THE EMSC TOOLS USED TO DETECT AND DIAGNOSE THE IMPACT OF GLOBAL EARTHQUAKES FROM DIRECT AND INDIRECT EYEWITNESSES' CONTRIBUTIONS

Abstract

This paper presents the strategy and operational tools developed and implemented at the Euro-Mediterranean Seismological Centre (EMSC) to detect and diagnose the impact of global earthquakes within minutes by combining « flashsourcing » (real time monitoring of website traffic) with social media monitoring and crowdsourcing. This approach serves both the seismological community and the public and can contribute to improved earthquake response. It collects seismological observations, improves situation awareness from a few tens of seconds to a couple of hours after earthquake occurrence and is the basis of innovative targeted real time public information services. We also show that graphical input methods can improve crowdsourcing tools both for the increasing use of mobile devices and to erase language barriers. Finally we show how social network harvesting could provide information on indirect earthquake effects such as triggered landslides and fires, which are difficult to predict and monitor through existing geophysical networks.

Full text

Bossu et al.
Citizen Seismology
Short Paper – Social Media Studies
Proceedings of the ISCRAM 2015 Conference - Kristiansand, May 24-27
Palen, Büscher, Comes & Hughes, eds.
THE EMSC TOOLS USED TO DETECT AND DIAGNOSE THE
IMPACT OF GLOBAL EARTHQUAKES FROM DIRECT AND
INDIRECT EYEWITNESSES’ CONTRIBUTIONS
Rémy Bossu
Euro-Mediterranean
Seismological Centre
CEA, DAM, DIF F91297
Arpajon France
bossu@emsc-csem.org
Robert Steed
Euro-Mediterranean
Seismological Centre
steed@emsc-csem.org
Gilles Mazet-Roux
Euro-Mediterranean
Seismological Centre
mazet@emsc-csem.org
Fréderic Roussel
Euro-Mediterranean
Seismological Centre
roussel@emsc-csem.org
Caroline Etivant
Euro-Mediterranean
Seismological Centre
etivant@emsc-csem.org
ABSTRACT
This paper presents the strategy and operational tools developed and implemented
at the Euro-Mediterranean Seismological Centre (EMSC) to detect and diagnose
the impact of global earthquakes within minutes by combining « flashsourcing »
(real time monitoring of website traffic) with social media monitoring and
crowdsourcing.
This approach serves both the seismological community and the public and can
contribute to improved earthquake response. It collects seismological
observations, improves situation awareness from a few tens of seconds to a couple
of hours after earthquake occurrence and is the basis of innovative targeted real
time public information services.
We also show that graphical input methods can improve crowdsourcing tools both
for the increasing use of mobile devices and to erase language barriers. Finally we
show how social network harvesting could provide information on indirect
earthquake effects such as triggered landslides and fires, which are difficult to
predict and monitor through existing geophysical networks.

Bossu et al.
Citizen Seismology
Short Paper – Social Media Studies
Proceedings of the ISCRAM 2015 Conference - Kristiansand, May 24-27
Palen, Büscher, Comes & Hughes, eds.
Keywords
Situation awareness, social media, crowdsourcing, flashsourcing, citizen science
INTRODUCTION
Seismology has always valued reports of felt experiences by laypersons because
they are often the only evidence of pre-instrumental period earthquakes. This
probably explains why seismology has been amongst the pioneers in citizen
science and crowdsourcing. For example, the « Did you feel it » system developed
by the US Geological Survey to massively collect felt experiences through online
questionnaires has been in operation since 1999 (Wald et al., 1999) before the
term « crowdsourcing » was even coined.
Today one of the main benefits for seismologists that result from the collection of
these eyewitnesses’ observations is to provide in-situ constraints to the
intrinsically uncertain earthquake damage scenario and improve situation
awareness (Bossu et al., 2015). We will also show through the different tools
developed and implemented at the EMSC that it also benefits the public since the
best way to optimize collection of in-situ observations is first to ensure a massive
and immediate convergence of eyewitnesses to the collection tools by associating
them with real time information services that meet eyewitnesses’ expectations
immediately after an earthquake occurrence. This efficient engagement strategy is
achieved by speeding up our information services, by focusing them on felt and
damaging earthquakes (the only ones that matter for the public) and, by digesting
user-generated information and compiling it into a more comprehensive
information service which covers both earthquake parameters (location,
magnitude, time) and descriptions of their effects (shaking level, pictures of
damage). We will illustrate these points by summarizing the main functions and
performances of LastQuake which is the name of our smartphone application, a
Twitterbot -in this case, it is also called a QuakeBot, i.e. a program that produces
automatic tweets on earthquakes- and web browser add-ons.
In this article, we show how the crowdsourced information (testimonies,
comments and geo-located pictures and videos) can be enhanced by 2
complementary indirect data contributions derived from the monitoring of activity
on Twitter and from traffic analysis of EMSC website (www.emsc-csem.org, our
popular website dedicated to global earthquake information). We then present
how social network harvesting is being tested to detect indirect effects of
earthquakes such as triggered landslides and fires and how such an integrated
approach can contribute to improved situation awareness after an earthquake.
MONITORING PUBLIC REACTIONS TO DETECT AND DIAGNOSE
EARTHQUAKE IMPACT
Up until the 1990’s, seismologists knew that an earthquake had been felt when
suddenly the different laboratory phones started to ring together. Today,
eyewitnesses are mostly turning to the Internet for searching for earthquake
information or for sharing their experiences after the shaking. The Internet acts as
a digital nervous system of our planet whose pulses offer a way to detect felt
earthquakes independently from seismic monitoring networks, and from
earthquake magnitude.
The EMSC uses 2 complementary approaches, Twitter earthquake detection TED
(Earle et al., 2010,
2011), performed by
the US Geological
Survey, and
flashsourcing (Bossu
et al., 2008, 2011a,
2011b), developed and
operated at EMSC.
TED monitors the
publication of 140-
characters Twitter
messages (tweets) and
applies place, time,
and key word filtering
to detect felt
earthquakes through
the surge in published
Figure 1. Example of a flashcrowd observed on the EMSC
website and caused by a M 4.9 earthquake in SE France. It
was automatically detected 96s after the earthquake
occurred.

