of 13
SMIP21 Seminar Proceedings
77
THE COMMUNITY SEISMIC NETWORK FOR DENSE, CONTINUOUS
MONITORING OF GROUND
AND STRUCTURAL STRONG MOTION
Monica Kohler
1
and the Community Seismic Network Team
1
Department of Mechanical and Civil Engineering
California Institute of Technology,
Pasadena, CA
Abstract
The Community Seismic Network (CSN) is a cloud
-based
, strong-
motion network of
seismic stations deployed in the greater Los Angeles area. The sensors report three
-component
acceleration time series data and peak acceleration scalar data for use in assessments of
earthquake shaking intensity in buildings and on the ground level
, monitoring structural health
of
instrumented buildings
, zonation maps of future shaking potential, and the ShakeAlert
earthquake early warning system.
The hardware and software behind CSN’s client and server
architecture are described, as well as network subarrays deployed at Los Angeles Unified School
Distri
ct campuses, the NASA
-JPL campus, and in mid-
rise
s and high-
rise
s.
Introduction
This paper describes the architecture of the Community Seismic Network (CSN), a
permanent strong
-motion seismic network. CSN has been developed over the past 10 years by a
team of scientists, listed in the Acknowledgments section of this paper, whose work is
represented here. CSN
hardware comprises commercially produced MEMS accelerometers that
are coupled with processors, external storage, and power supply. CSN consists of over 700
accelerometers that are deployed in mid
-rises
and high-
rises, school campuses, civic service
buildings, and homes in California (Fig. 1). The CSN project has increased the number of 3D
(ground level+all upper floors of buildings) seismic observations in greater Los Angeles by an
order of magnitude, by taking advantage of
advances in small
-form-
factor MEMS sensing
technologies, on-
site computing, and cloud infrastructure. The mission of CSN is to: 1) Provide
high spatial resolution assessments of
shaking intensity in buildings and on the ground following
major earthquakes; 2) Monitor the health and safety of structures through detection and location
of damage; 3) Create zonation maps of future shaking potential in populated areas; and 4)
Provide data for the ShakeAlert earthquake early warning system (Given et al., 2014, 2018;
Kohler et al., 2020).
The Community Seismic Network (CSN) currently comprises hundreds of stations
located in southern California, most of which are in the greater Los Angeles area (Clayton et al.,
2011, 2015, 2020; Kohler et al., 2013, 2014, 2018; Massari et al., 2017). The
accelerometers are
triaxial, and
capable of recording accelerations up to twice the level of gravity. The primary
product of the network is measurements of shaking of the ground as well as upper floors in
buildings, in the seconds during and following a major earthquake. Each sensor uses
a small,
dedicated ARM processor computer running Linux, and analyzes time series data in real time at
250 sps,
which
then is
downsampled to 50
sps for data storage purposes
.
SMIP21 Seminar Proceedings
78
Figure 1. Seismic
stations in southern California
. CSN stations: circles. SCSN stations: triangles.
CSN c
lient architecture
Hardware
CPU
CSN clients have primarily been deployed using low-power, "single-board-
computer"
(SBC) platforms
since 2012. Such platforms typically have a physically small form factor and
consume under 20 W of power
, making them suitable for use with low
-wattage battery backup
units
. They also have between one and four USB 1.0 ports, and at least one ethernet port with
10
Mbps or greater data transmission rates
. Some platforms have a USB serial port suitable for on
-
site connecti
vity
without disturbing an operating client
. Other platform
s have wireless radios,
such as Bluetooth or WiFi, that provide on-
site access without having to displace the active
network connection.
The most recent l
arge CSN deployment in early 2020 used
100 Raspberry Pi Model 4B
units, housed in passive cooling aluminum cases. The previous large CSN deployment in early
2019 used 200 Raspberry Pi Model 3B units, also housed in passive cooling plastic cases. Board
fail
ures have not been observed in the
~2.5 years after deployment, despite the reliance on small
chip-
size heat sinks used in the absence of aluminum cases to serve as heat sinks.
A small
number (~5) of Raspberry Pi Model 3B+ units have also been deployed; these were originally for
laboratory
use but were later
migrated into the field
to meet deployment goals. Each of the
Raspberry Pi units
allows for the use of SD cards
for a moderate a
mount (about three months) of
on-
site storage.
