A prototype system for earthquake early-warning
and alert management in southern Italy
Iannaccone G.1*, Zollo A.2, Elia L.3, Convertito V.1, Satriano C.3, Martino C. 3,
Festa G. 2, Lancieri M. 1, Bobbio A. 1, Stabile T.A. 3, Vassallo M. 3, Emolo A. 2
(1) Istituto Nazionale di Geofisica e Vulcanologia, Osservatorio Vesuviano,
via Diocleziano 328, 80124 Naples, Italy
(2) Università di Napoli “Federico II”, Dipartimento di Scienze Fisiche,
Complesso Universitario Monte Sant‟Angelo, via Cinthia, Naples, Italy
(3) AMRA scarl, via Nuova Agnano 11, 80125 Naples, Italy
Keywords: accelerometric network; earthquake early warning; real-time data
analysis; southern Apennines
*
Corresponding author: Giovanni Iannaccone, iannaccone@ov.ingv.it
1
Abstract
The Irpinia Seismic Network (ISNet) is deployed in Southern Apennines along
the active fault system responsible for the 1980, November 23, MS6.9 CampaniaLucania earthquake. It is set up by 27 stations and covers an area of about 100x70
km2. Each site is equipped with a 1-g full-scale accelerometer and a short-period
velocimeter. Due to its design characteristics, i.e. the wide dynamic range and the
high density of stations, the ISNet network is mainly devoted to estimating in realtime the earthquake location and magnitude from low- to high- magnitude events,
and to providing ground-motion parameters values so to get some insights about
the ground shaking expected. Moreover, the availability of high-quality of data
allows studying the source processes related to the seismogenetic structures in the
area. The network layout, the data communication system and protocols and the
main instrumental features are described in the paper. The data analysis is
managed by Earthworm software package that also provides the earthquake
location while homemade software has been developed for real-time computation
of the source parameters and shaking maps. Technical details about these
procedures are given in the article. The data collected at the ISNet stations are
available upon request.
Introduction
Over the last few centuries, the southern Apennines have been struck by several
strong earthquakes, the last of which occurred on 23 November, 1980 (Ms=6.9).
This resulted in more than 3,000 casualties and extensive damage throughout the
area. In terms of the scientific literature, this last earthquake has been one of the
most studied of those that have occurred in the Mediterranean area (Westaway
and Jackson, 1984; Bernard and Zollo, 1989). Similarly, many studies have
investigated the crustal structure underneath the Apennine chain, to provide
constraints for the geodynamic evolution in this sector of the Mediterranean
region, and specifically to obtain seismic-wave propagation models, which are the
2
key elements for all seismological studies (Chiarabba and Amato, 1996; Improta
et al, 2000; 2003).
At present, the southern Apennines, and in particular the Irpinia area, are
characterized by a background of continuous seismic activity that is probably
connected to the seismogenetic fault system that generated the 1980 main shock.
Magnitudes here are generally lower than 3, with occasional greater magnitude
events, like the earthquake on 3 April, 1996 (ML = 4.9). At the southern border of
the Irpinia seismogenetic fault system, these moderate energy events occur mainly
with strike-slip mechanisms, with a preferred fault plane oriented in the E-W
direction (i.e. the 1990-1991 seismic sequences, with M 5.2 and M 4.4,
respectively).
Finally, based on an analysis of historic and recent seismicity, and
according to a seismotectonic regionalization of the Italian peninsula, Boschi et al.
(1995) have indicated a probability in the range 0.22-0.41 for the occurrence of an
earthquake of M ≥5.9 in the Irpinia region in the next 20 years. Similarly, Cinti et
al. (2004) have provided a probability map of M >5.5 earthquakes predicted over
the next 10 years in Italy, and they indicate that the Campania-Lucania sector of
the southern Apennines has one of the highest probabilities of occurrence.
In 2005, with the financial support of the local government of Regione
Campania, the development of the local seismic network in the southern
Apennines started. This is known as ISNet, the Irpinia Seismic Network, and it is
designed around two main concepts: (i) to provide high quality data for studies
relating to seismogenic faults in the area; and (ii) to test a prototype system for
earthquake early warning and post-event warning for the protection of
strategically relevant infrastructure in the Campania region.
ISNet was set-up to acquire strong-motion records of large earthquakes
near to their source, along with very low magnitude local events, and records of
distant earthquakes (teleseisms). Consequently, each seismic station is equipped
with an accelerometer with a 1-g dynamic range and short-period seismometers;
furthermore, selected sites are equipped with broad-band sensors.
To realize an earthquake early-warning system that is reliable and as
robust as possible, we considered several constraints in the planning stage of the
hardware of the network. Examples here include redundancy in the
telecommunication pathways, so as to avoid data loss in the case of failure of a
3
radio link, and the storing and analysis of data, which are performed on different
sites distributed throughout the area of the network. This has been realized by
organizing the network into „sub-nets‟, each of which is managed by a data
concentrator (LCC, Local Control Center). Each node of the network can process
and analyze the seismic waveforms acquired in real-time, and can provide the
measured quantities to its closest LCC. As more stations record a seismic signal,
the new measurements are sent to and processed by the LCCs, which cross-check
the information coming from the different stations. This provides an output of
progressively refined estimations of the earthquake location and magnitude, along
with the associated uncertainties.
Similarly, to ensure the reliability of the final results, we have combined
different methodologies for the performing of the main analysis for early-warning
purposes, and we have developed software for real-time monitoring of the
functional status of the main components of this seismic network. This monitoring
will allow the early-warning system to be closely managed, to maintain its
functionality. The software is thus now in use by the staff of ISNet, to manage,
monitor and maintain the instrumentation, and by researchers, to access, analyze
and edit the seismic data that is being acquired. It also constitutes the means
through which the seismogram recordings and the data produced are made
available to scientific users.
This paper describes the characteristics of this earthquake early-warning
system that has been developed and is now under testing in southern Italy,
providing the technical aspects of its core infrastructure, the ISNet, and describing
its functional modalities.
ISNet layout and instruments
ISNet is a high dynamic range, dense seismographic network, that has been
deployed in southern Italy, along the Campania-Lucania Apennines (Weber et al.,
2007). The network covers an area of about 100 km × 70 km, over the active
seismic faults system that generated the 1980, M=6.9, Irpinia earthquake (Figure
1). It constitutes the core infrastructure for a regional Earthquake Early-Warning
System (EEWS) that remains under development today. ISNet is primarily aimed
at providing an alert to selected target sites in the Campania Region upon the
4
occurrence of moderate to large earthquakes (M>4), and to promptly compute
regional ground-shaking maps.
ISNet is currently composed of 27 seismic stations and five LCC data
storage and processing sites. All of the stations are equipped with a strong-motion
accelerometer (Güralp CMG-5T) and a three-component velocity meter (Geotech
S-13J), with a natural period of one second, thus ensuring a high dynamic
recording range. Five stations feature a broad-band velocity meter (Nanometrics
Trillium, 0.025-50 Hz), to record regional and teleseismic events and to provide
useful data for analysis of ambient seismic noise, which is aimed at obtaining a
shear-velocity model of the region. The full recording dynamic range is ±1g, and
the sensitivity is sufficient to record M 1.5 events at a distance of more than 40
km.