Bossu et al.
Citizen Seismology
Short Paper – Social Media Studies
Proceedings of the ISCRAM 2015 Conference - Kristiansand, May 24-27
Palen, Büscher, Comes & Hughes, eds.
tweets related to shaking experiences (Earle et al., 2010, 2011). TED data is
shared in real time with the EMSC, which automatically associates them with
seismic locations through time and spatial analysis using the geocoded tweets
(Earle et al., 2010).
Flashsourcing is based on flashcrowds, i.e. rapid and massive traffic increases
(Jung et al., 2002, Marnerides et al., 2008) generated by the natural convergence
of eyewitnesses looking for earthquake information on EMSC website
immediately after the shaking (Figure 1). The convergence is extremely rapid: it
was shown for the 2011 Mineral, Virginia earthquake that the arrivals times of
visitors on the website followed the seismic wave propagation, allowing one to
determine the epicentral
location with a 30 km accuracy
from only 2 minutes of EMSC
website traffic (Bossu et al.,
2014). This demonstrates that
the initial flashcrowd is
actually caused by
eyewitnesses rather than
referred visitors. In turn, it
implies, as discussed later, that
providing a recognized rapid
public information strategy is
an efficient strategy for
engagement with earthquake
eyewitnesses.
Detections of felt earthquakes
are typically within 2 minutes
of earthquake occurrence for both methods and they precede seismic locations in
the vast majority of cases (more than 90%) (Bossu et al. 2011a, Earle et al., 2011).
They are fully complementary, as only 10% of the detected earthquakes in 2014
were detected by both TED and flashsourcing (45 over a total of 429).
Figure 3: Interpretations of the different possible time evolutions of the number of
Internet sessions following an earthquake occurrence at T
0
. When the perception of
danger is significant, eyewitnesses are more likely first to flee to safety rather than
immediately browsing the Internet for information
Figure 2: Geographical origins (determined from IP locations) of the website visitors
within 5 min of the occurrence of the Nov. 22 2014 M 5.6 earthquake in Romania. Red
dots represent geographical origins of statistically significant increased traffic (for more
details see text). Yellow circles represent the geographical origins of website traffic with
no significant variations. Black triangles represent the geographical origins of website
visitors over the previous 12 months. Circle size is function of the difference between the
expected and observed number of unique IPs (more details in Bossu et al (2011a)

