Groundsource: Turning news reports into data with Gemini

a day ago

Introduction of Groundsource, a scalable methodology using Gemini to transform unstructured global news into actionable historical data.
First open-access dataset focuses on urban flash floods, comprising 2.6 million records for improved forecasting.
Historical data is crucial for climate research, hazard mitigation, urban planning, and emergency response.
Addresses global data scarcity in hydro-meteorological hazards like floods, which lack standardized observation infrastructure.
Existing archives like GFD and DFO have limitations such as cloud interference and focus on large disasters.
Groundsource leverages AI to analyze news reports, extracting and structuring flood data from 80 languages into English.
Uses Gemini LLM for classification, temporal reasoning, and spatial precision to verify and standardize data.
Technical validation shows 60% accuracy in location and timing, with 82% being practically useful for analysis.
Significantly expands coverage, capturing 85%-100% of severe flood events recorded by GDACS (2020-2026).
Enables near-global urban flash flood forecasts up to 24 hours in advance, integrated into Google’s Flood Hub.
Potential to apply methodology to other natural hazards like droughts, landslides, and avalanches.
Future goals include refining the model, expanding rural coverage, and integrating new data sources.

Hasty Briefsbeta