The seismic stations are housed in shelters, each of which is equipped with
two solar panels and two batteries. The data acquisition from the six channels is
performed by a Linux based embedded computer (74 MHz ARM CPU),
connected to a GPS receiver, and with a removable Compact Flash card (5 GB)
for local data archiving. The data logger from each station communicates with its
closest LCC through the Wi-Fi directional antennae and a wireless bridge. Sensor
data is thus continuously transmitted to remote servers too, for further archiving
and processing. Each station also houses a programmable device that is equipped
with a GSM modem, to send environmental data from the shelter (battery levels,
open door, fire alarm) in the form of text messages, either automatically or on
demand.
The stations are positioned within two imaginary concentric ellipses, about
10 km apart, with their major axes parallel to the Apennine chain. In the outer
ellipse, the average distance between stations is 20 km, in the inner ellipse it is 10
km. The network topology features multiple star-shaped sub-networks, with a few
stations and an LCC at their center. This ensures a fast and robust distributed data
analysis, through the multiple processing nodes, and a redundant and fully digital
communication infrastructure: a wireless radio link between each seismic station
and its nearest LCC; a higher bandwidth wireless backbone (under deployment)
between LCCs; redundant connections between the LCCs and the network control
center (NCC), located in Naples (Figure 1).
5
Finally, before installation, the sensor/ data-logger pairs were fully
calibrated for single-channel responses by an automated process. This calibration
covers the entire frequency spectrum, and uses the LabVIEW/MatLab software
package that provides the transfer function in graphical mode and in terms of
poles and zero.
Real-time data management
The real-time data management and analysis of ISNet is realized through several
levels that match the physical structure of the network (Figure 2). The first level is
the data logger, where the signal is digitized and time-stamped. From each single
physical channel, the data logger can provide several virtual channels, with
different sampling rates.
Each data logger uses the SeedLink protocol
(http://www.iris.edu/data/dmc-seedlink.htm) to send a real-time waveform data
stream to the associated LCC. This runs the SeisComP software (Hanka et al.,
2001), which acts as a hub for data collection and distribution. Indeed, external
users can obtain real-time data streams from ISNet stations by connecting to one
or more LCCs, using the SeedLink protocol. On top of SeisComP, each LCC runs
the Earthworm real-time analysis software (Johnson et al., 1995), which processes
data streams and performs filtering and automatic P-phase picking. The
permanent storage for data streams managed by Earthworm is performed at each
LCC using the Winston Wave Server software package
(http://www.avo.alaska.edu/Software/winston/W_Manual_TOC.html). This
software keeps a MySQL database of continuous waveforms and provides
segments of data on request. Moreover, Winston can serve a request for several
days worth of data as an image (helicorder), and for the day-to-day monitoring of
the stations. Since just an image is sent from the LCC to the requesting client, and
not the actual data, this feature helps save bandwidth.
An Earthworm installation running at the NCC performs the event
detection. This centralizes all of the phase readings coming from the LCCs and
performs phase association and event location using the “binder” module. The
6
binder computes the time difference between every pair of P arrivals and
performs a back-projection of this value, to search for a volume within a spatial
grid where the hypocenter is likely to be. When six or more consistent arrival
times are detected, a new event is declared. After its declaration each event is
relocated by an L1-norm, linearized algorithm, which uses the previously
determined hypocenter as its starting point. If new arrivals enter the binder, these
are first checked against the active events, or, should it be the case, used to declare
a new event.
The waveform and parametric data (source location and origin time) for
each event detected are stored in a database, the details of which are provided in
the next sections. The automatic event detection is at the basis of our near realtime analysis system, which will be discussed hereinafter.
The ISNet near real-time analysis system
The Earthworm seismic management software that runs at each LCC and at the
NCC is capable of real-time analysis. It provides a number of modules to perform
common tasks, like estimating the local magnitude or measuring the peak ground
values for ground shaking-map computation. However, implementing a new
feature as an Earthworm module is not a trivial task, since it requires a good
knowledge of the C programming language and a careful handling of the
input/output routines.
For this reason we decided to make use of Earthworm up to the automatic
event detection (performed at the NCC by the “binder_ew” module), while we
designed a custom, near real-time, system for computation of earthquake source
parameters and ground-shaking maps. The basic idea behind this system is that a
seismologist who is able to write the computer code to analyze off-line data could
easily make his work part of a near real-time processing chain, regardless of the
programming language he uses and without entering into the details of the
input/output strategies. We based our system on three key concepts: simplicity,
flexibility and extendibility.
An outline of the ISNet near real-time analysis system is shown in Figure
3. The system is structured as a processing chain, where each module is executed
once the previous one terminates. The chain is launched every 2 min: the next
7
instance of the chain can process a new event while the previous event is still
processed by the earlier instance. The modules can be logically divided in two
families:
Core modules. These are designed to interact with Earthworm, to: build a
list of events (00_parse_events); keep track of the P-arrival times used for
event association (01_parse_picks); and download event waveforms from
the Earthworm wave server and save them as sac files (02_get_traces and
06_get_full_traces). Core modules are connected to the underlying
network management system and need to be replaced by equivalent
modules if a different system is used.
User defined modules. These modules only rely on the existence of an
event file (with event id, and location, as reported by the binder), a pick
file, and the waveforms (in sac format) associated to each event.
All of the modules are written as Linux Bash shells, although this is not
mandatory. Several modules make use internally of sac macros, awk scripts
and/or custom Fortran code.
The results of the automatic analyses are published on an interactive web
page, called “ISNet Bulletin” (Figure 4). This page is designed around a Google
map, which covers the upper half of the page, and shows the event locations and
the stations. The default view is centered on ISNet, but it is possible to zoom in
and out. The second half of the page shows a table view of the events, with the
associated parameters. The fields are: event id, origin date, origin time, latitude,
longitude, depth, ML, MW, place (toponym), number of triggered stations, Sdisplacement spectra, and ground-shaking maps. The methodologies used to
compute all of these parameters are explained in the following paragraphs.
Some of these entries are clickable, and provide additional information.
For instance, clicking on the number of triggered stations pops up a window with
the recorded waveforms, while for a click on the place name, a balloon appears on
the map with detailed information of the event. This includes origin time, ML,
location, and focal mechanism if available. Finally, the controls in the last column
allow you to display the ground-shaking map on a Google map, for peak ground
8
acceleration (PGA), peak ground velocity (PGV) and instrumental intensity, or a
plot of the measured peak ground values.
Real-time analysis for early-warning purposes
For the real-time analysis for early-warning applications we are developing a
stand-alone software system, SeismNet Alarm, that is currently deployed at the
Network Control Center in Naples for testing the performance analysis. SeismNet
Alarm is implemented by a C++ application, and it can process the live stream of
the three-component acceleration recorded at all of the stations. Alternatively, it
can run in simulation mode, whereby it uses locally stored files that contain the
waveforms recorded by the stations for relevant events that have happened in the
past. In real-time mode, the application needs to retrieve the station data in
SeedLink format. Hence, for each station, it creates a processing thread that opens
a connection with the SeedLink server running at the relevant LCC, implemented
by SeisComP. Each thread is in charge of retrieving and buffering the data, and
carrying out the automatic P-wave-arrival detection. The main processing thread
takes care of binding picks from several stations to an event identifier, thus
detecting an event, locating the hypocenter and determining the event magnitude.
The main steps performed by this system are thus the following:
Arrival detection. We currently run a picking algorithm, based on that of
Baer and Kradolfer on each vertical component. This produces an arrival time
and its associated uncertainty for each station.