CSN's first SBC platform beginning in early 2012 was the Global Scale Technologies
SheevaPlug, which includes a single USB 2.0 port, a single RJ-45 1 Gbps ethernet port, 512 MB
SMIP21 Seminar Proceedings
79
internal flash drive, and an external SD card slot. It also provides a mini
-U
SB port for serial
console access. Over 400 of these are still deployed
and operating at CSN stations in the field
.
The SheevaPlug has a single USB mini-A port configured so that it can be connected to a laptop
or desktop USB A port
via USB serial interface. MacBook Pros and most Windows-
compatible
laptops need no special USB driver software to communicate with a SheevaPlug; however earlier
MacOS systems may need a dedicated
driver for the SheevaPlug.
Accelerometer
Through 2021, CSN has used only one
type of
accelerometer family to provide the base
acceleration data - the phidgets.com Model 1056, based on the STMicroelectronics analog
MEMS LIS344ALH triaxial
inertial sensor with ± 2g range.
Initial CSN stations used the
original 1056 model that included a compass and gyroscope, but most CSN stations
now use the
model 1056-
1 which only has the triaxial accelerometer.
Power
, battery backup, and power boards
A combination of power options attempt to provide CSN stations with a few hours of
backup power. In some cases, emergency power
is available from the sensor host enterprise
; in
other cases, local enterprise
-provided Uninterruptible Power Supply (
UPS
) units are available for
use. When reliable backup power is not available, small battery backup units are added to the
setup
.
The canonical UPS deployed to date is a CyberPower CP350SLG unit which typically
provides a couple of hours of standby power. This unit has a form factor that neatly fits into the
external CSN box packaging that has been
deployed since the outset, but the units
lack an
interface that would support status monitoring.
One issue with these units is the limited lifetime
(2-3 years) of the internal
sealed
lead
acid
battery. In addition,
while a single battery replacement
cycle works well, a second replacement cycle is less likely to succeed
. A
t the
third replacement
point, the unit is scrapped.
Early SheevaPlug
s were notorious for prem
ature failure of their internal 120 VAC
-to-5
VDC power supply board. Now, a
fter 5
-8 years of deployment, CSN’s SheevaPlugs are
experiencing increasing rates of power board failures. As of early 2021, these power boards are
no longer available from the vendor, so a suggestion from the user community was adopted in
which failed boards are replaced with a generic 2.5A
mp
power adapter wall wart
, and the factory
output plug is replaced with a harness obtained from a retired
SheevaPlug power board.
CSN s
erver architecture
Software
Operating s
ystem
The CSN client software is based on the Linux operating system running on the hardware
platform with USB and ethernet interfaces. The USB interface is primarily used to connect to the
accelerometer
. T
he ethernet interface is the primary means of sending locally collected
SMIP21 Seminar Proceedings
80
accelerometer
data
to the CSN server environment. The ethernet interface is also used to
maintain an open reverse-
ssh tunnel to the server environment. This permits remote access from
the ser
ver to the client even if client-
side enterprise firewalls are present.
Preparing the SBC’s o
perating system platform involves a combination of tasks that
include: ensuring that a base level of applications has been installed on the platform to support
CSN tools; setting
various configuration parameters specific files for each application of interest;
and in some cases disabling
conflicting applications.
CSN practice is to retain the operating
Linux system version originally installed on a hardware platform. This provides a degree of
stability and predictability that is valuable over many years. As of 2021, several 8-
year
-old
deployments are still in operation and are expected to continue in perpetuity. This also addresses
early versions of hardware that do not readily support operating system version upgrades.
However, the above practice also implies ongoing support for an ever-growing number of Linux
versions, each typically customized for a particular hardware platform family and model.
Client application
The CSN client application is a Python script, currently written to Python 2.7. It is thread-
based, in which the threads
are used to handle several different tasks:
Interfac
ing
with the a
ccelerometer over USB and receiv
ing triaxial sam
ples from the sensor
.
Processing
the
triaxial samples from the sensor (including decimat
ion, mean removal,
property assessment
).
Creat
ing picks from incoming sensor samples
.
