Content uploaded by Aldo Zollo
Author content
All content in this area was uploaded by Aldo Zollo on May 19, 2015
Content may be subject to copyright.
Earthquake Early Warning System for Schools:
A Feasibility Study in Southern Italy
by M. Picozzi, A. Emolo, C. Martino, A. Zollo, N. Miranda, G. Verderame,
T. Boxberger, and the REAKT Working Group
INTRODUCTION
Over the last few decades, metropolitan areas have experienced
a dramatic increase of exposure to earthquakes. Because of the
still very low probability level at which short-term earthquake
forecasting is feasible, earthquake early warning systems
(EEWS) currently represent an effective, pragmatic, and viable
means for the implementation of protective measures to reduce
the exposure of population to seismic risk. EEWS can be simply
defined as systems that integrate seismic networks and software
capable of performing real-time data telemetry and analysis to
provide alert messages to users within seconds of the beginning
of an earthquake and certainly before that the Swaves gener-
ated by the event reach the users.
Worldwide, a number of EEWS capable of rapidly per-
forming seismological analysis of ground motion during a
strong earthquake are currently operative or are under real-time
testing (e.g., in Japan, Taiwan, Mexico, Italy, Turkey, California,
etc.; Allen and Kanamori, 2003;Kanamori, 2005;Allen et al.,
2009). A tangible example of EEWS benefit is given by the ex-
perience of the OKI factory in the Miyagi prefecture in Japan
that, after having experienced severe damages in 2003 due to
two earthquakes, installed an EEWS, and at the following earth-
quakes experienced a significant reduction of losses (Allen et al.,
2009). Moreover, the Japan Meteorological Agency has shown
the effectiveness of a combined on-site and network-based ap-
proach to rapidly broadcast the rapid warning after a potential
damaging earthquake (Hoshiba, 2013).
The key parameter of EEWS is the lead time, which is the
time available for taking protective measures at distant targets
once an earthquake has been promptly detected and character-
ized and an alarm is issued. Obviously, lead times depend on
the distance between the earthquake source and the targets to
be protected, as well as the time necessary for the implemen-
tation of protective measures, hence may vary from no warning
at all (i.e., in the blind zone) to tens of seconds if the source is
at greater distance (e.g., further than 100 km).
As discussed by Goltz (2002), when training and prepar-
edness of the population programs are implemented, the public
safety benefit of early warning systems is in response readiness
so the population can take various protective measures (e.g.,
duck and cover, turn off gas burners, move away from windows
or equipment, etc.) within the available lead time and thus to
reduce the risk of injury and minimize damage.
Concerning the institutional sector of education, in the
framework of a study for the Governor’s Office of Emergency
Services in California, Goltz (2002) defined (in agreement
with end-users) the actions that might be taken in schools
when lead times of 10 and 50 s were available (Table 1).
The current Italian legislation (i.e., D.000M. 26.8.92;
D.Lgs. 626/94–242/96, and others deriving from them)
prescribes the establishment of emergency plans and security
actions to take in case of natural disasters. In summary, in case
of an earthquake, one teacher who previously was designated as
the emergency coordinator has the following duties: (1) con-
sider the necessity and the moment to evacuate the school by
the activation of a dedicated warning signal, (2) immediately
stop the gas and power distribution, (3) warn the individuals
who are responsible for evacuations at each school level so they
are ready for the evacuation, and (4) coordinate all the oper-
ations till the end of the emergency. During the emergency, the
other teachers must be always in contact with the emergency
coordinator to start the evacuation plan. The students must
(1) protect themselves under their desks from possible falling
objects, (2) calmly proceed to the safe zones as defined by the
emergency plan, and (3) follow the previously defined instruc-
tions for the evacuations.
This summary of tasks and actions to be included in an
seismic emergency plan highlights how difficult it can be for
the emergency coordinator during an earthquake to make the
right decision on what to do within seconds, often when the
S-wave ground motion has already started and with a dramatic
lack of real-time external information on what is occurring. In
this context, EEWS potentially represent a unique source of
information for assisting the emergency coordinators in their
duties, allowing them the valuable seconds needed to start the
emergency procedures before the Swaves generated by the
event reach the school.
In the framework of the Strategies and Tools for Real Time
Earthquake Risk Reduction (REAKT; www.reaktproject.eu;last
accessed January 2015) FP7 European project, we have been en-
gaged in a feasibility study on the application of earthquake early
warning procedures in the high school Istituto Tecnico Indus-
triale Somma Vesuviana (ITIS) (http://itismajoranasommaves.it;
last accessed January 2015) “E. Majorana,”Somma Vesuviana
(hereafter referred to as MAJI) located near the Irpinia region,
southern Italy (Fig. 1). MAJI is also part of the program Seis-
mology@School (www.sismoscholar.it;lastaccessedJanuary
398 Seismological Research Letters Volume 86, Number 2A March/April 2015 doi: 10.1785/0220140194
2015), developed in the framework of the Network of European
Research Infrastructures for Earthquake Risk Assessment and
Mitigation (NERA) project, which consists of installing profes-
sional seismic instruments in schools combined with suitable di-
dactic materials to increase both the scientific knowledge and the
consciousness about seismic hazard and risk.
Table 1.
Example of Actions That Might be Taken with 10 and 50 s Warning in Schools
Actions with 10 s Warning Actions with 50 s Warning
Get walkie-talkies
Notify teachers with walkie-talkies or speaker
Have custodian shut off gas
Alert custodial staff to secure building
Shut off machines, move away from laboratory
equipment
Notify security to be on alert
Move clear of falling objects
Shut off gas
Send out/gather emergency supplies
Contact fire department, district office, and police department
Secure laboratory equipment/evacuate laboratory
Sound alarms
Turn off computers
Evacuate to secure meeting points
Contact plant manager
Protection for students in the hallways, restrooms, etc., wandering
students
Get walkie-talkies, cell phones
Initiate emergency response plan
Notify security to be on alert
Focus on protecting younger children
Modified from Goltz (2002).
▴Figure 1. Locations of the high school Istituto Tecnico Industriale Somma Vesuviana (ITIS) (http://itismajoranasommaves.it; last ac-
cessed January 2015) “E. Majorana,”Somma Vesuviana (MAJI) and the Irpinia Seismic Network (ISNet) in Campania–Lucania Apennine
(southern Italy). Seismic stations (triangles) and cities (squares) are shown.
