Content uploaded by Jacob F Anderson
Author content
All content in this area was uploaded by Jacob F Anderson on Mar 21, 2022
Content may be subject to copyright.
Detecting geyser activity with infrasound
J.B. Johnson
a,
⁎, J.F. Anderson
b
, R.E. Anthony
b
, M. Sciotto
c
a
Department of Geosciences, Boise State University, 1910 University Drive, Boise, ID 83702, United States
b
Department of Earth and Environmental Sciences, New Mexico Tech, 801 Leroy Place, Socorro, NM 87801, United States
c
Dipartimento di Scienze Biologiche, Geologiche e Ambientali, Sezione Scienze della Terra, Università di Catania, Corso Italia 57, 95129 Catania, Italy
abstractarticle info
Article history:
Received 23 September 2012
Accepted 25 February 2013
Available online 13 March 2013
Keywords:
Geyser
Yellowstone
Infrasound monitoring
Array analysis
We monitored geyser activity in the Lower Geyser Basin (LGB) of Yellowstone National Park with dual four-
element microphone arrays separated by ~600 m. The arrays were independently used to identify incident
coherentplane wave energy, thenconjoint cross beam back-azimuthsfrom the two arrays wereused to precisely
locate signal sources. During a week in August 2011 we located repeating infrasound events, peaked in energy
between 1 and 10 Hz, originating from at least five independent geothermal features, including the episodically
erupting Great Fountain, Fountain and Kaleidoscope Geysers, as well as periodic infrasound from nearby Botryoidal
and persistent sound from Firehole Spring. Although activity from nearby cone-type geysers was not detected in
theinfrasoundbandupthrough50Hz,themajorfountain-type geysers (i.e., with columns greater than 10 m)
could be detected at several kilometers, and two minor geysers (i.e., a few meters in eruption height) could be
tracked at distances up to a few hundred meters. Detection of geyser activity was especially comprehensive at
night when ambient noise was low. We conclude that infrasound monitoring of fountain-type geysers permits
convenient tracking of geyser activity, episodicity, signal duration, energy content, and spectral content. These
parameters enable objective statistical quantification of geyser behavior and changes over time that may be due
to external forcing. Infrasonic study of geyser activity in an individual basin has great monitoring utility and can
be reasonably accomplished with two or more distributed sensor arrays.
© 2013 Elsevier B.V. All rights reserved.
1. Introduction
Geyser sound and volcano sound generation may be considered
analogous in a number of respects. In both systems, volatiles can
reach a liquid's free surface (water in the case of the geyser; silicate
melt in the case of the volcano) and burst with considerable overpres-
sure relative to the atmosphere. In volcanic systems both the distension
of the free surface due to sub-surface strains (Garces and McNutt, 1997;
Yokoo and Iguchi, 2010), and the expansion of gas following fragmenta-
tion (Ripepe and Gordeev, 1999; Jones et al., 2008), have been considered
as volumetric sources, which produce intense low-frequency sounds.
High-velocity emissions of gas and/or condensed phases are also respon-
sible for jetting sounds at volcanoes (Woulff and McGetchin, 1976;
Matoza et al., 2009), which may serve as analogs for certain geysers
that erupt as collimated jets of water and steam.
The style and vigor of a volcanic eruption generally dictate the
spectral content and intensity of the radiated sound. For relatively
low-energy explosive volcanic eruptions, often characterized as
strombolian or vulcanian, the radiated sound is most intense around
the near-infrasound band (and specifically in the frequency range of
a few seconds to a few Hz) (Johnson et al., 2004; Marchetti et al.,
2009). These low frequencies predominate because of the relatively
large physical dimension and long duration of source movements,
such as bubble oscillations or gas expansion (Vergniolle and Brandeis,
1996; Gerst et al., in review). Geysers, though smaller in physical scale
than volcanoes, are still capable of producing relatively large volume
fluid ejections with columns as wide as a few meters and as high as a
few tens of meters. Accordingly, fountain-type geysers radiate predom-
inantly low frequency acoustic energy in the near-infrasound band
(1–20 Hz).
Geophysical sources of infrasound, including volcanoes, earthquakes,
avalanches, thunder, bolides, and storms, are amenable to remote moni-
toring and tracking in large part because infrasonic frequencies attenuate
slowly with distance (Arrowsmith et al., 2010); however, geophysical
infrasound detection and interpretation are often obscured by unwanted
signals (e.g., human activity or microbaroms) or noise contributions from
atmospheric winds (Bowman et al., 2005; Fee and Garces, 2007). In order
to distinguish targeted signals from noise, microphone arrays are typical-
ly deployed to identify signal coherency and source direction (Rost and
Thomas, 2002). Toward the goal of locating and tracking geyser activity
at Yellowstone, we deployed two separated infrasound microphone
arrays in August 2011.
Although various seismic surveys have been carried out at geysers
to study ground-propagating elastic waves (e.g., Kieffer, 1984; Kedar
Journal of Volcanology and Geothermal Research 256 (2013) 105–117
⁎Corresponding author. Tel.: +1 208 426 2959; fax: + 1 208 426 4061.
E-mail addresses: jeffreybjohnson@boisestate.edu (J.B. Johnson), jfanders@nmt.edu
(J.F. Anderson), ranthony@nmt.edu (R.E. Anthony), mariangela.sciotto@ct.ingv.it
(M. Sciotto).
0377-0273/$ –see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jvolgeores.2013.02.016
Contents lists available at SciVerse ScienceDirect
Journal of Volcanology and Geothermal Research
journal homepage: www.elsevier.com/locate/jvolgeores
et al., 1996) this work is the first of its kind to investigate broadband
sound waves radiated from geysers into the atmosphere.
2. Background
Yellowstone National Park in Wyoming, USA hosts the world's
densest concentration of geysers with about 500 active in a typical
year, or more than half the world's total. Most of Yellowstone's geysers
are located in three basins, Upper Geyser Basin (UGB), Lower Geyser
Basin (LGB), and Norris Basin, which are extensive geographic regions
that comprise distinct groups of thermal features. For instance, the LGB,
which is the focus of this study, is 13 km
2
in area and has more than
1500 thermal features organized into about 13 distinct groups (Bryan,
2008). Classification as a geyser requires that a thermal feature exhibit
intermittent discharge of water accompanied by steam. According to
Bryan (2008) there are well over one hundred features that qualify as
geysers in the LGB alone.