Monitoring the system clock, obtain
ing
Network Time Protocol (NTP
) time from a time
server, comput
ing a regression, and providing other threads with timestamp
.
Uploading 10
-minute raw data files to Amazon S3, and uploading latest station configuration
data file to Amazon S3
.
Implementing
a w
eb server interface for remote users
to obtain data from the client
including: a) uptime, b) v
ersion, c) latest accelerometer sample, d) 10
-minute files for
arbitrary periods of time, e) latest 2 minutes of data in the form of Google Charts for each of
the three sensor axes.
Ensuring that sufficient space is available on local storage by deleting older files when
necessary
.
The main client program contains the credentials for accessing both the Amazon S3
service that will store the sensor data, and the
locati
on and credentials for the ActiveMQ broker
which will receive the
picks fro
m the client.
(ActiveMQ is open
-source messaging software that
is employed by the distributed algorithms and applications that require messaging, and the broker
is an application that validates, translates and routes a message from a sender to a receiver). The
CSN client file
is in a hum
an-readable format that
contains CSN
station metadata
including the
sensor’s latitude and longitude, building
identific
ation
, floor number, and client name. This is
routinely edited to contain the required details.
SMIP21 Seminar Proceedings
81
Timin
g an
d accuracy
Accurate tim
estam
ps are required in the
client
to ens
ure that the data poi
nts
are a
ll
synchronized w
ith accu
rate external clocks and
with those from
other
clients. T
his is especially
cr itical for deployments of several clients w
ithin the
same bui
lding,
for example all sensors from
sequential
floors i
n a
bu
ilding
f
or which in
ter
-story dr
ift or
propagating wave
property
calculations are de
sired.
To achieve
accurat
e timestam
ps, t
he client applies
a linea
r regressi
on to
the
reporte
d times
obt
a
ined from
one of
several
possible
NTP
servers on the
networ
k. Typically,
the server
s are at C
altech
, U
CLA, and U
SC, but
sometimes the loc
al router is us
e
d f or failover
capability.
In some cases, a CSN SBC in t
he field
ser
ves as the NTP ser
ver (
e.g.
a separate CSN
NTP s
erver
in a
bui
lding).
The client p
olls the NTP ser
ver every m
inute, an
d t
he offset between the
syst
em clock
and the
NTP
tim
e is added to a
sliding 10-
minute window
of
offsets. A lin
ear regression is the
n
performed on the
latest
10 minutes
of
data, whic
h allows an es
timate to be
made
fo
r the
true
time
at any d
ata
point
(i.e. t
ime series sample) over the
comi
ng
minute before
the NTP server is
polled again. An example of how
the
NTP
offsets
and predicted offsets look over a
1-
hour
period
is shown in Fig. 2.
Figure 2. E
xample of NTP offsets (blue
curve)
and predicted t
ime offsets (red line)
over
a 1-
hour
period
of
waveform data.
Cli
ent
serv
er architecture
CSN
provides
bot
h real-time
and ne
ar-real
-time access to client-generated
data. Limited
real
-time da
ta flow
s ins
tantly
from CSN c
lients to a C
SN c
loud
server data br
oker, to which tw
o
subs
cri
bers currently connect – one
for e
arthquake
earl
y warning test
applications and
the
other
for
a S
hakeMap (Wal
d et
al., 2008; Worden et
al., 2020)
service t
est i
nstance. N
ear
-real-
time
data
flows
in short
bur
sts
(currently, 10-
minute
-long
time series) from C
SN
clients
through a
CSN
cloud server to a
local se
rver
-based ar
chive
at C
altech
.
The
method by which t
hese parallel data flows are carried ou
t i s as
follows. T
he CS
N
client
runn
ing on the
Linux-
based processor
at each station has
two different
but c
oncurrent
SMIP21 Seminar Proceedings
82
modes for uploading data. In the first mode (referred to as “pick mode”), the station software
instantly detects anomalously large waveform events in the data stream, such as high
-amplitude
accelerations, and immediately sends the computed peak amplitude scalar values and their
associated time stamps to the Amazon cloud (AWS). In this mode, the CSN python application
assesses the incoming stream of samples from the phidget sensor device library, and
it forwards
the selected samples to
a server
running on the CSN cloud infrastructure.