Seismological Research Letters Volume 86, Number 2A March/April 2015 399
The Irpinia area has been affected by strong-magnitude
earthquakes, both in historic and recent times. The last de-
structive earthquake was the 23 November 1980 Ms6.9 Irpinia
earthquake, which caused more than 3000 casualties and pro-
duced huge and widespread damage (Ameri et al., 2011;
Chiauzzi et al., 2012). Since 2012, MAJI has been monitored
by a total of five accelerometric stations installed in different
parts of the school building. Furthermore, commercial asymmet-
ric digital subscriber lines (ADSL) provide transmission of real-
time data to an earthquake early warning (EEW)centerinNa-
ples, where data streams are acquired in real time in the PRob-
abilistic and Evolutionary early warning SysTem (PRESToPlus;
regional and on-site threshold-based early warning) software
platform (Satriano et al.,2011;Zollo et al.,2014). MAJI there-
fore has been provided with the means to receive alert messages
from the regional EEWS given by the integration of PRESToPlus
into the seismic network operating in the Irpinia area, the Irpi-
nia Seismic Network (ISNet; Iannaccone et al.,2010;Satriano
et al., 2011). They also benefit from warnings produced by the
on-site EEWS, which integrates PRESToPlus with the instru-
ments installed at the school itself.
After realizing that a low-cost tool capable of rapidly in-
terpreting the information coming from the EEWS in use and
translating it into a simple safe/unsafe scheme for supporting
the emergency coordinator during a seismic crisis was still miss-
ing, we used a low-cost intelligent electronic device (i.e., Ardu-
ino, http://www.arduino.cc; last accessed January 2015) to design
an actuator for EEW (hereafter referred to as the Sentinel). We
programmed the Sentinel to accomplish three main tasks: (1) lis-
ten and interpret messages delivered by the EEWS PRESToPlus
on the ground-motion severity expected at the target site, (2) pro-
vide different warnings to indicate different alert levels by the
control of different hardware (i.e., alarm bells, emergency lights,
and so on), and (3) declare the end of the most threatening con-
dition, which will assist the emergency coordinator in starting the
evacuation plan defined by the law. Similar to PRESToPlus,which
performs an evolutionary event characterization, the Sentinel
provides evolutionary real-time alerts to users.
In the following, we first describe the characteristics of the
accelerometers network installed at MAJI. Then we summarize
the main characteristics of both the PRESToPlus system and of
the EEW Sentinel. Finally, we present the preliminary results of
offline analyses carried out using both real and synthetic data to
test the feasibility of the early warning procedures and, in
particular, the delivery of warnings in real time to MAJI.
THE SEISMIC NETWORK INSTALLED AT THE
SCHOOL (MAJI)
The network of seismic stations installed at MAJI consists of
one standard accelerometric station and four low-cost wireless
accelerometers (Fig. 2). The high-quality accelerometric station
(hereafter referred to as MAJ3) consists of a 24-bit analog-to-
digital converter (ADC) Agecodagis Kephren data logger and a
Güralp CMG-5TC accelerometer with a 0:25gfull-scale, uses a
sampling of 125 Hz, and was installed in the school courtyard.
The four low-cost accelerometers are SOSEWIN stations
(Fleming et al., 2009;Picozzi et al., 2010,2014); these are low-
cost sensing units equipped with Micro-Electromechanical Systems
(MEMS) accelerometers with 0:2mgof resolution, ADCswith
effective 20 bits in high-resolution mode that sample the ground
motion at 100 Hz, and a wireless router applications platform (i.e.,
the PC Engines ALIX system board), which allows them to create
a self-organizing wireless mesh information network. The SOSE-
WIN stations have been installed in different parts of the school
(Fig. 2), which presents a rather complex structure (Emolo et al.,
2014); as shown by Picozzi et al. (2011), the strong motion of
earthquakeswillberecordedatdifferentfloorsandlocationsso
will allow a rapid assess of the state-of-health of the school.
The SOSEWINs communicate by wireless with the Age-
codagis Kephren data logger, which operates as a gateway and
provides transmission of real-time data to the EEW center in
Naples by commercial ADSLs. Data streams are acquired in real
time in the PRESToPlus platform.
The present configuration of the EEWS, which has
PRESToPlus running far from the school, is clearly not optimal
for an on-site approach but was selected for the sake of
simplicity during the test phase. Of course, in future, the im-
plementation of the EEWS will be with PRESToPlus running
directly at the school.
METHODS
Earthquake Early Warning Systems and PRESToPlus
Typically, EEWS follows two basic approaches: regional (or
network based), and on-site warning. Regional early warning
systems are based on the use of a seismic network located near
one or more expected epicentral areas, for which the aims are
to detect and locate an earthquake and to determine its mag-
nitude from the analysis of the first few seconds of the arriving
Pwaves at more stations (Satriano et al., 2011). The lead time
for a regional system is defined as the time difference between
the Pwaves recorded in the source area and the arrival of first S
waves at the target site, after accounting for the necessary com-
putation and data transmission times.
On-site early warning systems are designed to cope with sit-
uations in which target sites are located too close to a seismogenic
area and the analysis of data recorded at more stations of a regional
network determines a lead time too small to warn the target in
case of an event. For this reason, on-site systems rely on seismic
sensors installed directly at the target site and exploit only the in-
formationcarriedbythefasterearlyPwaves to infer the larger
shaking related to the incoming Sand surface waves (i.e., the lead
time is equal to the P-wave minus the S-wave arrival times).
Furthermore, Zollo et al. (2010) demonstrated the two
approaches can be profitably integrated within a unique system
that allow the early warning prediction of the potential damage
zone (PDZ) (i.e. the area in which the highest intensity levels
associated with an event are expected). Clearly, the integration
of regional and on-site approaches is particularly useful when-
ever target sites are threatened by more than a single seismic
source area, and these latter are placed at variable distances
400 Seismological Research Letters Volume 86, Number 2A March/April 2015
from the target sites. An exhaustive review of the concepts,
methods, and physical basis of EEWS has been presented by
Satriano et al. (2010).
PRESTo is a free and open source software platform for
earthquake early warning (www.prestoews.org; last accessed Janu-
ary 2015; Satriano et al., 2011;Zollo et al., 2014)thatwasde-
veloped at the University of Naples. Following the idea proposed
by Zollo et al. (2010), the new version of the system, PRESToPlus,
implements both a regional and an on-site approach. Since 2009,
PRESToPlus has been under real-time experimentation in south-
ern Italy, on the data streams of the ISNet. PRESToPlus has been
recently released to the scientific community through the website
www.prestoews.org (last accessed January 2015).