Picks binding. This phase determines whether new picks from the stations are
compatible with a new event that has just occurred, or with an ongoing event
already declared, rather than due to unrelated local phenomena, such as
anthropogenic or environmental noise. Several sets of information are
exploited to perform this step, such as the temporal coincidence of the picks at
several stations, the time sequence of the picks and the location of the
triggering sites.
Event location. This step is performed by the RTLoc algorithm (Satriano et
al., 2008), an evolutionary, real-time location technique based on an equal
9
differential time (EDT) formulation and a probabilistic approach for
describing the hypocenter. The location estimate is not only based on the
arrival times at the stations that are triggered, but also takes into account that
at the time of each computation some stations may not have been triggered.
With just one recorded arrival, the hypocentral location is constrained by the
Voronoi cell around the first triggering station, which is constructed using the
travel times to the not-yet-triggered stations. With two or more triggered
arrivals, the location is constrained by the intersection of the volume defined
by the Voronoi cells for the remaining, not-yet-triggered stations, and the EDT
surfaces between all pairs of triggered arrivals. As time passes, and more
triggers become available, the evolutionary location converges to a standard
EDT location.
Event magnitude estimation. The recorded acceleration is band-pass filtered
to focus on low frequencies, and converted to the overall peak displacement of
the ground. This is done over two temporal windows, starting at the measured
P-wave arrival and the estimated S-wave arrival, encompassing 2 s to 4 s of
signal. An empirical relationship that correlates the final event magnitude with
the logarithm of these quantities and the distance from the event to the station
is then used to yield a magnitude for each station. These are in turn combined
to produce an early estimate of the event magnitude, and of its uncertainty,
which evolves while the earthquake is occurring.
Each of the steps from event detection onwards triggers an alarm message that can
be sent over a dedicated network line to selected target sites. While the event
propagates at a speed of around 3.5 km/s from the its origin to the target, the
alarm messages can be sent almost instantly to front-end applications running at
the target site that can, for instance, initiate an automatic shut down procedure of
an infrastructure. For a destructive earthquake occurring in the Irpinia region, and
a target site in the city of Naples, this means that there is an interval of the order
of 20 s from when the alarm reaches the target, to when the destructive waves
arrive there.
For resilience to failures of the early-warning system or the network,
which will be somewhat more likely while an energetic event is occurring, a
future goal is to deploy several instances of the system within the network, at each
10
LCC, thus producing redundant sources of alarm. This will be possible due to the
decentralized architecture of ISNet, which provides several processing nodes, and
a redundant communication infrastructure.
It is worth noting that SeismNet Alarm is actually relatively neutral with
respect to the underlying seismic network. In fact, it uses the broadly available
SeedLink communication protocol to retrieve the seismic data. Furthermore, it can
be tailored for different network topologies, alarm thresholds, by altering its
configuration files. Of course, this requires a preliminary tuning phase for the
target network, achieved by testing the system with real-time and recorded data.
Magnitude estimations
For ISNet, different methods of estimating magnitudes are operative. We have
developed a local magnitude scale to provide external general information on the
seismicity of the area, and we routinely evaluate the moment magnitude for
seismological studies on the source properties of the recorded events. One
advantage of the Moment Magnitude scale is that unlike other magnitude scales, it
does not saturate at the upper end. That is, there is no particular value beyond
which all large earthquakes have about the same magnitude. For magnitudes
smaller than about 3, local magnitudes significantly underestimate the moment
magnitude (e.g. Deichmann, 2006), due to inaccurate distance attenuation effects
and instrumental corrections. Thus, according to the policy established by the
USGS (http://earthquake.usgs.gov/aboutus/docs/020204mag_policy.php), when it
is available, the moment magnitude is the preferred magnitude estimate for our
network.
Finally, for seismic early-warning applications, we have developed a realtime, probabilistic and evolutionary algorithm for estimation of magnitude, which
is aimed at predicting the ground-motion intensity at a given target site.
Local Magnitude
The local magnitude scale has been developed from synthetic Wood-Anderson
equivalent seismograms, using data recorded by ISNet (Bobbio et al, 2008).
Wood-Anderson displacements are synthesized from the waveforms recorded at
the ISNet seismic stations, by removing the response curve of the specific
11
instrument and by filtering according to the high frequency characteristic response
of the Wood-Anderson torsion seismograph, with eigenperiod T=0.8s, damping
factor 0.8 and magnification V=2800.
Data coming from horizontal components of short-period instruments and
accelerometers of ISNet are initially integrated to provide effective displacement.
The scaling law of amplitude, log A0 with the distance has been calibrated on a
dataset of events recorded at the ISNet stations from January 2006 to June 2008,
with the constraint that the magnitude of events with maximum amplitude of 1
mm is 3, at an epicentral distance of 100 km. Assuming a scaling with distance
with the following functional form:
log A0 n log R kR
where the logarithmic contribution mainly accounts for the geometrical spreading,
while the linear term is referred to the anelastic attenuation. Minimizing the L2
distance between observed amplitudes and predicted ones, according to the
Ricther law, we obtain the following relation that is valid for the southern
Apennines:
ML = log A + 1.79 log R - 0.58
where A is the maximum amplitude, in mm, and R, the hypocentral distance in
kilometers.
The local magnitudes of the earthquakes recorded at ISNet are computed
as the algebraic means of the magnitude values estimated at each station.
Generally, averaging over a larger number of stations (more than five) that
explore a broader distance range, the estimated error is about 0.2-0.3 (Bobbio et
al., 2008).
Moment Magnitude
The moment magnitude is derived from the estimation of the seismic moment
through the non-linear inversion of the S-wave displacement spectra obtained by
the spectral analysis of horizontal acceleration and the velocity records at the
ISNet stations. Only the stations that have been used for automatic event location
12
are included in the seismic moment determination. Based on the earthquake
location parameters, a window of 5 s bracketing the theoretical S-wave arrival is
selected on the horizontal ground velocity and acceleration records. The standard
signal processing chain is perfomed using SAC code and it includes: (i) the mean
and trend removal; (ii) the application of a cosine-taper; and (iii) a band-pass twopole, zero-phase shift, and Butterworth filtering in two frequency bands, 1-50 Hz
and 0.25-50 Hz, for the acceleration and velocity time series, respectively. The
parameters for signal processing were chosen after preliminary tests that were
aimed at optimizing the displacement spectral determination from acceleration
and velocity records. The Fourier acceleration and velocity spectra are therefore
obtained by Fast Fourier transform from which the displacement spectra are
obtained by double and single division for the term (i ) . The spectra obtained are
smoothed using a three-point moving window. The displacement spectra of the
horizontal components (NS and EW) are combined to build the spectral modulus:
D( ) NS ( )2 EW ( ) 2
where is the angular frequency. The displacement spectra obtained are fitted to
a theoretical model having the form (Boatwright,1980):
t
o
D( )
e 2
(1 ( / c ) 2
*
Where Ω0 is the low-frequency spectral level related to the seismic moment Mo
t*
(Aki and Richards, 1980), ωc = 2π fc , with fc the corner frequency and
TS
QS
the anelastic attenuation parameter, where T and Q are the S-wave travel-time and
quality factor. The parameters Ω0, fc and t*
are estimated by the non-linear
inversion of displacement spectra, using the Levenberg-Marquardt (Kenneth and
Levenberg,
1944)
algorithm
implemented
in
GNUPLOT
(http://www.gnuplot.info). This allows for best-fit estimations of parameters and
related uncertainties. For each station, an estimate of the seismic moment is
obtained assuming a homogeneous propagation medium (Aki and Richards,1980):
13
Mo
4 vS3 Ro
R FS
where R is the hypocentral distane, =2700 Kg/m3 is the medium density, vS
=3000 m/s, R = 0.62, as the average S-wave radiation pattern, and Fs=2, as the
free-surface correction factor. The final values of the seismic moment and the
uncertainties are computed by averaging the values obtained from accelerationand velocity-derived displacement spectra at each station analyzed. The average
moment magnitude and the standard deviation are obtained by seismic moment
estimates using the the relationship:
2
MW (log10 M o 9.1)
3
where Mo is expressed in N.m (Hanks and Kanamori, 1979). The spectral
parameters inferred from the displacement spectrum inversion also allow for the
simultaneous determination of the source radius (Brune,1970):
(a
2.34 vS
)
2 f c
and stress drop (Keilis-Borok,1959):
(
7 Mo
)
16 a3
Early-Warning Magnitude
The real-time and evolutionary algorithm for magnitude estimation is based on a
magnitude prediction model and a Bayesian formulation (Lancieri and Zollo,
2008). It is aimed at evaluating the conditional probability density function (PDF)
of magnitude as a function of ground motion quantities measured on the early part
of the acquired signals.