Because we anticipated that violent ejection of steam and water is
most likely to generate high signal-to-noise infrasound, we deployed
our microphone arrays within a few hundred meters of Great Fountain,
one the most prominent geysers of the LGB. Though Great Fountain
Geyser is located near the eastern edge of the LGB, we still anticipated
recording geyser activity from other nearby features. Table 1 provides
a list of selected LGB geysers, where plume height in excess of a few me-
ters is often reported (e.g., Bryan, 2008). A map showing these geysers
and our microphone arrays is provided in Fig. 1. Despite having fewer
major geysers than the UGB, the LGB provided an excellent test bed
for acoustic monitoring because of lower tourist traffic and associated
cultural noise.
3. Experiment
We deployed two four-element infrasound arrays in the LGB between
Aug. 8th (Julian Day 220) and Aug. 14th (Julian Day 226) of 2011. These
arrays consisted of four identi cal low-frequency microphones with flat re-
sponse between 0.02 Hz and a Nyquist frequency of 50 Hz. Linear
dynamic range of the instruments was +/−125 Pa and noise floor in
the 1 to 10 Hz band was ~2 mPa rms (Marcillo et al., 2012). Three of
the array elements were positioned at the vertices of an approximate
equilateral triangle and connected to the central datalogger by 30-m
cables. A fourth microphone was co-located at the center of the array
next to a 6-channel, 24-bit logger (Refraction Technol ogy RT-130) record-
ing continuously at 100 Hz. GPS timing of the loggers allowed coordina-
tion between the two arrays, and kinematic GPS surveying provided
sensor node locations accurate to within ~ 0.5 m in the horizontal and
~1 m in the vertical.
The array centers were separated from each other by 620 m. The
midpoint of the two arrays, or network center, was located at 110.802°
W, 44.537° N, and 2237 m above sea level, and is used as the coordinate
reference for mapped acoustic sources. The purpose of dual arrays was
to identify and locate sources producing coherent signals. We identify
source locations by first using each four-element array to independently
determine back-azimuth of coherent infrasound. Then we find the inter-
section region of the back-azimuth beams to identify the responsible
geyser. Owing to the distribution of the two arrays, location resolution
and errors are azimuthally and radially variable. We discuss location
uncertainties as part of our study's ‘network response’.The‘array re-
sponse’, a function of array geometry, is also examined as it influences
aliasing and back-azimuth uncertainty.
3.1. Array response and precision
The array response of a distribution of sensors characterizes the sus-
ceptibility of an array to aliasing. Such aliasing is problematic for arrays
with apertures that are large relative to incident plane wave wave-
lengths and is especially pronounced in four-element arrays with equal
spacing between sensor nodes (Christie and Campus, 2010). The normal-
ized theoretical wavenumber response of an n-element array is a func-
tion of 2-D wavenumber (k
x
and k
y
)(Rost and Thomas, 2002):
Rk
x;ky
¼1
n2X
n
i¼1
e−ffiffiffiffiffiffi
−1
pkxxiþkyyi
ðÞ
2
1
The array output is the convolution of the array response and the
horizontal wavefield defined by a propagation vector. An ideal array
response has a single peak at the origin (k
x
= 0 and k
y
=0)andnegli-
gible side lobe peaks.
Array responses with significant sidelobes (see Fig. 2b,e for the West
and East arrays respectively) are susceptible to possible aliasing. To
illustrate the potential ambiguity associated with an ~5 Hz infrasound
tone suppose that a recording on channel #1 of the West Array exhibits
a phase shift of half a cycle relative to channels #2–4. In the absence of
other information these observations could be attributed either to
horizontally propagating acoustic energy coming from either the
WNW or the ESE, corresponding to two different array response peaks.
Forourlocalgeysersourcesweassume that propagation must be sub-
horizontal (i.e., f¼c=2πðÞ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k2
xþk2
y
q;Fig. 2). This, coupled with the facts
that geyser infrasound is generally broadband (with frequencies less
Table 1
Location list of microphone arrays and various geysers, geyser groups, and geyser basins near LGB: Names and abbreviations used in figures, latitude, longitude, elevation, distance
from the center of the microphone network, geyser eruption repeat interval, duration of geyser activity, typical height of play, and type of geyser indicated as either (F)ountain, (C)
one, or (V)arious. Details are taken from Bryan (2008).
Name Latitude (degrees N) Longitude (degrees W) Elev. (m) Dist. (m) Interval Duration (minutes) Height (m) Type
Network Center (x) 44.536614 110.801662 2231 0
East Array (EA) 44.536765 110.797793 2238 309
West Array (WA) 44.536462 110.805531 2229 309
Great Fountain (GF) 44.536578 110.800026 2234 130 9–15 h 30–120 23–67 F
Firehole Spring (FS) 44.535141 110.801949 2235 165 Continuous Continuous b2F
Botryoidal Spring (BS) 44.534882 110.799529 2238 265 3–5 min 1 3 F
White Dome (WD) 44.539394 110.802823 2228 323 15–180 min 2 6–9C
Pink Cone (PC/PCG) 44.542893 110.796273 2235 819 18–25 h 90–120 b9C
Bead (B/PCG) 44.543418 110.794972 2239 924 27–33 min 2.5 7–8C
Narcissus (N/PCG) 44.544322 110.797004 2235 933 2–6h b15 4–6F
Labial (B/PCG) 44.543751 110.795304 2240 940 5–7h b2b8F
Steady (S/BWG) 44.544198 110.786705 2247 1455 Continuous Continuous b4F
Artesia (A/BWG) 44.544075 110.784056 2253 1623 Continuous Continuous b3F
Fountain (F/FG) 44.551205 110.808326 2228 1705 4–15 h 30 ~25 F
Kaleidescope Group (various/KG) 44.554275 110.813347 2212 ~2200 Various Various b45 V
Middle Geyser Basin (various/MGB) 44.525055 110.838148 2218 ~3200 Various Various b5V
Upper Geyser Basin (various/UGB) 44.466665 110.836993 2238 ~ 8700 Various Various 60 V
106 J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
Fig. 1. Detail map featuring LGB study area. Its position relative to the Middle Geyser Basin (MGB) and Upper Geyser Basin (UGB) is given by the red rectangle in the locator map.
Locations are shown for Great Fountain (GF), Firehole Spring (FS), Botryoidal Spring (BS), White Dome (WD), the Pink Cone Group (PCG) including Narcissus (N), Pink Cone (PC),
Bead (B), and Labial (L), the Black Warrior Group (BWG) including Steady (S) and Artesia (A), the Fountain Group (FG) featuring Fountain (F), and the Kaleidescope Group (KG).
