Google researchers have developed a new approach to predicting flash floods by combining artificial intelligence with historical news reports.

Flash floods remain one of the most dangerous natural disasters worldwide, claiming thousands of lives each year and often occurring with little warning.

Because flash floods happen suddenly and in localized areas, collecting consistent long-term data about them has been difficult. To address this challenge, Google researchers analyzed approximately five million news articles from around the world.

Using advanced language models, they identified more than 2.6 million reported flood events and transformed them into a structured dataset called “Groundsource.” This dataset allowed scientists to train a forecasting model that evaluates weather conditions and estimates the probability of flash floods in specific locations.

The system relies on a type of neural network capable of analyzing patterns in time-based data, enabling it to interpret weather forecasts and identify areas where flooding risks are increasing. The forecasting technology is now being used within Google’s Flood Hub platform, which provides flood risk information for urban areas across more than 150 countries.

Emergency response agencies can access this information to prepare for potential disasters and improve response times during extreme weather events. Although the model still has limitations and cannot match the precision of advanced radar-based systems used in some developed countries, researchers say it is especially useful for regions lacking expensive weather monitoring infrastructure.

The project highlights how artificial intelligence can transform unconventional data sources, such as news reports, into valuable tools for disaster prediction and public safety.

Source: TechCrunch