The predictive models are empirical relationships that correlate the final
14
event magnitude with the logarithm of quantities measured in the first 2 s to 4 s of
recording. In this application, we use the empirical relationship between low-pass
filtered, initial P-peak and S-peak displacement amplitudes, and moment
magnitude (e.g. Zollo et al, 2006). While the P-wave onset is identified by an
automatic picking procedure, the S-wave onset can be estimated from automatic
S-picking or from a theoretical prediction based on the hypocentral distance given
by the earthquake location. At each time step, progressively refined estimates of
magnitude are obtained from the P-peak and S-peak displacement data. Following
a Bayesian approach, the magnitude PDF computed at the previous step is used as
a-priori information.
Generation of the rapid ground-shaking map
In areas characterized by high seismic hazard and exposure, such as the southern
Apennines, the generation of strong ground-shaking maps soon after an
earthquake is a key tool to identify the areas that have suffered the greatest
damage and losses. This information is fundamental for emergency services, loss
estimation, and planning of emergency actions by the Civil Protection Authorities.
Ground-shaking maps are usually computed using an appropriate
weighting scheme, with interpolation of the peak ground motion recorded at
seismic stations with values estimated at a set of points (denoted as phantom or
virtual stations) located in areas where data are not available. At phantom stations,
the ground-motion parameters, such as PGA and PGV, are estimated using
attenuation relationships based on an empirical model of attenuation and a pointlike source, which are generally represented by the following formulation:
where PGX is the selected strong ground motion parameter (PGA or PGV), M is
the magnitude, R is a distance, h the depth of the hypocenter, and logPGx is the
standard error. The coefficients a, b and c have to be retrieved specifically for
each region. For the southern Apennines region, using the available seismological
data an ad-hoc attenuation relation has been deduced (Convertito el al., 2007).
The coefficients obtained for formula (1) are reported in Table I.
15
Using an attenuation relationship like formula (1), the predicted strongmotion field is isotropic around the epicentral area, while the observed field
shows a bi-dimensional distribution that depends upon both source-to-site
distance and the azimuth caused by fault geometry, focal mechanism, and
directivity effects. These effects are partially accounted for in a different way by
the existing techniques adopted for ground-shaking map computation. For
example, to account for the fault geometry, ShakeMap (Wald et al., 1999a) uses a
schematic representation of the fault, i.e. a box representing the surface fault
projection, and uses the minimum fault distance definition instead of the
epicentral distance.
Taking advantage of ISNet, a tool for the rapid estimation of groundshaking maps after moderate-to-large earthquakes has been developed (Convertito
at al, 2008). Named as GRSmap, its main features include:
-
The determination of peak parameters at phantom stations using
observed and predicted data at the same time, by the attenuation
relationship reported in Table I. In this way, the azimuthal properties
of the recorded peak-ground-motion field are preserved.
-
The automatic choice of the parameters controlling the distribution
of the phantom stations, mainly based on the density of the seismic
network.
This is obtained by dividing the area where the ground-shaking map has been
calculated into two zones: the area covered by the seismic network, denoted as the
data domain, and the external area to the seismic network. Different techniques
are then used in these two cases, both to define the location of the phantom
stations and to correct the estimated ground-motion values, to bring them into line
with the observations that implicitly contain source and propagation effects. In the
data domain, a triangulation scheme is used to obtain a uniform distribution of
stations covering the area of interest, while in the external area, a regular grid of
phantom stations is used.
The methodology used to develop the GRSmap software can be
schematically summarized as follow:
Triangulation of the data domain:
16
Recorded PGA and PGV values are reported to rock-site conditions using
ad-hoc corrective coefficients obtained using the same approach as
proposed by Borcherdt (1994) and Park and Ellrick (1998), and as
retrieved by Cantore et al. (2008).
The seismic stations correspond to the vertices of the triangles. For each
triangle, the barycenter is identified and used as a phantom seismic station.
The area of each triangle cannot exceed NAAave, where NA is an integer
that depends on the density of the seismic network, and Aave is the average
area of all of the triangles. The triangles with areas exceeding the fixed
threshold are recursively triangulated using the new barycenters as
additional vertices. At all of the new barycenters the ground-motion
parameters are assigned using the adopted attenuation relationships
corrected by the average residual calculated on a fixed number of real
seismic stations.
The epicenter is considered as an additional station where the peak
ground-motion values are estimated using the attenuation relationship at R
= 0 km. A correction is then applied, corresponding to an average residual
computed at a number of stations surrounding the epicenter below a
critical distance value (dc) that depends on the seismic network density.
For earthquakes located outside of the data domain area, triangulation of
the epicentral area is made denser and denser until a uniform station
distribution is obtained.
Ground-motion-residual estimation:
Given the optimal triangulation, the residuals are calculated at each vertex
of the triangles by comparing the observed and the predicted groundmotion values obtained by the attenuation relationship proposed by
Convertito et al. (2007) (see Table I). The average residual is then used to
correct the predicted value at each barycenter.
Once the earthquake location and magnitude have been fixed, the
attenuation relationship is used to obtain theoretical estimates at the
network recording sites. Considering the i-th triangle (Fig. 5, inset, panel
a), the vertices of which are labelled as P1, P2 and P3, the peak motion
residual term at the j-th vertex is computed as the difference between the
17
observed and the estimated peak-ground motion. The maximum
acceptable residual value is fixed on the basis of the estimated fault length
(L), obtained by using the Wells and Coppersmith (1994) relationships,
and an epicentral area is defined by a circle of radius L/2 centered on the
epicenter (panel a in Fig.5). The residuals cannot exceed NlogPGX, where
logPGX is the standard error of the selected attenuation relationship. The
value of N is generally fixed at 4 for sites located inside the epicentral
area, and otherwise at 3. If a single residual is outside the fixed range, the
datum is considered as an outlier and is not used in the map computation.
Otherwise, for a given triangle, the average peak-motion-residual term is
then obtained and used to estimate the peak-motion amplitude at the i-th
barycenter point Bi. The procedure is repeated for all of the triangles and
iterated until a uniform coverage of the data domain area is obtained. This
allows for a local correction, which accounts for azimuthal variations due
to source effects, like directivity and focal mechanisms.