PRESToPlus continually processes real-time streams of
three-component acceleration data for P-wave arrival detec-
tion. These data are normally streamed in real time from the
stations of a seismic network, but they can also be read from
files to provide a simulation mode whereby waveforms of past
events can be played back into the system. While an event (real
or simulated) is occurring, the software promptly performs the
event detection and provides location and magnitude estimates
as well as shaking predictions at target sites using a regional,
network-based approach (Fig. 3). The earthquake location is
obtained by an evolutionary, real-time probabilistic approach
based on an equal differential time formulation (Satriano et al.,
2011), which uses information from both triggered and not-
yet-triggered stations. Magnitude estimation exploits empirical
relationships that correlate this parameter to the filtered peak
displacement (Pd) measured over the first 2–4sofPand S
waves of high-pass (0.075 Hz) filtered signals (Lancieri and
▴Figure 2. Instruments deployed at MAJI. The station MAJ3 is installed in the school’s courtyard ground. The SOSEWIN stations are
installed in the school’s inner walls.
Seismological Research Letters Volume 86, Number 2A March/April 2015 401
Zollo, 2008;Satriano et al., 2011). Finally, peak ground-mo-
tion parameters at remote sites can be estimated through
ground-motion prediction equations (GMPEs) once location
and magnitude are available (e.g., for the Irpinia region, Emolo
et al., 2011). As shown by Zollo et al. (2010) and Satriano et al.
(2011),PRESToPlus provides robust early warning events char-
acterization by adopting the same set of parameters (e.g., time-
window length) for a broad range of magnitude (i.e., M≤7),
which covers the typical Italian seismicity (itaca.mi.ingv.it; last
accessed January 2015; Luzi et al., 2008). Of course, as sug-
gested by Colombelli et al. (2012) and others, a larger time
window for the early warning magnitude estimation should
be used for larger magnitude earthquakes (M>7). The imple-
mentation of expanding time windows in PRESToPlus is under
development.
Alarm messages containing the evolutionary estimates of
source parameters and ground motion expected at the target,
with the associated uncertainties, are sent over the Internet and
can thus also reach distant vulnerable infrastructures before the
destructive waves, enabling the initiation of automatic safety
procedures.
The regional approach to early warning is integrated with
an on-site, threshold-based method for the definition of inde-
pendent local alert levels at each station. To this end, the pre-
dominant period (τc;Wu and Kanamori, 2005) and the Pd in
a short time window after the first P-arrival time are simulta-
neously measured at each station, independently from the rest
of the seismic network. As shown by Zollo et al. (2010), Pd can
be correlated with the final peak ground velocity (PGV) and
consequently with the modified Mercalli intensity (IMM),
which is a measure of the expected damage, whereas τccan
be correlated with the earthquake magnitude. These two
parameters are compared with threshold values that define a
decisional table with four alert levels, correlated to the expected
on-site damage as well as the damage at a distance. The alert
level can be used to initiate safety measures at each site inde-
pendently of the regional processing. At the same time, at the
regional scale, the local alert levels (as they become available)
can be combined with the estimated source parameters to de-
fine the extent of the PDZ.
The Sentinel: A Low-Cost EEW Actuator for MAJI
Aiming at providing EEW information to the school users and,
in particular, the emergency coordinator we created an EEW
actuator, the Sentinel that is able to both listen and interpret
messages coming from PRESToPlus, as well as to assist the
coordinator starting the emergency procedures (e.g., alarms,
traffic-light systems). Among the low-cost programmable intel-
ligent electronic devices existing on the market, Arduino
(http://arduino.cc; last accessed January 2015) was selected
mainly for its properties of being easily programmed to sense
the PRESToPlus messages stream and to control in real-time
lights, ringing-bells, and other devices (e.g., Table 1lists mea-
sures designed to reduce the risk of injury and minimize dam-
▴Figure 3. PRESToPlus flow chart (modified from Zollo et al., 2010).
402 Seismological Research Letters Volume 86, Number 2A March/April 2015
age in case of an earthquake). Most important, the Arduino’s
simplicity created the right conditions for a deep interaction
between researchers and both school teachers and students,
which proceeded side-by-side in programming the EEW Sen-
tinel according to scientific principles and users requirements.
As shown in the schematic representation of Figure 4,the
EEWS at MAJI is designed so that PRESToPlus analyzes the
streaming of seismic data from both the on-site and the
regional network; and, in case of an earthquake, it provides the
Sentinel with an alert string containing the EEW information.
PRESToPlus is a self-contained system, thus it does not require
any other seismic software or platform to run, just the ground-
motion data from a seismic network. For this reason, it can run
both on a remote control center and on a dedicated computer
(PC) at the school. In the following sections, we present the
results of offline tests performed while running it in remote
at the Unit of Experimental and Computational Seismology
(RISSC)-Lab’sEEWcontrol center in Naples (see The Sentinel:
A Low-Cost EEW Actuator for Schools). Depending on the
site–event interdistance, and therefore on whether the regional
or the on-site has the fastest response, the Sentinel can receive
two different kind of alert messages (Fig. 5). First, it can receive
information from the regional system about the event, includ-
ing the date, the UTC time, latitude, longitude, magnitude,
and, most important, the prediction of the ground-motion
shaking in terms of PGV at the school obtained by GMPE.
Second, it can receive information from the on-site system,
including the date, the UTC time, the predominant period
(τc), the ground-motion peak associated with Pwaves (Pd),
and (derived from the latter) the expected ground shaking
for the incoming Swaves (PGV).
For regional EEW alerts (Fig. 5a), the Sentinel assigns an
alert level to the events by comparing the size of the PGV
predicted at the school and previously defined PGV thresholds
derived using, for this application, the PGV–intensity relation-
ship proposed by Faccioli and Cauzzi (2006), which translates
the PGV in different classes of perceived shaking and potential
damage conditions as in the European Macroseismic Scale
(EMS) intensity (Grünthal, 1998).
We defined the following classes, which in turn corre-
spond to different alert levels:
•The predicted PGV is smaller than 0:2cm=s, which cor-
responds to ground motion not felt from users and not
dangerous for structures (i.e., EMS intensities <IV). The
Sentinel alert level for this class is named “silent.”
