Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Titre : Review of Early Warning Dissemination in Media and Assessment of Flood Early Warning Systems in Media : A case study in West Africa

Auteur : Hohmann, Thierry

Etablissement de soutenance : Swiss Federal Institute of Technology (Zurich) ETHZ

Grade : Master 2021

Résumé
Climate change is projected to exacerbate flood hazards in the future. Around 20% of all floods worldwide occur in Africa. The impact of floods can be mitigated with Flood Early Warning Systems (FEWS). Effective FEWSs need early warning dissemination and sound forecasting modeling. Researchers found that media, and recently also social media, are key stakeholders in warning dissemination. Furthermore, researchers investigated the potential of social media in assessing the accuracy or improving early warning dissemination in FEWS. However, little research on these two aspects has been conducted in West Africa. In this study, we aim to address the current warning dissemination in media and social media in West Africa. Furthermore, we investigate if social media can be used to assess the performance of FANFAR, a FEWS in West Africa predicting streamflow. We queried archives of newspapers, radio, and TV and tweets based on the flood event disaster database EMDAT and identified if early flood warnings are issued and whether they differ between events of different magnitude and between countries. Furthermore, we compared the daily number of tweets to the daily forecasted flood risk by FANFAR on a country scale. We employed a flood event detection algorithm developed by de Bruijn et al., 2019, which uses machine learning to identify flood-related tweets. Results show that radio and TV archives that can be systematically queried are, to the best of our knowledge, not available. However, the assessed newspapers often disseminate the warning that has been issued by a governmental meteorological agency. We found, that in Nigeria, more early warnings are issued in newspapers and tweets with an increase in people affected during a flood. We did not observe this trend in Ghana and could not find any early warnings issued in the Ivory Coast or Burkina Faso. Nevertheless, we found that tweets contain links with potential information about flood warning dissemination. Results also show that tweets indicate flooding reliably. However, we found that the assessment of FANFAR based on only flood-related tweets is not adequate. Additional factors, such as rainfall, or systematic lags between tweets and forecasts, could play a role. These factors should be determined on a regional scale. Mostly the influence of rainfall on tweets limits their usage in FEWSs that predict streamflow, such as FANFAR. Nonetheless, in combination with an authoritative dataset, such as streamflow, tweets could provide additional information. Our findings imply that media and social media can be used as a tool for warning dissemination, a source for warning dissemination research, and as additional data to support the assessment of FEWSs. Especially the projected increase of social media in the future will facilitate the improvement of early warning research and FEWSs.

Présentation

Version intégrale (3,4 Mb)