of 10
GEOPHYSICS
An upper-crus
t lid over the Long
Valley
magma
chamber
Ettor
e Biondi*,
Weiqiang
Zhu,
Jiaxuan
Li,
Ethan
F. Williams,
Zhongw
en
Zhan
Geophy
sical characteriza
tion of calder
as is fundamental
in assessing
their potential
for future catastrophic
vol-
canic
eruptions.
The mechanism
behind
the unrest of Long
Valley Calder
a in California
remains
highly
debated,
with recent
periods
of uplift
and seismicity
driven either
by the release
of aqueous
fluids
from the magma
chamber
or by the intrusion
of magma
into the upper
crust. We use distributed
acoustic sensing
data recorded
along
a 100-kilometer
fiber-optic
cable
traversing
the calder
a to image
its subsurfa
ce structur
e. Our images
highlight
a definite
separ
ation between the shallo
w hydrothermal
system and the large
magma
chamber
located at ~12-kilometer
depth.
The combina
tion of the geological
evidence
with
our results
shows how
fluids
exsolved through
second
boiling
provide the source of the observ
ed uplift
and seismicity
.
Copyright
© 2023 The
Authors,
some
rights
reserved;
exclusive licensee
American
Associa
tion
for the Advancement
of Science.
No claim to
original
U.S. Government
Works. Distributed
under
a Creative
Commons
Attribution
License
4.0 (CC BY).
INTR
ODUCTION
Calder
as often
remain
active long after their
forma
tion as shown by
their surface activity
(
1
,
2
), such as fumar
oles and large-scale
hydro-
thermal
systems.
To evalua
te the risk of major
eruptions,
it is critical
to characterize
the connectivity
between surface features and sub-
surface structur
es, especially
to estimate the volume
of potentially
eruptible
material
(
3
,
4
). For example,
tomogr
aphic
images
of the
Yellowstone Calder
a show an upper-crus
t reservoir of 10,000
km
3
with
an estimated melt
fraction
varying
between 10 and 20% (
5
,
6
) that, under
certain
circums
tances,
can produce
eruptions
one
to two orders
of magnitude
larger
than historically
observ
ed volca-
nic events (
7
).
The Long
Valley Calder
a, located in the Eastern Sierra Nevada
mountains
of California,
is one of the larges
t calder
as in North
America
and was formed
approxima
tely 767 ka ago by a single
erup-
tive event that released
650 km
3
of rhyolitic
material
(
8
,
9
). Since
1978,
the area has experienced
multiple
periods
of pronounced
unrest (
10
).
During
these
periods,
the calder
a activity
included
crustal earthquak
e swarms
and sequences
(
11
13
),
long-period
vol-
canic
earthquak
es (
14
,
15
), surface deforma
tion (mainly
inflation)
(
16
18
),
and elevated efflux
of magma
tic gases
(
19
22
).
Despite
substantial
efforts
to unders
tand
the nature of the calder
a
s
unrest, the mechanism
is still debated (
23
). Two competing
hypoth-
eses have been
proposed:
(i) upper-crus
t magma
tic intrusion(s)
or
(ii) ascending
fluids
released
by second
boiling
of the rhyolitic
res-
ervoir terminally
crystallizing
at depths
greater than
10 km (
24
).
Second
boiling
occurs
when
a magma
body
has stopped
rising
toward the surface and is empla
ced in the upper
crust at a depth
influenced
by neutr
al buoyancy and roof-rock strength.
The crystal-
lization of such a body
reduces
the solubilities
of the contained
vol-
atiles within
the magma
mush
(e.g.,
H
2
O and CO
2
) and causes
them
to be released/e
xsolved in the form
of bubbles
or vesicles
that rise
toward the surface due to buoyancy. On the one hand,
relatively low
borehole
temper
atures and lack of observ
ed CO
2
or He anomalies
at
the surface indica
te a volcanic
system whose
activity
is driven by
second
boiling
(
19
,
20
,
25
). On the other
hand,
deforma
tions
con-
sistent with a dilating spher
oidal
body
at approxima
tely 8-km
depth
and regular
volcanic
earthquak
es sugges
t the potential
involvement
of magma
and volatile exsolution
from the magma
reservoir (
17
,
23
,
25
,
26
). Last, for large
rhyolitic
systems,
such
as the Long
Valley
Calder
a, eruptions
are typically
driven by melt last stored at upper
crust depths
(for the Long
Valley Calder
a, ~3 to 8 km) (
27
).
