Jones 2012 Decomposition Volcanic Tremor

Subband decomposition and reconstruction of continuous volcanic tremor
J.P. Jones
a,
, R. Carniel
b
, S.D. Malone
a
a
University of Washington, Department of Earth and Space Sciences, Seattle, WA 98195-1310, USA
b
Laboratorio di misure e trattamento dei segnali, DICA, Universit di Udine, Via delle Scienze, 206-33100 Udine, Friuli, Italy
a b s t r a c t a r t i c l e i n f o
Article history:
Received 29 January 2011
Accepted 12 July 2011
Available online 3 August 2011
A new method of analyzing volcanic tremor is presented, which uses properties of undecimated wavelet packet
transforms to lter, decompose, and recover signals from continuous multichannel data. The method preserves
manystandardproperties that are usedtocharacterizetremor, suchas waveeldpolarizationandseismic energy.
In this way, we can better understand the (potentially many) seismic sources that combine to form continuous
volcanic tremor, and we can specically address the problem of what causes changing tremor spectral content.
Tests on synthetic data suggest that SDR can recover multiple quasi-continuous signals that differ from one
another by an order of magnitude, even in noisy environments. Tests on real data recorded at Erta 'Ale in 2002
suggest that SDR can recover signals with geophysically meaningful interpretations, and corroborates existing
seismic and multiparametric work by Harris et al. (2005), Jones et al. (2006), and Harris (2008). We suggest that
this algorithm could effectively detect subtle changes in the time-frequency content of volcanic tremor, and
recover signals from real seismic sources that appear buried in background noise (and/or partly masked by one
another). Such an algorithm could allow volcanologists much greater insight into the dynamics of volcanic
systems, and could detect subtle signals that might help address the possibility of unrest.
2011 Elsevier B.V. All rights reserved.
1. Background and motivation
Continuous volcanic tremor is a persistent seismic signal observed
near active volcanoes, which generally lacks impulsive phase arrivals,
and can persist on timescales as long as several years (Chouet, 1996;
Konstantinou and Schlindwein, 2002). It has been observed since the
beginning of volcano monitoring: a few examples of volcanoes with
continuous tremor include Kilauea, Hawai'i (e.g. Goldstein and
Chouet, 1994), Etna, Italy (e.g. Di Lieto et al., 2007), Ambrym, Vanuatu
(Carniel et al., 2003), Stromboli, Italy (e.g. Ripepe et al., 2002), and
Erta 'Ale, Ethiopia (Harris et al., 2005; Jones et al., 2006). Recently, it
has been found that the spectral content of continuous volcanic
tremor can undergo relatively abrupt, characteristic changes on
timescales of hours to weeks, suggesting either a changing source
mechanism for the tremor, or additional seismic sources generating
signals that affect the seismic wave eld: this has been seen at e.g.
Stromboli, Italy (Ripepe et al., 2002), Ambrym, Vanuatu (Carniel et al.,
2003), Erta 'Ale, Ethiopia (Harris et al., 2005; Jones et al., 2006), and
Dallol, Ethiopia (Carniel et al., 2010). This is quite different from
gliding spectral lines observed in tremor data recorded at volcanoes
that erupt explosively, e.g. Montserrat (Powell and Neuberg, 2003)
and Lascar (Hellweg, 2000), in which spectral content changed over
relatively short timescales, but in a continuous (rather than abrupt)
way.
The changing spectral content of continuous volcanic tremor
presents a challenging problem, but is also a potentially useful
diagnostic tool, particularly if its relationship to eruptive behavior can
be quantied. Clearly, because spectral transitions are seen to some
degree at each station in each seismic network, it follows that the
tremor source must change somehow. However, it is difcult to
determine from modeling alone whether tremor has a single,
changing source mechanism, or is rather a composite of multiple
source processes with different spectral energies. It is useful to resolve
this ambiguity because continuous tremor has been shown to relate to
changes in eruptive behavior at some volcanoes (Ripepe et al., 2002;
Harris et al., 2005), and because this sort of changing spectral content
has been associated with changes in hydrothermal systems (Carniel
et al., 2010), hydrocarbon reservoirs (Dangel et al., 2003), and slow
slip in subduction zones (Obara and Hirose, 2006). Identifying the
physical processes that drive these observed spectral changes could
thus help volcanologists identify, understand, or even predict,
changes in volcanic activity at these systems.
The goal of this work is to develop a quantitative means of
determining whether changing spectral content in continuous volcanic
tremor represents a change in the tremor source mechanism, or an
introduction of new, superimposed, seismic sources with different
spectral peaks. More generally, we wish to introduce a quantitative
means of tracking the signal content of continuously recorded geophys-
ical data. Ideally, if the data are to be treated as a composite of several
Journal of Volcanology and Geothermal Research 213-214 (2012) 98115
Corresponding author. Tel.: +1 206 930 8065.
E-mail address: josh@ess.washington.edu (J.P. Jones).
0377-0273/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.jvolgeores.2011.07.006
Contents lists available at ScienceDirect
Journal of Volcanology and Geothermal Research
j our nal homepage: www. el sevi er. com/ l ocat e/ j vol geor es
signals, then any such method should recover some of the standard
parameters about each component signal, e.g. polarization (Montalbetti
and Kanasewich, 1970; Vidale, 1986; Jurkevics, 1988), and signal energy,
which can be used to locate a tremor centroid (e.g. Gottschmmer and
Surono, 2000; Patan et al., 2008).
2. Subband decomposition and reconstruction
It canbechallengingtoisolate what signals comprise the tremor data
without making ad hoc assumptions about the nature of the data. The
general concept of blind source separation methods is not new to
volcano seismology, and was previously used by e.g. Cabras et al. (2008)
to study low-frequency tremor at Merapi, Indonesia. Volcanic data can
be quite complex, often consisting of transients superimposed on a
continuous background signal. If LP or VLP events are present, classical
methods based on Fourier analysis can perform poorly, because the
notion of statistical stationarity may not apply. We wish to develop an
algorithm based on more robust wavelet methods, beginning with the
assumption that there exists a subband or subbands (f
1
, f
2
) of the
frequency spectrum (0, f
N
) where seismic energy at some stations is
dominated by energy from a single seismic source. From this, we will
introduce a simple, efcient, top down algorithm based on the
undecimated (a.k.a. maximal overlap) discrete wavelet packet
transform (Walden and Cristan, 1998; Percival and Walden, 2000).
Our algorithm makes use of subband selection methods similar to
the top-down best basis approach of Taswell (1996), based on
Coifman and Wickerhauser (1992), which generally allows for basis
selection without computing a full wavelet packet table. It has several
similarities to algorithms proposed for medical data by Oweiss and
Anderson (2007), though the subband selection criteria are dened in
a way more appropriate for geophysical data sets. We follow a simple
3-step procedure:
1.) Determine the wavelet decomposition in which a single
principal component of the multichannel data most complete-
ly dominates each subband comprising some part of the
frequency spectrum 0f f
N
.
2.) Hierarchically cluster the wavelet packets using quantitative
information about the signal content of each.
3.) Inverse transformthe wavelet packets of each cluster, thereby
ltering the input data to frequencies where the same signal
dominates the seismic energy.
3. Wavelet decomposition and representation
For purposes of this manuscript we use the following notation:
Assume that non-boldface variables (e.g. N) refer toscalars, andboldface
variables (e.g. h) refer to vectors. Assume that uppercase boldface
variables (e.g. X
t
) refer to N-length vectors of (or derived from) a single
data channel with N data points, and that uppercase boldface variables
with a bar (e.g. X
t
) refer to NK matrices of (or derived from) multi-
channel data, having K channels and N data points.
Let us rst reviewsome relevant principles of wavelet transforms for
a single time series X
t
, i.e. a single data channel at a single seismic
station. The discrete wavelet transform, or DWT, of an input time series
X
t
to some level J is an orthonormal transform that uses a sequence of
ltering operations to obtain wavelet coefcients W
j
and scaling
coefcients V
j
, associated with weighted differences on scales of
j
=2
j 1
. It captures information about both the frequency and
temporal contents of the input data; this is contrasted with Fourier
transforms, which characterizes the amplitude and phase of the
frequency content only, assuming a stationarity in time that is not
always satised in practice. Whereas each coefcient of a Discrete
Fourier Transform (DFT) is associated with a particular frequency, each
DWT coefcient W
j
(t) or scalingcoefcient V
j
(t) canbe thought of as the
difference in adjacent averages of X
t
on scales of
j
=2
j 1
, centered
about some time t. We can succinctly write W
j
and V
j
as circularly
lteredconvolutions of X
t
witha wavelet lter h
l
or scaling lter g
l
using
the recursion relations
W
j
t =
L1
t =0
h
l
V
j1;2t +1=l mod N
j
1
V
j
t =
L1
t =0
g
l
V
j1;2t +1=l mod N
j
1
1
and by denition V
0
X
t
. Here l is the index of the coefcient in a lter
(h
j
or g
j
), L is lter width, and Nis the length of the input vector X
t
. For a
more in-depth review of the meaning of wavelet and scaling
coefcients, see e.g. Chapter 4 of Percival and Walden (2000). A
complete descriptionof the DWTis giveninStrang (1993), withdetailed
discussion in Daubechies (1992) and Chui (1997). Wassermann (1997)
previously used the DWT to characterize and locate volcanic tremor at
Stromboli.
A common extension of the discrete wavelet transform is the so-
called maximal overlap discrete wavelet transform (MODWT), which
is equivalent to the undecimated, stationary, translation invariant, or
shift invariant DWT (Greenhall, 1991; Shensa, 1992; Percival and
Guttorp, 1994; Coifman and Donoho, 1995; Nason and Silverman,
1995; Liang and Parks, 1996; Percival and Mojfeld, 1997). The
MODWT carries out ltering steps nearly identical to the DWT, only
with no decimation (down sampling by 2) at each successive level j.
Several useful mathematical properties of the MODWT are described
in Walden and Cristan (1998) and Percival and Walden (2000). Here,
we list only those relevant to this research:
I. Because the MODWT does not down sample, the length N of an
input time series X
t
need not be a power of 2. (Percival and
Guttorp, 1994)
II. Inverse transforming the MODWT coefcients W
j
creates the so-
called detail coefcients or details D
j
and smooths S
J
that
form a multi-resolution analysis of X
t
; that is,
X
t
= S
J
+
J
j =1
D
j
III. D
j
and S
J
are associated with zero phase lters (Percival and
Mojfeld, 1997). Thus the pass band of the detail coefcients D
j
associated with each wavelet coefcient vector W
j
is a well-
dened subband of the frequency range [0,f
n
]. For data sampled
at Hz, the exact corner frequencies (in Hz) of the subband
whose detail coefcients are D
j
are given by *2
(j +1)
| f | *2
j
(Percival and Walden, 2000).
IV. The MODWT preserves energy. For a MODWT of an input time
series X
t
to any level J, V
J
2
+
J
j =1
W
j
2
= X
t
2
(Percival
and Walden, 2000). This property is not shared by the detail
coefcients, which follows from II. and the Schwarz inequality.
V. Wavelet transforms can estimate the time-varying cross-
correlations of 3-component seismic data in a subband of the
frequency spectrum, and hence its polarization and principal
components. (Lilly and Park, 1995; Anant and Dowla, 1997;
Oweiss and Anderson, 2007).
4. Maximal overlap discrete wavelet packet transform
We now briey review some relevant properties of the maximal
overlap discrete wavelet packet transform(MODWPT) of Walden and
Cristan (1998). This extension of the discrete wavelet transform,
while common in the signal processing literature, is relatively
unknown to volcano seismology. The MODWPT generalizes the
DWT by recursively ltering an input time series X
t
with all possible
combinations of the (rescaled) wavelet lter h
l
and scaling lter g
l
,
99 J.P. Jones et al. / Journal of Volcanology and Geothermal Research 213-214 (2012) 98115
without downsampling W
j,n
by 2 at each successive transformlevel j. A
sample selection of lters used to create the MODWPT to level j =2 is
given in Fig. 1, along with their squared gain functions at unit
sampling frequency. Observe that, for data sampled times per
second, each wavelet packet W
j,n
is now associated with frequencies
in the nominal pass band
n
2
j +1
bf

n + 1
2
j +1
: 2
It is clear from Fig. 1 that the MODWPT forms an overcomplete
representation of the frequency range [0,f
n
]. The idea behind the
MODWPT is thus to compute a generalized, highly redundant table of
non-decimated wavelet packet vectors

W
j;n
. For each successive level j
of the MODWPT, one lters each wavelet packet

W
j1;n
with (rescaled
wavelet or scaling) coefcients, u
n;l
h
n;l
=
2
p
or u
n;l
g
n;l
=
2
p
, to create
vectors

W
j;2n
and

W
j;2n+1
, respectively. At each successive level j, we
can write the equivalent recursive ltering operation to obtain

W
j;n
as
W
j;n
t =
L
j
1
l =0
u
n;l

W
j1;
n
2
j k
;t2
j1
l mod N
j1
3
where we dene

W
0;0
X
t
and u
n;l
as above.
Using this more general notation, the MODWT coefcients W
j
(and
corresponding detail coefcients D
j
) correspond to wavelet packets
W
j;1
for any level j. The scaling coefcients V
J
and corresponding
smooths S
J
correspond to wavelet packet

W
J;0
and its inverse
transformed detail coefcients

D
J;0
. For convenience, as the rest of
this manuscript deals exclusively with the MODWPT, we will write a
generalized wavelet packet

