Camille Cavaliere - Diagnosis and valorization of the actual benefits of geolocated tweets issued in response to naturel disasters. Application to extreme hydrometeorological phenomena in Texas

Organized by: 
Camille Cavaliere
Camille Cavaliere



  • Mme Sandrine Anquetin, DR CNRS, IGE, Université Grenoble Alpes, examinatrice
  • M. Thierry Joliveau, PR, Université Jean Monnet de Saint-Etienne, examinateur
  • M. Johnny Douvinet, MCF HDR, Université d'Avignon, rapporteur
  • Mme. Lena Sanders, DR CNRS, rapporteuse
  • Mme Paule-Annick Davoine, PR Université Grenoble Alpes, co-directrice
  • Mme Céline Lutoff, MCF HDR, Université Grenoble Alpes, co-directrice

Digital footprints have overwhelmed our daily lives: they are captured in real-time by mobile web-linked devices and they are often geolocated. Such footprints are considered as virtual markers witnessing the physical presence of any connected individual, in a given space, at a given time. For a decade, they have been integrated as a new research field in many disciplines, including geography. Furthermore, they are often considered as an opportunity to build a new individual-based knowledge about social phenomena, based on a bottom-up approach. This research focuses on geolocated tweets: because of the many violent phenomena that occurred in the early 2010s, natural disasters management quickly became a major field of research to evaluate the potential of these particular footprints as a new field and real-time information.
However, geographical studies based on geolocated digital footprints are now facing some underlying difficulties that data analysts do not encounter: as the use of mobile technologies is heterogeneous considering populations and places, what is the social and spatial representativeness of such footprints ? Are the traditional tools of cartography and spatial analysis adapted to those unstructured data that elude any professional standard? Can we transform geolocated tweets into geographical information? In this research, we explore these questions from the extreme hydrometeorological phenomena that occurred in Texas in spring 2016 and in August 2017.
First, we formalize tweets semantic and spatial retrieving methods to improve the step of building a crisis tweets dataset. The analyse of this crisis tweets dataset is based on two approaches: on one hand, we study statistic and spatial behaviors of virtual activity hotspots. On the other hand, we explore the questions of crisis tweets relevance and cartographic valorization: we evaluate crisis tweets as an indicator of the field and phenomenon properties (local-scale intensity, vulnerability of populations, level of damages), by cross-checking tweets with official ground-truth data.