Given
the comple
xity of this system, new upper-crus
t injections
and fluid
exsolution
by second
boiling
could
act in tandem
to induce
the ob-
served surface deforma
tion and seismicity
(
28
).
Therefore, the
ability
to exclude
the presence
of large-scale
upper-crus
t melt reser-
voirs would
enable
a better
characteriza
tion of the hazard
of this
volcanic
system. To summarize,
upper-crus
t magma
tic intrusions
would
indica
te an elevated potential
for eruptiv
e activity
, while
the second
boiling
nature of the unrest would
imply
a moribund
magma
tic system, still hazardous
but not as danger
ous.
RESUL
TS
Seismic
tomogr
aphy
is a valuable
tool to resolve this dichotomy
.
Recent
tomogr
aphic
studies
of the Long
Valley Calder
a highlighted
the presence
of a large
magma
tic body
in the middle
crust but were
limited
by the scale and resolution
of the seismic
arrays used to form
the tomogr
aphic
images
(
23
,
29
32
).
In this study, we use two dis-
tributed
acoustic sensing
(DAS)
units
to record
seismic
data along
telecommunica
tion fiber-optic
cables
(Fig.
1 and fig. S1). DAS
de-
ployed on fiber-optic
cables
provides
a novel methodology
to record
earthquak
es and other
seismic
signals
with unprecedented
tempor
al
and spatial resolution
(
33
,
34
), especially
in volcanic
areas (
35
37
).
Our two DAS arrays are composed
of more than 9000
channels
cov-
ering
an approxima
tely 100-km
north-south
transect
across the
calder
a, with
precise
channel
locations
obtained
using
a vehicle-
based
positioning
method
(
38
).
The total
apertur
e and channel
density
of the DAS
arrays enable
the imaging
of subsurfa
ce struc-
tures that could
not be resolved by previous
studies
that only relied
on conventional
networks.
Over a 12-month
period,
we detected
more than 6000
local
and
regional
earthquak
es that were also cataloged
by the Northern
Cal-
ifornia
Earthquak
e Data Center
(NCEDC)
(
39
).
We trained
a deep
neural network
model
to accurately pick more than
12 million
P
-
and
S
-w ave arrival
times
and incorpor
ate these
measur
ements
within
an efficient
tomogr
aphic
workflo
w. DAS
provides
a total
California
Institute of Technology
, Seismological
Laboratory, Pasadena,
CA, USA.
*Corresponding
author.
Email:
ebiondi@caltech.edu
SCIENCE
ADVANCES
|
RESEARCH
ARTICLE
Biondi
et
al.
,
Sci.
Adv.
9
, eadi9878
(2023)
18 October
2023
1 of
9
Downloaded from https://www.science.org at California Institute of Technology on October 18, 2023
number
of travel-time
measur
ements
that is two to three orders
of
magnitude
larger
than
conventional,
even dense,
seismic
arrays,
which
represents
a computa
tional
challenge
for existing tomo-
graphic
approaches.
Moreover, volcanic
areas present
subsurfa
ce
structur
es with
substantial
velocity
contr
asts resulting
in comple
x
seismic
raypaths. To properly
take into
account
comple
x ray
geometries
and handle
the large
number
of travel-time
measur
e-
ments,
we develop a double-differ
ence (DD)
Eikonal travel-time
to-
mogr
aphy
workflo
w based
on the adjoint-s
tate method.
The
nonlinear
iterative nature of our method
correctly
models
ray
bending,
while
the matrix-fr
ee formula
tion permits
the efficient
in-
version
of billions
of DD travel times.
Fig.
1.
Study
area
and
local
and
regional
events
from
DAS
arr
ay.
(
A
) Map of the study area in which
the distributed
acoustic sensing
(DAS) channels
(green line),
seismic
stations (blue triangles),
and earthquak
es (red dots) are indicated. The black dashed
line delinea
testhe limit of the Long Valley Calder
a. Thewhite arrows point to
the two events shown in the bottom
panels.