•The PGV is included in the 0.2 and 6:1cm=srange, which
is related to ground motion perceived by users and poten-
tially responsible of light damage levels (i.e., EMS inten-
sities V and VI). The Sentinel alert level for this class is
named “low.”
•The PGV is larger than 6:1cm=s(EMS Intensity ≥VII)
and corresponds to very high perceived shaking and from
moderate to very high damage. In this case, the Sentinel
has an alert level defined as “high.”
In case of EEW information coming from the on-site sys-
tem, the Sentinel is programmed to assign the event an alert
level on the basis of the Pd −τcalert level scheme (Fig. 5b),
which Zollo et al. (2010) proposed interpreting in terms of
potential damaging effects near the recording station and
far away from it. In particular, the maximum Pd −τcalert level
(level 3; i.e., τc≥0:6sand Pd ≥0:2cm) corresponds to an
earthquake with predicted magnitude M≥6and an expected
instrumental intensity IMM ≥VII. This means the earthquake
is likely to have a large size and to be located close to the record-
ing site; in this case, the Sentinel alert level is “high.”In the case
of Pd −τcalert level 2 (τc<0:6sand Pd ≥0:2cm), even if
the event is estimated to have a magnitude smaller than 6, it
should be located close to the target and thus responsive to a
ground motion having the potential to cause damage at the site.
In this case, the Sentinel alert level is set to “high.”On the
contrary, when the Pd −τcalert level is 1 (τc≥0:6s and
Pd <0:2cm), the event is likely larger than magnitude 6 but
at the same time is located so far from the target site as to be
only felt there without any damage. In this case, the Sentinel
alert level is “low.”Finally, for a recorded alert level equal to 0
(τc<0:6sand Pd <0:2cm), the event is likely to be small
and far from the site, thus no damage is expected either close or
▴Figure 4. Schematic overview of the earthquake early warning
system (EEWS) for the school MAJI, which comprises an on-site
system and a regional one managed by PRESToPlus, which in turn
sends alerts to the earthquake early Warning (EEW)Sentinelplaced
within the school that can be used to warn students and teachers.
Seismological Research Letters Volume 86, Number 2A March/April 2015 403
far away from the station; and the Sentinel alert level is in this
case set to “silent.”
In the current prototypical version, the three Sentinel alert
levels have been associated with three different traffic lights and
the school’s bell alarms. Therefore, whenever an alert is released
by PRESToPlus, regardless of its origin (on-site or regional), the
message (which has the size of few tens of kilobytes and typ-
ically a latency of few tens of milliseconds only) is taken in
charge by the Sentinel and exploited to provide a P-wave-
derived prediction of the ground-motion severity and damage
level at the school. It is worth noting that, as with PRESToPlus,
the Sentinel is evolutionary. Indeed, it keeps receiving and in-
terpreting the PRESToPlus messages, which are released every
second during the whole earthquake analysis. Therefore (as
shown in the examples presented in the following sections),
according to the EEW information received during the initial
phases of the event characterization, the Sentinel alert level can
change.
The last task of the Sentinel consists in declaring the end
of the event, so that the emergency coordinator can start the
evacuation plan defined by the law. This task is accomplished
by taking into consideration the relationship proposed by
Bommer et al. (2009) to compute the ground-motion duration,
given the magnitude and epicentral distance information. In
the application for the Irpinia area, the maximum event mag-
nitude inferred from seismic-hazard analysis is estimated to
be magnitude 7.3 (Barani et al., 2009). Hence, according to
Bommer et al. (2009), for such an event, the duration of the
ground motion for a site at about 70 km from the epicenter
should be of the order of 30–40 s. For this reason, the Sentinel
is programmed to declare the end of an event after 40 s unless it
receives a message from PRESToPlus that a new event occurred
(i.e., an aftershock).
RESULTS
On-Site EEW Feasibility at MAJI: The Analysis of the
False Event Detection Rate
On-site EEWS do not include very robust algorithms for the
real-time event location, which typically led them to misrecog-
nize P- and S-wave arrivals and thus be prone to high rates of
false detections. Moreover, in the case of civil infrastructure
with several users like schools (where the on-site EEW is most
needed), the number of false detections due to anthropic noise
is likely to become huge. Solutions to this issue are commonly
sought by filtering the high-frequency signals and imposing
▴Figure 5. Sentinel alert levels (a) based on the information from the Regional EEWS, and (b) based on the on-site EEWS.
404 Seismological Research Letters Volume 86, Number 2A March/April 2015
highly selective threshold parameters, which in turn can poten-
tially let the on-site EEWS be effective for extreme events only.
Therefore, one of the first analyses carried out at MAJI was
devoted to comparing the rate of false events detected by a
single station against the use of all the stations installed at the
school, considering them as a small array. For the test, we se-
lected 24-hour-long waveforms recorded by the five stations
during four different weekdays, with school activity but no
nearby earthquakes. Although the analyses were conducted off-
line, the picking was performed using the same picker as in
PRESToPlus (FilterPicker5, Lomax et al., 2012), adopting the
same set of parameters that have been optimized for the sta-
tions of the ISNet over approximately the last about three years
of seismic monitoring and early warning activities (Table 2). As
expected, due to the high noise caused by people inside the
building and the nearby infrastructures (e.g., a factory is located
a few tens of meters from the school), especially during the day
we observed a lot of false picking on the traces (Fig. 6), up to 1
pick every 5 min for the most sensitive station (top signal in
Fig. 6). For the chosen picking algorithm, the SOSEWIN
accelerometers showed a reduced number of false pickings be-
cause in this case the anthropic signals were masked by the
high-frequency instrumental noise of these sensors. These
results suggest that a single station EEW approach would be
Table 2
The parameters of FilterPicker5 used by PRESToPlus for the
Irpinia Seismic Network (ISNet) and SOSEWIN Stations
Parameter
ISNet Stations
(Including MAJ3)
SOSEWIN
Stations
Filter window (s) 1 1
Long-term
window (s)
55
Threshold 1 (S1) 15 11
Threshold 2 (S2) 15 15
tUpEvent (s) 0.1 0.1
Refer to Lomax et al. (2012) for an exhaustive description of
the parameters.
▴Figure 6. Twenty-four-hour ground-motion acceleration recorded during Julian day 067 (Saturday, 08 March 2014) at the five seismic
stations located at MAJI (see Fig. 2for the sensors disposition). Automatic picks from FilterPicker5 (Lomax et al., 2012) are represented as
vertical lines.
