Researchers have described a system called Groundsource that turns unstructured news reports about flooding into structured data by using Google tools and the Gemini Large Language Model (LLM). The method focuses on news articles where flooding is the primary subject, then isolates the main text from 80 languages, standardizes it into English, and extracts event details for analysis.
The process starts with the Google Read Aloud user-agent, which is used to isolate primary text. That text is then standardized into English via the Cloud Translation API.
The most critical step is handled by Gemini, which follows a strict verification process. The model classifies reports as actual, ongoing, or past floods, and excludes articles that only discuss future warnings, policy meetings, or general risk modeling. It also uses temporal reasoning to anchor phrases such as “last Tuesday” to an article’s publication date. For location data, the system identifies granular places such as neighborhoods and streets and maps them to standardized spatial polygons using Google Maps Platform.
In manual reviews, the researchers found that 60% of extracted events were accurate in both location and timing. They said 82% were accurate enough to be practically useful for real-world analysis, such as capturing the correct administrative district or pinpointing the event within a single day of its reported peak.
The researchers said Groundsource generated 2.6 million events, expanding coverage compared with records in traditional monitoring systems. Spatiotemporal matching showed that Groundsource captured between 85% and 100% of the severe flood events recorded by GDACS between 2020 and 2026.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
