Google has been leveraging artificial intelligence to enhance flood predictions, providing early alerts to over 2 billion people in 150 countries about potential riverine and urban flash floods. This initiative aims to bridge data shortages with a new method called Groundsource, which analyzes historical news reports to forecast floods even in areas without local sensors. Predictions are accessible through the Flood Hub tool, and data can also be used through an open-source API to support independent disaster resilience research.
Initially, in 2018, Google introduced a flood forecasting model in India, dependent on real-time river data to anticipate floods. Since then, AI advancements have expanded predictions to 150 countries, benefiting more than 2 billion people. In March 2026, Google unveiled Groundsource, an AI-driven method that compiles public disaster data into a comprehensive archive, starting with urban flash floods. Deborah Cohen, a senior staff research scientist at Google Research, explains how AI facilitates flood forecasting and its potential for predicting other natural disasters.
AI’s role in flood prediction
Google utilizes AI models to process vast amounts of global data related to rainfall, river levels, and ground conditions. These inputs help predict riverine floods up to seven days in advance and urban flash floods up to 24 hours before they occur. These predictions are displayed on the Flood Hub tool and appear in search results when users seek flood information. Additionally, a Floods API is available, allowing organizations to access flood predictions to issue warnings and assist communities before severe floods strike.
Flood Hub integrates global weather data with AI to generate alert maps, moving beyond traditional flood models requiring local historical data for calibration. Google’s AI draws from worldwide information, enabling predictions in regions lacking historical data, crucial for populations in data-scarce areas.
Challenges and data sourcing
Initially, Flood Hub focused on predicting river flooding, but recently expanded to urban flash flood predictions using Groundsource. This development was delayed due to the limited global data on urban flash floods compared to river floods. While river floods are monitored by numerous physical gauges around the world, urban flash floods lack such historical sensor data.
To address this, Google developed Groundsource, employing Gemini to analyze over 5 million news reports on flooding over 20 years. This process created a unique dataset of 2.6 million historical flood events across 150 countries, integrated into a new urban flash flood model now used in Flood Hub.
Community impact and future advancements
Aid organizations and researchers, such as Give Directly, utilize Flood Hub to assess flood risks and impact regions. In Nigeria, for instance, early cash distribution to affected areas facilitated evacuations, asset protection, and rebuilding efforts, significantly improving community resilience.
Looking ahead, the models powering Flood Hub are being refined to predict flash floods in rural and coastal areas, despite current data limitations. Research is ongoing to enhance model quality and explore Groundsource’s application to other natural disasters like heat waves and mudslides.
Google has open-sourced its hydrology framework, enabling meteorological services to integrate and tailor forecasts. The Groundsource dataset and Flood Forecasting API are also openly available, supporting future flood research and aligning with Google’s mission to use AI for accurate natural disaster predictions and public safety.