Details of these geysers are summarized in Table 1. West Array (WA) and East Array (EA) microphone sites are shown as blue triangles along with the network center indicated by
crosshairs. Array geometry detail is shown in Fig. 2.
Fig. 2. a,d) Detail plan view maps of West and East microphone arrays. Parenthetical coordinates are the east-west and north-south array center location relative to the center of the
network and map origin. b,e) Corresponding array responses calculated according to Eq. (1). Contours indicate wavenumbers for 5, 10 and 15 Hz horizontal acoustic plane waves. c,
f) Histograms of angular uncertainties in calculated back-azimuth for data digitally discretized to 0.01 s.
107J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
than 5 Hz) and often has transient pulses, limits our arrays' susceptibility
to aliasing.
Array back-azimuth precision is limited by small array dimension
and/or coarse timing resolution for correlated phases crossing the array
elements. For our digital data the precision of cross-correlation lag
times is discretized to the nearest sample, which is 0.01 s in our analysis.
Subsequent back-azimuth determination (see source localization section
below) is calculated by inverting these rounded phase lag times. To
anticipate the associated error due to time discretization we calculate
time of arrivals for incident rays crossing the arrays at a range of
azimuths and then round these arrival times to the nearest 0.01 s before
inverting for an inferred back-azimuth. For 360 different plane waves
crossing the arrays at 1° azimuthal increments the standard deviation dif-
ference between actual and calculated azimuths are 1.9 and 1.6 degrees
for the West and East arrays respectively (Fig. 2c,f).
3.2. Network response
Our two arrays separated by 620 m are used to locate infrasound
sources when source back-azimuths cross obliquely. The compass
azimuth (relative to true North, or 0°) connects the West array to
the East array at 87° and the azimuth connecting East to West array
is −93° (or 267°). As such, back-azimuth beams cross for
θW>θEwhen 93bθWb87 and 93bθEb87
or
θWbθEwhen 87bθWb267 and 87bθEb267 ð2Þ
where θ
W
and θ
E
are the compass bearing back-azimuths from the
West and East arrays to the source. Overlapping back-azimuth direc-
tions are indicated as colored regions in Fig. 3, which also show the
corresponding distance and azimuth to the crossing beams (Fig. 3a,b).
These parameters are determined by computing the locations of con-
verging beams (i.e., the inferred source location) for all possible permu-
tations of θ
W
and θ
E
(ranging from −93 to 267°).
Errors in source location distance (Fig. 3c) are calculated as the mag-
nitude of the gradient of Fig. 3a. At a distance rthe distance error per
degree of back-azimuth uncertainty is defined as:
εr¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
∂r
∂θW
2
þ∂r
∂θE
2
sð3Þ
Fig. 3. a) Distance (r) to cross-beam sources for conjointly computed back-azimuths from West and East arrays. b) Compass bearing (θ) to inferred sources. c) Radial error (ε
r
) per
degree of back-azimuth uncertainty. d) Azimuthal error (ε
θ
) per degree of back-azimuth uncertainty. All distance and source azimuths are relative to the network center (crosshairs
in Fig. 1). Blank regions correspond to non-converging back-azimuths. Expected bearing to those geysers and groups indicated in Fig. 1 and Table 1 are shown as white circles with
names annotated in panel c.
108 J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
For instance, the 100-m/° contour in Fig. 3c implies that the distance
to a source (e.g., the Fountain Group (FG)) is uncertain to ~100 m for a
back-azimuth uncertainty of one degree. An azimuthal error (Fig. 3d) is
transverse to the radial error and is computed from the azimuth to the
source (Fig. 3b) as:
εθ¼rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
∂θ
∂θW
2
þ∂θ
∂θE
2
sð4Þ
Generally, radial uncertainties are much larger than azimuthal un-
certainties and both uncertainties increase for greater source-receiver
distances.
4. Source localization
Our procedure to locate robust infrasound sources using dual arrays
involves identification of coherent energy arriving coincidentally at both
arrays. Coherency at a single array is established if timing of phase lags,
determined through cross-correlation of the array elements, is internally
consistent. If so, potential source back-azimuths may be calculated. For
coherent energy that traverses both arrays simultaneously a candidate
source is mapped as the intersection of two back-azimuths. This poten-
tial source is reliable if its position is in agreement with the phase delay
observed between beam-stacked waveforms at the WA and EA.
4.1. Back-azimuth determination
Phase lags between two elements of a microphone array are
determined through cross-correlation of pairs of sensors. For melements
in an array there are 1=2ðm2mÞunique sensor pair combinations
that can be cross-correlated. For a cross-correlation to be considered
significant it must exceed a normalized cross-correlation threshold, which
we fix in this study at the 95% confidence level for cross-correlated white
noise. For nsamples 2=ffiffiffi
n
pis the expected normalized cross-correlation
for Gaussian white noise. For our 4 node arrays and 20 s (2000 sample)
comparison windows a normalized cross-correlation threshold of 0.045
must be exceeded on all 6 station pairs. More stringent cross correlation
thresholds should probably be applied for three element arrays, which
have only 2 unique station pair comparisons.
In addition to correlation threshold, strict consistency criteria must
be met. Lag times of peak cross-correlation are calculated for sliding
windows and checked for internal consistency similar to that used in
the PMCC technique (Cansi, 1995). While Cansi (1995) searches for
consistency among unique triad pairs, our processing requires consis-
tency among all unique quad pairs. For our four-element array there
are 3 unique sequences of quad pair comparisons: ch1↔ch2↔ch3↔
ch4↔ch1, ch1↔ch3↔ch2↔ch4↔ch1, and ch1↔ch2↔ch4↔ch3↔
ch1. Internal consistency is met when the summed phase lags of the
quad pairs sum toward zero, i.e. |ε
ijkl
t
ij
+ε
ijkl
t
jk
+ε
ijkl
t
jk
+ε
ijkl
t
li
|≤χ.
Here the indices i,j,k, and lrefer to one of the 4 sensor array channels.
The variable t
ij
is the lag time associated with peak waveform cross-
correlation and ε
ijkl
is the Levi-Civita symbol, where only non-repeating
index permutations are non-zero, + 1 or −1, and sign is dependent
upon the order of indices. Because of digital signal discretization, which
rounds correlation phase lags to the nearest sample, we require the abso-
lute value of consistency to be less than or equal to χ= 4 samples.
