Google Develops New Flash Flood Prediction Method Using Old News and Large Model Gemini

Flash floods are among the world's deadliest meteorological disasters, but their suddenness and small scale make accurate prediction difficult. Traditional data shortages have limited AI model training. Google's team used the large model Gemini to extract 2.6 million flood events from 5 million news articles, building the Groundsource dataset with spatiotemporal markers to fill the data gap.

The LSTM model trained on this dataset now provides risk alerts for 150 countries via the Flood Hub platform, helping emergency agencies respond quickly. However, the system has limitations like low spatial resolution (20 square kilometers) and lack of real-time radar data. Future applications may extend to heatwaves and mudslides, aiding disaster response in data-scarce regions.