Seismological Research Letters Volume 86, Number 2A March/April 2015 405
problematic to be put in practice in such a context, because it
would require a specific and probably very complicated tuning
of the picking parameters to avoid the repeated occurrence of
false event identification.
On the contrary, we observed that when almost simulta-
neous picks (e.g., within a time window of 0.5 s) for at least
three stations was required for the event declaration, the de-
tection of false events decreased to just one to two in a day,
on average. Moreover, the occurrence of false events was totally
avoided when the coincidence of four or five station picks
within 0.5 s was required.
On-site and Regional EEW Analyses of the 2013 Matese
Earthquake Recordings
The lack of seismicity observed during the first two years of the
REAKT projects finally ended the 29 December 2013 (UTC
17:08:43) when an ML4.9, Mw5.0 (http://iside.rm.ingv.it; last
accessed January 2015) earthquake occurred in the Matese hills
area, approximately 55 km northwest of the MAJI site. The
event was recorded by only four stations installed at MAJI (un-
fortunately, one SOSEWIN was not working), as well as by
ISNET, and thus these data allowed us to test both the on-site
and the regional EEW approaches.
The on-site EEW analysis was carried out considering the
MAJ3 and SOSEWIN station. Because of the source-to-school
distance, the recordings of the less sensitive SOSEWINs sta-
tions (which are equipped with an internal MEMS sensor) were
perturbed by a high-frequency local noise. Therefore, for the
stations to correctly pick the event on Pwaves, we prefiltered
the SOSEWIN waveforms with a low-pass filter at 10 Hz; and
we decreased the picker threshold for them to 11 instead of the
value of 15 used for MAJ3 (Table 2). All the available stations
pick within 0.5 s of the arrival of the Pwaves (Fig. 7a). After
applying a high-pass filter at 0.075 Hz on these data, we esti-
mated the Pd on the 4 s P-wave window of the signal; the Pd
ranges from 1:3×10−4mfor MAJ3 (installed in the ground)
to a maximum of 3:1080 ×10−4mfor one of the SOSEWIN
(SE84C, which is installed on the wall on the second story of
the school; Fig. 2). Similarly, we estimated τc, which for all sta-
tions was around 0.4 s (Fig. 7b). As mentioned by Wu et al.
(2013), because of the low signal-to-noise ratio of waveforms in
the case of small and moderate earthquakes recorded by low-
gain instruments, the estimation of τcmight be affected by
noise. However, as shown in Fig. 7b for the 2013 M4.9 Matese
earthquake, the τcestimates obtained by SOSEWIN stations
were in very good agreement with the one obtained using the
station MAJ3. Indeed, the Pd and τcvalues observed at the
different stations (which have different characteristics and are
installed in different places) show a high level of coherency and
correctly assign to the event an alert level equal to zero. Users
are informed that a moderate event (i.e., smaller than M6)
occurred at a quite large distance from the target site and
no damage is expected. This result confirms that for a structure
that is rigid and has relatively small dimension, the influence of
the structural response on the high-frequency P-wave signals is
small in the case of moderate events. Therefore, Pd and τc
information coming from different stations can be beneficially
averaged for EEW purposes, resulting in a reduced uncertainty
associated with the PGV and magnitude estimates.
For the test with the regional approach, we considered all
the ISNET recordings. This means that among the stations
installed at the school, only MAJ3, which was already part
of ISNet, was considered.
Figure 8a shows a snapshot of PRESToPlus when the
system releases the first alert. At this moment, three stations
have triggered, one of which is the station at MAJI. However,
the 2 s P-wave window necessary to estimate the magnitude is
▴Figure 7. (a) Recordings of the 29 December 2013 Matese
earthquake at four of the five seismic stations at MAJI. Automatic
picks from FilterPicker5 (Lomax et al., 2012) are represented as
vertical lines. (See Fig. 2for the sensor disposition.) (b) Peak dis-
placement (Pd) and predominant period (τc) measured at the four
seismic stations at MAJI during the 29 December 2013 Matese
earthquake. The four quadrants represent the alert levels as de-
fined by Zollo et al. (2010).
406 Seismological Research Letters Volume 86, Number 2A March/April 2015
▴Figure 8. (a) Screenshot of the playback in PRESToPlus of the waveforms recorded by ISNet and the MAJ3 station (i.e., MAJI in the
subplot) during the M4.9 Matese earthquake. The MAJ3 recording is on the forefront in the left part, with S-waves (dark gray) evidenced.
The screenshot shows the moment when the first three stations, including MAJ3, detected the earthquake. (b) Same as (a) but at the end
of the playback in PRESToPlus. The alert levels at each site are shown in the geographic map; all alerts are 0 (no damage, M<6).
Seismological Research Letters Volume 86, Number 2A March/April 2015 407
available only for the other two stations, which were closer to
the epicenter. Hence, the magnitude estimate is constrained at
the beginning only by two stations and has a value of 4.3, which
is slightly underestimated with respect to the final value. It is
worth noting that when the first EEW magnitude estimate is
released by PRESToPlus, the school MAJI has a lead time of 6 s.
Figure 8b shows the results of the playback. Although the
final magnitude estimation of PRESToPlus (i.e., Mw5.0) is in
agreement with the value from the seismic bulletin, we observed
the signals recorded at the station MAJ3 led to a final magnitude
estimate equal to 5.1 and 5.4 from the 4 s P-wave and 2 s S-wave
windows, respectively. The slight magnitude overestimation,
especially for Swaves, with respect to the official magnitude
estimation and the value found by PRESToPlus considering the
ISNET stations, is most probably due to the presence of site
effects and directivity. This problem could be easily overcome in
the future by calibrating a station correction factor to be assigned
at MAJI and used in PRESToPlus for the local magnitude
estimation.
In addition to the event location and magnitude, the
level alert was computed for each station (Fig. 8b,geographic
map). As expected, all the stations were associated with
an alert level equal to zero, that is, the event was smaller than
magnitude 6 and a few tens of kilometers outside of the
network.
The Sentinel: A Low-Cost EEW Actuator for Schools
At the time of the 2013 Matese earthquake, neither the users
nor the Sentinel were within the school: the event occurred
during the Christmas vacancy period, and the Sentinel exper-
imentation started a few months later. Therefore, the perfor-
mance of the Sentinel was tested by the realization of offline
runs of PRESToPlus (i.e., playbacks) on the available earthquake
waveforms. The use of playbacks to assess the performance of
the PRESToPlus system in realistic conditions has been success-
fully demonstrated several times (e.g., Satriano et al., 2011;
Colombelli et al., 2012).