The pick mode data are
those that are currently designated for ShakeMap and the ShakeAlert earthquake early warning
system.
The CSN station uploads all waveform data in the second near
-real
-time
mode
(“continuous mode”). The station’s processor
accumulates 10 minutes of three-component,
minimally processed, unfiltered, acceleration data into a file. Once the file is complete, it is
compressed and uploaded to the cloud, while data are being accumulated into the next file. The
10-
minute data files are retained by the station’s system
until the local storage use reaches a
maximum threshold value, at which time the files are aged out; typically
this amounts to
a few
weeks of sensor data with the flash storage cards. This storage system would need to be accessed
for accumulated data in
the case of power outages or communications problems.
CSN’s software client computes peak accelerations which are reported as picks (the “pick
mode” introduced above) if values are > 0.5% g. The picks are computed from the time series’
deviations from a long-
term mean calculated over 10
-second sliding windows. Each orthogonal
axis is treated independently and the minimum repick i
nterval is 1 second on the same axis; thus
the maximum pick rate is 3 picks/second. Timestamps associated with the picks are calculated
using a continuous regression on the NTP offsets to the computer’s system clock
, as discussed
earlier
. Only ground-
level
station picks are sent to the cloud for the ShakeMap and ShakeAlert
applications.
I
n the current implementation of CSN’s pick distribution method, the CSN client
(running at the station) directly generates an ActiveMQ message for each pick locally on th
e
station’s processor, in the required format for ShakeMap or the earthquake early warning
applications
FinDer
(Böse
et al., 2012, 2015, 2018) and PLUM (Cochran et al., 2019)
. The CSN
server sends that message to an ActiveMQ broker running on an AWS virtua
l machine. The CSN
client includes an NTP
-based corrected pick timestamp in the ActiveMQ message. ShakeAlert
operates its own ActiveMQ brokers, whose topics are subscribed to by the various algorithms,
including FinDer and PLUM. A channel between the CSN broker and the ShakeAlert broker
used by development versions of FinDer and PLUM allows it to receive all CSN client picks.
Both FinDer and PLUM use all reported
CSN client picks
associated with an earthquake,
since
they always exceed 0.5% g. At the server side where PLUM is running, MMI values are
computed continuously for the incoming accelerometer measurements. MMI
is computed
from
the incoming
PGA pick values on all three components, and sent when an MMI threshold for
PLUM i
s exceeded. The maximum rate of MMI messages being sent by each client is one
message per second, for the duration of the shaking. Similar to
FinDer, these MMI values are
relayed to the ShakeAlert ActiveMQ broker to which the development version of PLUM is
subscribed
.
SMIP21 Seminar Proceedings
83
As of August 2021, the data stream provided by the CSN python client application to the
CSN ActiveMQ broker sufficiently matches the expectations of the ShakeAlert
infrastructure to
support a direct subscription. Future enhancements to the stream from the
CSN client to the CSN
ActiveMQ broker may require an additional level of processing and assessment within the cloud,
prior to making the data available to the
ShakeAlert production system.
The data flow architecture is illustrated in Fig.
3. The key motivation for this setup is to
prevent large latencies that can arise in part from the current ShakeAlert requirement that each
station sends full waveforms to the central processing site.
Figure 3. Data flow architecture
schematic. Sensors in the field (left side of diagram) send two
types of data to the message broker and storage servers in the Amazon cloud (AWS cloud). Data
are then passed on to development instances of FinDer and PLUM (right side of diagram).
The ShakeMap infrastructur
e requires a different format and content for peak amplitude
data than is provided by the CSN cloud ActiveMQ broker for ShakeAlert. Therefore, a service
running on the local server platform subscribes to the CSN ActiveMQ broker, provides very low
latency r
eformatting of the incoming peak amplitude (pick) data stream, and then sends those
data onward to the ShakeMap infrastructure.
A number of waveform data retrieval applications operate on the local server
to provide
continuous
waveform
time series files to researchers.