The red box in the map inset indicates the study area within
the United
States. (
B
and
C
) Strain recorded
by the DAS arrays
induced
by local events with Northern
California
Earthquak
e Data Center (NCEDC)
double-differ
ence (DD) catalog IDs 73482516
and 73491170,
respectiv
ely. The red and
blue curves in these panels
show the
P
- and
S
-w ave neural network
pick
ed travel times on these two events, respectiv
ely. M, Magnitude.
SCIENCE
ADVANCES
|
RESEARCH
ARTICLE
Biondi
et
al.
,
Sci.
Adv.
9
, eadi9878
(2023)
18 October
2023
2 of
9
Downloaded from https://www.science.org at California Institute of Technology on October 18, 2023
The
Long
Valle
y Calder
a shallo
w structur
es
Figur
e 2 shows our tomogr
aphic
images
of the Long
Valley Calder
a,
in a side-by-side
comparison
with
the latest tomogr
aphy
P
-w ave
(
V
P
) and
S
-w ave (
V
S
) velocity
model,
which
is also our initial
model,
based
on full waveform
inversion
of surface waves
between 6 and 20 s (
40
).
All plots
are shown as perturba
tions
with
respect
to an average one-dimensional
(1D)
Walker Lane
crust profile (fig. S5). With the improved data coverage from the
two DAS
arrays, we substantially
improve the model
resolution
in
the top 15 to 20 km. The heterogeneous
shallo
w structur
es within
Fig.
2. The
Long
Valle
y shear-w
ave anomalies.
The panels
on the left column
display the initial model
derived from surface-wave inversion,
while the panels
on the
right depict
the inverted
S
-w ave velocity
anomalies
obtained
by our tomogr
aphic DAS workflo
w. All perturba
tions are with respect
to a one-dimensional
Walker Lane
crust profile (obtained
byaveraging the initial model
along latitude and longitude).
(
A
and
B
) Depth
slices at
1.0 km elevation. The caldera and lakes
extents
are shown
by the black dashed
lines. (
C
to
H
) Model
profiles indicated in (A). The white
(A and B) and black (C to H) dashed
lines delinea
te the
20 and
15%
P
-w ave velocity
contours.
The white
solid lines separa
ting the shaded
areas denote
the resolvable
model
portions.
SCIENCE
ADVANCES
|
RESEARCH
ARTICLE
Biondi
et
al.
,
Sci.
Adv.
9
, eadi9878
(2023)
18 October
2023
3 of
9
Downloaded from https://www.science.org at California Institute of Technology on October 18, 2023
the calder
a, which
only appear
as a smooth
low-velocity
anomaly
in
the initial
model,
become
sharp
in our new tomogr
aphic
model
and
correlate well with
surface geology
. These
shallo
w velocity
reduc-
tions
are likely related to the filling
material
(e.g.,
volcanic
ashes
and alluvial)
deposited
in the depression
and the extensiv
e surface
hydrothermal
system (
20
,
41
). This hypothesis
is corroborated by
the high
V
P
/
V
S
ratio measur
ed (greater than
1.8) within
these
anomalies,
commonly
associa
ted with
highly
fractured and fluid-
permea
ted rocks
(Fig.
3) (
42
).
The shallo
w higher
velocity
and
low
V
P
/
V
S
ratio barrier
separ
ating the two anomalies,
visible
in
the cross section
AA
0
(see white
arrows in Figs.
2D and 3B), has
been
also observ
ed by other
local
tomogr
aphy
studies
(
30
,
31
) and
could
be composed
of less fractured crystalline
rocks compar
ed to
the surrounding
units.
This structur
e would
explain
the low temper-
ature measur
ed at the bottom
(appr
oxima
tely 2.5-km
depth
from
the surface) of the Long
Valley Explor
ation
Well (
43
),
which
could
inhibit
the transfer
of heat via convective movement
of mag-
matic volatiles by restricting
fluid
motion.