The first test consisted of the playback of the Matese
earthquake with the regional components of PRESToPlus run-
ning at the RISSC-Lab’sEEW control center in Naples, while
the Sentinel was working in real time at MAJI, where it kept its
idle state until the first warning message from PRESToPlus was
received. During the playback, teachers and students at the
school received real-time early warning information from the
Sentinel and, with the only exception of the P-wave ground
shacking, experienced a situation rather close to what would
occur during an earthquake. Figure 9a shows the time line of
the event characterization and in particular the three alert mes-
sages (which includes the lead time at the school and the M and
PGV estimates) that PRESToPlus transmitted to the Sentinel
via the Internet. We verified that the transmission of these
alerts between the RISSC-Lab and MAJI occurs in a few tens
of milliseconds, similar to the time required by a ping com-
mand. The Sentinel at MAJI checked in real time the content
of the alerting messages that it was receiving (Fig. 9b. Accord-
ing to the EEW rules we defined (Fig. 5), with the PGV at the
site predicted to be below the first threshold (0:2cm=s), it
switched on a green light (i.e., alert level “silent”), indicating to
the users that an earthquake occurred and within 4 s they were
going to feel a moderate-low shaking but that it would not
cause any damage at the school. The PGV actually measured by
the MAJ3 station was 0:21 cm=s. Hence, even if the difference
between predicted and observed PGV values was very small,
and certainly not perceivable from the eventual users experi-
encing the shaking, the message released to the users overesti-
mated the risk. It is worth noting, however, that after the first
warning message (Fig. 9a), the school users were correctly
alerted about the occurrence of a small/moderate earthquake,
and they had 7 s of lead time to duck and cover under their
desks to reduce their exposure to eventual nonstructural dam-
ages (e.g., falling piece of plaster or wall lamps). Moreover, like
every regional EEWS,PRESToPlus predicts the PGV at target
sites by a GMPE; and the observed small deviation between
predicted and observed PGV values (0:02 cm=s) at MAJI dur-
ing the 2013 Matese earthquake is totally in agreement with
the performance of standard GMPEs, which, as shown by Akkar
and Bommer, is generally within the range of 0:6 log units.
To verify the performance of the Sentinel for a large event,
we also ran the playback of the M6.9 1980 Irpinia earthquake.
In this case, we used synthetic recordings computed by the
Axitra code (Coutant, 1989) at the ISNet stations. MAJI was
at a distance of about 75 km from both the epicenter and the
fault planes activated during the earthquake, which corre-
sponds to a lead time of about 13 s. Macroseismic information
following the 1980 Irpinia earthquake indicates the Somma
Vesuviana village, where MAJI is located, experienced a damage
level corresponding to VI–VII on the Mercalli–Cancani–
Sieberg (MCS) scale.
Figure 10a shows the playback of the PRESToPlus and the
time line of alerts transmitted to the Sentinel at MAJI. In this
case, the evolutionary event characterization following the
progressive increase of the number of stations included in the
magnitude estimation led the estimated MAJI ground motion
to increase with the time. In fact, the PGV at the target site
increases in 7 s from 2:5cm=s(alert level “low”; following the
Faccioli and Cauzzi (2006) macroseismic scale, this corre-
sponds to MCS VI, or strong shaking and light damage) to
6:2cm=s, which corresponds to MCS VII (very strong shaking
and moderate damage). In this case, immediately after the first
two alerts were received, the Sentinel provided the users with
the information that a strong shaking would strike the school
in the next 18 s (i.e., yellow light, or alert level “low”; Fig. 10b).
However, due to the increase of the PGV expected at the site, at
the third alert the Sentinel switched on the red light (alert lev-
el“high”), indicating the lead time was then 13 s. The system
worked as predicted, and during drill tests (http://www.rissclab
.unina.it/en/experiments/710‑early‑warning‑applicationa‑at
‑school; last accessed January 2015) we verified that, in case of
an earthquake in the Irpinia region, it would allow the students
to duck and cover before the S-waves arrival at the school, thus
minimizing their risk of injury.
408 Seismological Research Letters Volume 86, Number 2A March/April 2015
CONCLUSION
This work presented the results of a feasibility study of an
EEWS for schools carried out in the framework of the REAKT
project. The system is composed of a small seismic network of
accelerometers, the PRESToPlus system, and a low-cost intel-
ligent electronic device (the Sentinel) that operates as actua-
tor. The Sentinel is fed data by PRESToPlus and, in the case of
an event, provides evolutionary alerts to users, starting differ-
ent traffic lights and bell alarms to indicate the level of
hazard.
The first tests were devoted to assess the rate of false alerts
of the EEWS. Our results showed the anthropic noise generated
in the school by the users is so high that a monitoring system
relying on a single station would be prone to one false event
declaration every few minutes. On the contrary, considering a
coincidence of at least four stations within a very short time
window allowed the complete elimination of false event detec-
▴Figure 9. Timeline of the playback of the 2013 M4.9 Matese earthquake. (a) An evolutionary event characterization by regional
PRESToPlus EEWS and messages received by the Sentinel at MAJI. Marker locations over the timeline are approximate only.
(b) EEW messages transmitted by PRESToPlus to the Sentinel via Internet and Sentinel’s alert levels.
Seismological Research Letters Volume 86, Number 2A March/April 2015 409
tion. Despite the simplicity of the analysis, we considered this
latter result very important from the practical point of view. In
fact, this test confirmed the setup of a small array of sensors,
including low cost ones, seems to be a promising way of design-
ing a monitoring on-site EEWS for infrastructures within cities.
Indeed, in countries like Italy, where the distance between seis-
mic sources and target sites are, in most cases, a few tens of kilo-
meters only; on-site EEWS like the one installed at MAJI might
represent a valid strategy to complement regional networks.
The very low seismicity level observed during the two years
of the system experimentation did not allow it to be tested as
extensively as it deserves. One test of the EEWS performance
▴Figure 10. Same as Figure 9but for the 1980 M6.9 Irpinia Earthquake.
410 Seismological Research Letters Volume 86, Number 2A March/April 2015
with real data using both the on-site and regional EEWS ap-
proaches was finally possible after the 29 December 2013
ML4.9 earthquake occurred in the Matese hills area. When
we ran the playback of the event following the on-site thresh-
old-based approach proposed by Zollo et al. (2010), we observed
the same alert level estimation for all stations of the array. This
result suggests, for the structure under study, that the on-site
EEWS might benefit from the average of the estimates obtained
by the different stations, which would correspond to a reduced
uncertainty associated with the PGV and magnitude estimates.
