 285 pp.
Ibanez, J.M., Carmona, E., Almendros, J., Saccorotti, G., Del Pezzo,
E., Abril, M., Ortiz, R., 2003. The 1998–1999 seismic series at
Deception Island volcano, Antarctica. J. Volcanol. Geotherm.
Res. 128, 65–88.
Jaupart, C., Vergniolle, S., 1988. Laboratory models of Hawaiian
and Strombolian eruptions. Nature 331, 58–60.
Jaupart, C., Vergniolle, S., 1989. The generation and collapse of a
foam layer at the roof of a basaltic magma chamber. J. Fluid
Mech. 203, 347–380.
Julian, B., 1994. Volcanic tremor: nonlinear excitation by fluid flow.
J. Geophys. Res. 99 (6), 11859–11877.
Klein, F.W., Koyanagi, R.Y., Nakata, J.S., Tanigawa, W.R., 1987.
The seismicity of Kilauea’s magma system. In: Decker, R.W.,
Wright, T.L., Stauffer, P.H. (Eds.), Volcanism in Hawaii, U. S.
Geol. Surv. Profess. Pap., vol. 1350, pp. 1019–1185.
Lahr, J.C., 1999. HYPOELLIPSE: a computer program for
determining local earthquake hypocentral parameters, magni-
tude, and first-motion pattern (Y2K compliant version). U.S.
Geological Survey Open-File 99-23.
Lahr, J.C., Chouet, B.A., Stephens, C.D., Power, J.A., Page, R.A.,
1994. Earthquake classification, location, and error analysis in a
volcanic environment: implications for the magmatic system of
the 1989–1990 eruptions at Redoubt Volcano, Alaska. J.
Volcanol. Geotherm. Res. 62, 137–151.
Lees, J., 1998. Multiplet analysis at Coso Geothermal. Bull.
Seismol. Soc. Am. 88 (5), 1127–1143.
McNutt, S.R., 2000. In: Sigurdsson, H., Houghton, B., McNutt,
S.R., Rymer, H., Stix, J. (Eds.), Volcanic Seismicity, Chapter
63 of Encyclopedia of Volcanoes. Academic Press, San Diego,
pp. 1015–1033.
McNutt, S.R., Jacob, K.H., 1986. Determination of large-scale
velocity structure of the crust and upper mantle in the vicinity of
Pavlof volcano, Alaska. J. Geophys. Res. 91, 5013–5022.
Miller, T.P., McGimsey, R.G., Richter, D.H., Riehle, J.R., Nye, C.J.,
Yount, M.E., Dumoulin, J.A., 1998. Catalog of the historically
active volcanoes of Alaska. U.S. Geological Survey Open-File
Report 98-482, 104 pp.
Minakami, T., 1960. Fundamental research for predicting volcanic
eruptions (I)-earthquakes and crustal deformations originating
from volcanic activities. Bull. Earthq. Res. Inst. Univ. Tokyo 38,
497–544.
Moran, S., 2003. Multiple seismogenic processes for high-
frequency earthquakes at Katmai National Park, Alaska:
evidence from stress tensor inversions of fault-plane solutions.
Bull. Seismol. Soc. Am. 93 (1), 94–108.
Moran, S.C., Stihler, S.D., Power, J.A., 2002. A tectonic earthquake
sequence preceding the April–May 1999 eruption of Shishaldin
Volcano, Alaska. Bull. Volcanol. 64, 520–524.
Morrissey, M.M., Chouet, B.A., 1997. A numerical investigation of
choked flow dynamics and its application to the triggering
mechanism of long-period events at Redoubt Volcano, Alaska. J.
Geophys. Res. 102, 7965–7983.
Neuberg, J., Baptie, B., Luckett, R., Stewart, R., 1998. Results from
the broadband seismic network on Montserrat. Geophys. Res.
Lett. 25 (19), 3661–3664.
Neuberg, J., Luckett, R., Baptie, B., Olsen, K., 2000. Models of
tremor and low-frequency earthquake swarms on Montserrat. J.