W
j;n
as W
j,n
and its corresponding detail
coefcients

D
j;n
as D
j,n
. Henceforth we assume non-decimation.
Percival and Walden (2000) showed that any complete partition of
the frequencies [0, f
N
] using wavelet packets W
j,n
is an orthonormal
transform. Thus, any MODWPT basis shares properties IV of the
MODWT. The collection of all wavelet packets for all levels 1j bJ is
called a wavelet packet (WP) table.
5. Wavelet basis selection for multichannel data using
the MODWPT
We canextract manydifferent orthonormal transforms froma WP table.
Many algorithms and cost functionals have been devised to determine
the best wavelet packet basis for single channel data (e.g. Coifman and
Wickerhauser, 1992; Taswell, 1996; Oweiss and Anderson, 2007). Such
algorithms select parts of the wavelet packet tree that evaluate the
characteristics of input data X
t
using some cost functional m(W
j,n
),
which is associated with each wavelet packet vector W
j,n
. The wavelet
basis that satises
min
C
j;n C
m W
j;n

4
is the best representation of the data in the wavelet domain.
Unfortunately, algorithms applicable to a single input time series
are not necessarily generalized to multichannel data in any straight-
forward way. For the example of geophysical data, the channels (i.e.
stations) nearest the source generally have the highest associated
cost, and therefore most signicantly affect the calculation. In the
ideal case of a single seismic source recorded by multiple receivers,
with no glitches, transients, or additive noise, wavelet packet vectors
are weighted for each station by the same power lawas the falloff rate
of the energy in the frequency range of Eq. (2). However, this means
that the best basis can be skewed by just one channel of bad data.
Additionally, with multiple sources having different relative energies
at each station, summing costs for each node of the wavelet packet
table over each data channel is not necessarily an appropriate method.
The problem of determining a best wavelet decomposition for
multichannel data was partially addressed by Oweiss and Anderson
(2007), who derived a multichannel cost functional relating the eigen
W(1,1)
0 1/16 1/8 3/16 1/4 5/16 3/8 7/16 1/2
Normalized Frequency [Hz]
W(1,2)
W(2,1)
W(2,2)
W(2,3)
W(2,4)
MODWPT Filter Squared Gain
W(1,1)
0 1/16 1/8 3/16 1/4 5/16 3/8 7/16 1/2
Normalized Frequency [Hz]
W(1,2)
W(2,1)
W(2,2)
W(2,3)
W(2,4)
MODWPT Filter Squared Gain
Fig. 1. Wavelet packet lters and corresponding normalized squared gain functions that create a MODWPT to level j =2. Left: The LA16 wavelet. Right: The Daubechies-2 or Haar
wavelet.
decomposition of each (multichannel) wavelet packet matrix W
j;n
to
that of the original multichannel data X
t
. Their approach was designed
toguarantee that the best basis is alsothe best t of the wavelet packet
transform to the input data, and is thus suitable for problems with
broadband sources whose spectra span [0,f
n
] and have additive noise.
However, geophysical data in general, and almost all volcano-seismic
data, have seismic sources that generate energy in a relatively narrow
frequency band. Due to the band-limited nature of volcanic tremor
sources (e.g. McNutt, 1996; Sherburn et al., 1998; Konstantinou and
Schlindwein, 2002), and attenuation and geometrical spreading, the
meaningful part of the frequency spectrumof the recorded data almost
never spans the frequency range [0,f
n
]. Thus, for volcano-seismic data
sets, it is not necessarily true or desirable that the eigen decomposition
of the covariance matrix of X
t
is a suitable match to each subband.
Rather than approaching the problem of wavelet basis selection
with the expectation that each W
j;n
will match the eigen decompo-
sition X
t
, we approach this problem with the more geophysically
appropriate expectation that we can nd a wavelet decomposition of
X
t
whose wavelet packets W
j;n
are each nearly dominated by a single
principal component, corresponding to a single physical source,
whose mathematical representation is the observed signal content.
This notion is widely assumed in volcano-seismological literature, as
suggested in many of the references above.
Here we dene a cost functional withthis goal in mind, for data from
K seismic stations. Whereas conventional cost functionals seek to
minimize an information cost, our goal is to select a basis using the
expectation that a single principal component dominates (each part of)
our observation matrix X
t
in some subband of [0, f
n
]. We can quantify
this expectation by making note of the following properties of the
principal components (cf. Pearson, 1901), i.e. eigenvalues
j,n,k
and
eigenvectors v
j,n,k
of the covariance matrix of each wavelet packet:
A. For Gaussian data, the eigenvectors of the covariance matrix of
each W
j;n
point in the direction (in K-dimensional space) of the
independent components, i.e. each eigenvector represents the
relative strength of each independent component at each
seismic station. Even for non-Gaussian or multi-modal Gaussian
data, principal components analysis (PCA) de-correlates the axes
of the independent components. (Hyvrinen et al., 2000)
B. The relative eigenvalues of the covariance matrix of each W
j,n
correspond to the relative energies of the independent
components. (see e.g. Shaw, 2003).
The use of these properties is best illustrated with two conceptual
examples. First, an eigenvector of one wavelet packet W
j,n
, which aligns
almost exactly in K-space to a single station, and whose corresponding
eigenvalue is very large, could represent a local transient recorded only
at that station, in the frequency range given by Eq. (2). On the other
hand, if the energies of the eigenvalues of one wavelet packet W
j,n
are
nearly equal, andtheeigenvectors appear randomly orientedinK-space,
then it follows that the subband is dominated by either local transients
at each receiver, or by incoherent noise.
Principal components analysis can produce erroneous results when
trace data are out of phase, as the maxima at one stationmay alignintime
with the minima of another station. For this reason, and following from
propertyVof wavelet coefcients, onecanalignthemultichannel wavelet
packet coefcients in a least-squares sense following the algorithm of
Vandecar and Crosson (1990), prior to computing their principal
components. This introduces some potential for cycle skipping, however,
which must be controlled by carefully choosing a maximum lag for each
pair of stations. We remark that this slows the algorithmslightly, because
unshiftedcoefcients arenecessarytocomputesuccessivelevels of W
j,n
in
iterative algorithms. However the added step is a necessary precaution.
From these properties of principal components analysis, we may
now dene a cost functional based on how well a single principal
component dominates a subband whose wavelet packet coefcients
are W
j,n
. For K stations, when
j,n,1
is large relative to
j,n,2
,
j,n,K
, we
expect the ratio
K
k=2
j;n;k
j;n;1
to be small. Recalling from Eq. (2) that each W
j,n
is associated with a
normalized bandwidth of approximately /2
j +1
, we can dene a cost
functional directly from the above ratio, which accounts explicitly for
this bandwidth:
M W
j;n

K
k=2
j;n;k
j;n;1
2
j +1
: 5
This cost functional behaves similarly to the entropy-based cost
functional of Coifman and Wickerhauser (1992), but is bounded by
0M(W
j,n
)b(K1)/2
j +1
. However, it must be noted that, like the
cost functional of Oweiss and Anderson (2007), this is not strictly an
optimization problem, as, by the very nature of geophysical data, the
principal components of the input data are not necessarily inherited
exactly by each subband.
Using this cost functional in traditional best basis type algorithms
determines those wavelet packets W
j,n
which are most dominated by a
single principal component. However, it is not necessarily true that the
same seismic source (or equivalently the same principal component) will
dominate each W
j,n
. In fact, due to the notorious complexity of tremor
sources, it is likely that some W
j,n
, and their associated passband frequen-
cies (2) will be dominated by very different seismic sources than others.
Thus, we constrain our basis selection algorithm in the following
way. The similarity of the most energetic principal component in each
wavelet packet is easily quantied by measuring the distance d(
)
between eigenvectors v
j,n,k
, v
j,n+1,k
, corresponding to adjacent wavelet
packets W
j,n
and W
j,n+1
. Note, however, that the distance between
eigenvectors must account for a possible sign change. Thus we compute
distance using the trigonometric formula
d v
j;n;k
; v
j;n+1;k

=
22 v
j;n;k
v
j;n+1;k
r
: 6
If this distance falls below some predetermined threshold , we say
that the dominant principal component of each wavelet packet vector is
the same. Thus, we apply the following constraint when selecting a basis.
It is clear that the wavelet packet table forms a tree of nodes, with
each parent node W
j,n
having two child nodes constructed via
circular convolution in Eq. (3), W
j +1,2n1
and W
j +1,2n
. Thus, at each
parent node W
j,n
in the wavelet packet tree, for wavelet levels 1j bJ,
we replace child nodes W
j +1,2n1
, W
j +1,2n
with their parent
node W
j,n
if the following two criteria are met:
1. M(W
j,n
)M(W
j +1,2n1
)+M(W
j +1,2n
)
2. d(v
j +1,2n1,1
, v
j +1,2n,1
)b.
By proceeding down (or up) the wavelet packet table, we use this
conditional basis selector to determine the wavelet decomposition in
which the subbands that span [0, f
n
] are most dominated by a single
principal component each.
6. Wavelet packet clustering and signal reconstruction
We now dene a recovered signal, using the properties of wavelet
transforms and principal components analysis. The wavelet packets W
j,n
in the best basis can be hierarchically clustered using Eq. (6). Because of
Property III of the MODWT, each W
j,n
in the best basis can be inverse
transformedto create zero-phase bandpass ltereddetail coefcients D
j,n
.
Because of Property II, summing the (inverse transformed) detail
coefcients D
j,n
in each cluster enables us to reconstruct our original
observation matrix X
t
, ltered exactly to those frequency bands where
each recovered signal dominates. Thus, we dene a recovered signal as
the sum of the detail coefcient vectors D
j,n
for all W
j,n
in one particular
group of wavelet packets in the best basis.
This nal step clustering, reconstruction, and summation
essentially treats the wavelet lters u
n,l
in Eq. (3) as orthogonal zero-
phase lter banks. To see why this is possible, recall from Walden and
Cristan (1998) that wavelet lters u
n,l
are orthogonal, and consider
the single-channel case X
t
. From Percival and Walden (2000) we have
a convenient equation to dene the wavelet detail coefcients D
j,n
for
a single channel X
t
in terms of the circular cross-correlation of wavelet
lters with their corresponding wavelet packets W
j,n
:
D
j;n
t
L
j
1
l =o
u
j;n;l
W
j;n;t +l mod N
7
where we dene
u
j;n;l
2
j =2

L1
k=0
u
n;k
u
j1;
n
2
j k
;l2
j1
k
8
i.e. u
j,n,l
is dened as the wavelet packet lter that directly creates W
j,n
by circular convolution with X
t
. It is already noted that each wavelet
detail vector D
j,n
is equivalent to X
t
convolved with a zero-phase band
pass lter. The idea that we can create a ltered time series merely by
summing detail coefcients follows from expressing X
t
as the sum of
those detail coefcients D
j,n
whose W
j,n
belong to the best basis;
expressed using Eq. (5), and noting from Eq. (6) that the detail
coefcients are created from orthonormal u
n,l
, we have
X
t
= 2
j =2