However, although the structural response on the high-
frequency P-wave signals for the Matese earthquake turned
out to be small, the structural behavior for larger ground motion
could be more complicated. Future modeling studies will be
dedicated to investigate this issue.
The preliminary tests carried out to assess the Sentinel per-
formance in cases of moderate and strong earthquakes showed
(1) the transmission of PRESToPlus alert messages requires only
a few tens of milliseconds; (2) the Sentinel allows promptly
initiation of an automatic procedure to warn school users;
and (3) in the case of the Irpinia scenario, the lead time for the
school (MAJI) would be 13 s. As discussed by Goltz (2002),
such an amount of lead time should be sufficient to take pro-
tective measures that minimize the risk of injuries for the
school users (e.g., duck and cover, turn off gas burners, move
away from windows or equipment, etc.). A future study will use
drill tests involving students and school staff to verify the fea-
sibility of these protection measures at MAJI and other schools
in the Irpinia region.
ACKNOWLEDGMENTS
The research leading to these results was realized in the frame-
work of the Strategies and Tools for Real Time Earthquake
Risk Reduction (REAKT, www.reaktproject.eu; last accessed
January 2015) project coordinated by Analisi e Monitoraggio
del Rischio Ambientale (AMRA S.c. a r.l) and Network of
European Research Infrastructures for Earthquake Risk Assess-
ment and Mitigation (NERA,nera-eu.org; last accessed January
2015) projects, and received funding from the European Com-
munity’s Seventh Framework Programme (FP7/2007–2013)
under Grant Agreement Numbers 282862 and 262330, respec-
tively.
We acknowledge the staff and students of the ITIS “E.
Majorana”for their effective collaboration during all the phases
of the project.
We also would like to thank the Associate Editor and two
anonymous reviewers for their comments and suggestions that
allowed us to significantly improve the manuscript.
The REAKT working group, who designed and tested the
earthquake early warning system at the high school ITIS “E.
Majorana,”Somma Vesuviana are A. Zollo, A. Emolo, M. Pi-
cozzi, C. Martino, L. Elia, G. Verderame, Sergio Del Gaudio, S.
Colombelli, O. Amoroso, P. Brondi, and M.T. De Risi (Uni-
versity of Naples Federico II, Italy; Unit of Experimental and
Computational Seismology (RISSC)-Lab, AMRA s.c. a r.l.,
Naples, Italy); S. Parolai, D. Bindi, and T. Boxberger (Helm-
holtz-Zentrum Potsdam–Deutsches GeoForschungsZentrum,
Germany); and N. Miranda, L. Buonaiuto, and A. Amelia
(ITIS “E. Majorana,”Somma Vesuviana, Italy).
REFERENCES
Allen, R. M., and H. Kanamori (2003). The potential for earthquake
early warning in southern California, Science 300, 786–789.
Allen, R. M., P. Gasparini, O. Kamigaichi, and M. Böse (2009). The sta-
tus of earthquake early warning around the world: An introductory
overview, Seismol. Res. Lett. 80, 682–693.
Ameri, G., A. Emolo, F. Pacor, and F. Gallovic (2011). Ground-motion
simulations for the 1980 M 6.9 Irpinia earthquake (southern Italy)
and scenario events, Bull. Seismol. Soc. Am. 101, no. 3, 1136–1151,
doi: 10.1785/0120100231.
Barani, S., D. Spallarossa, and P. Bazzurro (2009). Disaggregation of
probabilistic ground motion hazard in Italy, Bull. Seismol. Soc.
Am. 99, 2638–2661.
Bommer, J. J., P. J. Stafford, and J. E. Alarcón (2009). Empirical equations
for the prediction of the significant, bracketed, and uniform dura-
tion of earthquake ground motion, Bull. Seismol. Soc. Am. 99, no. 6,
3217–3233, doi: 10.1785/0120080298.
Chiauzzi, L., A. Masi, M. Mucciarelli, M. Vona, F. Pacor, G. Cultrera, F.
Gallovic, and A. Emolo (2012). Building damage scenarios based on
exploitation of Housner intensity derived from finite faults ground
motion simulations, Bull. Earthq. Eng. 10, no. 2, 517–545, doi:
10.1007/s10518-011-9309-8.
Colombelli, S., A. Zollo, G. Festa, and H. Kanamori (2012). Early mag-
nitude and potential damage zone estimates for the great Mw9
Tohoku-Oki earthquake, Geophys. Res. Lett. 39, doi: 10.1029/
2012GL053923.
Coutant, O. (1989). Program of numerical simulation AXITRA, Techn.
Rept. LGIT, Grenoble, France.
Emolo, A., V. Convertito, and L. Cantore (2011). Ground-motion pre-
dictive equations for low-magnitude earthquakes in the Campania–
Lucania area, southern Italy, J. Geophys. Eng. 8, no. 1, 46–60, doi:
10.1088/1742-2132/8/1/007.
Emolo, A., M. Picozzi, A. Zollo, C. Martino, L. Elia, S. Colombelli, P.
Brondi, G. Verderame, T. De Risi, O. Amoroso, S. Parolai, D. Bindi,
T. Boxberger, N. Miranda, L. Buonaiuto, and A. Amelia (2014).
Final report for feasibility studies on EEW: Application to schools,
REAKT, Strategies and Tools for Real Time Earthquake RisK Re-
ducTion, Deliverable 7.4b; http://www.reaktproject.eu/deliverables/
REAKT‑D7.4.pdf (last accessed January 2015).
Faccioli, E., and C. Cauzzi (2006). Macroseismic intensities for seismic
scenarios estimated from instrumentally based correlations, in Proc.
First European Conference on Earthquake Engineering and Seismol-
ogy (a joint event of the 13th ECEE & 30th General Assembly of the
ESC), Genève, Switzerland, Paper No. 569.
Fleming, K., M. Picozzi, C. Milkereit, F. Kuhnlenz, B. Lichtblau, J.
Fischer, C. Zulfikar, and O. Ozel (2009). The self-organizing seismic
early warning information network (SOSEWIN), Seismol. Res. Lett.
80, 755–771, doi: 10.1785/gssrl.80.5.755.
Goltz, J. D. (2002). Introducing earthquake early warning in California:
A summary of social science and public policy issues, Technical Re-
port, Governor’s Office of Emergency Services, Pasadena, California.