Bossu et al.
Citizen Seismology
Short Paper – Social Media Studies
Proceedings of the ISCRAM 2015 Conference - Kristiansand, May 24-27
Palen, Büscher, Comes & Hughes, eds.
Beyond the detection of felt earthquakes, flashsourcing also provides rapid
information on the local effects of earthquakes (Bossu et al., 2011a, 2014). For
instance, the area where shaking was felt can be automatically located by plotting
the geographical locations of statistically significant increases in traffic (Figure 2)
(Bossu et al., 2011b). More importantly, it can detect and map damaged areas in
certain cases through the concomitant loss of Internet sessions due to damaged
networks (Figure 3) (Bossu et al., 2015). This automatic damage detection system
was implemented in late 2014 although it has so far not produced definitive results
as only electricity black-outs have been observed and mapped so far.
THE IMPORTANCE OF CROWDSOURCED INFORMATION
EMSC collects testimonies, comments and geo-located pictures of earthquake
damage. In 2014, it received at least 3 testimonies (our criteria to confirm an
earthquake as being felt) for 507 earthquakes which had not been detected by
TED or flashsourcing; further supporting the hypothesis that a single method is
unlikely to detect all the events of interest.
Figure 4: Composite macroseismic maps from testimonies collected by EMSC. The
pace of collection has been increasing fast: 30% of the 58 000 questionnaires which
collection started in 2008 were collected in 2014, and 60% of the 16 000 thumbnails,
which collection started in 2011.
Testimonies are collected through online questionnaires available in 32 languages.
They are automatically converted in macroseismic maps depicting the local
shaking level (Figure 4) and made available online. Macroseismic data is more
detailed than flashsourced information but the latter is collected faster as it
generally takes a few tens of minutes to collect a significant number of
questionnaires. The conversion of testimonies to shaking level uses a statistical
approach excluding outliers due to error or misuses.
Geo-located pictures are essential for describing local damage and documenting
transient effects (Figure 5). They are manually validated by our seismologist on
call before being made available on the website. The seismologist checks their
pertinence and coherency with date, location
and the local expected shaking level.
CROWDSOURCING TOOLS IN A MOBILE
WORLD
Mobile devices (smartphones, tablets) have
significantly changed the way people access
information at EMSC. In 3 years, from 2011
to 2014, the average number of unique daily
visitors to EMSC mobile website
experienced a 3-fold increase while, during
the same period, accesses to the classical
website decreased by 10%. The change is
even more dramatic when it comes to
information access immediately after a
widely felt earthquake: in the first 30
minutes following the Aug. 24th 2014, M 6
Napa (California) earthquake, 2/3 of the visitors within 300km of the epicenter
accessed information with a mobile device (3656 over a total of 5579 individual
visitors) and were automatically directed to our mobile website (Figure 6).
Figure 5: Crowdsourced picture of
the 2013 Bohol (Philippines)
earthquake

Bossu et al.
Citizen Seismology
Short Paper – Social Media Studies
Proceedings of the ISCRAM 2015 Conference - Kristiansand, May 24-27
Palen, Büscher, Comes & Hughes, eds.
Figure 6: Traffic on EMSC websites in the minutes following the Napa earthquakes
(above) and percentage of visitors providing their testimonies (below). The continuous
line represents earthquake occurrence, the dashed one the publication of the first
information on EMSC website.
Such behavior is not unexpected, especially for an earthquake striking in the
middle of the night, when desktops are less likely to be on and mobile devices
offer a faster Internet access. In order to take into account this change and the
difficulty to fill questionnaires on a small screen, the online questionnaire used to
collect testimonies has been replaced on the website for mobile devices by
thumbnails depicting the 12 levels of the EMS98 macroseismic scale (Grunthal,
1998) (Figure 7). The ease of use of thumbnails, and the removal of language
hurdles are the likely cause of the significant
increase in the percentage of visitors offering
their testimonies from about 4% to more than 6%
(Figure 6). This ease of use compared to the
couple of minutes required to fill the online
questionnaire, may explain why the first
thumbnails were collected 5 minutes before the
first questionnaires (Figure 6) whereas both fixed
and mobile users hit their respective website with
the same swiftness. It may also be worth noting
that the first testimonies were only collected once
preliminary information was made available on
EMSC websites. This may indicate the
importance of meeting visitors’ expectations first
to initiate engagement and efficient
crowdsourcing.
Thumbnails were also implemented in the LastQuake application presented
below. With the same idea of meeting users’ expectations, users can share their
comments not only with EMSC but also on their Facebook and Twitter accounts.
The LastQuake application proved efficient for collecting geo-located pictures:
since its launch, all our collected pictures come from LastQuake. It is much more
convenient to share a picture taken by a smartphone using the app and this also
increases the number of GPS determined locations. Please note that page layout
may change slightly depending upon the printer you have specified
LASTQUAKE INFORMATION TOOLS
LastQuake is the name of a set of information tools developed by EMSC,
smartphone applications (iOS, Android), a twitter robot (or QuakeBot) and web-
browser add-ons which are all based on the same principle: providing rapid
information for the only earthquakes which matter to the public i.e. felt and
Figure 7: Example of
thumbnail used by EMSC to
collect testimonies on mobile
devices

