Google announced an expansion of its AI-driven flood forecasting capabilities, introducing a new approach that uses historical news reports and artificial intelligence to improve flash-flood predictions. The effort is designed to help communities, emergency responders, and governments better prepare for sudden flooding events, particularly in areas where traditional flood monitoring infrastructure is limited.
The initiative centers on a methodology called Groundsource, which converts narrative accounts of floods from historical news articles into structured datasets for training machine-learning models. Using its Gemini AI system, Google analyzed decades of public reporting to identify and map millions of flood events around the world.
According to the company, the dataset includes more than 2.6 million historical flood events spanning over 150 countries. This information helps address a long-standing challenge in flood forecasting: the lack of detailed historical data on flash floods, which often occur rapidly and with little warning.
Data generated by Groundsource has been used to train a new machine-learning model capable of predicting flash floods up to 24 hours in advance. These forecasts rely on meteorological data and geographic information to estimate where sudden flooding may occur in urban areas.
The predictions are delivered through Google’s Flood Hub platform, which provides publicly accessible flood forecasts and maps to support global disaster preparedness efforts. Google said the expansion into flash-flood prediction builds on its existing river-flood forecasting systems and aims to improve early warning capabilities in vulnerable regions.
KEY QUOTES:
“Our goal is to help communities better prepare for floods by improving the data that powers flood forecasting models.”
“By transforming decades of historical reports into structured flood data, we can train AI models to better predict where flash floods may occur.”
Google Research Team — Google Research