Grünthal, G. (1998). European macroseismic scale 1998 (EMS-98), in
Cahiers du Centre Européen de Géodynamique et de Séismologie,
Vol. 15, Centre Européen de Géodynamique et de Séismologie,
Luxembourg, 99 pp.
Hoshiba, M. (2013). Real-time correction of frequency-dependent site
amplification factors for application to earthquake early warning,
Bull. Seismol. Soc. Am. 103, 3179–3188, doi: 10.1785/0120130060.
Iannaccone, G., A. Zollo, L. Elia, V. Convertito, C. Satriano, C. Martino,
G. Festa, M. Lancieri, A. Bobbio, T. A. Stabile, M. Vassallo, and A.
Seismological Research Letters Volume 86, Number 2A March/April 2015 411
Emolo (2010). A prototype system for earthquake early-warning
and alert management in southern Italy, Bull. Earthq. Eng. 8,
no. 5, 1105–1129, doi: 10.1007/s10518-009-9131-8.
Kanamori, H. (2005). Real-time seismology and earthquake damage mit-
igation, Annu. Rev. Earth Planet. Sci. 33, 195–214.
Lancieri, M., and A. Zollo (2008). Bayesian approach to the real-time
estimation of magnitude from the early Pand Swave displacement
peaks, J. Geophys. Res. 113, no. B12, doi: 10.1029/2007JB005386.
Lomax, A., C. Satriano, and M. Vassallo (2012). Automatic picker devel-
opments and optimization: FilterPicker–A robust, broadband
picker for real-time seismic monitoring and earthquake early warn-
ing, Seismol. Res. Lett. 83, 531–540, doi: 10.1785/gssrl.83.3.531.
Luzi, L., S. Hailemikael, D. Bindi, F. Pacor, F. Mele, and F. Sabetta (2008).
ITACA (ITalian ACcelerometric Archive): A web portal for the
dissemination of Italian strong-motion data, Seismol. Res. Lett.
79, no. 5, 716–722, doi:10.1785/gssrl.79.5.716.
Picozzi, M., C. Milkereit, K. Fleming, J. Fischer, K.-H. Jaeckel, D. Bindi,
S. Parolai, and J. Zschau (2014). Applications of a low-cost, wireless,
self-organising system (SOSEWIN) to earthquake early warning
and structural-health-monitoring, in Early Warning for Geological
Disasters, Advanced Technologies in Earth Sciences, F. Wenzel and
J. Zschau (Editors), Springer, Berlin, Germany, 263–288, ISBN:
978-3-642-12232-3, doi: 10.1007/978-3-642-12233-0_14.
Picozzi, M., C. Milkereit, S. Parolai, K.-H. Jaeckel, I. Veit, J. Fischer, and
J. Zschau (2010). GFZ Wireless Seismic Array (GFZ-WISE), a
wireless mesh network of seismic sensors: New perspectives for seis-
mic noise array investigations and site monitoring, Sensors 10,
3280–3304.
Picozzi, M., S. Parolai, M. Mucciarelli, C. Milkereit, D. Bindi, R. Ditommaso,
M.Vona,M.R.Gallipoli,andJ.Zschau(2011).Interferometricanalysis
of strong ground motion for structural health monitoring : The example
of the L’Aquila (Italy) seismic sequence 2009, Bull. Seismol. Soc. Am.
101, no.2,635,doi:10.1785/0120100070.
Satriano, C., L. Elia, C. Martino, M. Lancieri, A. Zollo, and G. Iannaccone
(2011). PRESTo, the earthquake early warning system for southern
Italy: Concepts, capabilities and future perspectives, Soil. Dynam.
Earthq. Eng. 31, 137–153, doi: 10.1016/j.soildyn.2010.06.008.
Satriano, C., Y.-M. Wu, A. Zollo, and H. Kanamori (2010). Earthquake
early warning: Concepts, methods and physical grounds, Soil
Dynam. Earthq. Eng. 31, no. 2, 106–118, doi: 10.1016/j.soil-
dyn.2010.07.007.
Wu, Y. M., and H. Kanamori (2005). Experiment on an onsite early
warning method for the Taiwan early warning system, Bull. Seismol.
Soc. Am. 95, no. 1, 347–353.
Wu, Y. M., D. Y. Chen, T. L. Lin, C. Y. Hsieh, T. L. Chin, W. Y. Chang,
W. S. Li, and S. H. Ker (2013). A high density seismic network for
earthquake early warning in Taiwan based on low cost sensors, Seis-
mol. Res. Lett. 84, 1048–1054, doi: 10.1785/0220130085.
Zollo, A., O. Amoroso, M. Lancieri, Y. M. Wu, and H. Kanamori (2010).
A threshold-based earthquake early warning using dense accelerom-
eter networks, Geophys. J. Int. 183, 963–974.
Zollo, A., S. Colombelli, L. Elia, A. Emolo, G. Festa, G. Iannaccone, C.
Martino, and P. Gasparini (2014). An integrated regional and on-
site earthquake early warning system for southern Italy: Concepts,
methodologies and performances, in Early Warning for Geological
Disasters, Advanced Technologies in Earth Sciences, F. Wenzel and
J. Zschau (Editors), Springer, Berlin, Germany, 117–137, ISBN:
978-3-642-12232-3, doi: 10.1007/978-3-642-12233-0_7.
M. Picozzi
A. Emolo
C. Martino1
A. Zollo
G. Verderame
the REAKT Working Group1,2,3
Dipartimento di Fisica
Università Federico II
Naples, Italy
matteo.picozzi@unina.it
antonio.emolo@na.infn.it
martino.claudio@gmail.com
aldo.zollo@unina.it
verderam@unina.it
N. Miranda
ITIS “E. Majorana”
Somma Vesuviana
Italy
mirandanicola@gmail.com
T. Boxberger
Helmholtz-Zentrum Potsdam
Deutsches GeoForschungsZentrum
Germany
tobias.boxberger@gfz‑potsdam.de
Published Online 18 February 2015
1Also at Unit of Experimental and Computational Seismology (RISSC)-
Lab, AMRA s.c. a r.l., Naples, Italy.
2Also at ITIS “E. Majorana,”Somma Vesuviana, Italy.
3Also at Helmholtz-Zentrum Potsdam–Deutsches GeoForschungsZen-
trum, Germany.
412 Seismological Research Letters Volume 86, Number 2A March/April 2015