Bossu et al.
Citizen Seismology
Short Paper – Social Media Studies
Proceedings of the ISCRAM 2015 Conference - Kristiansand, May 24-27
Palen, Büscher, Comes & Hughes, eds.
damaging earthquakes. It aims at broadening the EMSC service portfolio to serve
a wider community.
LastQuake is based on the automatic discrimination of felt earthquakes by
merging different sources of information, mainly information collected directly or
indirectly from eyewitnesses described above (Figure 8). The full description of
the algorithm is beyond the scope of this article. It should however be underlined
that discriminating felt earthquakes, especially the low magnitude ones, is
challenging from seismological data only: variations of 1 or 2 km (i.e. well within
uncertainties of real time locations estimates) in relation to centres of population
can lead to a M 3 earthquake either being widely felt or only picked up by
instruments. The aim is not to locate all earthquakes, this is the role of monitoring
networks, but to identify the ones of societal importance because they affect the
population in one way or another. By focusing on the few thousands of widely felt
earthquakes a year, LastQuake prioritizes earthquake information aimed at public
and authorities and optimizes crowdsourcing of in-situ observations at little cost
(Bossu et al., 2015).
LastQuake is the prolongation of our engagement strategy with eyewitnesses
through additional channels. It creates a virtuous circle where collected data is
integrated in LastQuake information services to offer improved information on
both the earthquake itself and its consequences (macroseismic maps, pictures of
damage, comments from eyewitnesses…) which in turn should increase the
volume of collected data.
Since its launch in July 2014, the smartphone application has been downloaded
approximately 20 000 times.. There are 6 300 followers to the LastQuake Twitter
account which currently generates up to 40 different tweets that are automatically
published from 1 to 90 minutes of an earthquake occurrence. They include the
automatic detection of flashcrowds, epicentral and macroseismic maps and their
updates, the possible associated tsunami warning or alerts or links to additional
resources. New tweets are regularly implemented to cover additional cases, like
sequences of earthquakes, i.e. a number of earthquakes affecting the same area in
a short period of times.
Figure 8: Schematic description of LastQuake QuakeBot

Bossu et al.
Citizen Seismology
Short Paper – Social Media Studies
Proceedings of the ISCRAM 2015 Conference - Kristiansand, May 24-27
Palen, Büscher, Comes & Hughes, eds.
Web browser add-ons allow ‘pushed’ information like for the smartphone
notifications. It is less popular than the smartphone application or Twitter but it
contributes to increased damage detection capabilities through concomitant loss of
web sessions by increasing the baseline traffic (Figure 3). The add-on for the
Chrome browser has been released in January 2015. During the working hours
there are around 1 200 sessions generated by website visitors and 300 by the add-
on increasing damage detection capabilities by 25%.
CONCLUSION AND OUTLOOK
This paper focuses on the strategy and tools developed and implemented at the
EMSC to collect direct and indirect data contributions on global earthquake
effects from eyewitnesses for improved situation awareness and improve rapid
public earthquake information. It does not intend to review the use of Twitter or
other social tools to create earthquake detection or mapping system but to share
our specific experience with researchers and practitioners and present future
developments.
The engagement with eyewitnesses is based on targeted real time public
information services intending to meet their needs in the immediate aftermath of
the earthquake occurrence. Attracting more eyewitnesses increases the volume of
collected data which is then integrated to further improve the information
services, creating a virtuous circle.
Traditional online questionnaires have been replaced by a series of thumbnails to
fit the screen size of mobile devices, improve convenience and erase language
hurdles. Indirect data contributions are performed by 2 complementary methods
based respectively on the analysis of the use of Twitter and of the EMSC website,
one of the most popular rapid global earthquake information sources. They both
take advantage of the rapid onset nature of the earthquake phenomena; so they are
unlikely to be easily adapted to phenomenon with a slower dynamic.
There are currently several on-going or planned developments beyond the
constant evolution and improvements of the different components of our
information system. EMSC has been promoting and coordinating the deployment
of citizen operated seismological networks in the Euro-Med region in
coordination with the Quake Catcher Network (QCN) initiative (Cochran et al.,
2009) and ensures data collection on its own servers. Two prototype networks are
in operations in Patrai and Thessaloniki (Greece). The aim is to augment
seismological data to better map the spatial variations of the shaking level in
urban environment during an earthquake, a key parameter for damage scenarios.
We plan to test within a year in the same regions a smartphone application that
records earthquakes with the phone’s internal motion sensors. Another
development for the smartphone applications is to exploit the teachable moment
immediately after a felt earthquake by providing users with preparedness tips and
security guidelines for the ones subjected to damaging shaking levels. Social
network harvesting is a promising technique to identify possible indirect
earthquake effects such as triggered landslides and fires, which are difficult to
predict and monitor through global networks; a test of the Artificial Intelligence
for Disaster Response (AIDR) (Imran et al., 2014) platform is to begin to evaluate
the possibility of detecting weak signals associated to this phenomena.
In conclusion, our strategy has fully demonstrated its advantages for improved
public earthquake information and collecting information on earthquake effects at
little cost for the benefit of seismologists. The ultimate objective is to fully
demonstrate its operational benefit for improved earthquake response by
developing a fully functional, time evolving situation map integrating all available
data and sharing its results with first responders.
ACKNOWLEDGMENTS
EMSC wants to thanks first its members for data sharing and financial support,
the US Geological Survey and Paul Earle for sharing Twitter Earthquake
Detections, Digital Element for offering the IP location software, all the people
involved in the Quake Catcher Network experiments, in the US and in Greece.
Finally, we thank the public for its constant feedback on our tools which has been
a key driver for the development of citizen seismology.
References and Citations
1. Wald, D. J., Quitoriano, V., Dengler, L. and Dewey J. W. (1999). Utilization