The
Mono-Iny
o Craters
structur
es
With the DAS
arrays extending
approxima
tely 30 km north
of the
calder
a rim, our tomogr
aphy
closes
the gap in knowledge
of upper
crust structur
es across the broader
Long
Valley magma
tic system.
North
of the calder
a region
lies the Mono-Iny
o Craters, a north-
south
trend of lava domes
and volcanic
craters. These
craters are
the younges
t geological
features in the Long
Valley area, ranging
in age from 40,000
to only
250 years ago (
44
).
Previous
studies
have highlighted
the existence
of gravity and resistivity
anomalies
that have been
interpr
eted as part of the hydrothermal
network
of
this portion
of the volcanic
field (
45
,
46
). However, seismic
surveys
were not able to provide evidence
of any velocity
variations
due to
the sparsity
of station coverage outside
the calder
a (
32
). Our model
is in good
agreement
with the resistivity
and gravity observa
tions,
revealing
seismic
velocity
anomalies
associa
ted with
these
upper-
crust structur
es (AA
0
cross section,
Fig. 2D and fig. S4D).
A
basin-oriented
north-south
low-velocity
anomaly
is evident
below
the Mono-Iny
o Craters, the depth
of which
reaches approxima
tely 4
km below sea level. Moreover, two reductions
in seismic
velocity
are
located within
the Mono
basin,
with one center
ed below Mono
Lake
where the most recent
volcanic
eruptions
in the Long
Valley region
occurr
ed around
250 years ago (
2
) (CC
0
cross section,
Fig. 2H and
fig. S4H).
These
two anomalies
could
be again
linked with shallo
w
hydrothermal
systems
given their
relatively high
V
P
/
V
S
ratio (ap-
proxima
tely 1.8).
The depth
sensitivity
of our tomogr
aphy
is not
able to resolve the small
intruded
magma
tic bodies
that likely
formed
these
volcanic
centers
(
29
).
However, no conduit
is
evident
connecting
these
structur
es to a deeper
magma
tic source,
supporting
their
now hydrothermal
nature.
The
Long
Valle
y volcanic
upper-crus
t lid
In the cross-sectional
views of Long
Valley Calder
a, we observ
e a
clear
separ
ation between the large
magma
body
at depth
and the
upper-crus
t low-velocity
structur
es (Fig.
2 and fig. S4). The initial
V
P
and
V
S
models
sugges
ted the existence
of an approxima
tely 10-
to 15-km-wide
conduit
connecting
the deep magma
chamber
to the
shallo
w crust. This appar
ent connectivity
in previous
models
is an
artifa
ct of the limited
depth
resolution
of surface-wave inversion
methods
(
47
).
In our images,
the structur
e separ
ating the upper-
and mid-crus
t depths
is likely the remnant
of the roof block
that
collapsed
as part of the calder
a-forming
760-ka
eruption.
The top
interfa
ce of the magma
tic chamber,
located at approxima
tely 8-km
depth
from the mean
sea level, is in agreement
with previous
depth
estimates from reflection
studies
(
48
50
).
Furthermor
e, the crustal
block
above the magma
tic chamber
presents
a typical
crustal
V
P
/
V
S
value
(appr
oxima
tely 1.7), indica
ting that the magma
body
at depth
is disconnected
from the shallo
wer low-velocity
structur
es of the hy-
drothermal
system throughout
the calder
a (Fig.
3, B and D). The
lower boundary
of this structur
e represents
the transition
between
brittle
and ductile
rock behavior as sugges
ted by the concentr
ation
of the seismicity
mostly confined
in this layer (Fig.
3, B and D),
Fig.
3. The
Long
Valle
y
V
P
/
V
S
ratio
structur
es.
(
A
and
C
)
V
P
/
V
S
ratio derived from the initial wave-speed
models
for the cross sections
AA
0
and BB
0
in Fig. 1A. (
B
and
D
)
Same as the previous
panels
but obtained
from the final models.
In all these panels,
the red dots indicat
e the initial and relocated earthquak
es within
10 km of the cross
sections.
SCIENCE
ADVANCES
|
RESEARCH
ARTICLE
Biondi
et
al.
,
Sci.
Adv.
