Fusion of geospatial information from remote sensing and social media to prioritise rapid response actions in case of floods

Efficiently managing complex disasters relies on having a comprehensive understanding of the situation at hand. Immediately after a disaster strikes, it is crucial to quickly identify the most impacted areas to guide rapid response efforts and prioritise resource allocation effectively. Utilising early-stage estimations of impacted regions, derived from indicators such as building distribution, hazard zones or geo-social media reports, can aid in planning data collection initiatives to enhance situational awareness. Consequently, there is a need to improve the availability and accuracy of early-stage impact indicators and to integrate them into a coherent spatial and temporal analysis framework that enables identification of disaster-affected areas. In this study, a method is proposed that is tailored to quickly identifying disaster hotspots, especially in situations where detailed damage assessments or very high-resolution satellite images are not readily available. The approach leverages the H3 discrete global grid system and uses a log-linear pooling method coupled with an unsupervised hyperparameter optimization routine to fuse information on flood hazard extracted from medium-resolution satellite images with disaster-related data from Twitter and freely available supplementary geospatial data on exposed assets. The performance of the method is evaluated by comparing its outcomes against detailed damage assessments conducted during five real-world flood disasters. The results indicate that it is possible to determine the areas most affected by a flood solely based on readily available proxy information. Code and test data are available from: https://github.com/MWieland/h3h .