Identification of disaster-affected areas using exploratory visual analysis of georeferenced Tweets: application to a flood event

To enable decision makers to conduct a rapid assessment of the situation during the disaster response phase and improve situational awareness, we propose an approach to identify affected areas using geo-spatial footprints. These geo-spatial footprintssummarize information and threats and are derived from georeferenced social media messages and authoritative data sources. The combination of data mining techniques for data pre-processing and exploratory visual analysis is a promising approach for dealing with heterogeneous data under time pressure. This paper presents the first steps towards this objective by using georeferenced Tweets to define the geospatial footprint of a flood event that occurred in Italy in 2013. After cleaning the data, density-based clustering, distance-bounded spatio-temporal event clustering and data-driven territory tessellation techniques were applied; visual analysis was used to define the best parameters combination. A comparison between the results and ground-truth data was performed. The proposed methods showed positive results in the identification of areas affected by the flood at regional scale. The combination of data mining with visual analysis for parameters setting proved to be an intuitive and fast procedure that could help decision makers deal with geosocial media data and assist them with rapid assessment of the situation.