Extrapolation of peak motion in the external area
For the area not covered by the seismic network, the first problem is the definition
of the optimal grid spacing of the phantom stations. Another problem is
represented by the definition of the threshold distance to the closest station where
recorded data are available. This distance provides an empirical measure of the
extent to which the observed data can be extrapolated outside the data domain
area. In the proposed technique, the external area is covered with a uniform grid
of phantom stations, the spacing interval of which is fixed to a fraction of the
average distance between the stations and barycenters. The same value is used for
the threshold distance (dc) (Fig. 5, panel a).
Among all of the nodes of the grid, only those located at distances greater
than the threshold value from the closest recording station are retained (Fig. 5,
circles). At each retained node, the peak ground-motion parameter is then
predicted using the attenuation relationship, adding a mean residual weighted for
the epicentral distance, computed at seismic stations with an azimuth with respect
to the epicenter, comparable with that of the phantom station considered.
The estimated and recorded data are than integrated and used to generate
the ground-shaking map by re-interpolating onto a finer regular grid that is
18
uniformly spaced at an arbitrary spacing interval of 0.01 degree. This map is
finally corrected for site effects using the corresponding corrective coefficients
(Cantore et al., 2008).
Application to the 23 November, 1980, Irpinia earthquake (M 6.9)
The GRSmap software has been applied to compute ground-shaking maps of the
last destructive earthquake that occurred in the southern Apennines: the 23
November, 1980, Irpina M 6.9 earthquake. This was characterized by a complex
normal faulting that ruptured three different sub-parallel fault segments of the
southern Apennine belt chain (Westaway and Jackson, 1984; Bernard and Zollo,
1989). The parameters of the three faults are listed in Table 2. Figure 5 shows the
location of the accelerometers (triangles) of the local seismic network managed by
ENEA-ENEL (Berardi et al., 1981) at which data were available, and the
instrumental epicenter (grey star). The phantom stations (circles), the triangulation
scheme, and the barycenters (black dots) are also shown in Figure 5. The average
area of the triangles is about 473 km2, while the threshold distance and phantom
spacing of the grid is about 62 km.
To highlight the advantages of the technique proposed in the present study,
the ground-shaking maps were calculated using a version of the attenuation
relationships obtained by excluding the data of the Irpinia earthquake from the
dataset (Table I). The computed ground-shaking maps are shown in Figure 6. In
particular, Figure 6a shows the PGA maps expressed as percentages of the gravity
acceleration, Figure 6b shows the PGV map expressed in cm/s, and Figure 6c
shows the map of the instrumental intensity.
Note that although the predictive attenuation model was based on the
assumption of a point-like source, the maps reproduce the extension of the three
fault segments and the associated complex ground-motion pattern. This can be
attributed to the use of recorded data and corrected estimates at the barycenters
that provide improved coverage of the source area. Both the PGA and PGV maps
reproduce the directivity effect, which is towards the north-west for fault segment
F1 and towards the south-east for fault segment F2, and which is characterized by
the larger ground-motion values in those directions. Furthermore, as a result of
using the weighted average scheme proposed by Wald et al. (1999b) to convert
19
PGA and PGV into instrumental intensities, the instrumental intensity map is
directly connected to the PGA and PGV maps.
Software for the hardware and data management
To manage all of the hardware comprising ISNet, the software systems that are
running and the data produced by the network, we have developed a custom
application: SeismNet Manager. This application acts as a high level, web-based
graphical front-end to the network, for internal full management of ISNet, as well
as for external users who are interested in the seismological data acquired.
SeismNet Manager provides an instrumental and seismological database to
keep track of the hardware components that comprise the network, and of the data
they produce. The application fulfills the following needs:
to keep an inventory and to store the details of the components that
constitute a seismic network, including: station sites, sensors, loggers,
communication and generic hardware, and servers;
to keep a history of the installations and configurations of these
components, and of their mutual connections;
to perform real-time monitoring of the devices: retrieving their internal
variables, and detecting “health” problems and assessing the quality of
their output, thus producing alarms and information that complement the
seismic data;
to manage the seismic data produced by the network. These data are either
automatically retrieved, e.g. events from bulletins, automatically detected
events, and related waveforms, or manually inserted by the researchers,
e.g. arrival times, alternative event magnitudes and locations, and focal
mechanisms;
to perform some routine tasks on the seismic data, such as inspection,
filtering, picking and flagging;
to offer a graphical, web-based interface to the staff of the network for
inserting, editing, searching, downloading and displaying all of this
information (as tables, graphs, maps, waveform plots, 3D renderings).
20
SeismNet Manager is implemented through open technological components, and
can roughly be broken down into these main components:
the web application, that provides the user interface, controls the hardware
monitoring, and offers various tools to edit and display data. It is
composed of JavaServer Pages code (run by Apache Tomcat [1]), and Java
programs and applets;
a relational database for both instrumental and seismic data, implemented
in PostgreSQL [2];
several small programs, written in various languages, called agents. Each
agent is in charge of communicating with a different type of hardware that
is deployed as part of the network. The real-time hardware monitoring is
implemented through this plug-ins based approach;
procedures for the automatic acquisition of waveform data, from
heterogeneous data sources such as logger disks, Earthworm servers, and
FTP servers.
In the following paragraphs the management of the hardware forming ISNet is
initially described, followed by the management of the data produced.
Hardware Management
Through SeismNet Manager, it is possible to create a new object belonging to one
of several hardware types (e.g. logger, sensor, server) and to fill in the details of
that physical object. Some details are common to any hardware type, e.g. model
name, serial number, inventory number, vendor name. Other fields are specific to
each class of object e.g. number of channels of a data logger, physical quantity
recorded by a sensor. It is then possible to create stations and LCCs, and
communication lines between them, and to install devices at each site. A
hardware-specific configuration corresponds to each installed device, and a series
of connections with other nearby devices. A database of the entities mentioned,
each valid from a start date to an optional end date, records all of the details of the
ISNet hardware at any instant in time.
[1]
[2]
http://tomcat.apache.org
http://www.postgresql.org
21
To complement this “static” knowledge that is manually entered by the
administrators of the network, there exists a hardware monitoring layer, which
analyzes the internal state and working conditions of the hardware. Devices with
an IP address, such as loggers, servers and communication hardware, are routinely
queried for their most significant internal variables, to identify “health” problems.
Typical variables are the power supply voltage of a device or its internal
temperature, the disk space in a server, or the data-flow parameters from a logger.
The queried variables, as well as the communication protocol, in general
depend on the hardware type (and brand). For this reason, the interrogations are
carried out by several apposite external programs called agents, one for each
hardware class. The hardware monitoring is configured by choosing the target
devices, the agents to be used with them, and the starting times and frequencies of
the interrogations. Additionally, each station features a programmable GSM
phone terminal that is connected to several environmental sensors in the shelter,
e.g. the door, batteries level, smoke sensor. This sends a text message whenever
one of the thresholds is met.
All of the internal variables and the gathered information on the state of
the instrumentation are stored in the database, and can be shown as tables or
graphed directly in the browser.
The front page of SeismNet Manager (figure 7) is meant to convey the
state of the whole network at a glance. It consists of a map with stations, LCCs,
and communication lines. Overlaid on each station are: a color-coded overall
state; the installed sensors and their working conditions; icons for any problem
detected by the hardware monitoring agents or messages sent by the station.
