Google’s Groundsource AI Predicts Flash Floods with Unprecedented Accuracy

Google has launched Groundsource, a new AI-powered methodology leveraging its Gemini model to dramatically improve flash flood forecasting. The system analyzes decades of public reports, identifying over 2.6 million historical flood events across more than 150 countries, and predicts urban flash floods up to 24 hours in advance. Google announced the initiative on March 12, 2026.

Addressing a Critical Data Gap

For years, Google’s Crisis Resilience efforts have focused on providing early warnings for natural hazards. However, a significant obstacle remained: the lack of high-fidelity data for flash floods. This data gap hindered the development of effective AI models capable of predicting these events before they occur. Archyde highlights the historical scarcity of detailed data for flash floods, making accurate prediction challenging.

How Groundsource Works

Groundsource tackles this challenge by utilizing Gemini to analyze public reports and pinpoint the location and extent of past flood events. By combining this data with precise geographic boundaries determined using Google Maps, the system creates a robust dataset specifically focused on urban flash floods. This dataset is then used to train a new model capable of forecasting floods up to a day in advance. Decrypt notes that this is the largest-ever flash flood dataset created.

Expanding Google’s Flood Hub

The newly developed urban flash flood forecasts are now integrated into Google’s Flood Hub, expanding its coverage to an estimated 2 billion people in over 150 countries. Google already provides riverine flood forecasts through Flood Hub, and this expansion significantly enhances its disaster forecasting capabilities.

Open-Source and Future Applications

Groundsource provides a massive, open-source benchmark for researchers and scientists, particularly in urban regions lacking historical flash flooding data. Google plans to apply the same AI-driven approach to other natural disasters, such as landslides and heat waves, transforming verified reports into datasets that improve global resilience. Google emphasizes that this isn’t just about analyzing the past; it’s about building a more resilient future.