Google’s AI Model Predicts Flash Floods Using News Data
- Google created a dataset of 2.6 million historical flash floods using Gemini AI, covering over two decades of global news reports.
- The dataset powers an AI model that forecasts urban flash floods up to 24 hours in advance, now available on Google’s Flood Hub platform.
- The model uses LSTM neural networks to process weather forecasts and local factors like urbanization density and soil absorption rates.
- The system covers areas of about 20 square kilometers, signaling medium or high flood risk for urban areas with population densities above 100 people per square kilometer.
- A regional disaster authority in Southern Africa utilized a Flood Hub alert to confirm a flood and manage the response effectively.
Google’s new system addresses the lack of historical data for flash flood prediction by transforming news articles into actionable datasets, enhancing crisis preparedness and resilience globally.