9
, eadi9878
(2023)
18 October
2023
4 of
9
Downloaded from https://www.science.org at California Institute of Technology on October 18, 2023
whose
lower limit
varies
from 10- to 15-km
depth
outside
of the
calder
a to 5-km
depth
at the center
of Long
Valley. The west-east
section
of Fig. 3D reveals the thinning
of this layer at about
50
km along
the section
benea
th the resurgent
dome
at the center
of
the calder
a, with a concentr
ation of seismicity
at 4-km
depth.
The
extent
and position
of this structur
e correlate well with the geodetic
source of recent
uplift
(
51
). These
findings
exclude
the possibility
of
shallo
w intruded
magma
bodies
larger
than 2 km and support
the
interpr
etation of deforma
tion driven by the accumula
tion of ex-
solved fluids
at the center
of the calder
a that permea
te the preexist-
ing southern
moat and ring faults,
driving
the observ
ed seismicity
.
This interpr
etation is corroborated by the hypothesis-driv
en tests of
figs. S9 to S11 in which
any upper-crus
t velocity
reduction
larger
than 2 km in size can be correctly
detected
and estimated.
DISCUSSION
Ke
y impr
ovements
in
crus
tal
imaging
The spatial extent
and channel
density
of our DAS
arrays overcome
the inher
ent limita
tions
associa
ted with
conventional
broadband
networks
often
used
in body-w
ave tomogr
aphy
(
4
). Leveraging
the DAS
high-spa
tial sampling
capabilities,
we attain
exceptional
lateral resolution
in shallo
w depths
(0 to 8 km),
while
the wide
ap-
erture of our arrays enables
us to image
the middle
and lower por-
tions
of the subsurfa
ce (8 to 30 km) with a remarkable
level of detail.
Our findings
exhibit
resemblances
to earlier
studies
(
30
,
31
), such as
the high-v
elocity
barrier
at the center
of the resurgent
dome
(Fig. 2D) and the low velocity
and high
V
P
/
V
S
ratio shallo
w anom-
alies associa
ted with the hydrothermal
calder
a system (Figs.
2, B and
D, and 3B). However, thanks
to the aforementioned
advantages,
we
enhance
the current unders
tanding
by presenting
a compr
ehensiv
e
pictur
e of the entire volcanic
system, which
was missing
from pre-
vious
tomogr
aphic
results.
Similarly
, surface-wave inversion
strategies
can resolve large-
scale
velocity
anomalies
but lack the lateral resolution
to delinea
te
near-surfa
ce structur
es due to the commonly
consider
ed wave
periods
and station coverage (
43
).
This
limita
tion can be verified
by comparing
surface-wave dispersion
curves obtained
using
the
initial
and inverted
models.
To this end, we extract three velocity
profiles
from locations
placed within
the calder
a area (Fig.
4A)
and compute
the Rayleigh
phase-v
elocity
dispersion
curves (
52
).
The necessary
density
profiles
are obtained
using
an empirical
rela-
tionship
calibr
ated on crustal rocks
(
53
).
The left panels
in Fig. 4
depict
the
V
P
(red lines)
and
V
S
profiles
of the initial
(solid
lines)
and inverted
(dashed
line).
The corresponding
surface-wave disper-
sion curves are shown on the right
panels
in Fig. 4. The black dashed
vertical
lines bound
the range of periods
that were used to obtain
the
initial
velocity
models.
When
comparing
the dispersion
curves from
the initial
(solid
green lines)
and the final
dashed
green lines)
models,
only
minor
differ
ences
can be observ
ed, and,
even at
shorter
periods
than
5 s down to 1 s, the two curves do not
present
substantial
phase
velocity
differ
ences.
This
comparison
highlights
the consis
tency
of the inverted
wave speeds
in preserving
the surface-wave structur
es and the inability
of surface-wave inver-
sion methodologies
to refine
the near-surfa
ce structur
es due to an
intrinsic
nonuniqueness
of the inverse problem
for the consid-
ered periods.
The
upper-crus
t lid
confining
the
exsolv
ed
fluids
Our results
reinfor
ce the fluid-driv
en nature of the uplift
and unrest
occurring
in the Long
Valley Calder
a and represent
the first tomo-
graphic
evidence
supporting
the second-boiling
hypothesis
with a
lack of recent
upper
crust intrusions.