Consistent phase lags for unique quad sequences are used to com-
pute a back-azimuth by inverting for the horizontal projection of the
slowness vector s
→¼sx;sy
. Following the inversion procedure
outlined in Arechiga et al. (2011) time lags are related to the slowness
vector by
tij
tjk
tkl
tli
2
6
6
43
7
7
5¼
dxij dyij
dxjk dyjk
dxkl dykl
dxli dxli
2
6
6
43
7
7
5
sx
sy
ð5Þ
where dx and dy are the GPS surveyed east-west and north-south sep-
aration distances between pairs of sensor elements in an individual
array. The distance matrix, denoted as D,canberepresentedasa
two-column matrix because the vertical separation distance is assumed
zero (i.e., dz
ij
=dz
jk
=dz
kl
=dz
li
= 0) as all sensor nodes were
deployed on an approximately level surface to within ~1 m precision.
Because the solution to the slowness vector for t=Ds is overdeter-
mined we solve it by using a least squares solution with the generalized
inverse of D,whereD
−g
=(D
T
D)
−1
D
T
and the slowness vector is solved
as s=D
−g
t. A third (vertical) component of the slowness vector can be
computed assuming that the coherent arrival is an acoustic plane wave
with speed c,wheresz¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
c−2−s2
x−s2
y
q.
Imaginary values of s
z
imply impossibly low slowness values for
acoustic waves traversing the array, however near-horizontal acoustic
waves may potentially result in imaginary vertical slowness values
due to cross-correlation timing discretization, which leads to rounded
values of tand values of s
x
and s
y
, which may be rounded upwards.
For this reason we consider that horizontal slownesses, which exceed
the slowness amplitude (c
−1
) by less than 10%, may be treated as hor-
izontally propagating acoustic waves with zero degree elevation angles
(i.e., s
z
=0). We use the following conventions to calculate vertical
slowness:
sz¼imaginary for c−1b0:9ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s2
xþs2
y
q
sz¼0 for 0:9ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s2
xþs2
y
q≤c−1≤ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s2
xþs2
y
q
sz¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
c−2−s2
x−s2
y
qfor ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s2
xþs2
y
qbc−1ð6Þ
When s
z
is imaginary we consider the arrival to be spurious.
From the acoustic wave slowness vector the back-azimuth and inci-
dence are determined. Azimuth of the plane wave is calculated using
the trigonometric relations:
θ¼arctan sx=sy
for sy>0
θ¼arctan sx=sy
þ180for syb0ð7Þ
while plane wave elevation angle, as measured from the horizontal, is
ϕ¼arcsin cs
z
ðÞ ð8Þ
In this analysis of LGB local geyser sources propagation is expected
to be sub-horizontal. Thus, we ignore signals with values of ϕgreater
than 15°. We note that more steeply incident acoustic energy observed
during our study is often moving and attributable to aircraft.
Back-azimuths for internally consistent array detections are inde-
pendently calculated for the three unique permutations of sensor pair
correlations, i.e. ijkl = {1234,1243,1324} and then averaged. These
back-azimuths may then be plotted as a function of time to show the
temporal evolution of potential acoustic sourcedirections. The example
of Fig. 4 shows a one-hour period (starting August 10th at 10:00 PM
local time) when three distinct geyser sources were detected.
4.2. Cross beam source localization and validation
Together the back-azimuths from the West and East arrays are
used to locate potential geyser sources. Back-azimuth beams from
the two arrays converge under the conditions specified in Eq. (2).
Cross beam intersection then occurs at a location x
0
,y
0
where
x0¼xWþsinθW
yW−yE
ðÞ
sinθE−xW−xE
ðÞ
cosθE
sinθWcosθE−cosθWsinθE
and
y0¼yWþcosθW
yW−yE
ðÞsinθE−xW−xE
ðÞcosθE
sinθWcosθE−cosθWsinθE
ð9Þ
109J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
Here x
W
,y
W
and x
E
,y
E
correspond to the UTM coordinates of the
West and East arrays respectively.
A candidate source location is identified for converging beams when
coherent energy is conjointly identified on both arrays (i.e., during the
same 20 s sliding window period). In this case beam waveform stacks
(see Eq. (10) below) are produced for each array and a cross-correlation
time lag is calculated for the two beams. These inter-network lag times
indicate potential source locations lying along hyperbolic curves (Fig. 5).
If the hyperbolic curve for a given lag time coincides with the cross-
beam intersection locus x
0
,y
0
then we consider that source location to
be robust.
Source locations are plotted with footprints that scale with back-
azimuth uncertainty. An azimuthal uncertainty for each array is deter-
mined as the 95% confidence intervals for estimated errors (3.6° for
West Array and 2.5° for East Array; Fig. 2c,f). Error ellipses in Fig. 5 are
centered on the intersection of back-azimuths and have axes with di-
mensions of angular and radial uncertainties. It is evident that location
uncertainty increases markedly for more distant sources as predicted
by the network response (Fig. 3).Forinstance,fortheFountainGeyser
source, radial distance error is as great as half a kilometer. Locations of
geysers and other infrasound sources are shown in an animation that
is provided as auxiliary materials. This movie shows a 5-day sequence
of mapped sources, in the form of Fig. 5, for hourly time increments.
5. Results
5.1. Interpretation of beam stacks
Reliable source back-azimuths can be used to produce array beam
stacks δp
b
(t), which provide improved signal-to-noise over wave-
forms from individual channels. To create a beam stack the excess pres-
sure waveforms in an individual array δp
i
(t) are shifted by retardation
times corresponding to relative locations and incident slowness vector
and then stacked (Fig. 6):
δpbtðÞ¼1
mX
m
i¼1
δpitþdxijsxþdyij sy
ð10Þ
Fig. 4. (Upper left) Example 1-hour time series for West Array infrasound recording (filtered above 0.25 Hz) and calculated coherent back-azimuths for 20-s windows sliding at 1-s
increments. (Lower right) corresponding time series of East Array infrasound and calculated back-azimuths. (Upper right) map of conjoint back-azimuths from the two arrays.
Filled circles indicate conjoint sources with associated converging back-azimuths. Back-azimuths for various geyser basins, geyser groups, or specific geysers that are indicated
in Fig. 1 and Table 1 are shown and described in lower left text panel. Data shown are from a one-hour period starting August 10th at 10:00 PM local time (Julian Day 223 at
04:00). Corresponding detection source locations and example waveforms for this hour are shown in Figs. 5 and 6.