J
j =1
2
j
1
n=0
L
j
1
l =o
L1
k=0
u
n;k
u
j1;
n
2
j k
;l2
j1
k
0
@
1
A
W
j;n;t +l modN
0
@
1
A
:
9
Thedesiredresult follows fromtakingtheDFTof bothsides of Eq. (8),
multiplying each by its complex conjugate, and noting that cross terms
vanish due to orthogonality of u
n,k
. The optimal pass band of each detail
coefcient D
j,n
is, again, given by Eq. (2).
6.1. Algorithm description
We now describe conceptually the algorithm to compute an
iterative, top-down, wavelet decomposition of a multichannel obser-
vation matrix X
t
, clusters its wavelet packet vectors W
j;n
using the rst
principal component of each, andreturns X
t
lteredtothosefrequencies
where each recovered signal dominates (we'll call this X
t
). We rst
describe its use for single-component data, then discuss how to
generalize to the case of 3-component data.
6.1.1. Step 1. Iterative, top-down wavelet decomposition
For each level j, beginning with j =1 (and recalling fromabove that
we dened W
j;n
X
t
):
1. Compute wavelet packet coefcients W
j,n
for each channel at
each station. In practice, this is not computed directly using the
convolution of Eq.(3), but is obtained using more efcient
means, e.g. the pyramid algorithm of Mallat (1999).
2. Compute principal components v
j,n,1
and cost M(W
j,n
) of the
(possibly circularly shifted) wavelet packet coefcients.
a. For single-component data, compute the principal compo-
nents of the vertical component of each station.
b. For three-component data, polarization lter the data, and
computetheprincipal components of themost energetic (i.e. z)
component from each station.
3. Compare the costs of the sum of each pair of children nodes,
M(W
j +1,2n
) and M(W
j +1,2n+1
), with the cost of their parent
node, M(W
j,n
).
a. If M(W
j+1,2n
)+M(W
j+1,2n+1
)NM(W
j,n
), and d(v
j+1,2n,1
,
v
j+1,2n+1,1
) b, mark W
j,n
as a member of the best basis, and
do not compute wavelet packet coefcients for its children
nodes.
b. Otherwise, replace the cost of W
j,n
by the sumof the costs of
the children nodes, and continue down the wavelet packet
table.
4. Repeat steps 13 as we move down the WP table, computing
wavelet packet coefcients for those nodes whose parents do
not belong to the best basis.
5. Stop either when an arbitrary level j =J has been reached, or
when there are no further child nodes available.
6.1.2. Step 2. Hierarchical clustering
In this step, we cluster the wavelet packets W
j;n
belonging to the
best basis. We use the distance d() given in Eq.(6) to measure the
similarity between eigenvectors v
j,n,1
of each W
j;n
. To determine
whether the input data X
t
is a good match to some subband of the
data, the rst eigenvector of the original observation matrix (i.e. v
0,0,1
)
is included in clustering, though this is for comparison purposes only.
6.1.3. Step 3. Signal reconstruction
Inverse transform each W
j,n
in each cluster using Eq. (5) to create
its detail coefcients D
j,n
; summing D
j,n
over all j,n corresponding to a
cluster yields a recovered signal X
t
for a particular data channel.
With regard to volcanoseismic data in particular, the following
specic steps are performed to separate the subbands of K stations of
3-component data:
A.) Data are loaded and detrended. Instrument response is
deconvolved, then reconvolved with a common lter.
B.) For each subband n at wavelet level j:
1) Wavelet coefcients W
j;n
are computed via the pyramid
algorithm of Mallat (1999).
2) For 3-component data, the method of Jurkevics (1988) is
used to calculate the 33 covariance matrix from inner
products of the 3components. Asimilar methodwas usedin
Lilly and Park (1995) and specically in Anant and Dowla
(1997) in conjunction with wavelet transform methods.
3) Eigenvalues and eigenvectors are computed for each 3-
component station. Rectilinearity and planarity are com-
puted as dened in Jurkevics (1988).
4) Dene Z at a given station/subband as the (3 channel)
W
j;n
rotated into the eigenvector corresponding to the
largest eigenvalue. This eigenvector is multiplied by 1
(if necessary) to force the vertical component to be
positive.
5) Dene R and T following the convention of Jurkevics
(1988) by rotating into the azimuth of Z. This ensures that
detail coefcients D
j,n
formed from R and T are always
aligned identically relative to Z.
6) Align Z coefcients in time using a variant of the method
of Vandecar and Crosson (1990) if each Z wavelet packet
vector correlates to at least one other Z wavelet packet
vector at a level of r (1p)0.1. Here r is the maximum
cross-correlation computed over a range of lag times
determined from the estimated phase velocity and inter-
station distance. p is probability of that maximum arising
from random chance. A maximum of one under-con-
strained channel is allowed per subband.
7) The (possibly aligned) Z are nowused to calculate the KK
covariance matrix of the principal components. The total
cost of this subband n at wavelet level j is given by Eq.(5).
8) The eigenvector v
j,n,1
corresponding to the largest eigen-
value
j,n,1
is saved, as is the cost and polarization.
9) Successive child nodes of each W
j,n
are computed (if
necessary) from the (unaligned) unrotated wavelet co-
efcients corresponding to the parent node. The (un-
aligned) rotated wavelet coefcients are used to form the
detail coefcients D
j,n
.
For a matrix X
t
with K stations and N data points, and wavelet
packets grouped into P clusters, the algorithmreturns P ltered sets of
3KN outputs X
t
. Since this algorithm works from the top down, it is
almost never necessary to compute a full MODWPT. In an ideal case,
where one seismic source completely dominates a wide subband of
the frequency spectrum, and there are not many stations included,
real time implementation is possible on modern computing equip-
ment. However, the lack of decimation when computing the DWPT
does not allow the O(KN) efciency of Oweiss and Anderson (2007).
We remark that, despite the computational inefciency of computing
relative lags for each set of child nodes, forming D
j;n
by this method
preserves the phase of recovered signals X
t
, making it a potential
preprocessing technique for array processing.
Fig. 2. Sample synthetics and Fourier power spectra for three sinusoids at f =1.5 Hz,
f =1.8 Hz, and f =3.6 Hz, randomly phase shifted and masked by Gaussian white noise
of 6 unit amplitude. Signal scaling is described in the text.
Fig. 3. Shaded intensity plot of log
10
(RMS) of recovered signals as a function of noise
amplitude scaling factor. Input signals are sinusoids whose median amplitude is unity.
Fig. 4. Plot of log
10
(RMS) of recovered signals as a function of number of channels used
in SDR. Solid line indicates mean RMS of recovered signals in each set of simulations.
Dashed lines indicate 2 errors for each set of simulations. Dotted lines indicate
maxima and minima of each set of simulations. Noise amplitude is held constant at a
scale factor of 4. Input signals are sinusoids whose median amplitude is unity.
Fig. 5. Erta 'Ale, Ethiopia, and stations from the 2002 Erta 'Ale experiment,
superimposed on an aerial photograph. Notable features of summit caldera are labeled.
Short-period sensors (L22 and LE3D) are indicated by small white triangles. Broadband
sensor (CMG-40 T) is indicated by large white triangle. Position of lava lake is
consistent with that of the aerial photograph, left and under indicative text.
7. Tracking signal invariance using principal components
Implicit in this method is that the invariance of recovered signals X
t
can be tracked over time by examining the principal components of their
constituent subbands. If all principal eigenvectors v
j,n,1
of the wavelet
packets W
j;n
whose detail coefcients D
j;n
form a recovered signal X
t1
cluster to all principal eigenvectors v
j,n,1
of a recovered signal X
t2
from a
later period, and the frequencies of their nominal passbands in Eq. (2)
overlap, then it follows from the denition of a recovered signal that X
t1
and X
t2
are mathematically identical. This property is potentially useful
for determining the difference between whether a recovered signal X
t
changes over time, or whether there are merely different secondary
signals superimposed upon a signal whose principal components do not
change.
8. Applicability and limitations
A much more detailed discussion of the limitations and capabilities
of the above algorithm is found in Jones (2009). Here, we wish to
discuss some important limitations.
Implicit in the use of PCA to select an optimal data transformation is
the assumption that each pair of stations records some coherent source
energy in some subband(s). Recall that, following the notation of Aki
and Richards (2002), a measured seismic signal can be written as the
ltering of a source time series S(t) by a path effect lter P(t) and a
receiver effect lter R(t), i.e. X(t)=S(t)*P(t)*R(t). This method works
best when the energy of one (possibly rotated) channel of each pair of
stations has a cross-correlation that is statistically signicant. However,
even if the receiver functions R(t) can be neglectedor deconvolved from
eachseismogram, patheffects P(t) change the frequencycontent of each
station. Path effects at real volcanic systems can be very complex (see
e.g. Harrington and Brodsky, 2007). Thus, this method will be most
effective when the seismic source S(t) has an isotropic component.
It is also true that this approach becomes less appropriate as intrinsic
attenuation and scattering increasingly affect the frequency content and
waveforms recorded at each station. In a sense this method can be
thought of as the complement to coda wave interferometry (Pacheco and
Sneider, 2005; Brenguier et al., 2008), which favors array geometries
where scattering dominates the recorded multichannel data X
t
.
This feature is inherently useful to detect subbands of the frequency
spectrum dominated by transients and/or by path effects. For example,
follows from Eq.(5) that a subband whose cost is low, but which is
nearly aligned with the direction (in K-dimensional space) of station k,
could be dominated by transients at station k. Furthermore, a subband
whose cost is high, and which does not align well with the direction
(in K-dimensional space) of a station k, could be dominated by path
effects. In this way, the method introduced here could also be used to
discriminate between regions of the frequency spectrum dominated by
source effects, and regions dominated by path effects.
Now, recall that a non-volumetric source function S(t) (or even a
volumetric source functioninaninhomogeneous medium) generates at
least two distinct phase arrivals (P and S) in the far eld, and even a
volumetric source has a transverse near-eld term (Aki and Richards,
2002). In the most general case, PCA decouples the axes of the
mathematically independent inputs; however, these independent
components are any signal whose mathematical properties (e.g. arrival
time, frequency content) are different. Now, because the radiation
patterns of P andS differ e.g. the orientationof the maximumenergy of
each phase is rotated 45 their relative amplitudes at each station also
differ. Therefore one feature of this algorithm is that it could create two
Fig. 6. Representative samples of trace data (top) and corresponding spectrograms (bottom) from the Erta 'Ale experiment. Amplitudes of trace data are normalized to illustrate
detail. All data have been detrended, downsampled to 50 Hz, preprocessed using a 3 s cosine taper, and ltered to the instrument response of a Lennartz MarsLite (f
0
=0.2 Hz). Data
from station L22 are further highpass ltered using a 4 pole Butterworth lter at f =0.4 Hz. Spectrogram scaling is in dB computed from ground velocity. a. (Left) 17 m 35 s of raw
data beginning 15 Feb 2002, 16:28:46 GMT, during the low convective regime. Spectrogram corresponds to the vertical component of station L22 (Fig. 5). b. (Right) 10 m 54 s of
raw data beginning 15 Feb. 2002, 19:53:23 GMT, during the high convective regime. Spectrogram corresponds to the vertical component of station L22 (Fig. 5). Longer, more
detailed spectrograms from each convective regime can be seen in Harris et al. (2005) and Jones et al. (2006).
(or more) recoveredsignals for each unique seismic source one for the
region of the frequency spectrumwhere P dominates (where possible),
another for S.
8.1. Performance testing with synthetic data
Because this algorithm has been developed with continuous vol-
canic tremor in mind, it is instructive to illustrate its ability to recover
quasi-continuous signals under various circumstances that mimic real