Figur
e 5 shows a schema
tic
model
based
on the structur
es highlighted
in the BB
0
cross
section.
The resurgent
dome
presents
lower
V
P
/
V
S
and higher
veloc-
ities than
the surrounding
region,
as observ
ed by previous
studies
(
30
,
31
). The higher
V
P
/
V
S
values
and lower velocities
in the eastern
portion
of the calder
a correlate well with the location of hot springs
and ash-rich
sediments.
In our interpr
etation, the Sierran basement,
which
was part of the pre-calder
a magma
tic roof block,
covers the
contempor
ary magma
chamber
and isolates the magma
body
from
the shallo
w crust. Our new observa
tions
place tighter
constraints
on
the melt region,
which
exhibits
an overall VS anomaly
of
15%
and
a total volume
of 6400 km
3
, which
is in the same
order
of magnitude
as other
large
volcanic
systems
(
5
7
).
By using
experimental
melt-
fraction
curves (
32
),
the melt
fraction
varies
from 21 to 23% and
corresponds
to a total
storage of 1350
km
3
of melt,
which
agrees
with
previous
estimates (
32
).
Within
sill-lik
e structur
es, inferr
ed
from the estimated seismic
anisotr
opy in this volcanic
system
(
54
),
the melt fraction
might
be slightly
under
estimated compar
ed
to the one obtained
from the average inverted
V
S
values.
Such
melt-
fraction
values
are not close
to the critical
porosity
of a magma
tic
mush
[appr
oxima
tely 40%;
(
4
)] required to induce
the mobiliza
tion
of magma
toward the surface. Thus,
the retriev
ed velocity
anomalies
sugges
t a current textural equilibrium
(as a distributed
melt
or as
small
melt-rich
pockets) of the crystal mush,
which
implies
the stag-
nation and crystalliza
tion of the mid-crus
t chamber
associa
ted with
subsequent
exsolution
of fluids.
Fluids
released
from the apex of the
crystallizing
chamber
are then trapped
at the bottom
of the Sierran
basement
providing
the pressure source of the observ
ed uplift.
Last,
the accumula
ted fluids
migrate laterally toward the southern
segment
of the ring-fault
zone
and drive the south-moa
t observ
ed
seismicity
. This interpr
etation does
not preclude
the possibility
of
new mantle
injections
that would
perturb
the textural equilibrium
of the magma
chamber,
which
could
result
in the revitaliza
tion of
this moribund
volcanic
system.
MA
TERIALS
AND
METHODS
Picking
P
-
and
S
-w
ave arrivals
on
DAS
data
using
ma
chine
learning
Figur
e S1 depicts
the local
and regional
seismicity
used
within
this
study. A total
of 2173
events (red dots
in fig. S1) from the DD
catalog
of the NCEDC
(
39
) have been
employ
ed to form
our tomo-
graphic
images.
These
earthquak
es were recorded
by both
conven-
tional
stations
(blue
triangles
in fig. S1) and DAS
arrays (green lines
in the same
figure) and occurr
ed between November
2020
and No-
vember
2021.
The minimum
magnitude
consider
ed is 0.1, while
the
maximum
is 4.96.
From these
events,
we select
843 earthquak
es
with
an average signal-to-noise
ratio (SNR)
equal
to or above 40
dB on the DAS
data (examples
of the selected
events are shown in
Fig. 1). We estimate the SNR of each event by computing
the noise
and signal
energy
using
2- and 0.8-s
windo
ws before and after the
initially
predicted
P
-w ave travel time,
respectiv
ely. We obtain
the
necessary
P
- and
S
-w ave observ
ed travel times
by using
a neural
network
model
designed
to accurately pick DAS
data called
Phase-
Net-DAS
(
55
).
This
model
is based
on PhaseNet
(
56
),
which
is a
SCIENCE
ADVANCES
|
RESEARCH
ARTICLE
Biondi
et
al.
,
Sci.
Adv.
9
, eadi9878
(2023)
18 October
2023
5 of
9
Downloaded from https://www.science.org at California Institute of Technology on October 18, 2023
modified
U-Net
architectur
e (
57
) with 1D convolutional
layers for
processing
1D time
series
of seismic
waveforms.