110 J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
In our study we calculate a center node beam array stack where jis
channel 1.
Similarity between the beam stack waveforms of the two arrays is
variable and depends upon signal strength, background noise, and
frequency band. For the four time windows displayed in Fig. 6 the signal
correlation is indicated for both broadband infrasound and four narrow-
band overlapping frequencies. Signal similarity and relative delay times
are quantified from the peak normalized cross-correlation of band-
passed waveforms. Higher signal-to-noise waveforms, such as the ones
displayed in Fig. 6a–b, are more highly correlated than smaller tran-
sients, such as those shown in Fig. 6c–d.
For the featured data in Fig. 6 cross-network correlation is generally
greatest in near-infrasoundand low audio band (1–32 Hz) although this
varies somewhat depending upon the particular source. For example,
Fig. 6a corresponds to infrasound originating from the Northeast and
external to the LGB. For this event peak signal correlation between
arrays occurs in the band 0.25 to 2 Hz and candidate source types
could include earthquakes, bolides, thunder, or cultural signal (such as
aircraft or explosions) (Arrowsmith et al., 2010). In this particular case,
we feel that the most probable signal source is distant thunder owing
to the signal shape and amplitude, intermittency (many events from
this direction occurring over tens of minutes), and spectral content sim-
ilar to that previously observed for thunder (e.g., Assink et al., 2008;
Arechiga et al., 2011).
Geyser sources including Fountain Geyser (Fig. 6b), Botryoidal Spring
(Fig. 6c), and Firehole Spring (Fig. 6d) are also identified during the hour
starting at 10:00 PM local time on August 10th. Fountain Geyser and
Botryoidal Spring signal correlation is greatest in the 0.25–2Hzbands
while Firehole Spring is best identified in the 1–8 Hz band. Correlation
lag times are consistent with sources at Fountain Geyser, Botryoidal
Spring, and Firehole Spring and corroborate cross beam locations of
the geyser sound sources. Notably, the low-amplitude correlated signal
from both Firehole and Botryoidal Spring is not clearly evident through
visible inspection of the time series data. For these geysers relatively
high levels of ambient infrasound noise are indicated by similarities
between the spectra for the events and pre-event noise windows
(Fig. 8c). Filtering above ~ 0.25 Hz coupled with array and/or network
analysis is thus vital to identify and track activity from ‘quieter’geysers.
5.2. Geyser detection
Dual array cross beaming and validation through inter-array lag time
delays enable robust identification of geyser and/or other signals. If the
source coincides with a known geyser feature, e.g. referenced in Bryan
(2008), we consider it to be a geyser signal. During the week-long mon-
itoring interval in August 2011 we identified at least five repeating
geyser sources and potential activity from several others. In general,
geyser detection was affected by levels of wind, which contribute to am-
bient noise throughout the near infrasound band. Obfuscation of geyser
signal in the LGB wasparticularly pronounced during windy afternoons,
however nighttime recordings had much improved signal-to-noise
(refer to summary of 5-day record in Fig. 7). The five primary identified
LGB sources were Great Fountain (130 m), Firehole Spring (165 m),
Botryoidal Spring (265 m), Fountain Geyser (1706 m), and at least one
source from Kaleidoscope Group (2171 m). Descriptions of their activi-
ty, including episodicity and eruption duration, are given below.
Fig. 5. Map of infrasound sources occurring during the-one hour period shown in Fig. 4. Locations of candidate geysers (from Table 1 and Fig. 1) are marked by red circles while red
arrows indicate direction to geyser basins located off the map. Map origin, indicated by crosshair, is the center of the two arrays indicated by blue triangles. Contours indicate
expected time lag delays between the East Array and West Array. Ellipses designate those conjoint back-azimuth intersections, which have been validated by inter-network lag
time delays and for which incidence is nearly horizontal (i.e., elevation angles less than 15°). Sources located off map are indicated with yellow arrows. Numbered source epicenters
correspond to featured events shown in Fig. 6. Events #2–4 correspond to geyser activity.
111J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
5.2.1. Great fountain
Ten eruptions of Great Fountain were detected during the 5-day
study interval shown in Fig. 7. Activity of main events is separated by
11 to 19-hour intervals and duration of detected events ranges from
45 to 75 min with one event lasting 130 min. This long-duration event
precedes an exceptionally long 19-hour quiescent interval suggesting
that more voluminous eruptions may require longer recharge intervals.
Great Fountain infrasound records corroborate anecdotal observations
that most events are composed of 4 or more pulses of 5 to 20-minute du-
ration separated by up to 15 min of quiet (Fig. 7). A detailed example of
a typical event from Great Fountain, along with its normalized power
spectrum, is provided in Fig. 8a.
Characteristic event durations and intervals between events can easily
be quantified from the overview records of Figs. 7 and 9a, which both
show detections as a function of time. For our 5-day monitoring interval
the eruption durations Dappear correlated with inter-eruption intervals
I,i.e.I=4.9×D+7.1hours (r-squared value of 0.73). Despite our
short observation period these observations exhibit similarities with
previous eyewitness observations cited in Bryan (2008) where I=
7×D+6hours(Fig. 9b).
Due to the relative intensity of the Great Fountain infrasound source
and its proximity to both infrasound arrays the signal from this geyser
is often identifiable through visual inspection of the time series after fil-
tering above the microbarom band. Peak-to-peak amplitude of the largest
pulses from a Great Fountain sequence occasionally exceed 2 Pa recorded
at the East Array (276 m distant). These peak-amplitude pulses are bipo-
lar ~ 2 Hz wavelets, with asymmetrically larger compression than rarefac-
tion. They accompany explosive bursts from the geyser's vent, which
manifest voluminous vapor and water columns reaching 30 to 50 m
high. Smaller explosive bursts, as seen in video records, are correlated
with less intense infrasound pulses. Impulsivity, frequency content, and
bimodal pulse shape are reminiscent of explosion infrasound N-wave sig-
nals accompanying explosions of pressurized gas at many erupting volca-
noes (Johnson and Ripepe, 2011).