volcanogenic signals. To this end, two sets of Monte-Carlo simulations
are performed using synthetic data buried in Gaussian white noise. A
sample of such data is given in Fig. 2. Both tests use the LA-16 wavelet,
which offers a good balance between linear phase and compact length
(Daubechies, 1992), and whose detail coefcients are nearly perfect
bandpass lters (Fig. 1). MODWPT coefcients are computed to level
J =7. All tests use a clustering threshold of =0.3 to group the
principal eigenvectors v
j,n,1
of each W
j,n
. This is equivalent to a maxi-
mum average angle in K-space of =17 between members of any
two wavelet packet clusters.
In both sets of Monte-Carlo simulations, the inputs are three
sinusoids of unit amplitude. The rst test evaluates the algorithm's
ability to recover signals as background noise becomes more energetic.
The second test evaluates the algorithm's sensitivity to the number of
available channels, or stations. Inbothtests, the inputs are 3 sinusoids at
f =1.5 Hz, f =1.8 Hz, and f =3.6 Hz, sampled at 50 Hz for 81.92 s. The
rst two sources differ in frequency by only slightly more than the pass
band width of each wavelet packet at level J =7 (from Eq. (2),
~0.195 Hz). Their closely spaced spectral peaks test the algorithm's
sensitivity, while the third sinusoid tests whether the algorithm falsely
detects harmonics instead of two unique sources.
For all tests, input amplitudes at each station (i.e. channel) are
scaled by a randommultiplier chosen from a normal distribution with
=1 and =0.5. Sinusoids are phase shifted randomly in each
channel by 2 to 2 radians. Thus the randomly generated param-
eters of each simulation are Gaussian white noise and amplitude and
phase of each of 3 inputs in each data channel. We restrict our K3
matrix S of input scale factors so that max(rms(cov(S)I))b0.3. Thus
the amplitude falloff of one input is never proportional to another.
For the rst set of simulations, 6 data channels (equivalent to
stations) are generated. Noise in each channel is multiplied by a
scaling factor that increases from 0.1 to 100 in increments of 10
0.2
. 100
Monte-Carlo simulations are performed for each scale factor, making
1600 total simulations. The algorithm recovers each sinusoid indepen-
dently in 1589/1600 tests (i.e. 99.3% success rate), without clustering
themtogether. However, we remark that control tests of pure Gaussian
white noise, containing no sinusoidal input, also recover each band
independently in N95%of tests. Afar more appropriate measure of SDR's
signal recovery ability is the RMS error betweenthe output signal Y
t
that
contains each sinusoid, and each (scaled, shifted) input sinusoid X
t
.
With 6 channels and 3 signals, eachsimulation produces 18 RMS values.
A shaded intensity plot of log
10
(RMS) vs. noise scaling factor is given in
Fig. 3. Because the medianamplitude of each input sinusoid is unity, this
plot suggests that SDR recovers signals even when noise amplitude is
almost an order of magnitude larger than input signal amplitude.
It is similarly important to investigate how the number of data
channels affects the ability to resolve recovered signals. Thus, a second
set of simulations follows a similar process to the rst for generating
noise, signal scaling, and phase shifts, but varies the number of chan-
nels from 3 up to 10. 100 simulations are performed for each number
of channels. In each simulation, background noise is held constant at a
scale factor of 4, which (from Fig. 3) produces typical RMS values of
0.360.08 with 6 data channels. Fig. 4 shows a plot of log
10
(RMS) vs.
number of channels. The mean RMS of the recovered signals, and
variations in RMS, are virtually unchanged when 5 or more channels
are used. However, the mean RMS of recovered signals is factor of two
greater (0.670.41) when the number of data channels is reduced to
3. Jones (2009) found empirically that choosing a value for in the
range 0.250.4 produced nearly identical results, with a slight
increase in all values of RMS.
It must be re-emphasized here that sampling interval and sinusoid
frequencies were carefully chosen so that wavelet coefcients on the
nest scale (J =7) had nominal passbands narrower than the dif-
ference between the two closest spectral peaks. SDR cannot be
expected to recover signals whose spectral peaks are closer together
than its nominal pass band width. In a hypothetical worst-case
scenario, multiple signals with nearly identical spectral peaks could be
indistinguishable.
8.2. Performance testing with real data: Erta 'Ale, Ethiopia
To illustrate the use and limitations of this algorithm with a real
volcano-seismic example, we present analysis of two samples of data
recorded at Erta 'Ale, Ethiopia, a basaltic shield volcano in the Danakil
Fig. 7. Decomposition of the data sample in Fig. 6a using the SDR algorithm. Top
Dendrogram of subband clustering. Horizontal dashed line indicates distance threshold
distance =0.3 for clustering of principal components eigenvectors. Labels give the
nominal passband in the interval [0,f
n
] associated with each wavelet packet. Bottom
Wavelet basis selected by subband clustering. Selected wavelet packets are shaded in
grayscale on a scale-frequency representation of the full wavelet packet table. Wavelet
packets belonging to the same recovered signal are the same shade of gray. Y-axis
indicates wavelet scale j, with j =0 corresponding to the wideband (input) signal by
convention. X-axis shows the position in the wavelet packet table (and nominal
passband) for each wavelet packet in the interval [0,f
n
].
Table 1
Recovered signals from a data sample recorded during the low convective regime in
2002, beginning 15 Feb 2002, 16:28:46 GMT (Fig. 6a). Signal names use the convention
L
(n)
, described in the text. Values for eigenvectors v
j,n,1
correspond to the largest
eigenvalue
j,n,1
of the principal components of each wavelet packet W
j,n
. The quantities
v
j,n
, energy (En), azimuth (Az), incidence angle (In), rectilinearity (Rc), and planarity
(Pl) are arranged in columns by station. Az and In are given in degrees. Wavelet packets
are designated W
j,n
and nominal passbands f
min
f
max
for each W
j,n
are given in Hz. Cost
functional M(W
j,n
) is renormalized (multiplied by normalized bandwidth 2
j +1
) so that
0M(W)b3. Spectral leakage is tabulated in the form log
10
(E
out
/E
in
), the base 10
logarithm of spectral energy outside the passband to spectral energy inside the
passband. Relative lags of each subband are given in seconds for subbands whose
wavelet cross-correlations are well constrained.
Signal name Station L22 CMG MAR
Wavelet packet v
wavelet
f
min
f
max
(Hz) En (A.U.)
M(W
j,n
) Lag (s)
log
10
(E
out
/E
in
) Az (deg)
In (deg)
Rc (0rc1)
Pl (0pl 1)
L(1)
W
7,7
v
7,7,1
0.90 0.44 0.02
f
min
f
max
1.371.56 En 1.4e07 5.9e08 1.7e08
M(W
j,n
) 0.313 Lag 0.22 0.12 0.34
log
10
(E
out
/E
in
) 0.194 Az 160.31 45.96 56.54
In 30.36 74.04 84.95
Rc 0.90 0.57 0.66
Pl 0.97 0.90 0.85
W
7,8
v
7,8,1
0.76 0.64 0.02
f
min
f
max
1.561.76 En 1.9e07 1.5e07 5.1e08
M(W
j,n
) 0.123 Lag 0.10 0.12 0.22
log
10
(E
out
/E
in
) 0.081 Az 166.97 53.89 116.45
In 28.01 71.45 87.99
Rc 0.97 0.79 0.86
Pl 0.96 0.83 0.97
W
6,5
v
6,5,1
0.80 0.53 0.27
f
min
f
max
1.952.34 En 2.5e07 1.2e07 5.4e08
M(W
j,n
) 0.303 Lag 0.02 0.02 0.04
log
10
(E
out
/E
in
) 0.544 Az 173.09 44.29 73.91
In 24.75 69.19 89.27
Rc 0.98 0.74 0.86
Pl 0.98 0.81 0.97
W
7,12
v
7,12,1
0.82 0.57 0.07
f
min
f
max
2.342.54 En 4.4e08 2.9e08 7e09
M(W
j,n
) 0.375 Lag 0.04 0.12 0.16
log
10
(E
out
/E
in
) 0.139 Az 160.73 163.14 69.21
In 33.73 87.01 71.88
Rc 0.90 0.78 0.79
Pl 0.97 0.90 0.85
W
7,24
v
7,24,1
0.86 0.52 0.01
f
min
f
max
4.694.88 En 3.8e09 2.8e09 5.4e10
M(W
j,n
) 0.615 Lag 0.18 0.10 0.28
log
10
(E
out
/E
in
) 0.040 Az 179.80 8.05 55.94
In 45.68 80.52 77.74
Rc 0.70 0.77 0.48
Pl 0.90 0.89 0.55
L(2)
W
6,0
v
6,0,1
1.00 0.05 0.00
f
min
f
max
0.000.39 En 1.1e08 4.9e09 2.1e09
M(W
j,n
) 0.007 Lag 0.02 0.16 0.14
log
10
(E
out
/E
in
) 0.708 Az 158.09 20.01 122.18
In 8.23 86.36 80.80
Rc 0.62 0.65 0.63
Pl 0.59 0.88 0.63
W
7,14
v
7,14,1
1.00 0.00 0.00
f
min
f
max
2.732.93 En 1.5e09 5.9e10 1.8e10
M(W
j,n
) 0.367 Az 149.44 111.71 116.15
log
10
(E
out
/E
in
) 0.418 In 28.95 85.83 83.82
Rc 0.54 0.75 0.68
Pl 0.44 0.79 0.78
W
4,2
v
4,2,1
1.00 0.00 0.00
f
min
f
max
3.124.69 En 4.6e10 8.4e11 3.4e11
M(W
j,n
) 0.201 Az 161.15 65.41 103.52
log
10
(E
out
/E
in
) 0.355 In 39.93 88.80 82.91
Rc 0.54 0.76 0.60
Table 1 (continued)
Wavelet packet v
wavelet
Pl 0.55 0.70 0.88
W
7,31
v
7,31,1
1.00 0.00 0.00
f
min
f
max
6.056.25 En 4.3e10 3.3e11 1.8e11
M(W
j,n
) 0.614 Az 166.25 69.87 91.47
log
10
(E
out
/E
in
) 0.045 In 34.78 85.75 83.46
Rc 0.66 0.80 0.56
Pl 0.67 0.82 0.90
W
2,1
v
2,1,1
0.99 0.14 0.01
f
min
f
max
6.2512.50 En 2.6e07 3.8e08 1.9e08
M(W
j,n
) 0.641 Lag 0.10 0.18 0.28
log
10
(E
out
/E
in
) 0.259 Az 157.88 164.44 68.02
In 31.66 88.38 89.43
Rc 0.93 0.54 0.67
Pl 0.96 0.74 0.75
W
2,2
v
2,2,1
1.00 0.00 0.00
f
min
f
max
12.5018.75 En 8.6e08 3.4e10 2.3e10
M(W
j,n
) 0.506 Az 95.26 65.96 68.87
log
10
(E
out
/E
in
) 0.260 In 87.61 85.68 76.79
Rc 0.31 0.61 0.21
Pl 0.46 0.50 0.32
W
3,6
v
3,6,1
1.00 0.06 0.00
f
min
f
max
18.7521.88 En 4.2e08 9.4e09 6.1e09
M(W
j,n
) 0.258 Lag 0.04 0.26 0.30
log
10
(E
out
/E
in
) 0.494 Az 164.27 40.42 54.76
In 37.15 74.83 85.66
Rc 0.87 0.70 0.80
Pl 0.97 0.63 0.89
W
3,7
v
3,7,1
0.98 0.20 0.03
f
min
f
max
21.8825.00 En 1.1e09 4.9e10 2.3e10
M(W
j,n
) 0.120 Lag 0.06 0.16 0.10
log
10
(E
out
/E
in
) 0.909 Az 20.72 14.20 128.37
In 10.72 84.66 75.98
Rc 0.77 0.81 0.73
Pl 0.74 0.89 0.67
L(3)
W
7,2
v
7,2,1
0.59 0.75 0.29
f
min
f
max
0.390.59 En 6.7e09 8.5e09 2.2e09
M(W
j,n
) 0.323 Az 170.34 79.89 53.33
log
10
(E
out
/E
in
) 0.061 In 55.42 87.45 65.77
Rc 0.59 0.95 0.87
Pl 0.47 0.94 0.84
W
7,6
v
7,6,1
0.47 0.86 0.21
f
min
f
max
1.171.37 En 6.2e08 1.2e07 1.1e08
M(W
j,n
) 0.288 Lag 0.20 0.02 0.18
log
10
(E
out
/E
in
) 0.170 Az 167.14 67.01 22.29
In 30.11 77.43 87.98
Rc 0.92 0.88 0.54
Pl 0.94 0.89 0.93
L(4)
W
7,3
v
7,3,1
0.43 0.71 0.55
f
min
f
max
0.590.78 En 9.1e09 2.4e08 1.8e08
M(W
j,n
) 0.368 Lag 0.12 0.10 0.22
log
10
(E
out
/E
in
) 0.170 Az 179.54 97.63 53.46
In 43.90 87.27 53.25
Rc 0.89 0.88 0.94
Pl 0.86 0.97 0.91
L(5)
W
7,4
v
7,4,1
0.28 0.86 0.42
f
min
f
max
0.780.98 En 1.3e08 3.1e08 1.7e08
M(W
j,n
) 0.568 Lag 0.14 0.18 0.04
log
10
(E
out
/E
in
) 0.435 Az 173.40 121.89 52.51
In 29.05 80.35 49.56
Rc 0.89 0.85 0.88
Pl 0.95 0.93 0.96
L(6)
W
7,5
v
7,5,1
0.54 0.84 0.00
f
min
f
max
0.981.17 En 4.3e08 9.4e08 7.2e09
M(W
j,n
) 0.107 Lag 0.12 0.10 0.00
log
10
(E
out
/E
in
) 0.655 Az 176.25 93.68 92.81
In 36.12 87.88 59.06
Rc 0.92 0.83 0.56
depression of northeast Ethiopia (Fig. 5). The summit caldera features
two pit craters, the southernmost of which held a persistent, active
lava lake from (at least) 1967 through late 2004 (Martini, 1969;
Oppenheimer and Francis, 1998; Bardintzeff and Gaudru, 2004).
During a pilot study in February, 2002, seismic, thermal, and video
data were collected for 5 days, to better understand the dynamics of
the shallow magma system that fed the southern crater's persistent
lava lake (Harris et al., 2005; Jones et al., 2006). The three-station
seismic array geometry of the 2002 experiment is shown in Fig. 5.
The 2002 campaign found that the lava lake uctuated between
two convective regimes, characterized by low (0.010.08 m s
1
) and
high (0.10.4 m s
1
) velocities of cooled crust on the lava lake
surface, which corresponded to sluggish and vigorous convection,
respectively (Harris et al., 2005). The persistent, continuous tremor
was described by Jones et al. (2006), in which we found distinct
spectral characteristics corresponding to each convective regime
(Fig. 6). Because active lava lakes can be considered the exposed
upper surface of a convecting magma column (Swanson et al., 1979;
Harris et al., 1999), these changes could be explained by cooling and
degassing processes in the shallowpart of the exposed conduit (Harris
et al., 2005). The modeling work of Harris (2008) and the observations
of Harris et al. (2005) strongly suggest that magma feeds the
convecting lava lake at a relatively steady supply rate and constant
viscosity, and the changing rate of convection was driven by shallow
processes within the lava lake.
Because the model of Erta 'Ale's conduit convection is well
supported by observations, thermal modeling, and existing analysis,
it is a suitable test of the SDR method to examine whether or not the
method corroborates these results. If the modeling work of Harris
(2008) and the observations of Jones et al. (2006) are correct, then we
