How to improve rapid public earthquake information? Listen to eyewitnesses!

by Rémy Bossu, Laure Fallou, Sylvain Julien-Laferrière, Matthieu Landès,
Julien Roch, Fréderic Roussel, Sébastian Soriat and Robert Steed

I set the area of interest, and then tried to adjust the magnitude threshold. Either the threshold was too high and I did not get information for the earthquakes that I felt, or it was too low and I got tens of notifications a day for earthquakes that nobody noticed. I deleted all earthquakes apps”

“I set the area of interest, and then tried to adjust the magnitude threshold. Either the threshold was too high and I did not get information for the earthquakes that I felt, or it was too low and I got tens of notifications a day for earthquakes that nobody noticed. I deleted all earthquakes apps”.  This experience was related online during the 2009 l’Aquila aftershock sequence. The conclusion was crystal clear: felt earthquakes are the only ones that really matter for the public. This was a primary motivation for LastQuake, the public earthquake information system developed by the European-Mediterranean Seismological Centre (EMSC). Still today, every earthquake or earthquake sequence, like in 2018 Lombok, Indonesia or the on-going Mayotte sequences (more info here), is an opportunity to improve our understanding of public’s expectations and to pave the way for more efficient public communication and information. These events also cause us to question the extent of the role and social responsibility of seismologists to give the public information.

Figure 1. Time evolution of the traffic to the EMSC websites (blue for desktop, green for nomad devices) and of LastQuake smartphone app launches (red) following a felt earthquake of M 3.1 in North Macedonia. The observed surges result from eyewitnesses’ information seeking behavior immediately following a tremor. Monitoring such surges is a way to automatically detect felt earthquake regardless of seismic data. These methods are named “crowdsourced earthquake detections”.

Felt earthquakes and information on demand

This may surprise non-seismologists, but the identification of felt earthquakes from only seismological data is challenging, especially for small magnitude quakes where the slightest variation in location, the time of the day, or local geological conditions can make the difference between being widely felt or not perceived at all! In the pre-Internet era, eyewitnesses would phone the local lab to check whether an earthquake had been reported. Today, people jump to their smartphone, some for an Internet search, others to launch their app or to share their experience on social networks. The Internet acts as the digital nervous system of our planet and now the fastest way to detect many felt earthquakes is to monitor these online reactions.

The first of the “crowdsourced earthquake detections” methods to be implemented is based on EMSC website traffic monitoring (Figure 1). The EMSC, being one of the top global earthquake information centers, has eyewitnesses converge rapidly on its website after a tremor and thanks to the ability to use IP (Internet Protocol) addresses for the determination of geographical origin of visitors, the area where the tremor was felt can be identified even before seismic location is available. In other words, eyewitnesses become real time sensors. This is how the epicenter of the Mineral (Virginia) 2011 earthquake was located with 30 km accuracy using only the first 2 min. of EMSC website traffic by the mere retro-propagation of the hit times of the visitors (Bossu et al., 2014)!

Figure 2. The main list of the LastQuake app displays only felt and damaging earthquakes. Crowdsourced earthquake detections are automatically published on the different components of the LastQuake information system: our websites, the Twitter quakebot @LastQuake and the eponymous smartphone app. The tweet contains the word “earthquake” in the local language to increase visibility to searches engines and a link allows eyewitnesses to report their experience on our website. Eyewitnesses visiting our websites are invited to download the app thus linking the different components together.

More importantly, when dealing with rapid public information, the amplitude of the traffic surge is a proxy measurement of the public desire for information. This allows the EMSC to switch from the traditional broadcasting of all available information to showing only the information in demand. In other words, rather than simply making publicly available the 50 000 earthquakes located per year, LastQuake highlights the few thousand (i.e. few percent) which arouse public interest and are identified as widely felt (Figure 2). On December 27, 2018, a small M 2.5 earthquake in Cyprus led to the collection of more than 150 felt reports from eyewitnesses. This neatly illustrates that the earthquake magnitude is not the gold standard for evaluating the social importance of an earthquake. We believe that not providing the information that eyewitnesses are looking for could reduce trust level towards the seismological community.

Figure 3. Earthquake safety tips are presented to LastQuake app users within minutes of strong shaking. Their intention is to limit inappropriate behaviors after earthquakes.