Band-limited acoustic power radiated from a monopole into a homo-
geneous hemisphere may be quantified from filtered infrasound record-
ings according to Dowling (1998):
PtðÞ¼2πr2
ρc∫
tþΔt
t
δp2τþr=cðÞ
Δtdτð11Þ
where Δtis the averaging interval over which power is calculated,
sethereat1s,andρis the atmospheric density, approximated as
1.0 kg/m
3
. Cumulative energy can then be calculated as the time in-
tegrated acoustic power:
EtðÞ¼2πr2
ρc∫t
0δp2τþr=cðÞdτð12Þ
For the most intense Great Fountain pulses, power can exceed
300 W averaged over a 1-s interval (Fig. 8a). Cumulative energy over
the course of an hour-long event (t= 3600 s) is several thousand
Joules, or slightly less than 1 W averaged over a Great Fountain event.
For some Great Fountain events the infrasound network identifies
potential short (minute-long) precursors occurring an hour to several
hours prior to the main eruption. Geyser observers at Great Fountain
have commonly reported this activity as ‘pre-play’that accompanies
boiling and overflow from the vent occurring on average 85 min prior
to the main event (Bryan, 2008). Of the ten events detected in our
infrasound records half of them show these precursory infrasound
signals occurring prior to the main event (see indicated red arrows in
Fig. 9a). Infrasound monitoring has the potential to identify pre-play
and quantify how long and how often it precedes Great Fountain erup-
tions. As with statistics relating event duration and quiescent intervals
we suggest that longer monitoring periods will facilitate robust statisti-
cal relationships.
5.2.2. Fountain geyser and Kaleidoscope Group
Fountain Geyser, not to be confused with Great Fountain, is the major
geyser of the regularly performing features in the Fountain Group, located
about 1700 m from the center of our twin infrasound arrays. During
nighttime periods of relative low wind and quiet (9:00 PM through
10:00 AM) events from Fountain were routinely recorded. During our
five-day observation period 13 events were identified and 11 proba-
ble events were missed due most likely to ambient wind noise. Non-
detected events were inferred from the extrapolation of the excep-
tional regularity of Fountain Geyser eruptions.
From our infrasound records we detect eruptions with regular in-
tervals of 5.9 +/−0.3 h that appear to correlate with ‘long mode in-
tervals’discussed by Bryan (2008) for Fountain Geyser. Infrasonic
waveforms from Fountain Geyser events were also remarkably similar,
beginning and terminating abruptly, with detections lasting between
27 and 36 min (for 11 events). Signal envelope was substantially differ-
ent from that of Great Fountain with less intense pulses, but more
sustained amplitude reminiscent of a stationary volcanic broadband
tremor (Fig. 8b). At the distance of the East Array peak-to-peak tremor
amplitudes occasionally exceeded 0.1 Pa, which would reduce to ~ 1 Pa
at 100 m invoking a 1/rpressure decay for a homogeneous atmosphere.
The nature of the infrasound is suggestive of the descriptions of typical
Fountain activity, which reportedly begins abruptly and then plays in a
sustained fashion with splashing and a wide column up to 15 m in height
(Bryan, 2008). Cumulative infrasound power radiated from Fountain to-
tals several thousand Joules and is comparable to infrasound from Great
Fountain events.
Sporadic infrasound originating from the vicinity of Fountain
Group, but with a slightly more westerly back-azimuth (335°), was
intermittently recorded during our week-long survey. Although this
back-azimuth is close to that of Fountain Geyser (342°) the location
ellipsoids from this source are spatially distinct from Fountain Geyser
and we conclude that they represent a separate source occurring at
slightly greater distant range (~ 2000 m) than the Fountain Group
(~1700 m). We speculate this is Kaleidoscope Group geyser activity
that is characterized by infrasound with a few bursts that last just a
few minutes (e.g., on Julian Day 223 at 07:20 UTC). Based upon the
infrasound character the most likely candidate geyser source is the
namesake Kaleidoscope Geyser, which hosts short-duration activity
(20 to 120 s) that suddenly shoots water jets 15 to 35 m (Bryan,
2008).
5.2.3. Botryoidal and Firehole Springs
Infrasound radiation from Botryoidal Spring is routinely identified
during periods of low background noise, i.e. when other ‘louder’geysers
are quiet and when wind-induced noise is low. During these conditions
the activity from Botryoidal, which is 338 m from the East Array, is
periodic. During our observational period infrasound bursts occurred
with remarkable regularity at intervals of 4.5 +/−0.5 min (Figs. 8c
and 10). Interval times between successive eventdetections are consis-
tent with a normal distribution (Fig. 10).
Transientsignal amplitudes from Botryoidal are invariably small and
short in duration, typically only 0.05 Pa and composed of only one or a
couple of 2.5 Hz oscillations. This spectral content is somewhat higher
Fig. 6. Detail beamstack waveforms filtered into five frequency bands using 2-pole Butterworth filters with the indicated cornerfrequencies. Peak-to-peak signal amplitudes, normalized
correlation coefficients, and associatedpeak correlationlag times are shown for each waveformand each band. Featuredevents correspond to the bestcorrelated signalsoccurring in Fig. 4
from four representative source regions including: a) probable distant thunder source(s) to the Northeast of the LGB, b) Fountain Geyser ~1700 m, c) Botryoidal Spring ~250 m, and d)
Firehole Spring ~150 m from the network center.
112 J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
than the peak acoustic frequencies of the larger Great Fountain and
Fountain Geysers, which were both peaked in the 0.5 to 1.5 Hz band
(Fig. 8a, b). In general the infrasound observations of Botryoidal are con-
sistent with anecdotal reports. In recent years reported periodicity has
ranged between 2.5 and 5.5 min. Eruptions consist of single steam bub-
bles distending the surface of the spring before bursting and throwing
water to heights of 4 to 6 m (Bryan, 2008).
Nearby Firehole Spring is another geothermal feature, which pro-
duces prodigious (but even lower-amplitude) infrasound. Owing to
its near-continuous activity Firehole Spring is sometimes considered
a‘perpetual spouter’with play typically reaching only a few meters
in height (Bryan, 2008). Its corresponding infrasound is manifested
as a persistent infrasonic tremor of such low intensity that it is gener-
ally not detected except during the most quiet of intervals, when it is
registered as a quasi-continuous source. At a source–receiver dis-
tance of ~ 350 m Firehole Spring infrasound is not visually apparent
in time series records and it is at the limits of signal detection using
array analysis techniques. The spectral content of Firehole Spring
Fig. 7. Infrasound detections from a 5-day interval in August 2011. Featured geysers and their detected activity include Great Fountain (GF; blue), Firehole Spring (FS; green),
Botyroidal Spring (BS; red), Fountain and Kaleidescope (FG/KG; cyan), and other sources (other; mauve). Each detection corresponds to coherent energy identified on the EA.
