Today I Learnt (TIL): Groundsource shows how Public Reports can become GeoSpatial Evidence

[Amit Kumar]

Hi,

Source: https://research.google/blog/introducing-groundsource-turning-news-reports-into-data-with-gemini/

Today I learnt about Groundsource, a Google Research effort that uses Gemini to turn unstructured flood reporting into a usable geospatial dataset. Its first open release contains 2.6 million historical urban flash-flood records across more than 150 countries, built from public reports in 80 languages since 2000. The validation details are encouraging too: while 60% of events were exact in both place and time, 82% were still accurate enough for practical analysis. That is a powerful reminder that public narratives, if processed carefully, can become scientific evidence rather than anecdote.

This feels especially relevant to India, where many flood, waterlogging, crop-loss, and infrastructure-failure events first surface in district newspapers, local-language reporting, and community accounts long before they enter formal databases. A Groundsource-like approach could help convert those scattered signals into GIS-linked evidence for better disaster response, seasonal planning, and long-term ecological monitoring. It also suggests a path for validating satellite-detected events with human reporting.

For CoRE Stack, the most relevant technical idea is not just collecting reports, but spatializing them well. A useful onboarding entry point is `extract_coordinates()` in `dpr/utils.py`, which shows how field records are converted into latitude-longitude pairs inside our workflows: https://github.com/core-stack-org/core-stack-backend/blob/main/dpr/utils.py#L150-L160. That small function captures the larger promise here: grounded observations, once structured and geotagged, can strengthen public-good geospatial intelligence.

If this sparks an idea, contribute to CoRE Stack and help turn local knowledge into open geospatial public infrastructure.

- Amit