Google is using artificial intelligence to forecast flash floods by analyzing millions of news reports, an approach researchers say could improve disaster warnings in regions lacking traditional weather-monitoring infrastructure.
Flash floods kill more than 5,000 people globally each year and remain among the hardest disasters to predict because they occur rapidly and often in highly localized areas. Conventional forecasting systems rely on sensors, river gauges and dense meteorological networks — resources that are limited or absent in many countries.
To address this gap, Google researchers turned to an unconventional data source: decades of public reports documenting flood events.
Using its AI model Gemini, the company analyzed roughly five million news articles and other public reports, identifying more than 2.6 million historical flood events across more than 150 countries. Those reports were then converted into a geotagged dataset known as Groundsource, providing a large archive of past flood signals.
The dataset was combined with global weather forecasts and processed through a Long Short-Term Memory neural network, a type of machine learning model designed to analyze time-based patterns. The resulting system estimates the probability of flash floods in urban areas.
The model now powers the company’s Flood Hub platform, which highlights flood risks in cities across more than 150 countries and is designed to help communities prepare before disasters strike.
“We trained a new flood forecasting model designed to predict flash floods in urban areas up to 24 hours in advance,” Google CEO Sundar Pichai said in a post on X.
“To help address a flash floods data gap, we created Groundsource: a new AI methodology using Gemini to identify 2.6M+ historical events across 150+ countries. We’re open-sourcing this dataset to advance global research, and urban flash flood forecasts are live now in Flood Hub to help communities stay safe.”
According to Yossi Matias, Google Vice President & Head of Google Research, the absence of reliable historical data has long hindered efforts to develop predictive models for flash floods.
“High-fidelity data for certain disasters like flash floods simply did not exist,” Matias said. “This data gap has long prevented our ability to train AI models to predict flash floods before they happen — until now.”
Groundsource addresses that gap by transforming public information into structured scientific data. Researchers used mapping technology to determine precise geographic boundaries for each flood event identified in the news reports, creating a detailed global dataset focused on urban flash floods.
Urban flash flood forecasts are now available on Flood Hub alongside Google’s existing river flood forecasting system, which already covers about two billion people worldwide.
Researchers say the same AI approach could eventually be used to build datasets for predicting other disasters such as landslides, mudslides and heat waves.
By turning written reports and public records into structured data, the project suggests that vast archives of journalism and historical documentation could become valuable resources for scientific research and disaster forecasting.
Discover more from Pluboard
Subscribe to get the latest posts sent to your email.