expect to recover at least one, nearly identical seismic signal, from
both convective regimes, which represents the response of a
subsurface conduit to the steady ow of fresh, hot, gas-rich magma
from a deeper reservoir. We expect further that during a period of
high-velocity convection in the lava lake, we will recover signals
whose measurable properties correspond to shallow sources within
the lava lake itself.
Prior to analysis, all data were downsampled to 50 Hz for computa-
tional efciency, and a 3 s cosine taper was applied. As in Jones et al.
(2006), the instrument response of all stations was corrected to match
that of an LE3D sensor (f
0
=0.2 Hz). Finally, to prevent low-frequency
artifacts that might result from this convolution, all data recorded by the
L22 geophone was high-pass ltered using a 4-pole Butterworth lter
with corner frequency 0.4 Hz.
8.2.1. 2002 tremor recorded during slow convection
We begin with the quiescent, low convective regime described in
Harris et al. (2005), i.e. those periods characterized by lava lake
convection of 0.010.08 ms
1
. Fig. 6a shows the raw data sample and a
spectrogram from station L22. Observe that the signal's spectral energy
contains few transients and is concentrated mostly below 5 Hz. Fig. 7
shows a dendrogram of the chosen wavelet decomposition, with the
nominal corner frequencies of eachsubbandassociatedwitheachwavelet
packet W
j,n
used to label the appropriate nodes. Fig. 7 shows the wavelet
basis that best represents this decomposition, with wavelet packets W
j,n
positioned according to wavelet level j andband n. Eachwavelet packet is
arranged to ll the associated passband, and shaded so that all wavelet
packets with the same shade belong to the same cluster. From Fig. 7, we
see that our algorithm recovers 9 signals, to which we henceforth refer
using the conventionL
(n)
, nbeing anarbitrarily assignednumber for each
recovered signal. From Fig. 7, we see that most of these are narrowband
signals whose energy is concentrated below3.13 Hz, i.e. in the regions of
the spectrum where seismic energy is highest.
We wish now to discuss this decomposition in detail, and focus in
particular on its implications for the tremor sources. The frequency
content, wavelet polarization, subband energy, cost M(W
j,n
), spectral
Table 1 (continued)
Wavelet packet v
wavelet
Pl 0.94 0.94 0.85
W
7,25
v
7,25,1
0.38 0.91 0.13
f
min
f
max
4.885.08 En 1.2e09 2.4e09 5.1e10
M(W
j,n
) 0.564 Lag 0.10 0.20 0.30
log
10
(E
out
/E
in
) 0.040 Az 36.49 3.92 58.76
In 30.54 87.48 64.04
Rc 0.53 0.83 0.61
Pl 0.68 0.86 0.69
L(7)
W
7,9
v
7,9,1
0.82 0.50 0.28
f
min
f
max
1.761.95 En 9.5e08 5.7e08 2.9e08
M(W
j,n
) 0.473 Lag 0.14 0.22 0.08
log
10
(E
out
/E
in
) 0.228 Az 179.88 97.73 73.19
In 16.43 69.55 88.19
Rc 0.89 0.58 0.85
Pl 0.97 0.85 0.96
W
7,29
v
7,29,1
0.82 0.57 0.08
f
min
f
max
5.665.86 En 1.1e09 9.3e10 2.1e10
M(W
j,n
) 0.814 Lag 0.02 0.06 0.04
log
10
(E
out
/E
in
) 0.155 Az 65.97 162.16 121.90
In 25.66 89.20 87.91
Rc 0.65 0.82 0.62
Pl 0.68 0.95 0.65
W
7,30
v
7,30,1
0.79 0.61 0.06
f
min
f
max
5.866.05 En 1e09 9.2e10 2e10
M(W
j,n
) 0.779 Lag 0.16 0.00 0.16
log
10
(E
out
/E
in
) 0.056 Az 64.57 15.21 135.41
In 12.07 87.69 68.38
Rc 0.69 0.87 0.63
Pl 0.63 0.94 0.62
L(8)
W
7,13
v
7,13,1
0.89 0.44 0.13
f
min
f
max
2.542.73 En 3e08 1.3e08 6.4e09
M(W
j,n
) 0.356 Lag 0.14 0.14 0.26
log
10
(E
out
/E
in
) 0.237 Az 176.35 169.59 70.10
In 60.36 80.20 79.14
Rc 0.87 0.54 0.83
Pl 0.97 0.49 0.81
W
7,15
v
7,15,1
0.96 0.27 0.00
f
min
f
max
2.933.12 En 5.8e08 1.4e08 6.1e09
M(W
j,n
) 0.253 Lag 0.12 0.14 0.26
log
10
(E
out
/E
in
) 0.204 Az 155.27 30.83 140.20
In 31.85 83.79 84.67
Rc 0.94 0.63 0.83
Pl 0.98 0.83 0.84
W
7,27
v
7,27,1
0.95 0.32 0.01
f
min
f
max
5.275.47 En 1e09 6.1e10 2e10
M(W
j,n
) 0.715 Lag 0.14 0.06 0.08
log
10
(E
out
/E
in
) 0.140 Az 30.67 21.86 104.19
In 34.67 84.17 81.52
Rc 0.67 0.77 0.52
Pl 0.77 0.85 0.75
W
7,28
v
7,28,1
0.97 0.21 0.10
f
min
f
max
5.475.66 En 1.3e09 8.7e10 2.5e10
M(W
j,n
) 0.816 Lag 0.06 0.06 0.12
log
10
(E
out
/E
in
) 0.163 Az 50.20 22.70 128.22
In 26.45 87.82 81.33
Rc 0.74 0.84 0.60
Pl 0.78 0.92 0.75
L(9)
W
7,26
v
7,26,1
0.02 0.99 0.16
f
min
f
max
5.085.27 En 9.6e10 1.5e09 2.3e10
M(W
j,n
) 0.769 Lag 0.08 0.20 0.28
log
10
(E
out
/E
in
) 0.223 Az 27.81 179.59 92.89
In 51.57 89.04 69.32
Rc 0.57 0.87 0.42
Pl 0.68 0.90 0.56
Fig. 8. Trace data, spectrograms, and azimuths of recovered signal L
(1)
fromthe lowconvective regime. a. (Top left) The recovered signal L
(1)
. Amplitudes of each trace are normalized
to illustrate detail. b. (Top right) Spectrograms of the input Z data at station L22 (above) and the Z component of recovered signal L
(1)
at L22 (below). Intensity scaling is in dB. Color
scaling of input data is computed from true ground velocity. c. (Bottom) Azimuths of subbands W
j,n
that form recovered signal L
(1)
are superimposed on an aerial photograph
showing station locations and relevant physical features in the Erta 'Ale summit caldera. Azimuths are scaled according to relative energy of each W
j,n
at each station. The largest
azimuth vector corresponds to W
7,8
at L22, whose nominal passband is 1.561.76 Hz. Boxes on spectrograms indicate the nominal passbands associated with the wavelet coefcients
whose detail coefcients form the recovered signal.
leakage, and principal eigenvectors for select recovered signals are
given in Table 1.
We begin with the recovered signal whose spectral energy is
highest. Signal L
(1)
(Fig. 8a) is formed from 5 clustered wavelet
packets: W
7,7
(nominal passband 1.361.56 Hz), W
7,8
(1.561.76 Hz),
W
6,5
(1.952.34 Hz), W
7,12
(2.342.54 Hz), and W
7,24
(4.694.88 Hz).
It contains the most energetic spectral peak of the input data at
stations L22 and CMG (e.g. Fig. 6a). Spectrograms of the Z component
of the reconstructed signal, and the Z component of the input data, are
shown in Fig. 8b. The recovered signal's subbands are highly
rectilinear at station L22, with rectilinearity (cf. Jurkevics, 1988)
Rc=0.900.98 for frequencies below2.53 Hz. Its polarizations at MAR
and CMG are also rectilinear for some subbands. The spread in
azimuths of this recovered signal's subbands, noting the 180
ambiguity, is only 33. Fig. 8c shows a plot of the azimuths of the
subbands that form this recovered signal, in which azimuth vectors
Fig. 9. Spectrograms and azimuths of recovered signal L
(3)
fromthe lowconvective regime. a. (Upper left) Spectrograms of the input N data at station L22 (top) and the Z component
of recovered signal L
(4)
(bottom). b. (Upper right) Spectrograms of the input E data at station CMG (top) and the Z component of recovered signal L
(4)
(bottom). Scaling intensity is
in dB computed from true ground velocity. c. (Bottom) Azimuths of W
j,n
that form recovered signal L
(3)
are superimposed on an aerial photograph showing station locations and
relevant physical features in the Erta 'Ale summit caldera. Azimuths are scaled according to relative energy of each W
j,n
at each station.
are rescaled to the relative energy of that subband at that station; thus
the largest azimuth vector corresponds to W
7,8
, whose nominal
passband is 1.561.76 Hz. Notably this subband is also the most
rectilinear.
The computed incidence angles of these subbands are 2433 from
vertical at L22. If we assume that the source underlies this part of the
crater, and that the azimuth points toward the source, then simple
trigonometry (from Fig. 5) constrains the maximum depth to about
100 masl, or about 420 m below the crater oor. If we further assume
that the source of this tremor is related to fresh, hot magma upwelling
roughly in the center of the lava lake (Harris et al., 2005), then the
source is probably 40100 mbelowthe lava lake surface. This range of
depths corresponds to estimates of the lava lake depth in Harris et al.
(2005) and Oppenheimer and Francis (1998).
Unfortunately, these source depths contraindicate expanding on or
recreating the location work of Jones et al. (2006). 3 stations cannot
adequately constrain a tremor centroid in 3 dimensions, and neglecting
depths comparable to (or greater than) the nearest epicentral distance
canintroduce grievous errors tothe calculation. However, we canobtain
some additional clues about this recovered signal's source by examining
its spectrogram. From Fig. 8b, observe that the peaks of this signal
correspond roughly to two sets of harmonics: One set of spectral peaks
(i.e. fundamental and rst overtone) at 1.8 and 3.6 Hz, one at 2.4 and
4.8 Hz. We note that the narrowbandenergy centered at 3.6 Hz is not an
aliasing phenomenon, but a result of wavelet lters not being ideal
bandpass lters (see e.g. Percival and Walden, 2000, Ch. 4). It could be
that these are real harmonics, and that their spectra lack very sharp
peaks merely because many other signals are superimposed on themat
similar frequencies.
It is not the case that all recovered signals appear to originate in
the lava lake. An excellent counter-example of this is shown in Fig. 9
for signal L
(3)
, formed from wavelet packets W
7,2
(0.390.59 Hz) and
W
7,6
(1.171.37 Hz). The spectrogram of this recovered signal
(Fig. 9b) suggests the recovered signal consists of three harmonics
(1st, 3rd, and 5th) with a fundamental overtone f =0.45 Hz. It may be
that these are true harmonics mixed with other signals that comprise
the observed tremor, while the other harmonics are masked by the
intense energy of L
(1)
. Observe from Table 1 that the energy of this
recovered signal's subbands is greatest at station CMG, which is also
where the subbands are most rectilinearly polarized, and that the
azimuths of this signal's constituent subbands point toward the old
(north) crater, which was a source of fumarolic degassing in 2002
(Harris et al., 2005). Incidence angles at station CMG (Fig. 8b) range
from 77 to 87, suggesting that its source is shallow. At station L22,
azimuths and incidence angles of the recovered signal Z component
point to the lava lake; this is consistent with a weak source in the
north crater being masked by the intense energy of tremor from the
lava lake. Notably, recovered signals L
(4)
, L
(5)
, and L
(6)
have similar
features, though obviously lack the apparent harmonics of L
(3)
.
One additional, important conclusion can be drawn from analysis of
Table 1. During the slow lava lake convection, there is no clear
evidence of multiple signals whose sources are in or near the lava lake.
The signals L
(1)
and L
(9)
, each of which appear to originate near station
L22, cannot beunambiguously associatedwithuniquesources. It maybe
that these signals all originate from a single tremor source, and
following the surface morphology described in e.g. Oppenheimer and
Francis (1998) undergo phase conversions in the complex structure
that underlies the Erta 'Ale summit caldera. There is, however, good
evidence that four signals (L
(3)
throughL
(6)
) containenergy at CMGthat
originates elsewhere.
8.2.2. 2002 tremor during rapid convection
We now turn to tremor related to more rapid convection of the
lava lake in 2002. This convective regime was characterized by Harris
et al. (2005) by more rapid surface velocities of the lava lake, ranging
from 0.1 to 0.4 ms
1
, corresponding to vigorous overturn of cooled
crust on the lava lake surface, with frequent episodes of very small
lava fountains. Fig. 5b shows sample data and a spectrogram for the
high convective regime at station L22. Note the increase in t high
frequency energy (f N5 Hz) during the high convective regime.
Fig. 10 shows SDR analysis of the data sample of Fig. 6b. Table 2
tabulates the frequency content, wavelet polarization, subband
energy, cost M(W
j,n
), spectral leakage, and principal eigenvectors for
selected recovered signals from this convective phase. Note that SDR
recovers 16 signals (designated H
(n)
), suggesting more seismic
sources are present in the high convective regime. This is consistent
with the modeling work of Harris et al. (2005) and Harris (2008).
Fig. 10 shows a dendrogram of the chosen wavelet decomposition,
with the nominal corner frequencies of each subband associated with
each wavelet packet W
j,n
used to label the appropriate nodes.
We wish to focus on which signals have changed, and which signals
persist, between the convective regimes. As noted above, our denition
of a recovered signal enables us to track signal persistence in a very
straightforward way. Comparing the principal eigenvectors v
j,n,1
of
Table 1withthose of Table 2 shows several examples: L
(1)
, L
(2)
, L
(3)
, L
(4)
,
and L
(6)
persist (and are renamed in the high regime) as H
(1)
(and
Frequency [Hz]
W
a
v
e
l
e
t

