2019 FOSS4G Bucharest Talks speaker: Barbara Pernici
Fast insight about the severity of hurricane impact with spatial analysis of Twitter posts
Social media have shown significant contribution in disaster reliefs. It could be very valuable source of the on-site information shared by the affected citizens. Particularly, Twitter is currently one of the most popular social media used for the exchange of information connected to the disasters. If this type of source is considered as a real-time crowdsourcing of crisis information, the spatial distribution of geolocated tweets related to an event can represent an early indicator of the severity of impact. This raises a question if rapid mapping teams could use additional information from Twitter before mapping. Would it be possible to estimate the outcome, to understand the affected zones and approximate level of impact?
The aim of this paper is to explore the spatial distribution of the Twitter posts related to a disaster and to analyse their potential in providing fast insight regarding the impact. The focus of the analysis was on the tweets related to the hurricane Michael that happened in Florida, in the United States on October 2018. The crisis maps produced by Copernicus Emergency Management service were used as reference data and obtained results were compared with them. Copernicus EMS have produced twenty-five delineation maps over the coast of Florida. Six maps were delivered on the 11th of October and the rest of crisis maps were published on the 12th of October. The focus of this study is to explore the potential of Twitter’s crisis posts in providing information before the delivery of maps. The available message dataset consisted 8169 tweets posted from 10th until 15th October. The tweets published before the delivery of crisis maps that are inside of the crisis maps’ area, in total 30% of the available dataset, were analysed. Weights have been assigned to each tweet, on a base of the date of posting; i.e. newer posts were considered as more relevant.
Spatial statistics have been performed with QGIS and GeoDa. For example, the QGIS plugin Hotspot was used to identify where statistically significant spatial clusters were present, more precisely, the zones with significant concentration of relevant posts were corresponding to areas with high impact of the hurricane. Comparisons of the results with the reference data have been performed and discussed, showing the potential value of these data for crisis mapping.