Google's flash flood forecasting system
16 views
Skip to first unread message
Aaditeshwar Seth
Mar 14, 2026, 11:25:08 PMMar 14
to learning, ac...@googlegroups.com, CoRE stack: NRM, core-st...@googlegroups.com
They used news articles to identify flash flood events in urban areas: https://research.google/blog/introducing-groundsource-turning-news-reports-into-data-with-gemini/. It's a generic methodology and can be applied to heat waves or other events to create an occurrence dataset at scale. This is another important project idea, to build such datasets. Google seems to have used their crawlers to get the news articles. That itself is probably not open but they used to run a system called GDELT to provide open access to news articles, not sure if it's still operational, worth checking: https://www.gdeltproject.org/
And then used this data to train a model to forecast flash floods. This itself is a very simple pixel-level model that just takes static Alpha Earth embeddings and weather forecast data. Not sure why it works! Maybe because urban areas so complex, it's not worth taking the broader spatial context into account which hydrology process based models otherwise do. https://research.google/blog/protecting-cities-with-ai-driven-flash-flood-forecasting/
These links above have further links to the papers.
Adi
Co-founder, Gram Vaani; Co-founder, CoRE Stack
Reply all
Reply to author
Forward
0 new messages