A rapid description of the LastQuake multichannel information system

LastQuake is a multichannel information system comprising of websites, a smartphone app and a ‘quakebot’ on Twitter and Telegram (Figure 2). Two other crowdsourced earthquake detection methods complement the real time monitoring of website traffic: one developed by P. Earle based on the number of tweets containing the keyword “earthquake” in various languages and more recently, the time evolution of the number of LastQuake smartphone app launches (Bossu et al., 2018). Detection times typically range from 20 to 100s after the earthquake’s occurrence and precede seismic location in the vast majority of cases. Crowdsourced detections are automatically published on the different components of the system and eyewitnesses are invited to share their experience through felt reports, comments and geo-located pictures or videos (Figure 2). Each of LastQuake components presents different functionalities and different strengths and weaknesses, notably in terms of crowdsourcing efficiency or reach. One point which is worth mentioning is that Twitter alone can significantly complement websites in terms of visibility with more than 50 M views in 2018 and some individual tweets being viewed 500 000 times. A more detailed description can be found in Bossu et al. (2018).

Beyond public information, the LastQuake system also aims at heightened situational awareness and reducing the intrinsic uncertainties involved in estimating damage scenarios for improved efficiency of response (Bossu et al., 2016).

How user experience drives LastQuake evolution

The Gorkha (Nepal) 2015 earthquake was the first heavily damaging event after the initial launch of the LastQuake app in July 2014. It was massively downloaded immediately after the mainshock, and more than 7 000 felt reports and 100 geo-located pictures were collected during the sequence (Bossu et al., 2015). A few weeks later, we performed an online survey of its Nepalese users to identify its strengths and weaknesses. Notably no-one asked for higher level seismological products, the main request was simply “Tell us what we should do after an earthquake!”. This led to the design, online validation and implementation of visual earthquake safety tips (Figure 3) which are now integrated with the general menu as well as in specific app notifications to users that occur after they have been through high shaking levels.

Figure 4.  Example of how the app is used and feedback received on Twitter during the Lombok (Indonesia) earthquake sequence.

During the summer of 2018, Lombok (Indonesia) was struck by a sequence of earthquakes – two of them deadly (more info here) leading to about 12 000 additional LastQuake app downloads in the country by tourists and nationals. It also led to extensive questions and exchanges on Twitter, notably related to the possible evolution of the earthquake sequence, whether one should cancel planned holidays in the neighboring island of Bali, etc. Many of the questions, mainly from tourists, are those for which science has no final answer, but for which we did our best to provide scientifically-based replies.

Three main lessons were learnt from this sequence (Figure 4):

►   in many countries and regions, informing public in the national language(s) is not enough because of tourism and foreign communities. This very same issue was discussed last May at the Earthquake Research Institute in Japan, which has the best public earthquake information system in the world, but is exploring the ways to make it more easily accessible to foreigners ahead of the 2020 Olympic games

►   if the people directly affected by an earthquake sequence, (i.e. onsite at the moment of the quakes) are the primary target of rapid information; the diaspora, or more generally anyone having an emotional connection to the affected area, may also be looking for information and should be taken into account

►   rapid information reduces anxiety during these critical moments of crisis and high emotion

The latest point is well recognized by sociologists (e.g., Saathoff and Everly, 2002) and confirmed by the feedback received. Seismologists have a role to play here, although the practice of answering questions on social media does not seem to be widespread and it is difficult for a layperson to get answers. It confirmed for EMSC that, even when no final answer can be offered, questions should not remain ignored and direct exchanges with eyewitnesses should be promoted as much as possible. This is a reminder that social networks are a two-way communication system and have a role to play in emotional support.

Figure 5. In its latest release, LastQuake preserves access to information (felt reports and comments) related to felt earthquakes even when the magnitude and location has not been determined. If and when seismic location and magnitude become available, the tan coloring of the event entry is replaced by the usual color code.

In May 2018, an earthquake sequence started in Mayotte in the Mozambique Channel and is still active at the time of writing (April 2019). Its cultural context and implications for communication was the topic of the previous post of this blog (Fallou and Bossu, 2019).

Despite recent instrumental deployment efforts to better understand this sequence, it remains difficult to rapidly locate those earthquakes in the region which are not teleseismically recorded (ie. those earthquakes below M4.5-4.8). As a consequence for many of the felt earthquakes, although the eyewitnesses’s online reaction was detected, the EMSC was unable to provide a rapid seismic location and magnitude estimate. The situation had not been anticipated in the initial development of the app. In practice, eyewitnesses could initially see the message of the crowdsourced detection on the app (Figure 1), provide their felt report, but without seismic confirmation within 15 min, the message disappeared and felt reports were no longer accessible. This reoccurring situation was not understood by many users and some believed the earthquakes were hidden on purpose!

