Google has developed a system for predicting river floods and flash floods using artificial intelligence, and says its forecasts are currently available in 150 countries and benefit more than 2 billion people. The company provides this information through the Flood Hub tool; some alerts also appear in search results and are available to organizations and institutions through the Flood Forecasting API.
The project began with an experiment Google launched in India in 2018, which relied on real-time river data to predict floods. Since then, the company has expanded the models’ scope to predict river floods up to seven days before they occur and urban flash floods up to 24 hours in advance.
How Does the Flood Hub System Work?
The models process a large set of global data, including rainfall, river water levels, land-surface conditions, and weather data. River-flood forecasting relies on two main models: the hydrological model calculates the amount of water that will flow in a river based on weather and land conditions, while the inundation model estimates the areas that may be affected by flooding based on water-flow data.
Traditional models usually require local historical data, such as water-level measurements, to be calibrated. However, monitoring-station networks are not equally available everywhere in the world, particularly in areas where the need for warnings is greatest. Google says its models can draw on data from different regions to provide forecasts in locations that lack sufficient local records.
Groundsource Fills the Urban Flash-Flood Data Gap
Adding urban flash floods to Flood Hub was more difficult than predicting river floods because of the scarcity of sensors and historical records for this type of event. To address this gap, Google announced the Groundsource methodology in March 2026, using the Gemini model to analyze more than 5 million news reports about floods over a period spanning 20 years.
The process produced a database containing 2.6 million historical flood events in more than 150 countries. These data were then integrated into a new model for urban flash floods, and the model became available through Flood Hub.
Why Does This Development Matter?
The value of early forecasts is not limited to issuing an alert; they give communities and humanitarian organizations time to take practical measures before water levels rise. Google cites the experience of the GiveDirectly organization in Nigeria’s Kogi State, where it used the Flood Forecasting API to transfer money to people at risk of flooding before water levels rose. According to the results reported by the company, early support helped families evacuate, protect their assets, and rebuild; incomes more than doubled, food insecurity fell by 90%, and 93% of beneficiaries said they were better prepared for future floods.
Next Steps and Data Availability
Flood Hub’s flash-flood models remain limited to urban areas, where the best current data are available. Google is exploring expanding them to include flash floods in rural areas and coastal flooding, and is also examining the possibility of using the Groundsource methodology to predict other risks, such as heat waves and mudslides.
The company says it has made its hydrological framework open source, allowing national meteorological and hydrological agencies and relevant authorities to integrate their own data and produce customized forecasts. It has also made the Groundsource dataset and the Flood Forecasting API available to support flood research and disaster-resilience applications.