Publications Office of the EU
Social media image analysis in the immediate aftermath of the 2020 Beirut blast
DisplayCustomHeader
MyPortal
slovenščina
BasketSummary
Web Content Display (Global)
Publication Detail Actions Portlet

Nastavitveni pripomočki

Publication Detail Actions Portlet
 Moja lestvica
custom-survey-notification
Publication Detail Portlet

Publication detail

Home

Social media image analysis in the immediate aftermath of the 2020 Beirut blast

Exploratory activity for enhancement of Copernicus emergency management services

Metapodatki publikacije

On the afternoon of August 4th 2020, two massive explosions devastated the port of Beirut and the surrounding area. DG ECHO immediately activated Copernicus Rapid and Risk & Recovery Mapping services, whose first product, based on Earth Observation (EO) data, was made available about 30h after the event. In order to fill in this gap, the CEMS Validation and Early Warning Services (EWS) teams (JRC.E1) started a test to understand how social media could be used to provide initial situational awareness to central and local emergency managers. A request was sent to the Europe Media Monitor team (EMM) team (JRC.I3) to collect the tweets posted during the 72h following the event and the outcome was analysed initially with a manual procedure to evaluate the validity of the approach before initiating the development to automate the required processing. The initial analysis consisted in selecting, across over 500k tweets containing the keywords “Beirut” “BeirutExplosions” “BeirutBlast” in 3 different languages and media content, which were geo-tagged with coordinates near the city of Beirut. This selection provided a manageable amount of data that we used for a preliminary study. The research was restricted to the first 24h, when, as also demonstrated by the processing done in this work, relevant content has a higher probability to be detected, and noise generated by indirect public in social media is lower. The following step consisted of processing the images contained in the tweets and classifying them as relevant to damage assessment, relevant to the event or not relevant. Those included in the first two classes where then displayed in an interactive map allowing the visualization of the positions from where the tweets were posted, the timestamps, the images and text of the post. This proof of concept for a potential enhancement of the Copernicus Emergency Services (CEMS), was shown to 28 practitioners who joined the workshop organized by JRC on December 1st 2020 “Social Media for Disaster Risk Management – Researcher meet Practitioners”. 79% said they would find such product useful in the initial phases of the crisis management, while the remaining 21% was unsure. This uncertainty was mainly due to concerns related to the reliability of social media. The outcome of this initial analysis shows that the potential of the social media in disaster management is significant. Images showing the impact of the event on assets and population are available from few seconds after the beginning of the crisis. Relevant information was concentrated in the initial phases with a peak between the 2nd and the 3rd hour, after which it starts decreasing. Contemporarily the probability of media content to belong to the category of “not relevant” escalates constantly, showing how the engagement of indirect public generates more and more noise reducing the confidence in the social media information. In conclusion this work proposes a framework to strengthen the use of social media for crisis management that could complement the current Copernicus Mapping service in the initial phases of the crisis. It introduces, as first development, an automatic system based on machine learning that, thanks to an automated image analysis, could improve the decision-making process in disaster monitoring, impact analysis and early-warning. Further developments aim at the automated detection of relevant media, their localization and classification providing an initial impact assessment starting from a few moments after the disaster. Its achievements may further help emergency managers to proactively implement specific actions enhancing their decision-making. While the methodologies presented can be applied to any social network, the experiment has been conducted using data extracted from Twitter platform because it offers the possibility to access freely the public stream of messages in real time and it is widely used worldwide.
Prikaži več
Prikaži manj
ISSN 1831-9424
ISBN 978-92-76-32151-4
DOI 10.2760/944555
Kataloška številka KJ-NA-30632-EN-N
PDF
ISSN 1831-9424 ISBN 978-92-76-32151-4 DOI 10.2760/944555 Kataloška številka KJ-NA-30632-EN-N
  • Released on EU publications website:
Publication Viewer

Prikaz dokumentov

Pop up window annotations