Gray records correspond to 20-s averaged absolute signal amplitudes, analogous to real-time seismic amplitude measurements (Murray and Endo, 1992).
114 J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
activity is peaked notably higher (> 4 Hz) than the other frequently
detected geysers.
6. Discussion
We recorded infrasound radiation associated with activity from at
least five fountain-type geysers. Fountain-type geysers erupt steam and
water from open pools and in the process they accelerate large volumes
of the overriding atmosphere, efficiently generating low-frequency
acoustic waves. These acoustic waves are dominated by 1–8Hzinfra-
sound most likely because the time scales of surface accelerations occur
during tenths of seconds. Corresponding wavelengths of 1–8Hzinfra-
sound is 40 m or longer, much larger than the vent dimensi on of the stud-
ied geysers. As such, sound generation may reasonably be considered as a
compact, point-source volumetric signal, or a monopole. These sounds
carry efficiently for hundreds of meters to several kilometers.
We did not observe any definitive infrasound signal produced by
nearby cone-type geysers. Cone-type geysers are generally erupted as
collimated jets of steam and water from a narrow orifice (~ 0.1 m) that
is often located at the summit of a mound of sinter (or geyserite). The
nearest cone-type geyser to our dual arrays was White Dome, a regular
performer located only 323 m from the center of the microphone
network. Although White Dome is frequently active with 9 min to
hour-long quiescent intervals and produces a lofty jet up to 10 m it was
not detected by our infrasound surveillance. Unsurprisingly, Pink Cone,
another similar-sized cone-type geyser located farther away (940 m),
was never definitively detected.
We speculate that cone-type geysers do not produce significant
amounts of infrasound because their volumetric, or monopole, contribu-
tions are small. Instead th ey erupt multi-phase fluid jets, which are often
modeled as dipole or quadrupole sources (Woulff and McGetchin, 1976;
Lighthill, 1978; Matoza et al., 2009), and are much less efficient at
ensonifying the atmosphere, especially in the infrasonic band. White
Dome's jet is narrow and fairly low-energy. Larger cone-type geysers,
such as Old Faithful and Lone Star, are more energetic and more likely
to produce intense sounds. Short-duration infrasound surveys of these
geysers during our August 2011 experiment indeed revealed lower
infrasound spectral power and enhanced higher frequency sound, com-
pared to fountain geysers with eruptions of similar height.
Our survey of the LGB confirmed that major fountain-type geysers
are reliably identified with dual microphone arrays at distances of up
to several kilometers. However, we note that we did not detect reliable
signal from the major fountain geyserslocated in the UGB, located more
than 8.5 km away. This suggests a limit of somewhere between 3 and
8 km for infrasound detection of major fountain-type geysers. We also
note that we did not detect activity from any minor fountain type
Fig. 8. One-hour pressure time series, acoustic power, and corresponding power spectral density for select events at: a) Great Fountain, b) Fountain Geyser, and c) Botryoidal Spring.
Data are shown for band-pass filtered (0.25–20 Hz) signal. Acoustic power is calculated according to Eq. (11) using Δt = 1 s intervals. Total energy for the hour-long interval is
shown in each panel. Power spectral density shows a combination of ambient infrasonic noise, centered at the 0.25 Hz corner frequency (dashed line), as well as generally higher
frequency geyser signal. Power spectrum from a low-noise one-hour nighttime period is indicated for comparison.
115J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
geysers farther than about 500 m distant, such as those found in the
Pink Cone Group or the Black Warrior Group. Based upon our observa-
tions of the minor geyser activity at Firehole and Botryoidal Springs we
conclude that to reliably track smaller features it is necessary to deploy
sensors within a few hundred meters of their sources.
7. Conclusion
Dual acoustic arrays separated by approximately 600 m can be used
to identify and track activity from individual geysers out to several kilo-
meters. As such, future monitoring of geyser activity with non-intrusive
acoustic arrays in the infrasound band is warranted. Acoustic monitor-
ing can also complement ongoing efforts to track geyser activity, such
as those measuring thermal flux in geyser outflow channels in Norris
Basin (Perry, 2011). Further, acoustic monitoring can facilitate compre-
hensive records of geyser eruption statistics, including repose periods
between eruptions, eruption duration, and style of eruption (e.g., pulsing,
spasmodic or continuous, intense or benign), which will enable a better
understanding of hydrologic controls including exchange of function
with neighboring features (Marler, 1951). Relationships between erup-
tion duration and repose time can be also be robustly studied given con-
tinuous and long-duration monitoring of a system of geysers and should
aid in a better understanding of periodicity controls (Ingebritsen and
Rojstaczer, 1993). Finally, we anticipate that changes in geyser activity,
due to seasonal effects or dynamic strains from transient earthquake
waves (Husen et al., 2004),canbemorerobustlyquantified and studied
with long-term acoustic monitoring.
We have shown that the ability to comprehensively monitor
geysers is affected by both the intensity of the geyser infrasound source,
which appears greater for fountain-type geysers than for cone-type
geysers, and the level of background noise. During local daytime periods
(e.g., 10:00 AM to 9:00 PM) wind was often so intense in LGB as to
obscure all activity except from the nearby Great Fountain. More compre-
hensive monitoring of geyser activity will be facilitated with better strate-
gic deployment of arrays closer to the smaller geothermal features and
utilization of a greater number of arrays. In the future, local monitoring
of the UGB with its incredible population of major geysers will be partic-
ularly illuminating. We believe that infrasound monitoring is an effective
and non-intrusive tool for tracking activity for a cluster of geysers and can
be substantially less work-intensive than relying upon eyewitness or
video observations.
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.jvolgeores.2013.02.016.
Acknowledgments
Fieldwork was carried out with the help from a team of students in a
volcano geophysical field methods course held in Yellowstone National
Park in 2011. The students involved included the co-authors as well as
A. Curtis, N. Iverson, R. Johnson, D. Krzesni, and A.Quezada-Reyes. We
are indebted to GPS surveying of the sensor arrays by M. Murray. We
also thank the National Park Service research staff for their assistance
with permitting and logistical advice. NSF EAR grant #1151662 sup-
ported this work.
