During the last decade, Twitter has become a robust platform for distributing messages (tweets) among numerous subscribers worldwide. Tweets tend to increase significantly during and around the occurrence of natural hazards. While Twitter is used for near real-time alerts, processes for extracting reported damage from tweets and resolving their geographical spread in high resolution are under development. In this study, we attempt to examine what was the spatiotemporal distribution of the tweets associated with the November 2016 fire in Haifa (Israel). The acquired tweets were classified and filtered using topic modeling and RCNN (Recurrent Convolutional Neural Network), a portion of them was georeferenced, and their hyperlocal spatiotemporal patterns were examined. It was found that the tweets' sentiment corresponds to the fire's cascading events, while their spatial and temporal distribution is equivalent to most of the actual reports. Despite large uncertainties in the process of examining tweets, the results indicated that Twitter could serve as another layer of near real-time information to assist decision-makers and emergency agencies during and after cascading catastrophes striking a small-scale city.
ASJC Scopus subject areas