Google has introduced a new artificial intelligence-based system designed to forecast flash floods in urban areas, a category of extreme weather event that has long proved difficult to predict with precision.
The company’s new approach is built around Groundsource, a dataset created from public reporting and geospatial information. According to Google Research, the system analyses global news coverage of flood events and converts that material into a structured archive of localised flood incidents. Google says the dataset covers more than 150 countries and spans records from 2000 to the present.
This archive has then been used to train a forecasting model aimed specifically at flash floods in cities. Google says the model can provide warnings up to 24 hours in advance, extending the company’s existing flood forecasting work beyond river flooding and into rapid-onset urban flood events.
Flash floods are among the most dangerous forms of flooding because they can develop within minutes after intense rainfall, often leaving little time for evacuation or emergency response. Their speed and localised nature have historically made them harder to forecast than riverine floods, which tend to evolve over a longer period and with more established monitoring systems.
The new forecasts are being integrated into Flood Hub, Google’s public flood information platform. The company says the service now highlights flash flood risk in urban areas in roughly 150 countries, presenting map-based information intended for communities, public authorities and emergency planners.
Google has presented the system as a practical tool for disaster preparation rather than a purely research exercise. In its public statements, the company says the expansion is meant to help improve climate resilience, strengthen emergency planning and support local decision-making in areas exposed to flood risk.
A central feature of the project is the use of Google’s Gemini models to extract structured information from unstructured reporting. Google Research says Gemini was used to identify flood events from large volumes of news material and turn them into a training resource for forecasting. This was intended to address a longstanding data shortage, particularly in regions where official flood records are limited or incomplete.
That point is significant because disaster forecasting is often constrained by patchy historical records. Google argues that by using publicly available reports as a source of event data, the model can be extended to places where conventional hydrological information is weak. The company has also said the Groundsource dataset is being open-sourced for wider scientific and partner use.
The development also reflects a broader push by large technology firms into AI-assisted weather and climate forecasting. Google has already used machine learning in flood prediction, and the company’s wider climate-related research includes agricultural forecasting tools. In September 2025, Google said AI-powered monsoon forecasts had helped 38 million farmers in India anticipate the onset of the rainy season.
In practical terms, more accurate warnings could help local authorities make earlier decisions on evacuation planning, road closures and emergency deployment. For businesses and infrastructure operators, advance notice may reduce disruption and improve contingency planning. For city administrations, repeated risk mapping may also support longer-term planning on drainage, land use and resilience measures.
The significance of Google’s latest announcement lies in its attempt to solve a narrow but persistent forecasting problem. Flash floods remain difficult to model because they emerge quickly, vary sharply by location and have often lacked a sufficiently detailed global event database. Google’s claim is that Groundsource fills part of that gap by transforming large volumes of public information into a usable predictive tool.
Whether the system proves consistently reliable in operational use will depend on forecast accuracy, local uptake and the ability of emergency systems to act on warnings. Even so, the launch marks a further expansion of artificial intelligence into real-world weather risk forecasting and disaster preparedness.