S
c
a
l
e
3.12 6.25 9.38 12.50 15.62 18.75 21.88 25.00
0
1
2
3
4
5
6
7
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
.
0
0
.
4
1
6
.
0
1
6
.
4
1
8
.
8
2
5
.
0
2
.
7
3
.
1
7
.
8
9
.
4
1
8
.
0
1
8
.
8
1
4
.
1
1
4
.
3
3
.
1
6
.
2
1
5
.
4
1
5
.
6
6
.
6
6
.
8
1
6
.
8
1
7
.
0
2
.
0
2
.
3
1
5
.
6
1
5
.
8
1
7
.
0
1
7
.
2
7
.
2
7
.
4
7
.
4
7
.
6
1
7
.
2
1
8
.
0
1
2
.
3
1
2
.
5
1
5
.
0
1
5
.
2
1
6
.
4
1
6
.
8
6
.
2
6
.
4
1
2
.
5
1
4
.
1
6
.
8
7
.
0
1
5
.
8
1
6
.
0
1
2
.
1
1
2
.
3
1
5
.
2
1
5
.
4
1
1
.
7
1
2
.
1
I
n
p
u
t

D
a
t
a
2
.
3
2
.
7
1
.
4
1
.
6
1
.
6
1
.
8
6
.
4
6
.
6
7
.
0
7
.
2
7
.
6
7
.
8
9
.
4
9
.
6
9
.
8
1
0
.
0
9
.
6
9
.
8
1
.
8
2
.
0
1
4
.
6
1
4
.
8
0
.
4
0
.
6
0
.
8
1
.
0
1
.
0
1
.
2
1
.
2
1
.
4
0
.
6
0
.
8
1
0
.
0
1
0
.
2
1
0
.
9
1
1
.
1
1
0
.
7
1
0
.
9
1
1
.
1
1
1
.
3
1
0
.
2
1
0
.
4
1
0
.
4
1
0
.
5
1
1
.
5
1
1
.
7
1
0
.
5
1
0
.
7
1
4
.
8
1
5
.
0
1
1
.
3
1
1
.
5
1
4
.
3
1
4
.
5
1
4
.
5
1
4
.
6
F
i
r
s
t