The main application is based on the
Seismogram Transfer Program (
STP
) client (
STP, 2007) and
can deliver data for all stations
rapidly for recent months
, and with a small latency for data older than that. Customized
CSN
STP-
based clients
se
rve a
subset
of sensors such as all
stations
at Los Angeles Unified
School
District campuses, NASA-JPL, and several
instrumented mid
-rise and high-rise buildings. The
applications
rely on three file types for their operation: 1) the station file which contains the
metadata
information for each station in CSN, 2) the waveform metadata
files which contain
metadata
information about each waveform segment, and 3) the wave
form
segment files
SMIP21 Seminar Proceedings
84
themselves which contain the digital samples in SAC format. The data used by the applications
is
refreshed by restarting them
; this is done on a regular basis several times a day.
Los Angeles Unified School District (LAUSD) stations
The majority of CSN stations are deployed at campuses of the Los Angeles Unified
School District
(LAUSD)
. LAUSD is a public, general-community stakeholder and partner, with
approximately 1000 campuses that span the City of Los Angeles (2000 km
2
). The school campus
buildings are typically 1
-3 story wood frame or reinforced concrete structures built after 1950.
These structures include types that are known to be prone to severe damage due to seismic
hazards, especially for older construction and soft first-
story construction. Approximately 400 of
the LAUSD campuses are instrumented with CSN sensor
s (majority of red circles shown in Fig.
1). Average CSN station spacing
at LAUSD campuses is about 0.5 km.
An
experimental ShakeMap
-ShakeCast
-like
setup has been
developed for the LAUSD
campuses at
which
CSN has deployed a sensor (Kohler et al., 2018). The setup is generally based
on features of ShakeCast (Wald et al., 2008), including the use of the Hazus Earthquake Model
Loss Estimation Methodology (H
azus, 2020) to classify structures and supply fragility curves.
However, key differences are that it uses CSN data recorded at the actual structure as shaking
intensity input
into the fragility functions for the structure
, and CSN
-developed web-
based tools
for its user interface. The LAUSD campuses used in this installation consist of only low
-rise
structures across a lateral dimension spanning about 20 km. All sensors are located in
communication or utility closets; none are in classrooms. The ShakeCast application is installed
in the central LAUSD office in downtown Los Angles and communication is modeled on a
centralized decision engine setup in which information could be subsequently sent via formal
channels to local principals and campuses.
As mentioned above, CSN’s current software client computes broadband peak
accelerations which are reported to ShakeMap if values are > 0.5% g, obtained from the time
series’ deviations from the long
-term mean, on any axis. Many CSN stations on school campu
ses
are in locations with frequent human activity that influences noise levels. For example, many
LAUSD stations exhibit noticeably higher noise levels during school hours. In future work,
station
-specific noise models taking into account time of day and d
ay of the week could
be
trained
, allow
ing for more reliable picking and signal-
to-noise estimation at stations with
predictable human-
generated noise.
Mid
-rise and high-
rise instrumentation
Several mid
-rise and high-
rise
buildings are currently instrumented by CSN
with at least
one triaxial
sensor deployed on most floors. A
ll are located in the downtown or greater Los
Angeles region
. The
buildings include a 52-
story dual system (concentrically braced steel frames
at core with outrigger moment frames (with 6
3 sensors); 15-
story steel moment frame and
concrete shear wall (with 34 sensors); 9
-story reinforced concrete (
with 10
sensors); and two 9-
story steel moment
-frame with trusses and girders (one with 31
sensors and the other with 15
sensors
).
SMIP21 Seminar Proceedings
85
Several of the instrumented buildings have two
or three
triaxal sensors deployed on most
floors, for the purpose of measuring torsion, rotations or rocking. Although the sensors are not
usually located at the edges of the floors, their locations relative to the center of mass on the
building floor can
be determined from structural engineering drawings obtained for most of the
instrumented buildings.
The majority of sensors are located in electrical or IT closets. One of the
CSN-instrumented buildings – the 52-
story high
-rise
– also has CSMIP instrumentation that has
recorded significant earthquakes since the 199
2 M7.3 Landers
earthquake, and could be used for
data comparison
of the July 2019 M7.1 and M6.4 Ridgecrest earthquakes, the September 2020
M4.5 South El Monte earthquake
, and
the April 2021 M4.0 Lennox earthquake.