Volcanol. Geotherm. Res. 101, 83–104.
Nye, C.J., Keith, T., Eichelberger, J.C., Miller, T.P., McNutt, S.R.,
Moran, S.C., Schneider, D.J., Dehn, J., Schaefer, J.R., 2002. The
1999 eruption of Shishaldin Volcano, Alaska: monitoring a
distant eruption. Bull. Volcanol. 64, 507–519.
Petersen, T., Caplan-Auerbach, J., McNutt, S.R., 2002. Temporal
distribution and rates of repetitive low-frequency earthquakes at
Shishaldin Volcano, Alaska. Eos Trans.-Am. Geophys. Union
83 (47) (Fall Meet. Suppl., Abstract V21A-1173).
Petersen, T., Caplan-Auerbach, J., McNutt, S.R., 2004. The
continuously high level of long-period seismicity at Shishaldin
Volcano, Unimak Island, Alaska, 1999–2004, abstract, IAVCEI
General Assembly, November, 2004.
Ripepe, M., Gordeev, E., 1999. Gas bubble dynamics model for
shallow volcanic tremor at Stromboli. J. Geophys. Res. 104 (5),
10639–10654.
Rowe, C.A., White, R.A., Thurber, C.H., White, R.A., 2003.
Precise, correlation-based seismic event locations at Sourfriere
Hills volcano: insights into magma extrusion behavior through
J. Caplan-Auerbach, T. Petersen / Journal of Volcanology and Geothermal Research 145 (2005) 151–172172
detailed mapping of seismic energy release. Seismol. Res. Lett.
74 (2), 214.
Stephens, C.D., Chouet, B.A., 2001. Evolution of the December 14,
1898 precursory long-period event swarm at Redoubt Volcano,
Alaska. J. Volcanol. Geotherm. Res. 109, 133–148.
Thompson, G., McNutt, S.R., Tytgat, G., 2002. Three distinct
regimes of volcanic tremor associated with the eruption of
Shishaldin Volcano, Alaska 1999. Bull. Volcanol. 64, 535–547.
Vergniolle, S., Brandeis, G., 1994. Origin of the sound generated
by Strombolian explosions. Geophys. Res. Lett. 21 (18),
1959–1962.
Vergniolle, S., Brandeis, G., 1996. Strombolian explosions: 1. A
large bubble breaking at the surface of a lava column as a source
of sound. J. Geophys. Res. 101 (9), 20433–20448.
Vergniolle, S., Brandeis, G., Mareschal, J.-C., 1996. Stromblian
explosions: 2. Eruption dynamics determined from acoustic
measurements. J. Geophys. Res. 101, 20449–20466.
Vergniolle, S., Boichu, M., Caplan-Auerbach, J., 2004. Acoustic
measurements of the 1999 basaltic eruption of Shishaldin
Volcano, Alaska: 1. Origin of Strombolian activity. J. Volcanol.
Geotherm. Res. 137 (1–3), 109–134.
Wessel, P., Smith, W.H.F., 1998. New, improved version of the
Generic Mapping Tools released. EOS Trans.-Am. Geophys.
Union 79, 579.
White, R.A., Miller, A.D., Lynch, L., Power, J., 1998. Observations
of hybrid seismic events at Soufriere Hills volcano, Montserrat:
July 1995 to September 1996. Geophys. Res. Lett. 25 (19),
3657–3660.
Williams, H., McBirney, A.R., 1979. Volcanology. Freeman,
Cooper. 397 pp.
Wohletz, K., 1986. Explosive magma–water interactions: thermo-
dynamics, explosion mechanisms, and field studies. Bull.
Volcanol. 48, 245–264.
Wohletz, K., 2002. Water/magma interaction: some theory and
experiments on peperite formation. J. Volcanol. Geotherm. Res.
114, 19–35.
Wolfe, C.J., Okubo, P.G., Shearer, P.M., 2003. Mantle fault zone
beneath Kilauea volcano, Hawaii. Science 300, 478–480.