A new version of the app has now been released which publishes the map of the felt reports and the comments when magnitude and location are not yet available (Figure 5). It has also been an opportunity to study how earthquake and seismic risk is perceived and understood by the population, an approach that will be applied in Haiti, in a project aiming at complementing national seismic network with RaspberryShake cheap sensors and improving public earthquake information and communication (Calais et al., 2019).

Finally, on September 28, 2018, the scientific community was reminded by the earthquake and tsunami in Palu, Indonesia (more info here) that at short distances from the epicenter, the earthquake’s shaking is the only tsunami early warning. When shaking is violent and/or long, the distance to the epicenter is likely to be small and if possible, people should evacuate to higher grounds without waiting for a possible official alert which is likely to arrive too late. We are currently working, thanks for a funding from Foundation MAIF, on visual tsunami safety tips which will complement the current earthquake safety tips.

Conclusion

LastQuake is a means for the EMSC to rapidly crowdsource information on the effects of earthquakes. This can only be achieved by a large adoption of these tools by the public which itself depends on LastQuake’s capacity to fulfill their expectations. The joint analysis of public reactions with sociologists has already led to several changes and improvements and this iterative process is likely to continue in the future.

From our point of view, the main lesson is that each felt earthquake triggers a public desire for information and each damaging one or significant earthquake sequence leads to tremendous expectations that may not be always met by the seismological community.

In practice, especially after damaging earthquakes, many questions may remain unanswered, emergency services do not always having the right expertise, and many seismologists have little time and not always the willingness and appetite to engage with the public. One can argue that for the layperson, the aftermath of felt and damaging earthquakes is the only time when they can directly benefit from seismology. These moments are then essential to build trust, teach, raise awareness and potentially attract interest in our field and recruit the future generation of seismologists. Contrarily, failing to meet these public demands leaves space and opportunity for charlatans who rarely hesitate to exploit population anxiety to spread their harmful earthquake predictions. Seismologists could help with alleviating this problem, if they want to.

______________
References

Bossu, R., Lefebvre, S., Cansi, Y., & Mazet‐Roux, G. (2014). Characterization of the 2011 Mineral, Virginia, earthquake effects and epicenter from website traffic analysis. Seismological Research Letters, 85(1), 91-97.

Bossu, R., Laurin, M., Mazet-Roux, G., Roussel, F., & Steed, R. (2015). The importance of smartphones as public earthquake-information tools and tools for the rapid engagement with eyewitnesses: A case study of the 2015 Nepal earthquake sequence. Seismological Research Letters, 86(6), 1587-1592.

Bossu, R., Steed, R., Mazet-Roux, G., Roussel, F., Etivant, C., Frobert, L., & Godey, S. (2016). The key role of eyewitnesses in rapid impact assessment of global earthquakes. In Earthquakes and their Impact on Society (pp. 601-618). Springer, Cham.

Bossu, R., Steed, R., Roussel, F., Landès, M., Fuenzalida, A., Matrullo, E., … & Fallou, L. (2018a). App Earthquake Detection and Automatic Mapping of Felt Area. Seismological Research Letters, 90(1), 305-312.

Bossu, R., Roussel, F., Fallou, L., Landès, M., Steed, R., Mazet-Roux, G., … & Petersen, L. (2018b). LastQuake: From rapid information to global seismic risk reduction. International journal of disaster risk reduction, 28, 32-42.

Calais, E., Boisson, D., Symithe, S., Momplaisir, R., Prépetit, C., Ulysse, S., Etienne, G. P., Courboulex, F., Deschamps, A., Monfret, T., Ampuero, J.-P., Mercier de Lépinay, B., Clouard, V., Bossu, R., Fallou L. and E. Bertrand Can a Raspberry Shake Seismic Network Complement a National Seismic Network? A case study in Haitï (accepted by EOS) https://eartharxiv.org/nyp7h/

Fallou L. and Bossu R. Taking into account the cultural context to improve scientific communication – Lessons learnt from earthquakes in Mayotte. March8, 2019 https://blogs.egu.eu/divisions/sm/2019/03/

Steed, R. J., Fuenzalida, A., Bossu, R. Bondár, I., Heinloo, A., Dupont, A., Saul, J. and A. Strollo. (2019). Crowdsourcing triggers rapid, reliable earthquake locations Science Advances 03 Apr 2019. http://advances.sciencemag.org/content/5/4/eaau9824

Saathoff, G., and Everly, G. (2002). Psychological challenges of bioterror: containing contagion, Int. J. Emerg. Ment. Health 4, no.4 (Fall 2002), 249.

Authors
Rémy Bossu
Laure Fallou
Sylvain Julien-Laferrière
Matthieu Landès
Julien Roch
Fréderic Roussel
Sébastian Soriat
Robert Steed