Identifying disaster impacted areas with machine learning and geospatial analysis using social media data

Authors: Clemens Havas*, University of Salzburg, Bernd Resch, University of Salzburg
Topics: Hazards, Risks, and Disasters, Applied Geography
Keywords: Social media, disaster management, machine-learning, semantic topic analysis, geospatial analysis
Session Type: Poster
Presentation File: No File Uploaded


Emergency management relies on up-to-date information about the impact of a disaster in an area. However, current emergency systems rely on data sources that do not provide a complete geospatial and temporal view of the disaster or have a temporal lag due to activation or orbital constraints. To fill these important information gaps, georeferenced social media posts are analysed in near-real time. The Evolution of Copernicus Services (E2mC) - project aims at demonstrating the technical and operational feasibility of the integration of social media analysis and crowdsourcing by developing a prototype of the innovative Copernicus Witness, a new Copernicus Emergency Management Service component. One module of this component focuses on analysing the social media data stream and providing a “big picture” of the area of interest. Semi-supervised topic models enable the automatic interpretation of topics to identify disaster-relevant information as the social media corpus includes a high degree of noise. A hot spot analysis is applied on the extracted posts and allows the visualisation of hot and cold spots in the area of interest. Results of multiple hurricanes and other disasters in the last years demonstrate the provided information of this methodology in real use cases. The trajectory of the hurricanes as well as the impacted areas could be visualised on maps by distinctly highlighting the impacted areas.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login