Bluesky as a social media data source for disaster management: investigating spatio-temporal, semantic and emotional patterns for floods and wildfires

Abstract Social media has become a key data source for near-real-time disaster monitoring and response, with Twitter playing a central role for over a decade. However, recent Application Programming Interface (API) changes on Twitter (now: X) have restricted academic data access, creating an urgent need to identify viable alternatives. This study investigates the suitability of the decentralised social media platform Bluesky for disaster-related geo-social media analysis, aiming to evaluate whether it can serve as a viable alternative microblogging platform for spatio-temporal disaster monitoring. Using a keyword-based crawling pipeline, we collected 676,337 posts related to two major natural disasters: the September 2024 Central Europe floods and the January 2025 Southern California wildfires. We applied a multilingual analysis pipeline covering semantic, emotional, geospatial, and temporal modalities. It includes disaster-relatedness classification, emotion detection, geoparsing and subsequent spatio-temporal aggregation. Our results show that disaster-related content on Bluesky surged in direct response to the disasters, peaking at up to 80% of daily posts during the main impact phases. Emotional expressions, particularly fear and anger, rose sharply alongside event progression. Geospatial analysis of the geoparsed data revealed heightened disaster-related posting activity in affected areas, demonstrating the platform’s utility for geographic disaster monitoring. However, large differences between urban and rural regions, as well as between different countries, were identified. Furthermore, we demonstrate current platform limitations such as user penetration, API constraints and sensitivity to keyword selection.