Google's AI Predicts Flash Floods in 150 Countries — by Reading Historical News Reports
Google has expanded its Flood Hub platform to provide flash flood risk forecasting for urban areas across 150 countries, and the key innovation is counterintuitively low-tech: large language models that convert decades of qualitative newspaper reports and emergency records into quantitative training data. The approach was designed specifically for regions that lack expensive weather sensors or extensive meteorological archives — covering an estimated 5,000 deaths annually from flash flooding that previous models could not predict. The system currently covers 20-square-kilometer resolution areas and shares live risk alerts directly with emergency response agencies. Southern African Development Community officials who piloted the system reported faster flood response times in initial field trials.
Sofia Reyes4 min read