E
i
g
e
n
v
e
c
t
o
r

D
i
s
t
a
n
c
e
Fig. 10. Decomposition of the data sample in Fig. 6b using the SDR algorithm. Top
Dendrogram of subband clustering. Horizontal dashed line indicates distance threshold
distance =0.3 for clustering of principal components eigenvectors. Labels give the
nominal passband in the interval [0,f
n
] associated with each wavelet packet. Bottom
Wavelet basis selected by subband clustering. Selected wavelet packets are shaded in
grayscale to showclustering. Y-axis shows wavelet scale j, with j =0 corresponding to the
wideband (input) signal. X-axis shows the position in the wavelet packet table (and
nominal passband) for each wavelet packet in the interval [0,f
n
].
Table 2
Recovered signals from a data sample recorded during the high convective regime in
2002, beginning 15 Feb 2002, 19:53:23 GMT (Fig. 6b). Signal names use the convention
L
(n)
, described in the text. Values for eigenvectors v
j,n,1
correspond to the largest
eigenvalue
j,n,1
of the principal components of each wavelet packet W
j,n
. The quantities
v
j,n
, energy (En), azimuth (Az), incidence angle (In), rectilinearity (Rc), and planarity
(Pl) are arranged in columns by station. Az and In are given in degrees. Wavelet packets
are designated W
j,n
and nominal passbands f
min
f
max
for each W
j,n
are given in Hz. Cost
functional M(W
j,n
) is renormalized (multiplied by normalized bandwidth 2
j +1
) so that
0M(W)b3. Spectral leakage is tabulated in the form log
10
(E
out
/E
in
), the base 10
logarithm of spectral energy outside the passband to spectral energy inside the
passband. Relative lags of each subband are given in seconds for subbands whose
wavelet cross-correlations are well constrained.
Wavelet packet v
wavelet
f
min
f
max
(Hz) En (A.U.)
M(W
j,n
) Lag (s)
log
10
(E
out
/E
in
) Az (deg)
In (deg)
Rc (0rc1)
Pl (0pl 1)
H(1)
W
6,6
v
6,6,1
0.94 0.33 0.02
f
min
f
max
2.342.73 En 6.7e08 2.4e08 1.1e08
M(W
j,n
) 0.398 Lag 0.02 0.12 0.14
log
10
(E
out
/E
in
) 0.307 Az 166.32 169.90 60.62
In 35.59 83.39 79.27
Rc 0.88 0.64 0.73
Pl 0.96 0.76 0.65
H(2)
W
6,0
v
6,0,1
1.00 0.01 0.01
f
min
f
max
0.000.39 En 1.4e09 1.7e10 1.8e10
M(W
j,n
) 0.009 Az 160.14 112.65 61.93
log
10
(E
out
/E
in
) 0.613 In 25.71 88.93 89.25
Rc 0.68 0.74 0.60
Pl 0.57 0.69 0.76
W
6,5
v
6,5,1
0.99 0.13 0.02
f
min
f
max
1.952.34 En 4.4e07 6.1e08 7.1e08
M(W
j,n
) 0.608 Lag 0.16 0.14 0.28
log
10
(E
out
/E
in
) 0.451 Az 167.73 95.96 52.38
In 39.20 87.27 53.20
Rc 0.82 0.83 0.92
Pl 0.75 0.94 0.90
W
6,7
v
6,7,1
0.99 0.04 0.11
f
min
f
max
2.733.12 En 1.5e08 2.3e09 5.3e09
M(W
j,n
) 0.352 Lag 0.12 0.06 0.08
log
10
(E
out
/E
in
) 0.162 Az 148.24 37.29 115.99
In 25.49 87.96 75.82
Rc 0.81 0.53 0.59
Pl 0.71 0.89 0.67
W
3,1
v
3,1,1
0.99 0.13 0.10
f
min
f
max
3.126.25 En 1.9e10 7.7e11 7.1e11
M(W
j,n
) 0.281 Lag 0.02 0.06 0.08
log
10
(E
out
/E
in
) 0.812 Az 166.09 114.96 73.58
In 5.98 85.42 85.92
Rc 0.45 0.78 0.65
Pl 0.41 0.74 0.71
W
7,32
v
7,32,1
1.00 0.05 0.08
f
min
f
max
6.256.45 En 1.4e10 4.7e11 6.8e11
M(W
j,n
) 1.032 Az 163.38 116.12 75.19
log
10
(E
out
/E
in
) 0.028 In 27.82 87.38 80.09
Rc 0.44 0.70 0.78
Pl 0.46 0.61 0.83
W
7,34
v
7,34,1
1.00 0.00 0.00
f
min
f
max
6.646.84 En 4.7e08 2.4e10 1.6e10
M(W
j,n
) 0.799 Az 136.24 127.93 147.50
log
10
(E
out
/E
in
) 0.168 In 26.97 22.63 32.54
Rc 0.65 0.72 0.65
Pl 0.60 0.70 0.57
W
7,37
v
7,37,1
0.99 0.15 0.06
f
min
f
max
7.237.42 En 9.7e08 3.8e08 2.4e08
M(W
j,n
) 0.840 Lag 0.10 0.08 0.04
log
10
(E
out
/E
in
) 0.003 Az 163.77 71.24 127.05
In 31.35 67.27 85.79
Rc 0.90 0.53 0.79
(continued on next page)
Table 2 (continued)
Wavelet packet v
wavelet
Pl 0.96 0.89 0.82
W
7,38
v
7,38,1
1.00 0.07 0.02
f
min
f
max
7.427.62 En 7.7e08 1.7e08 1.1e08
M(W
j,n
) 0.618 Lag 0.12 0.08 0.20
log
10
(E
out
/E
in
) 0.232 Az 158.35 39.39 45.30
In 33.44 82.76 87.99
Rc 0.90 0.65 0.78
Pl 0.96 0.74 0.82
W
4,5
v
4,5,1
1.00 0.00 0.00
f
min
f
max
7.819.38 En 1.9e10 7.9e11 7.9e11
M(W
j,n
) 0.510 Lag 0.14 0.06 0.18
log
10
(E
out
/E
in
) 0.171 Az 142.85 115.02 73.15
In 29.99 83.29 87.42
Rc 0.71 0.74 0.69
Pl 0.56 0.74 0.67
W
7,63
v
7,63,1
0.95 0.32 0.03
f
min
f
max
12.3012.50 En 1.7e10 9e11 6.7e11
M(W
j,n
) 1.142 Lag 0.04 0.18 0.14
log
10
(E
out
/E
in
) 0.048 Az 138.15 122.75 58.16
In 26.16 82.09 87.94
Rc 0.66 0.84 0.66
Pl 0.58 0.82 0.70
W
7,72
v
7,72,1
0.96 0.26 0.02
f
min
f
max
14.0614.26 En 2e09 1.1e09 1.1e09
M(W
j,n
) 1.198 Lag 0.08 0.20 0.28
log
10
(E
out
/E
in
) 0.546 Az 74.98 29.96 151.35
In 23.90 76.28 45.53
Rc 0.68 0.77 0.78
Pl 0.68 0.91 0.87
W
7,77
v
7,77,1
0.99 0.08 0.07
f
min
f
max
15.0415.23 En 2.7e09 1e09 1.2e09
M(W
j,n
) 1.164 Lag 0.14 0.12 0.02
log
10
(E
out
/E
in
) 0.215 Az 166.52 38.99 151.05
In 19.18 85.87 45.07
Rc 0.59 0.69 0.62
Pl 0.70 0.92 0.79
W
7,79
v
7,79,1
1.00 0.07 0.05
f
min
f
max
15.4315.62 En 2.1e09 6.4e10 1.1e09
M(W
j,n
) 0.813 Lag 0.06 0.02 0.04
log
10
(E
out
/E
in
) 0.277 Az 159.50 21.30 72.85
In 38.32 88.93 88.67
Rc 0.55 0.70 0.66
Pl 0.67 0.91 0.65
W
7,80
v
7,80,1
0.99 0.08 0.07
f
min
f
max
15.6215.82 En 2e09 4.3e10 8.5e10
M(W
j,n
) 0.817 Lag 0.02 0.14 0.12
log
10
(E
out
/E
in
) 0.404 Az 174.55 17.01 131.72
In 43.90 85.34 64.28
Rc 0.65 0.67 0.58
Pl 0.63 0.91 0.58
W
6,41
v
6,41,1
0.98 0.19 0.03
f
min
f
max
16.0216.41 En 9.1e11 6.8e11 3.8e11
M(W
j,n
) 0.813 Az 138.24 81.63 71.97
log
10
(E
out
/E
in
) 0.166 In 42.48 89.59 84.30
Rc 0.39 0.80 0.34
Pl 0.27 0.84 0.53
W
6,42
v
6,42,1
1.00 0.01 0.06
f
min
f
max
16.4116.80 En 2.1e10 7.6e11 1.8e10
M(W
j,n
) 0.809 Lag 0.00 0.18 0.16
log
10
(E
out
/E
in
) 0.012 Az 128.80 108.13 114.80
In 33.38 84.57 80.31
Rc 0.72 0.75 0.87
Pl 0.69 0.83 0.83
W
7,86
v
7,86,1
0.99 0.16 0.02
f
min
f
max
16.8016.99 En 9.4e11 4.5e11 6.6e11
M(W
j,n
) 0.680 Lag 0.10 0.06 0.06
log
10
(E
out
/E
in
) 0.039 Az 157.88 70.59 74.32
In 23.63 88.03 87.67
Rc 0.70 0.72 0.81
Pl 0.61 0.76 0.82
W
7,87
v
7,87,1
0.98 0.19 0.03
f
min
f
max
16.9917.19 En 1.1e10 4.8e11 4.8e11
M(W
j,n
) 0.585 Lag 0.10 0.08 0.00
log
10
(E
out
/E
in
) 0.084 Az 147.84 56.41 113.83
Table 2 (continued)
Wavelet packet v
wavelet
In 29.20 88.77 79.42
Rc 0.73 0.77 0.75
Pl 0.59 0.81 0.69
W
5,22
v
5,22,1
0.99 0.14 0.10
f
min
f
max
17.1917.97 En 1.1e10 4.6e11 4.8e11
M(W
j,n
) 0.738 Lag 0.14 0.20 0.06
log
10
(E
out
/E
in
) 0.118 Az 145.01 121.83 112.37
In 28.48 89.92 77.28
Rc 0.73 0.75 0.76
Pl 0.58 0.79 0.68
W
5,23
v
5,23,1
0.98 0.11 0.13
f
min
f
max
17.9718.75 En 8.2e11 3.3e11 2.6e11
M(W
j,n
) 0.821 Az 157.10 128.56 115.84
log
10
(E
out
/E
in
) 0.151 In 12.87 86.24 81.83
Rc 0.58 0.80 0.79
Pl 0.63 0.84 0.77
W
2,3
v
2,3,1
0.98 0.13 0.13
f
min
f
max
18.7525.00 En 8.2e11 2.8e11 2.3e11
M(W
j,n
) 0.240 Lag 0.16 0.12 0.28
log
10
(E
out
/E
in
) 0.709 Az 155.87 111.01 112.80
In 11.52 84.38 80.03
Rc 0.65 0.84 0.67
Pl 0.61 0.85 0.71
H(3)
W
7,2
v
7,2,1
0.56 0.77 0.31
f
min
f
max
0.390.59 En 4.4e09 6.3e09 1.7e09
M(W
j,n
) 0.277 Az 176.73 78.63 53.95
log
10
(E
out
/E
in
) 0.017 In 53.17 87.16 65.11
Rc 0.58 0.93 0.86
Pl 0.48 0.93 0.83
H(4)
W
7,3
v
7,3,1
0.46 0.66 0.59
f
min
f
max
0.590.78 En 7.1e09 1.4e08 1.3e08
M(W
j,n
) 0.432 Lag 0.12 0.08 0.22
log
10
(E
out
/E
in
) 0.116 Az 167.73 95.96 52.38
In 39.20 87.27 53.20
Rc 0.82 0.83 0.92
Pl 0.75 0.94 0.90
H(5)
W
7,4
v
7,4,1
0.28 0.90 0.32
f
min
f
max
0.780.98 En 1.2e08 2.4e08 1.5e08
M(W
j,n
) 0.810 Lag 0.16 0.16 0.00
log
10
(E
out
/E
in
) 0.585 Az 168.33 127.51 52.40
In 25.36 82.38 46.80
Rc 0.77 0.77 0.82
Pl 0.67 0.85 0.74
H(6)
W
7,5
v
7,5,1
0.54 0.84 0.05
f
min
f
max
0.981.17 En 3.2e08 6.4e08 8.9e09
M(W
j,n
) 0.201 Lag 0.10 0.10 0.02
log
10
(E
out
/E
in
) 0.597 Az 176.88 99.79 86.55
In 33.72 88.97 55.52
Rc 0.86 0.74 0.61
Pl 0.77 0.89 0.62
W
7,6
v
7,6,1
0.51 0.86 0.09
f
min
f
max
1.171.37 En 4e08 6.8e08 1.4e08
M(W
j,n
) 0.452 Lag 0.16 0.06 0.10
log
10
(E
out
/E
in
) 0.183 Az 164.67 67.25 101.22
In 27.80 77.09 84.97
Rc 0.88 0.79 0.70
Pl 0.89 0.86 0.83
H(7)
W
7,7
v
7,7,1
0.77 0.62 0.14
f
min
f
max
1.371.56 En 1.3e07 9e08 4.2e08
M(W
j,n
) 0.270 Lag 0.02 0.00 0.02
log
10
(E
out
/E
in
) 0.122 Az 167.75 54.15 80.08
In 28.02 71.87 87.75
Rc 0.96 0.77 0.80
Pl 0.96 0.74 0.93
W
7,8
v
7,8,1
0.80 0.52 0.29
f
min
f
max
1.561.76 En 1.7e07 7.6e08 4.8e08
Table 2 (continued)
Wavelet packet v
wavelet
M(W
j,n
) 0.160 Lag 0.04 0.00 0.04
log
10
(E
out
/E
in
) 0.197 Az 173.58 44.62 87.67
In 25.26 70.31 86.48
Rc 0.97 0.72 0.84
Pl 0.97 0.73 0.95
H(8)
W
7,9
v
7,9,1
0.78 0.53 0.33
f
min
f
max
1.761.95 En 6.4e08 4.2e08 4.1e08
M(W
j,n
) 0.755 Lag 0.12 0.06 0.18
log
10
(E
out
/E
in
) 0.380 Az 176.36 111.31 112.15
In 13.21 71.37 86.27
Rc 0.85 0.59 0.88
Pl 0.95 0.83 0.91
W
7,75
v
7,75,1
0.80 0.35 0.49
f
min
f
max
14.6514.84 En 9.5e11 7.7e11 1e10
M(W
j,n
) 1.639 Lag 0.16 0.12 0.26
log
10
(E
out
/E
in
) 0.264 Az 178.47 100.60 75.71
In 41.67 82.51 82.02
Rc 0.58 0.79 0.86
Pl 0.54 0.85 0.81
H(9)
W
7,33
v
7,33,1
0.79 0.12 0.61
f
min
f
max
6.456.64 En 1.5e09 9.6e10 1.2e09
M(W
j,n
) 0.862 Lag 0.04 0.14 0.10
log
10
(E
out
/E
in
) 0.073 Az 69.27 29.16 136.95
In 31.72 82.59 58.61
Rc 0.37 0.74 0.72
Pl 0.62 0.89 0.72
W
7,36
v
7,36,1
0.76 0.16 0.63
f
min
f
max
7.037.23 En 2.3e09 5.9e10 1.9e09
M(W
j,n
) 0.702 Lag 0.06 0.04 0.08
log
10
(E
out
/E
in
) 0.112 Az 172.09 30.25 82.60
In 32.55 84.86 85.66
Rc 0.64 0.67 0.79
Pl 0.57 0.84 0.77
H(10)
W
7,35
v
7,35,1
0.95 0.14 0.27
f
min
f
max
6.847.03 En 1.4e09 5.8e10 9.9e10
M(W
j,n
) 0.959 Lag 0.00 0.16 0.18
log
10
(E
out
/E
in
) 0.207 Az 136.24 105.89 115.65
In 26.97 87.16 84.17
Rc 0.65 0.75 0.81
Pl 0.60 0.84 0.73
W
6,30
v
6,30,1
0.95 0.13 0.27
f
min
f
max
11.7212.11 En 4.3e10 1.3e10 2.4e10
M(W
j,n
) 0.776 Lag 0.06 0.02 0.08
log
10
(E
out
/E
in
) 0.007 Az 127.93 108.23 111.72
In 22.63 88.71 89.17
Rc 0.72 0.71 0.79
Pl 0.70 0.84 0.74
W
7,62
v
7,62,1
0.91 0.16 0.38
f
min
f
max
12.1112.30 En 2.2e09 6.4e10 1.7e09
M(W
j,n
) 0.757 Lag 0.12 0.06 0.06
log
10
(E
out
/E
in
) 0.032 Az 147.50 14.41 79.82
In 32.54 79.44 89.84
Rc 0.65 0.65 0.73
Pl 0.57 0.86 0.83
W
4,8
v
4,8,1
0.90 0.06 0.42
f
min
f
max
12.5014.06 En 1.6e10 5.2e11 1e10
M(W
j,n
) 1.073 Az 131.02 69.81 71.44
log
10
(E
out
/E
in
) 0.041 In 13.84 88.57 88.72
Rc 0.68 0.74 0.78
Pl 0.58 0.85 0.68
W
7,78
v
7,78,1
0.90 0.06 0.43
f
min
f
max
15.2315.43 En 1.2e10 4.2e11 8.4e11
M(W
j,n
) 0.878 Lag 0.08 0.08 0.00
log
10
(E
out
/E
in
) 0.219 Az 157.75 114.74 71.14
In 28.36 89.72 82.29
Rc 0.73 0.72 0.83
Pl 0.62 0.76 0.73
W
7,81
v
7,81,1
0.93 0.12 0.35
f
min
f
max
15.8216.02 En 1.4e10 4.8e11 7.5e11
H
(3)
), H
(2)
, H
(6)
, H
(4)
, and H
(6)
, respectively (note the ambiguity of H
(6)
meeting the mathematical criterion for persistence with two signals
fromthe low regime). These are signals most strongly associated with
the lava lake (L
(1)
), the signal whose principal eigenvectors most
strongly suggest transients at station L22 and scattered energy
elsewhere (L
(2)
), and several signals where station CMG appears
dominated by body waves from a source in the north crater. The
polarizationof eachrecoveredsignal remains roughly constant between
the two regimes, at those stations where the recovered signal energy is
largest. The signals H
(1)
and H
(3)
have similar principal eigenvectors,
and eachclusters to signal L
(1)
. That they donot cluster to one another is
simply a mathematical consequence of the hierarchical clustering
algorithm.
We also wish to discuss a few new recovered signals during the
high convective regime. Recovered signals H
(12)
H
(16)
may be
artifacts of transients at station MAR, a limitation which was already
discussed. However, signals H
(8)
H
(11)
have different principal
Table 2 (continued)
Wavelet packet v
wavelet
M(W
j,n
) 0.755 Lag 0.06 0.02 0.02
log
10
(E
out
/E
in
) 0.353 Az 154.02 120.32 75.16
In 31.01 88.94 87.37
Rc 0.78 0.73 0.84
Pl 0.65 0.78 0.78
H(11)
W
7,39
v
7,39,1
0.96 0.03 0.27
f
min
f
max
7.627.81 En 2.4e09 5.2e10 1.2e09
M(W
j,n
) 0.643 Lag 0.02 0.06 0.04
log
10
(E
out
/E
in
) 0.243 Az 173.28 151.61 117.92
In 27.76 89.68 74.10
Rc 0.77 0.70 0.64
Pl 0.75 0.90 0.79
W
7,48
v
7,48,1
0.90 0.08 0.43
f
min
f
max
9.389.57 En 6.4e10 1.7e10 3.8e10
M(W
j,n
) 0.684 Lag 0.08 0.06 0.02
log
10
(E
out
/E
in
) 0.015 Az 147.96 2.63 114.06
In 31.11 77.97 63.85
Rc 0.75 0.55 0.81
Pl 0.66 0.90 0.79
W
7,49
v
7,49,1
0.89 0.08 0.46
f
min
f
max
9.579.77 En 6.3e10 2e10 4.9e10
M(W
j,n
) 0.914 Lag 0.12 0.08 0.04
log
10
(E
out
/E
in
) 0.043 Az 170.10 86.57 109.67
In 33.34 87.48 86.00
Rc 0.76 0.69 0.80
Pl 0.63 0.84 0.79
W
7,50
v
7,50,1
0.90 0.06 0.44
f
min
f
max
9.779.96 En 3.8e10 1.6e10 3e10
M(W
j,n
) 1.050 Lag 0.08 0.14 0.04
log
10
(E
out
/E
in
) 0.051 Az 178.05 90.32 114.37
In 29.33 88.89 82.30
Rc 0.65 0.72 0.71
Pl 0.57 0.86 0.75
H(12)
W
7,51
v
7,51,1
0.54 0.18 0.82
f
min
f
max
9.9610.16 En 2.5e10 1.4e10 3.1e10
M(W
j,n
) 1.018 Lag 0.08 0.02 0.10
log
10
(E
out
/E
in
) 0.022 Az 179.56 83.18 72.38
In 82.28 86.71 89.35
Rc 0.53 0.77 0.73
Pl 0.72 0.88 0.77
W
7,55
v
7,55,1
0.57 0.02 0.82
f
min
f
max
10.7410.94 En 3.5e10 1.4e10 5e10
M(W
j,n
) 0.569 Lag 0.08 0.10 0.02
log
10
(E
out
/E
in
) 0.206 Az 100.37 86.44 114.38
In 46.92 84.66 75.71
Rc 0.64 0.75 0.88
Pl 0.68 0.83 0.88
W
7,56
v
7,56,1
0.43 0.23 0.87
f
min
f
max
10.9411.13 En 2.9e10 1.3e10 4.8e10
M(W
j,n
) 0.640 Lag 0.02 0.08 0.08
log
10
(E
out
/E
in
) 0.019 Az 105.62 85.51 112.14
In 53.22 81.84 75.28
Rc 0.57 0.76 0.88
Pl 0.64 0.83 0.85
H(13)
W
7,52
v
7,52,1
0.02 0.34 0.94
f
min
f
max
10.1610.35 En 1.9e10 1.7e10 4.2e10
M(W
j,n
) 0.719 Lag 0.12 0.16 0.28
log
10
(E
out
/E
in
) 0.091 Az 12.19 83.45 75.65
In 88.25 89.67 78.24
Rc 0.32 0.79 0.81
Pl 0.53 0.90 0.81
H(14)
W
7,53
v
7,53,1
0.15 0.19 0.97
f
min
f
max
10.3510.55 En 2.5e10 1.5e10 5.3e10
M(W
j,n
) 0.689 Lag 0.04 0.06 0.10
log
10
(E
out
/E
in
) 0.345 Az 139.03 64.49 72.55
In 44.01 80.20 79.31
Rc 0.54 0.74 0.83
Table 2 (continued)
Wavelet packet v
wavelet
Pl 0.58 0.88 0.84
W
7,54
v
7,54,1
0.31 0.11 0.94
f
min
f
max
10.5510.74 En 2.6e10 1.3e10 5.5e10
M(W
j,n
) 0.593 Lag 0.10 0.18 0.26
log
10
(E
out
/E
in
) 0.378 Az 107.14 98.68 113.46
In 46.86 89.29 82.61
Rc 0.58 0.72 0.86
Pl 0.59 0.83 0.88
W
7,59
v
7,59,1
0.15 0.24 0.96
f
min
f
max
11.5211.72 En 1.9e10 5.4e11 2.1e10
M(W
j,n
) 1.048 Lag 0.04 0.06 0.00
log
10
(E
out
/E
in
) 0.070 Az 165.42 14.31 110.45
In 27.21 78.00 85.58
Rc 0.61 0.47 0.81
Pl 0.65 0.86 0.83
W
7,76
v
7,76,1
0.43 0.25 0.87
f
min
f
max
14.8415.04 En 7.4e11 7.5e11 8e11
M(W
j,n
) 1.695 Lag 0.10 0.18 0.26
log
10
(E
out
/E
in
) 0.308 Az 173.16 98.59 75.06