Events, waveforms and seismic data
SeismNet Manager contains a seismological database that keeps track of the
seismic events detected by the network, with the associated metadata and
waveforms recorded by the sensors. The main source of events is the automatic
earthquakes detection system that runs at each LCC, implemented in Earthworm.
Upon detecting an event, the earthquake metadata (location, magnitude
estimation, focal mechanism) is sent to SeismNet Manager. For events that are not
automatically detected, such as regional and teleseismic ones, SeismNet Manager
22
makes use of alert messages and bulletins produced by national and international
seismological agencies.
New events signaled to SeismNet Manager are first tested against some
rather conservative magnitude and distance thresholds, to filter out too distant or
too weak earthquakes. Then the procedures for automatic waveform data retrieval
from the stations are activated. These procedures exploit several waveforms
sources (e.g. Earthworm servers, mass storage in the data loggers, FTP servers
with manually obtained data) to retrieve the sensor signals, in a time window that
includes the expected recording of the event at each site. To determine the seismic
stations and time window to retrieve data from, the procedures take into account:
the sensor type (e.g. only broad-band sensors for teleseismic events); the P-wave
arrival time at each station, estimated using the IASPEI travel-time tables
[3]
for
regional and teleseismic events, or a custom velocity model for local events; the
earthquake time length duration, computed through a regression law between
magnitude and duration (for local and regional events), or other criteria based on
distance and magnitude (teleseisms).
Each waveform entering the system is converted into a uniform file
format. We chose the SAC [4] file format (Seismic Analysis Code, from Lawrence
Livermore National Laboratory), with the file header filled with the details of the
associated event, the estimated arrival time, the originating site and instruments
that recorded the data. A data quality parameter is also assigned to each
waveform, automatically computer-evaluating the signal to noise ratio of the
signal level of the recorded earthquake compared to the noise level before the
event. Users can then search events and waveforms by defining multiple search
criteria on a web page (figure 8). Events can be filtered for time and location,
magnitude value and type, and epicentral distance. Waveforms can be filtered for
station, sensor type and model, and component and quality. Waveforms recorded
by a sensor flagged as having issues can be filtered out. Additionally, it is possible
to filter out all of the three components from a sensor, whenever even a single
component has a quality below that requested.
The waveforms matching the search criteria, and the associated metadata,
can be downloaded as a compressed archive, or viewed and manipulated through
[3]
[4]
http://www.iris.edu/pub/programs/iaspei-tau
http://www.iris.edu/software/sac
23
the SeisGram2K [5] Java applet (figure 9). The matching events can be rendered as
an interactive 3D scene, using a browser plug-in for VRML (Virtual Reality
Modeling Language) Files (figure 9). All of the data associated with a seismic
event can be displayed and edited through either web pages or java applets. In a
typical session, a logged in user will:
-
display the waveforms associated to a seismic event with SeisGram2K,
including the estimated arrival times, as computed when the waveform was
inserted;
-
manually revise these picks;
-
submit changes to the system (by clicking a button). This will automatically
compute a new earthquake location, magnitude, and focal mechanism. The
system retains the previous values and keeps track of the author and
timestamp of each change, making it easy to choose among several authors, or
to revert to previous solutions.
A demonstration tour of SismNetManager can be found here:
http://dbserver.ov.ingv.it:8080
Discussion and conclusions
The system presented here has been developed in the framework of an ongoing
project financed by the Regional Department of Civil Protection, with the idea
that Regione Campania can be considered as a potential EEWS target-site for
experimenting innovative technologies for acquisition, rapid processing,
management and diffusion of data based on ISNet. Indeed, with about six million
inhabitants and a large number of industrial plants, the Campania Region
(southern Italy) is a zone of high seismic risk due to moderate to large magnitude
earthquakes on active fault systems in the Apennine belt. Considering an
earthquake warning window ranging from tens of seconds before to hundred of
seconds after an earthquake, many public infrastructures and buildings of strategic
relevance (hospitals, gas pipelines, railways, railroads) in the Regione Campania
can be considered as potential EEWS target-sites for experimenting innovative
technologies for data acquisition, processing and transmission, based on ISNet.
[5]
http://alomax.free.fr/seisgram/SeisGram2K.html
24
The expected time delay to these targets for the first energetic S-wave train is
more than 20 s at about 100 km from a crustal earthquake occurring in the source
region.
At present, several EEWS have been implemented worldwide. In Japan,
since 1965, the JNR (Japanese National Railways) has developing and is
operating the Urgent Earthquake Detection and Alarm System (UrEDAS) system,
which is an on-site warning system along the Shinkansen railway. UrEDAS is
based on seismic stations deployed along the Japanese Railway at average
distances of 20 km, and an alert is issued if the horizontal ground acceleration
exceeds 40 cm/s2 (Nakamura, 2004). Furthermore, an innovative EEWS started
nationwide in Japan at the end of 2007, managed by the Japan Meteorological
Agency (JMA) using data from more than 1,000 seismic stations (Hoshiba et al.,
2008). After a quick determination of the hypocenter and magnitude using records
from the closest stations, a predicted arrival time of shear waves is provided for
districts where the seismic intensity is predicted to be equal to 4 or more on the
JMA scale. A step-by-step procedure is adopted to improve the accuracy of the
estimation as the available data increase with elapsed time. The information are
automatically disseminated by the JMA to the final users, who are classified as
limited or general users. The limited users are organizations (railway companies,
elevator companies, manufacturing industries) who can carry out an automatic
check of their system. For the general users, the earthquake early-warning alarms
are provided by various means, such as television, radio, cellular phone and the
Internet (Hoshiba et al., 2008).
In Taiwan, the Taiwan Central Weather Bureau (CWB) has developed an
early-warning system based on a seismic network consisting of 79 strong-motion
stations (Wu and Teng, 2002). Since 1995, the network has been able to report
event information (location, size, strong-motion map) within 1 min of the
earthquake occurrence (Teng et al. 1997). To reduce the reporting time, Wu and
Teng (2002) introduced the concept of a virtual sub-network: as soon as an event
is triggered by at least seven stations, the signals coming from the stations that are
less then 60 km distant from the estimated epicenter are used to characterize the
event. This system was operating from December 2000 to June 2001 (7 months),
and it successfully characterized all of the 54 events that occurred, with an
average reporting time of 22 s.
25
Other systems are under development in Mexico, Turkey, Romania and
California. An extended review of the existing early-warning systems is reported
in a special volume of “Seismic Early Warning”, edited by Gasparini et al. (2006),
and by the study of Zollo et al. (2008a).
However, how can it be verified whether an EEWS is functioning
correctly? The main test would be to wait until a significant number of
earthquakes have been recorded, also of medium to large energy, and to verify the
number of alarms that have correctly been sent, along with the number of false
alarms and alarms missed. Moreover, it is necessary to verify the significance of
each alarm, including the useful time before the arrival of the destructive seismic
wave, and the predicted amplitude at a site with respect to that which is actually
recorded. For instance, the EEWS operating in Japan by JMA was tested for 29
months, starting in February 2004. During this period, the JMA sent out 855
earthquake early warnings, with only 26 recognized as false alarms due technical
problems or human error (Hoshiba et al., 2008).
