Google has introduced Groundsource, a new AI-powered methodology designed to improve the prediction of urban flash floods. Announced on March 12, 2026, Groundsource uses the Gemini model to analyze decades of public reports and Google Maps data, creating a high-quality historical dataset of flood events. The system identified over 2.6 million historical flood incidents across more than 150 countries and trained a new model capable of forecasting urban flash floods up to 24 hours in advance. These forecasts are now available through Google’s Flood Hub platform.

The initiative is part of Google’s broader Crisis Resilience efforts, which aim to provide early warnings for natural hazards. Groundsource addresses a long-standing data gap that had limited the ability to predict flash floods before they occurred. By transforming public information into structured datasets, the project enhances global preparedness and supports scientists and partners with an open-source benchmark for further research.

Google noted that the same AI-driven approach could be extended to other natural disasters such as landslides or heat waves, contributing to improved global resilience and disaster readiness.