In 40.52 82.11 85.58
Rc 0.54 0.82 0.84
Pl 0.60 0.87 0.79
H(15)
W
7,57
v
7,57,1
0.32 0.17 0.93
f
min
f
max
11.1311.33 En 1.7e10 1.1e10 2.8e10
M(W
j,n
) 0.867 Lag 0.04 0.22 0.18
log
10
(E
out
/E
in
) 0.098 Az 134.09 100.11 119.46
In 47.32 85.49 66.55
Rc 0.30 0.79 0.83
Pl 0.55 0.83 0.81
H(16)
W
7,58
v
7,58,1
0.33 0.12 0.94
f
min
f
max
11.3311.52 En 1.7e10 1e10 2.7e10
M(W
j,n
) 0.894 Lag 0.10 0.14 0.24
log
10
(E
out
/E
in
) 0.177 Az 153.85 93.86 116.89
In 32.15 88.33 79.07
Rc 0.39 0.74 0.89
Pl 0.55 0.83 0.90
W
7,73
v
7,73,1
0.63 0.04 0.78
f
min
f
max
14.2614.45 En 1.5e10 8e11 1.6e10
M(W
j,n
) 0.993 Lag 0.08 0.14 0.20
log
10
(E
out
/E
in
) 0.514 Az 128.97 108.63 112.01
In 32.81 85.73 86.19
Rc 0.69 0.79 0.86
Pl 0.64 0.84 0.83
W
7,74
v
7,74,1
0.51 0.05 0.86
f
min
f
max
14.4514.65 En 1.2e10 5.9e11 1.4e10
M(W
j,n
) 1.154 Lag 0.14 0.04 0.10
log
10
(E
out
/E
in
) 0.380 Az 161.93 66.56 72.18
In 33.46 89.01 81.81
Rc 0.66 0.75 0.85
Pl 0.57 0.81 0.82
eigenvectors from any signal seen during the low convection, and
are narrowband signals at high frequencies, where little energy exists
during low convection. Each shows primarily planar polarization,
though Z azimuths at L22 still suggest a source in or beneath the
active lava lake. These recovered signals are therefore consistent with
the model suggested by Harris (2008), in which cooled, sinking, brittle
crust scrapes against the lava lake sides.
We can now conrm that the SDR technique is able to reproduce
the work of Harris et al. (2005), Jones et al. (2006), and Harris (2008).
Because we observe that (part of) the most energetic recovered signal
(L
(1)
) is unchanged, and that this signal is persistent, it follows that
the source mechanism for (the most energetic part of) the tremor
remains virtually unchanged, and instead there are secondary sources
superimposed. Its polarization suggests that this is, indeed, the
persistent resonance of a conduit in response to a steady rate of
magma ow. Thus this interpretation corroborates the earlier results.
This is further supported by several new recovered signals seen only
during the high convective regime (e.g. H
(8)
H
(11)
, Table 1), which
have different principal eigenvectors from any signal seen during the
low convection, and are narrowband signals at high frequencies,
where far less energy exists during low convective regime.
9. Conclusions
We present a top-down, adaptive preprocessing algorithmthat lters
multichannel volcano-seismic data sets by selecting and grouping sub-
bands of data in the wavelet domain, based on the similarity and relative
energy of their principal components. This algorithm, called Subband
Decomposition and Reconstruction, or SDR, is designed to select regions of
the frequency spectrum that are strongly dominated by a single seismic
phase or input signal. Because undecimated wavelet packet coefcients
are constructed from orthogonal lters, and because the wavelet packet
transformpreserves energy, this method for subband decomposition and
reconstruction preserves polarization and signal amplitude of each
recovered signal. Thus, it allows different signals in various regions of
the spectrum to be examined in a quantitative, semi-automated way.
Tests on synthetic data suggest that SDR can recover multiple quasi-
continuous signals that differ fromone another by anorder of magnitude,
eveninnoisy environments, providedthat their spectral peaks are further
apart than the bandwidth of wavelet coefcients W
j,n
at the highest
wavelet level J. Tests onreal data recordedat Erta 'Alein2002suggest that
SDR can recover signals with geophysically meaningful interpretations,
and corroborates existing seismic and multiparametric work. In this
particular example, we see that SDR detects a persistent signal associated
with a resonating conduit responding to the ow of fresh, hot magma
from a deeper reservoir, signals associated with surface features (e.g.
degassing at fumaroles) and transients. We suggest that this algorithm
could effectively detect subtle changes in the time-frequency content of
volcanic tremor, and thereby recover signals from real seismic sources
whose spectral peaks are otherwise buried in background noise (and/or
masked by one another). Such an algorithm could allow volcanologists
much greater insight into the dynamics of volcanic systems, and could
detect subtle signals that might help address the possibility of unrest.
Acknowledgments
We thank Dr. Ken Creager, Dr. Don Percival, and Joshua D. Carmichael
of the University of Washington for many insightful mathematical
discussions, without which this paper would not have been possible, and
Dr. Chris Newhall (Earth Observatory of Singapore) and Dr. Andy Harris
(Laboratoire Magmas et Volcans, Universit Blaise Pascal) for advice and
geological insight. We thank OGS-CRS, Udine (Italy) for providing seismic
stationMAR, the PNSN, Seattle (WA) for seismic stationCMGandthe DAS
for stationL22, andDr. TomYelin(USGS, Seattle, WA) for generous loanof
the L22 sensor. R. Carniel's 2002 eldwork was partially funded by the
MULTIMO project (Energy, Environment and Sustainable Development
Program, EU Contract n. EVG1-CT-2000-00021). We thank Philippe
Lesage and another anonymous reviewer for their excellent suggestions,
which have hopefully led to signicant improvement of this paper.
References
Aki, K., Richards, P., 2002. Quantitative Seismology, 2nd ed. University science books,
New York.
Anant, K.S., Dowla, F.U., 1997. Wavelet transform methods for phase identication in
three-component seismograms. Bull. Seismol. Soc. Am. 87, 15981612.
Bardintzeff, J.-M., Gaudru, H., 2004. On 45 December 2004 visitors noted active
hornitos but solidied lava lake. B. Glob. Volc. Net. 29, 11.
Brenguier, F., Shapiro, N.M., Campillo, M., Ferrazzini, V., Duputel, Z., Coutant, O.,
Nercessian, A., 2008. Toward forecasting volcanic eruptions using seismic noise.
Nat. Geosci. 1, 126130.
Cabras, G., Carniel, R., Wassermann, J., 2008. Blind source separation: an application to
the Mt. Merapi Volcano, Indonesia. Fluctuation Noise Lett. 8 (34), L249L260.
Carniel, R., Di Cecca, M., Rouland, D., 2003. Ambrym, Vanuatu (JulyAugust 2000):
spectral and dynamical transitions on the hours-to-days timescale. J. Volcanol.
Geotherm. Res. 128, 113.
Carniel, R., Muoz Jolis, E., Jones, J., 2010. A geophysical multi-parametric analysis of
hydrothermal activity at Dallol, Ethiopia. J. Afr. Earth Sci. 58 (5), 812819.
Chouet, B., 1996. Long-period volcano seismicity: its source and use in eruption
forecasting. Nature 380, 309316.
Chui, C.K., 1997. Wavelets: A Mathematical Tool for Signal Processing. SIAM, Philadelphia,
PA, USA.
Coifman, R.R., Donoho, D.L., 1995. Translation-invariant de-noising. In: Antoniadis, A.,
Oppenheim, G. (Eds.), Wavelets and Statistics (Lecture Notes in Statistics, v. 103).
Springer-Verlag, New York, pp. 125150.
Coifman, R.R., Wickerhauser, M.V., 1992. Entropy-based algorithms for best basis
selection. IEEE Trans. Inf. Theory 38, 713718.
Dangel, S., Schaepman, M.E., Stoll, E.P., Carniel, R., Barzandji, O., Rode, E.-D., Singer, J.M.,
2003. Phenomenology of tremor-like signals observed over hydrocarbon reser-
voirs. J. Volcanol. Geotherm. Res. 128 (13), 135158.
Daubechies, L., 1992. Ten Lectures on Wavelets. SIAM, Philadelphia, PA, USA.
Di Lieto, B., Saccorotti, G., Zuccarello, L., Rocca, M.L., Scarpa, R., 2007. Continuous
tracking of volcanic tremor at Mount Etna, Italy. Geophys. J. Int. 169, 699705.
doi:10.1111/j.1365-246X.2007.03316.x.
Goldstein, P., Chouet, B., 1994. Array measurements and modeling of sources of shallow
volcanic tremor at Kilauea Volcano, Hawaii. J. Geophys. Res. 99 (B2), 26372652.
Gottschmmer, E., Surono, I., 2000. Locating tremor and shock sources at Bromo
Volcano. J. Volcanol. Geotherm. Res. 101, 199209.
Greenhall, C.A., 1991. Recipes for degrees of freedom of frequency stability estimators.
IEEE Trans. Instrum. Meas. 40, 994999.
Harrington, R.M., Brodsky, E.E., 2007. Volcanic hybrid earthquakes that are brittle-
failure events. Geophys. Res. Lett. 34, L06308. doi:10.1029/2006GL028714.
Harris, A.J.L., 2008. Modeling lava lake heat loss, rheology, and convection. Geophys.
Res. Lett. 35, L07303. doi:10.1029/2008GL033190.
Harris, A.J.L., Flynn, L.P., Rothery, D.A., Oppenheimer, C., Sherman, S.B., 1999. Mass ux
measurements at active lava lakes: implications for magma recycling. J. Geophys.
Res. 104 (B4), 71177136.
Harris, A.J.L., Carniel, R., Jones, J., 2005. Identication of variable convective regimes at
Erta Ale Lava Lake. J. Volcanol. Geotherm. Res. 142, 207223.
Hellweg, M., 2000. Physical models for the source of Lascar's harmonic tremor.
J. Volcanol. Geotherm. Res. 101, 183198. doi:10.1016/S0377-0273(00)00163-3.
Hyvrinen, A., Karhunen, J., Oja, E., 2000. Independent Component Analysis. John Wiley
and Sons, New York.
Jones, J.P., 2009. Subband investigation of continuous volcanic tremor. University of
Washington: Ph.D. thesis.
Jones, J.P., Carniel, R., Harris, A.J.L., Malone, S., 2006. Seismic characteristics of variable
convection at Erta 'Ale lava lake, Ethiopia. J. Volcanol. Geotherm. Res. 153, 6479.
Jurkevics, 1988. Polarization analysis of three-component array data. Bull. Seismol. Soc.
Am. 78, 17251743.
Konstantinou, K.I., Schlindwein, V., 2002. Nature, waveeld properties and source
mechanism of volcanic tremor: a review. J. Volcanol. Geotherm. Res. 119, 161187.
Liang, J., Parks, T.W., 1996. A translation-invariant wavelet representation algorithm
with applications. IEEE Trans. Signal Process. 44, 225232.
Lilly, J.M., Park, J., 1995. Multiwavelet spectral and polarization analyses of seismic
records. Geophys. J. Int. 122, 10011021.
Mallat, S.G., 1999. A Wavelet Tour of Signal Processing. Academic Press, New York.
Martini, M., 1969. Studio di prodotti fumarolici di alcuni vulcani della catena dell'Erta
Ale (Etiopia). Rend. Soc. Min. Ital. 25, 7992.
McNutt, S.R., 1996. Seismic monitoring and eruption forecasting of volcanoes: a review
of the state-of-the-art and case histories. In: Scarpa, T. (Ed.), Monitoring and
Mitigation of Volcanic Hazards. Springer, Berlin, pp. 100146.
Montalbetti, J.F., Kanasewich, K.R., 1970. Enhancement of teleseismic body phases with
a polarisation lter. Geophys. J. R. Astron. Soc. 21, 119129.
Nason, G.P., Silverman, B.W., 1995. The stationary wavelet transform and some statistical
applications. In: Antoniadis, A., Oppenheim, G. (Eds.), Wavelets and Statistics (Lecture
Notes in Statistics, v. 103). Springer-Verlag, New York, pp. 281299.
Obara, K., Hirose, H., 2006. Non-volcanic deep low-frequency tremors accompanying slow
slips in the southwest Japan subduction zone. Tectonophysics 417 (12), 3351.
Oppenheimer, C., Francis, P., 1998. Implications of longeval lava lakes for geomorpholog-
ical and plutonic processes at Erta 'Ale volcano, north Afar. J. Volcanol. Geotherm. Res.
80, 101111.
Oweiss, K.G., Anderson, D.J., 2007. Tracking signal subspace invariance for blind
separation and classication of nonorthogonal sources in correlated noise. EURASIP
J. Adv. Signal Process. 2007 (37485).
Pacheco, C., Sneider, R., 2005. Time-lapse travel time change of multiply scattered
acoustic waves. J. Acoust. Soc. Am. 118 (3), 13001310. doi:10.1121/1.2000827.
Patan, D., Di Grazia, G., Cannata, A., Montalto, P., Boschi, E., 2008. The shallow magma
pathway geometry at Mt. Etna volcano. Geochem. Geophys. Geosyst. 9. doi:10.1029/
2008GC002131.
Pearson, K., 1901. On lines and planes of closest t to systems of points in space. Philos.
Mag. 2 (6), 559572.
Percival, D.B., Guttorp, P., 1994. Long-memory processes, the Allan variance and wavelets.
In: Foufoula-Georgious, E., Kumar, P. (Eds.), Wavelets in Geophyics. Academic Press,
San Diego, pp. 325344.
Percival, D.B., Mojfeld, H., 1997. Analysis of subtidal coastal sea level uctuations using
wavelets. J. Am. Stat. Assoc. 92, 868880.
Percival, D.B., Walden, A.T., 2000. Wavelet Methods for Time Series Analysis. Cambridge
University Press, Cambridge, U.K.
Powell, T.W., Neuberg, J., 2003. Time dependent features in tremor spectra. J. Volcanol.
Geotherm. Res. 128, 177185. doi:10.1016/S0377-0273(03)00253-1.
Ripepe, M., Harris, A.J.L., Carniel, R., 2002. Thermal, seismic and infrasonic evidences of
variable degassing rates at Stromboli volcano. J. Volcanol. Geotherm. Res. 118 (34),
285297.
Shaw, P.J.A., 2003. Multivariate Statistics for the Environmental Sciences. Hodder-Arnold.
Shensa, M.J., 1992. The discrete wavelet transform: wedding the a trous and Mallat
algorithms. IEEE Trans. Signal Process. 40, 24642482.
Sherburn, S., Scott, B.J., Nishi, Y., Sughihara, M., 1998. Seismicity at White Islandvolcano, New
Zealand: a revised classication and inferences about source mechanism. J. Volcanol.
Geotherm. Res. 83, 287312.
Strang, G., 1993. Wavelet transforms versus Fourier transforms. Bull. Amer. Math. Soc.
28 (2), 288305.
Swanson, D.A., Dufeld, W.A., Jackson, D.B., Peterson, D.W., 1979. Chronological narrative
of the 196971 Mauna Ulu eruption of Kilauea volcano, Hawaii. USGS Prof. Pap. 55 pp.
Taswell, C., 1996. Satiscing search algorithms for selecting near-best bases in adaptive
tree-structured wavelet transforms. IEEE. Trans. Sig. Proc. 44 (10), 24232438.
Vandecar, J.C., Crosson, R., 1990. Determination of teleseismic relative phase arrival
times using multi-channel cross-correlation and least squares. Bull. Seismol. Soc.
Am. 80 (1), 150169.
Vidale, J., 1986. Complex polarization analysis of particle motion. Bull. Seismol. Soc. Am.
76 (5), 13931405.
Walden, A.D., Cristan, A.C., 1998. The phase-corrected undecimated discrete wavelet
packet transform and the recurrence of high latitude interplanetary shock waves.
Proc. R. Soc. London, Ser. A 454, 22432266.
Wassermann, J., 1997. Locating the sources of volcanic explosions and volcanic tremor
at Stromboli volcano (Italy) using beam-forming on diffraction hyperboloids. Phys.
Earth Planet. Inter. 104, 271281.

Jones 2012 Decomposition Volcanic Tremor

Diunggah oleh

Informasi Dokumen

Deskripsi Asli:

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Jones 2012 Decomposition Volcanic Tremor

Diunggah oleh

Hak Cipta:

Format Tersedia

Subband decomposition and reconstruction of continuous volcanic tremor

Anda mungkin juga menyukai