For the area of the southern Apennines, because of the scarcity of
relatively large earthquakes, this means that it is difficult to experimentally test
this EEWS based on ISNet. Many tests have been performed using low energy
earthquakes, with magnitudes of about 3, but we believe these tests are actually
not fully significant. Therefore, we have decided to use synthetic seismograms
that have been computed at all of the recording sites of our seismic network to
evaluate the performance of the implemented EEWS. We have considered several
cases of earthquakes of M 6 and M 7 occurring inside or at the border of ISNet,
and we have performed a massive computation of seismograms for a large number
of characteristic earthquake scenarios (Zollo et al., 2008b). By using the
computational methodologies previously described, we have retrieved early
estimates of source parameters and we have predicted the peak ground motions
(PGA, PGV) at selected sites. In this way, we have investigated the system
performances in cases of complex, extended rupture processes, and the seismic
source characteristics such as directivity, rupture velocity distribution and nearfield contributions have been considered. Two parameters are used to define the
system performance: Effective Lead Time (ELT), i.e. the time at which the
probability of observing the true PGV, within one standard deviation, becomes
stationary; and the Probability of Prediction Error (PPE), which provides a
26
measure of the PGV prediction error. The geographical distribution of ELT and
PPE for the southern Apennines shows a significant variability up to large
distances around the fault, thus indicating that the ability of the system to
accurately predict the observed peak ground motions strongly depends on the
distance and azimuth from the fault. Assuming an earthquake with similar source
characteristics to that of the November, 1980, Ms=6.9 earthquake for the
metropolitan area of Naples (see Figure 1), the ELT ranges between 8 s and 16 s,
and the PPE between 50% and 60%, indicating that several mitigation actions
could be effective before S-waves shake the town (Zollo et al, 2008b).
ISNet is thus set up to acquire strong-motion records of large earthquakes
near to their source, along with very low magnitude local events, and records of
distant earthquakes (teleseisms). The data recorded are inserted into our database,
which now comprises more than 1,050 events with 0.1 ≤ ML ≤ 3.0, with more
than 23,000 three-component traces. This dataset grows at the rate of about 30-35
events with an ML ≤ 3 per month, providing us with an outstanding tool for the
analysis of the microseismicity in the area.
AVAILABILITY OF DATA
All of the seismic waveform data archived by the ISNet-Irpinia Seismic Network
are available upon request directly at info@isnet.amracenter.com. Alternatively,
waveform data can be retrieved from the SeismNetManager (SAC format). To
access SeismNet Manager, it is necessary to register an account and to
authenticate this first. The form to request access to the ISNet data can be found at
the following address: http://dbserver.ov.ingv.it:8080.
27
References
Aki K, Richards P (1980) Quantitative Seismology, Theory and Methods, vols. I
and II
WH Freeman, New York
Berardi RA, Berenzi, Capozza F (1981) Campania-Lucania earthquake on 23
November 1980: accelerometric recordings of the main quake and relating
processing. Technical report Ente Nazionale per l‟Energia Elettrica (ENEL),
Rome
Bernard P, Zollo A (1989) The Irpinia (Italy) 1980 earthquake: detailed analysis
of a complex normal fault. J Geophys Res 94:1631-1648
Bobbio A, Vassallo M, Festa G (2008) A Local Magnitude Scale for the Southern
Italy. Bull Seismol Soc Am, submitted.
Borcherdt RD (1994) Estimates of site-dependent response spectra for design
(methodology and justification). Earthquake Spectra 10:617-654
Boschi E, Gasperini P, Mulargia F (1995) Forecasting Where Larger Crustal
Earthquakes Are Likely to Occur in Italy in the near future. Bull Seismol
Soc Am, 85:1475-1482
Brune JN (1970) Tectonic stress and the spectra of seismic shear waves from
earthquakes. J Geophys Res 75:4997-5009
Cantore L, Convertito V, Zollo A, Elia L (2008) Site-condition map and site
amplification corrections for early-warning applications and seismic hazard
assessment in the Campania-Lucania Region (southern Apennines), Italy, in
preparation
Cinti, F R, L Faenza, W Marzocchi, P. Montone (2004), Probability map of the
next M 5.5 earthquakes in Italy, Geochem. Geophys. Geosyst., 5, Q11003,
doi:10.1029/2004GC000724
Chiarabba C, Amato A. (1996) Crustal velocity structure of the Apennines (Italy)
from P-wave travel time tomography. Annali di Geofisica, 34, N.6, 11331148.
Convertito V, De Matteis R, Romeo A, Zollo A, Iannaccone G (2007) Strong
motion relation for early-warning applications in the Campania Region
(southern Apennines), Italy. In Gasparini et al (eds) Earthquake early
warning systems, Berlino, Springer-Verlag.
28
Convertito V, De Matteis R, Zollo A Iannaccone G (2008) Rapid ground motion
prediction for emergency management in the Campania-Lucania region,
Southern Italy by using ISNet seismic network. Natural hazards, in press.
Deichmann N (2006) Local Magnitude, a Moment Revisited. Bull Seismol Soc
Am 96:1267-1277
Gasparini P, Manfredi G, Zschau J (2006) Earthquake early warning systems,
Berlino, Springer-Verlag.
Hanka W, Heinloo A, Jaeckel K (2001) Networked Seismographs: GEOFON
Real-Time
Data
Distribution,
OHP/ION
symposium
"Long-Term
Observations in the Oceans: Current Status and Perspectives for the Future",
Mt. Fuji, Japan, January 21 - 27, 2001.
Hanks TC, Kanamori H (1979) A moment magnitude scale. J Geophys Res 84:
2348–2350
Hoshiba M, Kamigaichi O, Saito M, Tsukada S, Hamada N (2008),
Earthquake Early Warning Starts Nationwide in Japan. Eos Trans. Vol.89,
8, 73-80
IASPEI new manual of seismological observatories practice (NMSOP) (2002).
Ed. Peter Bormann, Geoforschungs Zentrum Postdam.
Improta L, Iannaccone G, Capuano P, Zollo A, Scandone P (2000) Inferences on
the upper crustal structure of Southern Apennines (Italy) from seismic
refraction investigations and subsurface data. Tectonophysics, 317, 273-297.
Improta L, Bonagura MT, Capuano P, Iannaccone G (2003) An integrated
geophysical investigation of the upper crust in the epicentral area of the
1980, Ms=6.9, Irpinia earthquake (Southern Italy). Tectonophysics 361:
139-169.
Johnson CE, Bittenbinder A, Bogaert B, Dietz L, Kohler W (1995) Earthworm: A
flexible approach to seismic network processing. IRIS Newsletter 14(2), 14.
Keilis-Borok V (1959) On the estimation of the displacement in an earthquake
source and of source dimensions. Ann. di Geofisica 12:205-214
Kenneth Levenberg (1944) A Method for the Solution of Certain Non-Linear
Problems in Least Squares. The Quarterly of Applied Mathematics 2: 164–
168.
29
Lancieri, M, Zollo A (2008) A Bayesian approach to the real-time estimation of
magnitude from the early P and S wave displacement peaks. J. Geophys.
Res. 113- B12302
Nakamura Y (2004) Uredas, Urgent Earthquake Detection and Alarm System,
Now and Future 13th World Conference on Earthquake Engineering 908
Pantosti D and Valensise G (1990) Faulting mechanism and complexity of the 23
November, 1980, Campania-Lucania earthquake inferred form surface
observations, J. Geophys. Res. 95:319-341
Park S, Elrick S (1998) Predictions of shear wave velocities in southern California
using surface geology. Bull Seismol Soc Am 88:677-685
Reiter L (1990) Earthquake hazard analysis - issues and insights. Columbia
University Press, New York, 254 pp.