NASA-JP
L stations
A total of 220
CSN triaxial accelerometers are deployed on the ~ 1 km × 1 km NASA-
JPL campus. This subset of sensors could be considered an “array within an array” due to their
smaller but approximately equidistant station spacing
. Sensors are installed on both ground-
level
and upper-level floors of several buildings. The ground-
level stations have an average spacing of
about 100 m.
The NASA-
JPL sensor deployment can be viewed and tested as a prototype mini-c
ity
strong-motion deployment, as there are one or more CSN accelerometers installed in about 90
buildings (mostly single or two-story structures) on the campus. The building types comprise
wood frame, steel sheds, modular trailers, steel-
moment frame, and
reinforced concrete. Of the
total 220 stations deployed at JPL, only the 100 ground-
level stations are contributing maximum
shaking peak acceleration pick data for
the experimental ShakeMap
and earthquake early
warning algorithms
. As with the LAUSD subarra
y, a ShakeMap-
ShakeCast
-like setup has been
configured for JPL (Massari et al., 2017). Each of the buildings uses fragility
curves supplied by
the Hazus Earthquake Model Loss Estimation Methodology (H
azus, 2020). The ShakeCast
configurations for the JPL sites
are set up so that they use the CSN ShakeMap as input for
localized and customized building performance assessment.
Several buildings have either two or
three sensors located on the ground level floors because the structur
es are long or they contain a
significant element joint halfway down the longitudinal axis of the building.
2019 Ridgecrest earthquake
The July 2019 Ridgecrest earthquake sequence
that occurred in southeastern California,
was recorded on hundreds of CS
N sensors in the Los Angeles basin
(Kohler et al., 2020;
Filippitzis et al., 2021). In particular,
CSN captured variations in ground-level motion and upper
floor deformation within mid
- and high-rise buildings and showed unexpected patterns of large
spatial variations in shaking amplification, as was envisioned as a primary purpose of CSN.
Work with CSN recordings of the M7.1 mainshock revealed amplified shaking in the
instrumented 52
-story high-rise in downtown Los Angeles lasting over two minutes. In addition,
ground-
level accelerations showed increases in the amplification of long
-period motions (> 1 s)
from the northern Los Angeles sedimentary basin (Fig. 4). In Fig 4, several locations show
nearly co
-located CSN and SCSN or CSMIP instrumentation (circle
s and diamond symbols in
close proximity), indicating the consistent response between stations of the different networks
for this earthquake
. High
-rises experienced unusually strong long-
period shaking in the east
-west
SMIP21 Seminar Proceedings
86
direction as a result of excitation b
y a complex train of scattered shear waves inside the basin,
including surface waves propagating in the basin. The density of the CSN observations
demonstrated that the behavior of structures (e.g. buildings) with long natural periods does not
follow long-
standing expectations for how sedimentary basins affect amplification
(Filippitzis et
al., 2021).
Figure 4. Pseudo-
spectral acceleration amplitudes using 5% damping for the 2019 M7.1
Ridgecrest earthquake in southern California. Periods shown are: (a,b) 2 s, and (
c,d) 5 s. Left
column: Greater urban Los Angeles region. Right column: Blow-up of the region inside the
marked squares on the left, showing detail. CSN stat
ions=circles. SCSN+CSMIP
stations=diamonds.
The ground-
level accelerations recorded by CSN from the Ridgecrest earthquakes also
exhibited coherent, gradational variations in the spectral amplitudes of high
-frequency
motions
across the NASA
-JPL campus that suggest correlations with geomorphological features (ridges,
canyons, and foothills).
The variations in spectral amplitudes are most pronounced for
frequencies between 1 and 3 Hz. For the M7.1 mainshock, the overall maximum amplification
occurred in the highest elevation, on top of the bedrock mesa bounding the campus to the north.
The M6.4 and M5.4 foreshocks show a similar amplification pattern. T
he amplification pattern
changes with the frequency as energy components of various wavelengths interact with surface
and subsurface features of different characteristic lengths. For this higher-frequency range, the
SMIP21 Seminar Proceedings
87
incoming waves likely
interacted
with numerous small-
size features at the site
to generate a
complex, rapidly-
varying amplification pattern.
