Google has unveiled a new artificial intelligence model aimed at predicting flash floods in urban areas up to 24 hours in advance, a move expected to strengthen disaster preparedness in cities around the world.
Sundar Pichai announced the development in a recent post on X, stating that the system is designed to identify flood risks earlier and provide communities with more time to respond to potential emergencies.
The forecasts are now accessible through Google Flood Hub, the company’s public platform that offers free global flood risk data. Unlike river floods, which can often be monitored days in advance through water-level gauges, flash floods tend to occur suddenly and unpredictably, especially in densely populated urban regions.
To address this challenge, researchers at Google developed a new artificial intelligence methodology known as Groundsource. The system uses the company’s large language model, Gemini, to analyse large volumes of unstructured data such as news reports and public records.
By scanning millions of articles across multiple languages, the system identified more than 2.6 million historical flood events across over 150 countries, creating one of the largest datasets on urban flash floods to date.
The dataset was then used to train a machine-learning model that combines global weather forecasts with historical flood patterns to estimate the likelihood of flash flooding in a specific location within the next 24 hours.
These AI-driven forecasts are now integrated into Flood Hub, enabling governments, emergency responders and residents to monitor potential flood risks in their regions.
The system maps risk zones using grid areas of approximately 20 by 20 kilometres, helping authorities plan evacuations and coordinate emergency responses ahead of possible flooding events.
Google has also decided to open-source the Groundsource dataset, allowing researchers, policymakers and disaster-response organisations to use the information to enhance flood modelling and climate risk analysis worldwide.