Richter C (1935) An instrumental earthquake magnitude scale. Bull Seismol Soc
Am 25:1-32.
Satriano C, Lomax A, Zollo A (2008) Real-Time Evolutionary Earthquake
Location for Seismic Early Warning. Bull Seismol Soc Am 98:1482-1494
Teng, T. L., Y-M. Wu, , T. C. Shin, Y.B. Tsai and W.H.K. Lee (1997) One
minute after: strong-motion map, effective epicenter, and effective
magnitude. Bull. Seism. Soc. Am., 87, 1209-1219.
Wald DJ, Quitoriano V, Heaton TH, Kanamori H, Scrivner CW, Worden CB
(1999a) TriNet shakemaps: rapid generation of instrumental ground motion
and intensity maps for earthquakes in southern California. Earthquake
Spectra 15:537-555.
Wald DJ, Quitoriano V, Heaton TH, Kanamori H (1999b) Relationship between
peak ground acceleration, peak ground velocity, and modified Mercalli
intensity for earthquakes in California. Earthquake Spectra 15:557-564.
Westaway RWC, Jackson J (1984) Surface faulting in the southern Italian
Campania-Basilicata earthquake of 23 November 1980. Nature 312:436438.
Weber E, Convertito V, Iannaccone G, Zollo A, Bobbio A, Cantore L, Corciulo
M, Di Crosta M, Elia L, Martino C, Romeo A, Satriano C (2007) An
advanced seismic network in the southern Apennines (Italy) for seismicity
investigations and experimentation with earthquake early warning:
Seismological Research Letters 78, 622-534
30
Wells DL, Coppersmith KJ (1994) New empirical relationships among
magnitude, rupture length, rupture width, rupture area, and surface
displacement. Bull Seismol Soc Am 84:974-1002.
Wu Y-M and Teng T (2002) A virtual subnet approach to earthquake early
warning. Bull Seismol Soc Am 92, 2008-2018
Zollo A, Lancieri M, Nielsen S (2006) Earthquake magnitude estimation
from peak
amplitudes
of
very early seismic
signals
on
strong
motion records. Geophysical Research Letters 33- L23312
Zollo A., Iannaccone G., Convertito V., Elia L., Iervolino I., Lancieri M., Lomax
A., Martino C., Satriano C., Weber E. and Gasparini P., (2008a), The
Earthquake Early Warning System in Southern Italy. Encyclopedia of
Complexity and System Science - Springer Verlag.
A.Zollo A, Iannaccone G, Lancieri M, Cantore L, Convertito V, Emolo A, Festa
G, Gallovic F, Vassallo M, Martino C, Satriano C and Gasparin P (2008b) The
Earthquake Early Warning System in Southern Italy : Methodologies and
Performance Evaluation, submitted to Geophysical Research Letters
31
Tables
Pgx
a
b
c
h
Pga (m/s2)
-0.559
0.383
-1.4
5.5
0.155
Pgv (m/s)
-3.04
0.552
-1.4
5.0
0.154
Pga (m/s2)a
-0.514
0.347
-1.4
5.5
0.145
Pgv (m/s)a
-3.13
0.570
-1.4
5.0
0.185
Table 1: Regression coefficients and standard errors of the regional attenuation
relationship used to compute the ground-shaking maps (Convertito et al., 2007).
The superscript a indicates the coefficients of the same attenuation relationships
obtained without introducing the PGA and PGV values of the 23 November 1980
Irpinia earthquake into the dataset.
Parameter
F1
F2
F3
Length
35 km
20 km
20 km
Width
15 km
15 km
10 km
Depth of the
2.2 km
10 km
2.2 km
top
Strike
315°
300°
124°
Dip
60°
20°
70°
Slip
-90°
-90°
-90°
Seismic
21019 Nm
41018 Nm
31018 Nm
moment
Table 2: Fault parameters of the 23 November 1980 Irpinia earthquake (after
Bernard and Zollo, 1989)
32
FIGURES:
Figure 1: The ISNet network in Campania-Lucania Apennine (Southern Italy).
Green squares indicate seismic stations. Yellow lines symbolize wireless radio
links between each seismic station and its nearest Local Control Center (LCC,
blue circles). Gray lines represent higher bandwidth, wireless connections among
LCCs and the Network Control Center (red star). The latter transmission system is
conceived as a redundant double ring.
33
Figure 2: Real-Time data management at the ISNet is organized into three logical
layers, which follow the physical structure of the network. The base layer is the
data-logger, where the ground motion signal is digitized, time stamped and sent
over a network connection. The middle layer is the Local Control Center (LCC)
where real-time analysis is performed on data from the attached stations.
Furthermore each LCC maintains a waveform database for local stations. The top
layer is the Network Control Center (NCC), where phase association and event
detection is performed and where the network-wide database is kept. Also the
NCC provides facilities for other applications (seismic early warning, near-real
time processing, etc.) and for end users.
34
Figure 3: Scheme of the near real-time analysis procedure at the ISNet. The
procedure is organized as a chain where each module is activated after the
previous one. The whole chain is run every 2 minutes; several chains can run in
parallel. The modules are logically divided into two families: "Core system",
which comprises modules that interact with the underlying Earthworm system,
and "User defined modules".
35
Figure 4: The “ISNet Bulletin” interactive web page. Circles in the Google map
on the upper half of the page represent events detected by the system. The events,
with the associated parameters, are reported in the interactive table on the second
half of the page. Additional information for each event is reported in the map or
in a pop-up page by simply clicking on one or more parameters of the event. As
an example, the instrumental intensity and the detailed information, including
focal mechanism, for a ML=2.8 event are displayed.
36
a)
b)
Figure 5: (a) Schematic representation of the main parameters used to triangulate
the data domain area and to cover the area external to the seismic network. (b)
Location of the stations of ENEL-ENEA network and triangulation scheme used
to compute the ground shaking map of the 23 November 1980 Irpinia earthquake.
Figure 6: Ground shaking maps of the 23 November 1980 (M 6.9) Irpinia
earthquake. Panel (a) shows the ap of PGA, panel (b) shows the map of PGV and
panel (c) shows the map of Instrumental Intensity. Triangles correspond to the
recording stations, red dots correspond the virtual stations obtained from the
triangulation procedure while empty circle correspond to the phantom stations
used to cover the area external to the seismic network. The labels F1, F2 and F3
identify the three fault segments which ruptured during the Irpinia earthquake.
37
Figure 7: The map page of SeismNet Manager, showing the current overall state
of ISNet. We can see: the Local Control Centers (in cyan), the stations (with a
color coded working condition), the installed sensors (evidenced by a red outline
if they have problems), the data links (high bandwidth ones are thicker), the
alarms sent by the stations (mail icons), the internal state of the hardware (tick
sign if all is well, blinking alert or no-connection icon otherwise).
38
Figure 8: Events and waveforms can be searched using this interface. Events can
be filtered for origin time, location, magnitude and distance to the stations.
Waveforms can be filtered for seismic network, station, sensor and data quality.
39
Figure 9: Events can displayed as an interactive 3D rendering in the web browser
(left). Waveforms can be viewed and processed through the SeisGram2K Java
applet (right). The parametric information associated to the waveforms (e.g.
picks) can also be edited through this applet.
40