Groundsource is a high-resolution historical flood event dataset automatically constructed from global news data. Released by Google in 2026, it aims to provide high-resolution global flood event data. Related research papers include:Groundsource: A Dataset of Flood Events from News .

This dataset contains 2.6 million flood records, covering more than 150 countries. During data processing, the research team used Gemini Large Language Models (LLMs) to systematically extract structured information such as the time and location of flood events from unstructured news texts, achieving automated construction of large-scale historical disaster events.