Bossu et al.
Citizen Seismology
Short Paper – Social Media Studies
Proceedings of the ISCRAM 2015 Conference - Kristiansand, May 24-27
Palen, Büscher, Comes & Hughes, eds.
of the Internet for rapid community intensity maps. Seismol. Res. Lett. 70:87–
102.
2. Bossu, R., Steed, R., Mazet-Roux, G., Roussel, F. Etivant, C., Frobert, L. and
Godey, S. (2015) The Key Role of Eyewitnesses in Rapid Impact
Assessment of Global Earthquakes. Earthquakes and Their Impact on Society
(Springer). In press.
3. Earle, P., Guy, M., Buckmaster, R., Ostrum, C., Horvath S. and Vaughan, A.
(2010). OMG earthquake! Can Twitter improve earthquake response?,
Seismol. Res. Lett., 81, 246-251.
4. Earle, P., Bowden, D. and Guy M. (2011) Twitter earthquake detection:
earthquake monitoring in a social world. Annals of Geophysics, 54 (6), 708-
715
5. Bossu, R. Mazet-Roux, G., Douet, V., Rives, S., Marin, S. and Aupetit, M.
(2008). Internet users as seismic sensors for improved earthquake response.
EOS, Transactions, 89(25), 225-226
6. Bossu, R., Gilles, S., Mazet-Roux, G., Roussel, F., Frobert, L. and Kamb L.
(2011a). Flash sourcing, or rapid detection and characterization of earthquake
effects through website traffic analysis. Annals of Geophysics, 54(6) 716-727
7. Bossu, R., Gilles, S., Mazet-Roux, G.. and Roussel, F. (2011b). Citizen
Seismology or How to Involve the Public in Earthquake Response.
Comparative Emergency Management: Examining Global and Regional
Responses to Disasters. Editors: D. M. Miller and J. Rivera.
Auerbach/Taylor and Francis Publishers. Chapter 11, 237-259
8. Jung, J., Krishnamurthy, B. and Rabinovich, M. (2002). Flash crowds and
denial of service attacks: Characterization and implications for CDNs and
web sites. Proceedings of the 11th international conference on World Wide
Web, 293-304
9. Marnerides, A., Pezaros, D., and Hutchison, D. (2008) Flash crowd detection
within the realms of an Internet service provider (ISP). The 9th Annual
Postgraduate Symposium on the Convergence of Telecommunications,
Networking and Broadcasting, June 23-24, 2008, Liverpool, UK.
10. Bossu, R., Lefebvre, S., Cansi, Y. and Mazet‐Roux, G. (2014).
Characterization of the 2011 Mineral, Virginia, Earthquake Effects and
Epicenter from Website Traffic Analysis. Seism. Res. Letters, 85(1), 91-97
11. Grunthal, G. (1998). European Macroseismic Scale EMS-98. Cahier du
Centre Européen de Géodynamique et de Séismologie, 15
12. Cochran, E. S., Lawrence, J. F., Christensen, C., and Jakka, R. S. (2009). The
quake-catcher network: Citizen science expanding seismic horizons. Seism.
Res. Letters, 80(1), 26-30.
13. Imran, M., Castillo, C., Lucas, J., Meier, P., and Vieweg, S. (2014, April).
AIDR: Artificial intelligence for disaster response. In Proceedings of the
companion publication of the 23rd international conference on World wide
web companion (pp. 159-162). International World Wide Web Conferences
Steering Committee.