The integration of site
-specific and structure
-specific instrumentation provided by the
Community Seismic Network
enables the deployment of large numbers of seismic sensors for
dense spatial sampling. The developed framework offers a path forward for city-
scale and
regional-
scale seismic network operations, and could serve as a scalable and reconfigurable tool
for monitoring stru
ctures such as tall buildings, bridges and dams, as well as lifeline
infrastructure.
Acknowledgements
The Community Seismic Network is the product of years of work by many people who
have made up the CSN team over the years. The author
is a part of this
team, and in
represent
ing
this team wishes to acknowledge their past and ongoing contributions to the CSN project. The
current CSN team members are Robert Clayton
(Seismological Lab; California Institute of
Technology), Thomas Heaton (Department of Mechan
ical and Civil Engineering, and
Seismological Lab, California Institute of Technology
), Richard Guy (
Department
of Civil and
Environmental Engineering, University of California, Los Angeles; and Seismological
Lab,
California Institute of Technology
), Julia
n Bunn (
Center for Data
-Driven Discovery
, Division of
Physics, Math and Astronomy, California Institute of Technology
), Filippos Filippitzis
(Department of Mechanical Engineering, University of Western Macedonia, Kozani, Greece
; and
Department
of Mechanical and Civil Engineering, California Institute of Technology
), Yousef
Bozorgnia (
Department
of Civil and Environmental Engineering, University of California, Los
Angeles) and E
rtugrul Taciroglu
(Department of Civil and Environmental Engineering,
University of California, Los Angeles). The CSN team is grateful to
Caltech
, UCLA, the Conrad
N. Hilton Foundation, and Computers & Structures Inc., the Gordon and Betty Moore
Foundation, and the National Science Foundation for providing support for the Community
Seismic Network
over the past decade.
References
Böse, M., T. Heaton, and E. Hauksson (2012). Real-
time finite fault rupture detector (FinDer) for
large earthquakes,
Geophys. J. Int., 191
, 803-812.
Böse,
M., C. Felizardo, and T. H. Heaton (2015). Finite-
fault rupture detector (FinDer): Going
real
-time in Californian ShakeAlert Warning System,
Seis. Res. Lett.,
86(6), 1692-1704.
Böse, M., D. E. Smith, C. Felizardo, M.-A. Meier, T. H. Heaton, and J. F. Clinton (2018). FinDer
v.2: Improved real-
time ground-
motion predictions for M2
-M9 with seismic finite
-source
characterization,
Geophys. J. Int.,
212
, 725-742.
Clayton, R., T. Heaton, M. Chandy, A. Krause, M. Kohler, J. Bunn, R. Guy, M. Olson, M.
Faulkner, M. H. Cheng, L. Strand, R. Chandy, D. Obenshain, A. Liu, and M. Aivazis (2011).
Community Seismic Network,
Annals of Geophysics
, 54, 6; doi: 10.4401/ag-5269.
SMIP21 Seminar Proceedings
88
Clayton, R. W., T. Heaton, M. Kohler, M. Chandy, R. Guy, and J. Bunn. (2015). Community
Seismic Network: a dense array to sense earthquake strong motions,
Seis. Res. Lett
., 86, 1354-
1363, doi: 10.1785/0220150094.
Clayton, R., M. Kohler, R. Guy, J
. Bunn, T. Heaton, and M. Chandy (2020). CSN/LAUSD
netwo
rk: A dense accelerometer network in Los Angeles schools,
Seis. Res. Lett.,
91(2A), 622-
630, doi:10.1785/0220190200.
Cochran, E. S., J. Bunn, S. E. Minson, A. S. Baltay, D. L. Kilb, Y. Kodera, M. Hoshiba (2019).
Event detection performance of the PLUM ear
thquake early warning algorithm in southern
California,
Bull. Seis. Soc. Am.,
109 (4), 1524–1541, doi:10.1785/0120180326.
Filippitzis, F., M. D. Kohler, T. H. Heaton, R. W. Graves, R. W. Clayton, R. G. Guy, J. J. Bunn,
and K. M. Chandy, Ground motions in urban Los Angeles from the 2019 Ridgecrest earthquake
sequence,
Earthquake Spectra
, version available online, doi:10.1177/87552930211003916,
2021.
