A framework for assessing the credibility of flood-inundation locations derived from social media using multi-source data.

In recent years, social media data has been widely applied in disaster management due to its large data volume, diverse content, low acquisition cost, and immediacy. However, the primary challenge in leveraging social media data for flood disaster management lies in the inherent uncertainty of social media data quality. Based on multi-source data (meteorological, topographic, and urban infrastructure data), this study proposed a framework for assessing the credibility of flood-inundation locations derived from social media (FLLSs). From the perspective of flood formation mechanisms, the framework aims to enhance the application quality of social media data in flood disaster emergency management. First, the most probable inundated coordinate was selected from multiple geocoded results for each ambiguous location name. Second, FLLS credibility was assessed by the combined weighting method based on multi-source data. Finally, the correlation mechanisms between the assessment indicators and the FLLS credibility were analyzed using an interpretable machine learning method. The proposed framework addressed the geocoding problem, where a single location name maps to multiple coordinates, resulting in a 22.81% increase in localization accuracy. FLLS credibility effectively enhances the performance of utilizing social media data for flood risk zone identification. Rainstorm factors and topographic wetness index significantly impact FLLS credibility. There is an inequality in the attention given to flood disasters on social media, with more attention focused on urban areas, and less on suburban and rural areas. Therefore, social media data should be integrated with other data sources to comprehensively analyze flood disasters.