221 221.5 222 222.5 223 223.5 224 224.5 225 225.5 226
0
5
10
15
20
25
30
# detections per minute
1 2 3 4 5 6 7 8 9
a) Great Fountain detections
Julian day
050 100 150
8
10
12
14
16
18
20
event duration (minutes)
event interval (hours)
b) interval statistics
1
2 3
4
5
6
7
8
Aug. 2012 data fit (R2 = 0.73)
Bryan (2008) observations
Fig. 9. a) Five-day record of Great Fountain event detections showing episodicity of events. Arrows are drawn 85 min prior to select main events and coincide with some detected
precursors. b) Duration of event detections plotted against subsequent quiescent interval for 8 events (black circles). Event interval is measured as the time between a main event's
last detection and the next event's first detection.
Fig. 10. Distribution of event intervals for detected Botryoidal Spring eruptions. Inter-
vals are defined as time differences between center times of successive event detec-
tions. Normal distribution fit to data is shown.
116 J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117
References
Arechiga, R., Johnson, J.B., Edens, H., Thomas, R.J., Rison, W., 2011. Acoustic localization
of triggered lightning. Journal of Geophysical Research 116 (D09103).
Arrowsmith, S.J., Johnson, J.B., Drob, D.P., Hedlin, M.A.H., 2010. The seismo-acoustic
wavefield: a new paradigm in studying geophysical phenomena. Reviews of Geo-
physics 48, RG4003.
Assink, J.D., Evers, L.G., Holleman, I., Paulssen, H., 2008. Characterization of infrasound
from lightning. Geophysical Research Letters 35 (L15802).
Bowman, J.R., Baker, G.E., Bahavar, M., 2005. Ambient infrasound noise. Geophysical
Research Letters 32 (9).
Bryan, T.S., 2008. The Geysers of Yellowstone, 4th ed. University Press of Colorado,
Niwot, CO (463 pp.).
Cansi, 1995. An automatic seismic event processing for detection and location; the
P.M.C.C. method. Geophysical Research Letters 22 (9), 1021–1024.
Christie, D.R., Campus, P., 2010. The IMS infrasound network: design and establishment
of infrasound stations. Infrasonic Monitoring for Atmospheric Studies. Springer,
Heidelberg, pp. 29–75.
Dowling, A.P., 1998. Steady-state radiation from sources. In: Crocker, M. (Ed.), Hand-
book of Acoustics. John Wiley & Sons, New York, pp. 99–117.
Fee, D., Garces, M., 2007. Infrasonic tremor in the diffraction zone. Geophysical Research
Letters 34, L168 26.
Garces, M.A., McNutt, S.R., 1997. Theory of the airborne sound field generated in a reso-
nant magma conduit. Journal of Volcanology and Geothermal Research 78 (3–4),
155–178.
Gerst, A., Hort, M., Aster, R.C., Johnson, J.B., in review. The first second of a volcanic
eruption - energies, pressures, mechanisms. Journal of Geophysical Research.
Husen, S., Wiemer, S., Smith, R.B., 2004. Remotely triggered seismicity in the Yellow-
stone National Park region by the 2002 Mw 7.9 Denali Fault earthquake, Alaska.
Bulletin of the Seismological Society of America 94 (6B), 317–331.
Ingebritsen, S.E., Rojstaczer, S.A., 1993. Controls on geyser periodicity. Science 262
(5135), 889–892.
Johnson, J.B., Ripepe, M., 2011. Volcano infrasound: a review. Journal of Volcanology
and Geothermal Research 206, 61–69.
Johnson, J.B., Aster, R.C., Kyle, P.R., 2004. Volcanic eruptions observed with infrasound.
GeophysicalResearch Letters 31 (L14604). http://dx.doi.org/10.1029/2004GL020020.
Jones, K., Johnson, J.B., Aster, R., Kyle, P., McIntosh, W., 2008. Infrasonic tracking of large
bubble bursts and ash venting at Erebus volcano, Antarctica. Journal of Volcanology
and Geothermal Research 177, 661–672.
Kedar, S., Sturtevant, B., Kanamori, H., 1996. The origin of harmonic tremor at Old
Faithful geyser. Nature 379, 708–711.
Kieffer, S.W., 1984. Seismicity at Old Faithful Geyser; an isolated source of geothermal
noise and possible analogue of volcanic seismicity. Journal of Volcanology and Geo-
thermal Research 22, 59–95.
Lighthill, M.J., 1978. Waves in Fluids. Cambridge University Press, New York (504 pp.).
Marchetti, E., Ripepe, M., Harris, A.J.L., Delle Donne, D., 2009. Tracing the differences be-
tween Vulcanian and Strombolian explosions using infrasonic and thermal radiation
energy. Earth and Planetary Science Letters 279 (3–4), 273–281.
Marcillo, O., Johnson, J.B., Hart, D., 2012. Implementation, characterization, and evalu-
ation of an inexpensive low-power low-noise infrasound sensor based on a
micro-machined differential pressure transducer and a mechanical filter. Journal
of Atmospheric and Oceanic Technology 29, 1275–1284.
Marler, G.D., 1951. Exchange of function as a cause of geyser irregularity. American
Journal of Science 249, 329–342.
Matoza, R., etal., 2009. Infrasonicjet noise from volcanic eruptions. Geophysical Research
Letters 36 (L08303).
Murray, T.L., Endo, E.T., 1992. A real-time seismic-amplitude measurement system (RSAM).
In: Ewart, Swanson (Eds.), Monitoring Volcanoes: Techniques and Strategies Used by
the Staff of the Cascades Volcano Observatory, 1980–1990: USGS Bulletin, pp. 5–10.
Perry, J., 2011. Tracking Yellowstone's activity. Earth Magazine, April.
Ripepe, M., Gordeev, E.I., 1999. Gas bubble dynamics model for shallow volcanic tremor
at Stromboli. Journal of Geophysical Research 104 (B5), 10639–10654.
Rost, S., Thomas, C., 2002. Array seismology: methods and applications. Reviews of
Geophysics 40 (1008) (27 pp.).
Vergniolle, S., Brandeis, G., 1996. Strombolian explosions 1. A large bubble breaking at
the surface of a lava column as a source of sound. Journal of Geophysical Research
101 (B9), 20433–20447.
Woulff, G., McGetchin, T.R., 1976. Acoustic noise from volcanoes: theory and experiment.
Geophysical Journal of the Royal Astronomical Society 45, 601–616.
Yokoo, A., Iguchi, M., 2010. Using infrasound waves from eruption video to explain
ground deformation preceding the eruption of Suwanosejima volcano. Japan Journal
of Volcanology and Geothermal Research 196 (3–4), 287–294.
117J.B. Johnson et al. / Journal of Volcanology and Geothermal Research 256 (2013) 105–117