Google Uses Gemini to Predict Flash Floods in 150 Countries
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google researchers used Gemini to sort through 5 million news articles, identifying 2.6 million flood reports and converting them into a geo-tagged time series called "Groundsource" — the first time the company has used language models for this kind of work, per Google Research product manager Gila Loike.
- Groundsource was used to train a Long Short-Term Memory (LSTM) neural network that ingests global weather forecasts and generates the probability of flash flooding in a given area.
- Flood Hub now displays flash flood risk for urban areas in 150 countries, with Google sharing its data with emergency response agencies worldwide.
- António José Beleza, an emergency response official at the Southern African Development Community, said the forecasting model helped his organization respond to floods more quickly during a trial with Google.
- The model has limitations: it identifies risk across 20-square-kilometer areas, is not as precise as the U.S. National Weather Service's flood alert system, and does not incorporate local radar data for real-time precipitation tracking.
- Juliet Rothenberg, a program manager on Google's Resilience team, said the approach targets regions where local governments cannot afford expensive weather-sensing infrastructure, and the team hopes to apply LLM-derived datasets to forecasting heat waves and mudslides.
Why it matters: Flash floods kill more than 5,000 people annually, and the countries most affected often lack the weather-sensing infrastructure that powers existing U.S. alert systems. Google is now filling that gap with an active forecasting tool in 150 countries, though its 20-square-kilometer resolution is far coarser than the National Weather Service's radar-based system.
