Citizens acting as "sensors" for a better grasp of what actually happens on the ground during natural disasters: this is the idea the BRGM is pursuing with the Suricate-Nat platform, which reports and analyses tweets in real time on natural disasters in France.
7 February 2019
Suricate-Nat intercepts key words (earthquake, flood, place names, etc.) used in tweets to pinpoint risk situations

Suricate-Nat intercepts key words (earthquake, flood, place names, etc.) used in tweets to pinpoint risk situations.

© BRGM

The reactions of social network users, geolocated tweets, the number of messages sent and their precision are all indicators that can serve as "sensors" to measure the scale of a disaster

The reactions of social network users, geolocated tweets, the number of messages sent and their precision are all indicators that can serve as "sensors" to measure the scale of a disaster.

© BRGM

Citizen involvement, cooperation and participatory science are key levers that the BRGM is attempting to promote through its digital tools and platforms in the field of the Earth Sciences.

For scientists, social networks are at once a vast source of information on what is happening on the ground and a very modern way of involving citizens and cultivating risk awareness among the public. This is the BRGM’s aim with its Suricate-Nat platform on natural risks, designed to capture information and inform or warn citizens through social networks, particularly Twitter. The BRGM and the Troyes University of Technology (UTT), with support from the MAIF Foundation, have uploaded the platform to the Web so that social networks can become a new source of citizens’ information for natural risk monitoring.

Citizens as "sensors"

The idea, first with earthquakes and moving on to other hazards such as floods and ground movements, is to capture tweets about events, automatically recognise posts from eye-witnesses bringing first-hand information and send tweets to these eye-witnesses with a link to the platform where they can post more information.

The situation when a natural disaster strikes is typically very difficult to describe. There is always a critical phase when the main problem is to make use of as much information from the ground as possible to build up a realistic picture of the situation.

The speed of eye-witness responses on social networks holds out prospects for citizens to act as real-time “sensors” during natural disasters.

Diagram showing how multisensor surveillance (technological or human sensors) can be used to distribute the real-time information critical to crisis management to rapidly estimate the scale of a disaster

Diagram showing how multisensor surveillance (technological or human sensors) can be used to distribute the real-time information critical to crisis management to rapidly estimate the scale of a disaster.

© BRGM

Participatory science and AI

In practice, Suricate-Nat, with Version 1 launched at the end of 2017, collects “primary” information from citizens on natural disasters, then processes that information to inform and warn the public about the risks. This is a new kind of tool to help geoscientists and the public authorities to closely monitor natural disasters and their effects.

Tweets in French that include key words associated with earthquakes, for example, are collected and automatically processed by artificial intelligence algorithms. The messages are then analysed to build up instantaneous indicators as to the intensity of the phenomena. The messages are then sent to web users who can in turn contribute to the analyses to supplement and improve predictive models.

Twitter is very well suited to this kind of application, thanks to its immediacy (short 280-character messages) and coverage (over 6 million users in France). However, analysing the tweets is by no means a simple task, hence the use of AI to process the huge volumes of data, distinguish rumour from hard information, analyse the history of tweets and process messages that are not necessarily informative or descriptive.

The automatic analysis tool developed by the BRGM will delete duplicates and messages posted by robots; the system also uses a seismology-inspired algorithm to detect when thresholds are exceeded. The tweets are sorted into “eye-witness” and “hearsay” reports, then geolocated and enriched with witness responses to questionnaires, which are manually sorted by volunteer web users prior to processing via the platform.

The algorithms were developed with Twitter from the Barcelonnette earthquake of 7 April 2014, which is still the strongest recorded in France in over ten years. The Suricate-Nat platform has already recorded some thirty events, including the January 2018 floods in the Paris region and some of the innumerable earthquakes that have been sensed in Mayotte on a daily basis since May 2018.

The main goal, then, is for “connected citizens” to become actively involved when a natural disaster strikes. The project therefore makes a significant contribution to the cultivation of risk awareness in the population.