Google pushes Flood Hub into urban flash flooding

On March 12, 2026, Google Research announced that it is rolling out urban flash flood forecasts on Flood Hub, extending the company’s flood-warning system beyond slower riverine events. Google says the new service can provide up to 24 hours of advance notice for rapid-onset floods in cities. The company cites World Meteorological Organization data saying flash floods account for roughly 85% of flood-related fatalities worldwide and more than 5,000 deaths annually. In that context, even a modest increase in warning time can materially change evacuation, traffic, and emergency response decisions.

A new way to create training data

The central technical problem is that flash floods are hard to observe consistently at global scale. River forecasts can be trained from stream gauges, but urban flash floods often happen away from gauge networks and leave behind weak structured records. Google says it addressed this by using Groundsource, an AI-powered method that turns publicly available news reports into a dataset of historical flood events. Gemini was used to confirm event details such as time and location, creating enough ground truth to train and evaluate a new urban flood model. That makes the announcement noteworthy not only as a product rollout, but also as a data pipeline for climate AI.

How the model works

According to Google, the model combines global weather products from NASA IMERG, NOAA CPC, ECMWF HRES, and Google DeepMind’s AI-based medium-range weather forecast with static local features such as urbanization density, topography, and soil absorption. The architecture is based on a recurrent neural network with an LSTM unit, optimized for time-series prediction. Google says the initial launch targets urban areas with population density above 100 people per square kilometer and currently operates at 20x20 kilometer resolution. That is a compromise shaped by the limits of globally available data, but it allows the system to scale where dense local sensor networks do not exist.

Why the rollout matters

Google says performance in parts of South America and Southeast Asia is comparable to results in wealthier countries that already have better instrumentation and expert forecasting support. The company also notes that accuracy remains harder to measure in parts of Africa because reliable ground truth is still scarce. Even with those limitations, the release matters because it shows a practical path toward expanding early-warning coverage without waiting for every city to build expensive local hydrology infrastructure. For Google, Flood Hub becomes more than a river forecast product; it becomes a broader AI-enabled warning layer for climate resilience.

#forecasting

Google expands Flood Hub with urban flash flood forecasts up to 24 hours ahead