Groundsource: A Dataset of Flood Events from News

High-quality historical flood data is critical for disaster risk management, infrastructural planning, and climate change attribution, however, existing global archives are constrained by sparse geographical coverage, coarse spatial resolution, or reliance on prolonged satellite observation.To address this gap, we introduce Groundsource, an open-access global dataset comprising 2.6 million high-resolution historical flood events, curated from the automated processing of over 5 million news articles across more than 150 countries.Our methodology leverages Gemini large language models (LLMs) to systematically extract structured spatial and temporal data from unstructured journalistic text.Comprehensive technical validation demonstrates that the pipeline achieves an 82% practical precision rate in manual evaluations.Furthermore, spatiotemporal matching against established external databases reveals recall capturing 85% to 100% of severe flood events recorded in the Global Disaster Alert and Coordination System (GDACS) between 2020 and 2026.By transforming unstructured global news media into a structured, localized event archive, Groundsource provides a massive-scale, extensible resource to support the training of predictive hydrological models, quantify historical exposure, and advance global disaster research.