We extend
this
model
using
2D convolutional
layers to fully
exploit
both
spatial
and tempor
al informa
tion of 2D DAS
data.
PhaseNet-DAS
obtains
accurate travel-time
picks
when
high-
SNR
DAS
data are fed into the neural network.
Figur
e 1 (B and
C) shows two representa
tive events overlaid
with
the
P
- and
S
-
wave picked travel times.
Both
curves closely
follow the
P
- and
S
-
wave onsets
clearly
observable
in these
two panels.
All the other
events present
a similar
behavior in terms
of onset
travel-time
matching
quality
. We also quantita
tively estimate the picking
accu-
racy using
a cross-corr
elation methodology
. We cut a 4-s windo
w
around
the arrivals
obtained
by PhaseNet-DAS,
apply
a band-
pass filter
between 1 and 10 Hz, and calcula
te the cross-corr
elation
between event pairs.
We estimate the differ
ential
time by picking
the
peaks
within
the cross-corr
elation profiles
for each channel.
To
further
improve the accuracy of the differ
ential
travel-time
measur
e-
ments,
we use a multichannel
cross-corr
elation strategy (
58
,
59
) to
extract the peaks
across multiple
cross-corr
elation profiles.
For our
analysis, we choose
event pairs
whose
average cross-corr
elation co-
efficients
are higher
than
0.4 for
P
wave and 0.6 for
S
wave. This
choice
allows us to retriev
e 34,193,571
P
-w ave and 3,944,318
S
-
wave differ
ential
travel-time
measur
ements
from 7583
(fig. S2A)
and 1095
event pairs,
respectiv
ely. The histograms of these
differ
en-
tial travel-time
values
are shown in fig. S2 (B and C). If we assume
Gaussian-dis
tributed
travel-time
picking
errors, then
the differ
en-
tial time
measur
ements
obtained
by waveform
cross-corr
elation
can be used
to estimate the error distribution.
The differ
ential
travel-time
measur
ements
have a mean
of
1 ms and an SD of 70
ms for
P
waves and a mean
of 3 ms and an SD of 140 ms for
S
waves.
These
values
correspond
to SDs of 49.5 and 99 ms for
P
- and
S
-w ave
picking
errors made
by the PhaseNet-DAS
model,
respectiv
ely. For
comparison,
the absolute
arrival-time
errors of the PhaseNet
archi-
tectur
e compar
ed with manual
picks
on conventional
stations
have
Fig.
4.
Surfa
ce-w
ave dispersion
curv
es
from
initial
and
inv
erted
models.
(
A
)
Map showing the locations for which
the surface-wave dispersion
curves are com-
puted.
(
B
,
D
, and
F
)
V
P
(red lines) and
V
S
profiles of the initial (solid lines) and in-
verted (dashed
line) models
extracted at the corresponding
points
in (A). (
C
,
E
, and
G
) Rayleigh
dispersion
curves for the corresponding
panels
on the left evalua
ted
using
the initial
(solid
green line) and inverted
(dashed
green lines)
velocity
profiles.
The black vertical
lines indicate the period
range used to construct the
initial velocity
models
[i.e., 6 to 20 s periods;
(
40
)].
Fig.
5. Schema
tic
model
of the
Long
Valle
y magma
tic
system
interpr
eted
from
the
tomogr
aphic
sections.
Theorienta
tionofthisinterpr
etationisalongthecross
section
BB
0
in Fig. 3D. The shallow formations are based
on the geologic
section
in
(
24
). The location of the ash-rich
sediments
with a high
V
P
/
V
S
ratio and reduced
wave speeds
correlates well with the hydrothermal
activity
present
in the
eastern caldera area (
20
). The boundaries
of the Sierran basin follow the 1.74
V
P
/
V
S
ratio bottom
contour
in Fig. 3D.
SCIENCE
ADVANCES
|
RESEARCH
ARTICLE
Biondi
et
al.
,
Sci.
Adv.
9
, eadi9878
(2023)
18 October
2023
6 of
9
Downloaded from https://www.science.org at California Institute of Technology on October 18, 2023