Given, D., E. Cochran, T. Heaton, E. Hauksson, R. Allen, M. Hellweg, J. Vidale, and P. Bodin
(2014). Technical implementation plan for the ShakeAlert production prototype system—An
Earthquake Early Warning system for the West Coast of the United States,
U.S. Geological
Survey Open
-File Report 2014–1097
, 25 pp.
Given, D. D., R. M. Allen, A. S. Baltay, P. Bodin, E. S. Cochran, K. Creager, R. M. de Groot, L.
S. Gee, E. Hauksson, T. H. Heaton, M. Hellweg, J. R. Murray, V. I. Thomas, D. Toomey, and T.
S. Yelin (2018). Revised technical implementation plan for the ShakeAlert system—
An
earthquake early warning system for the West Coast of the United States,
U.S. Geological Survey
Open
-File Report 2018–1155
, 42 pp., doi:10.3133/ofr20181155.
Hazus (2020).
Hazus Earthquake Model Technical Manual.
Federal Emergency Management
Agency, Washington, D.C. https://www.fema.gov/sites/default/files/2020
-
10/fema_hazus_earthquake_technical_manual_4-2.pdf
.
Kohler, M. D., T. H. Heaton, and M. H. Cheng (2013). The Community Seismic Network and
Quake-Catcher Network: enabling structural health monitoring through instrumentation by
community participants,
Proceedings of the SPIE Smart Structures/Non-
destructive Evaluation
Conference
, San Diego, CA, March 10-14, 2013.
Kohler, M. D., T. H. Heaton, M. H. Cheng, and P. Singh (2014). Structural health monitoring
through dense instrumentation by community participants: the Community Seismic Network and
Quake-
Catcher Network,
10th U.S. National Conference on Earthquake Engineering
(
10NCEE
),
Anchorage, Alaska, July 21-25, 2014.
Kohler, M. D., R. Guy, J. Bunn, A. Massari, R. Clayton, T. Heaton, K. M. Chandy, H.
Ebrahimian, and C. Dorn (2018). Community Seismic Network and localized earthquake
situational awareness,
11
th
U.S. National Conference on Earthquake Engineering
(
11NCEE
), Los
Angeles, CA, June 25-29, 2018.
SMIP21 Seminar Proceedings
89
Kohler, M. D., D. Smith, J. Andrews, A. Chung, R. Hartog, I. Henson, D. Given, R. de Groot,
and S. Guiwits (2020). Earthquake Early Warning ShakeAlert 2.0: public rollout,
Seis. Res. Lett
.,
91(3), 1763–1775, doi:10.1785/0220190245, 2020.
Kohler, M. D., F. Filippitzis, T. H. Heaton, R. W. Clayton, R. G. Guy, J. J. Bunn, and K. M.
Chandy (2020). 2019 Ridgecrest earthquake reveals areas of Los Angeles
that amplify shaking of
high-
ris es,
Seis. Res. Lett
., 91(6), 3370–3380, doi:10.1785/0220200170.
Massari A., M. Kohler, R. Clayton, R. Guy, T. Heaton, J. Bunn, K. M. Chandy, D. Demetri
(2017). Dense building instrumentation application for city-
wide structural health monitoring
and resilience,
16th World Conference on Earthquake Engineering
(
16WC
EE
), Santiago, Chile,
January 9-13, 2017.
Seismogram Transfer Program (STP) Reference Manual (2007). Prepared by the Southern
California Earthquake Data Center, Seismological Laboratory, California Institute of
Technology, Version: 1.01, https://scedc.cal
tech.edu/data/stp/STP_Manual_v1.01.pdf.
Wald, D., K.
-W. Lin, K. Porter, and L.
Turner (2008). ShakeCast: Automating and improving the
use of ShakeMap for post-earthquake decision-making and response,
Earthquake Spectra
, 24, 2,
533-553.
Worden, C. B., E. M. Thompson, M. Hearne, and D. J. Wald (2020). ShakeMap Manual Online:
Technical Manual, User’s Guide, and Software Guide, U. S. Geological Survey,
http://usgs.github.io/shakemap, doi: 10.5066/F7D21VPQ.