Flash floods kill more than 5,000 people each year and remain among the hardest weather events to predict. The core problem is straightforward: these events are too short-lived and localised to be captured by traditional weather monitoring networks. Unlike river flooding, which can be tracked through sensor-equipped gauges across river basins, flash floods erupt and vanish before conventional data collection systems can respond.
The lack of historical data has been the fundamental barrier to prediction. Traditional flood models rely on physical gauges in rivers that have collected water levels for decades, but no equivalent infrastructure exists for city streets and urban waterways. News articles present a largely untapped source to fill this gap.
Google's response is Groundsource, which uses Gemini to analyse decades of public reports and identify over 2.6 million historical flood events spanning more than 150 countries, then uses Google Maps to determine precise geographic boundaries for each event. The process works by instructing Gemini to extract specific details from millions of news stories; the system learns to identify where and when floods occurred from journalistic accounts that humans wrote about them.
Using this historical dataset, researchers trained a new model that makes progress towards predicting flash floods in urban areas up to 24 hours in advance. The urban flash floods forecasts are now available in Google's Flood Hub, alongside existing riverine flood forecasts which cover 2 billion people in more than 150 countries.
Early validation shows practical value
Google validated Groundsource by cross-checking it against manual annotations of news reports and existing flood databases, finding that about 82% of events were labelled with the correct location and timeframe. Spatiotemporal matching shows Groundsource captured between 85% and 100% of the severe flood events recorded by GDACS between 2020 and 2026.
In practical deployment, the system has already shown its worth. A regional disaster authority in Southern Africa caught a flash flood alert while the tool was still in beta, confirmed the flood on the ground, and then deployed a humanitarian worker to oversee the response. According to Juliet Rothenberg, the product director for Google's crisis resilience work, "that chain of events from a prediction in Flood Hub to boots on the ground is exactly what Flood Hub was built for."
Clear limitations acknowledge the gap
The system is not a panacea. It identifies risk across 20-square-kilometre areas and is not as precise as the U.S. National Weather Service's flood alert system, in part because Google's model doesn't incorporate local radar data, which enables real-time tracking of precipitation.
The tool is somewhat crude, simply indicating whether there is medium or high likelihood of a flash flood occurring in the next 24 hours in a given area. It only covers urban areas and doesn't tell you how severe the flood could be. These constraints are deliberate: the project was designed to work in places where local governments can't afford to invest in expensive weather-sensing infrastructure or don't have extensive records of meteorological data.
The broader significance lies not in what Groundsource does today, but in what it signals about AI's application to data scarcity. Marshall Moutenot, CEO of Upstream Tech, a company using similar deep learning models to forecast river flows, said Google's contribution is part of a growing effort to assemble data for deep learning-based weather forecasting models. "Data scarcity is one of the most difficult challenges in geophysics," Moutenot said. "This was a really creative approach to get that data."
The same AI-driven approach of Groundsource has potential to be applied to other natural disasters, like landslides or heat waves, turning verified reports from around the world into datasets that enable improved global resilience. For regions that lack the infrastructure and institutional capacity of wealthy nations, that capability matters enormously. The system won't replace sophisticated weather infrastructure where it exists, but it may save lives where it never existed at all.