Google Uses Gemini to Predict Flash Floods

Google Uses Gemini to Predict Flash Floods Worldwide via the Groundsource Tool

Google announced the launch of a new AI-powered tool called Groundsource, aimed at predicting flash floods on a global scale. The tool relies on the Gemini model to analyze millions of old news reports and extract precise historical data about past flood events in various countries.

The Major Challenge: Predicting Flash Floods

Flash floods are considered among the most difficult natural disasters to predict, as prediction models typically rely on extensive historical data that is often unavailable in many regions. To overcome this challenge, Google developed a methodology that relies on AI to convert textual content in old news articles into a scientific database that can be used in climate forecasts.

The Gemini model analyzed more than 5 million global news reports to extract flood-related events and convert them into a precise geographical timeline containing more than 2.6 million flood incidents around the world.

Integrating Current Data for Higher Accuracy Predictions

After building the historical database, researchers trained a new prediction model that combines current weather forecasts with Groundsource historical data to estimate the probability of flash floods occurring in specific areas with greater accuracy.

Google currently uses this data to show flood risk in urban areas across 150 countries via the Flood Hub platform, and also provides information to emergency and disaster response agencies to improve the speed of responding to natural disasters and reduce human and material losses.

Technical Limitations of the Tool

Despite its high effectiveness, there are some technical limitations, as the system can identify risks within a range of up to approximately 20 square kilometers only, and its accuracy is lower compared to the flood warning system operated by the US National Weather Service.

Nevertheless, the tool was designed to work efficiently in areas that lack advanced weather monitoring infrastructure, enabling developing countries to better increase their preparedness for natural disasters.

Future Horizons for Climate Disaster Prediction

Google plans to use this technology in the future to predict other complex climate phenomena, such as heat waves, mudslides, and heavy rainfall. Researchers confirm that analyzing millions of news reports helps in inferring precise climate patterns even in data-scarce regions.

This marks the first time Google has used a large language model such as Gemini in the field of weather prediction — having previously developed other AI-based models, most notably DeepMind WeatherNext 2, which proved highly accurate in weather forecasts.