Social media-driven multi-scale flood sensing: A systematic review of applications from global monitoring to community resilience

With the acceleration of climate change and urbanization, floods have become increasingly frequent. Social media has emerged as a valuable data source for flood monitoring. However, current social media flood research is confined to a single spatial scale, overlooking the heterogeneity and methodological challenges of using such data across different spatial levels. This study proposed a multi-scalar analytical framework to systematically assess the current applications and future directions of social media in flood-related research. We adopted a semi-automated method supported by large language models, combining the PRISMA protocol with KeyBERT for keyword extraction, and analyzed 105 relevant publications. The framework covered four spatial levels: global, regional, urban, and community, and integrated extracted insights using GPT-4o. Results showed that: (1) the global and national scales emphasized cross-platform disaster identification and policy feedback mechanisms; (2) the regional scale highlighted variations in risk perception and interregional information coordination; (3) the urban scale focused on real-time monitoring and fine-grained spatial modeling; and (4) the community scale concentrated on individual evacuation behavior and neighborhood network analysis. Social media data demonstrated strong timeliness, high public engagement, and significant complementarity with traditional data sources in flood research. Nevertheless, key challenges remained, including limited cross-linguistic and cultural modeling capabilities and insufficient mechanisms for multi-source data integration. In addition, the underrepresentation of digitally disadvantaged populations and the lagging ethical and privacy governance further constrained the research. This study provides theoretical and methodological guidance for developing more resilient multi-scale urban disaster information systems.