Google Research has developed an innovative AI-powered flood forecasting solution called Flood Hub that provides real-time river forecasts in more than 80 countries, with an average of five days in advance. The team at Google Research has published an article in the journal Nature, discussing the importance of this technology in improving global access to flood forecasts, especially in developing countries that lack dense flow measurement networks.
Floods are one of the most common natural disasters with a disproportionate impact on developing countries. Accurate and timely flood alerts are crucial in mitigating risks, and AI technology like Flood Hub can improve predictions of extreme river events and provide early flood warnings.
The machine learning models powering Flood Hub have been developed over years of research, and the AI model is based on long-term neural networks. By ingesting historical and forecasted weather data, the model can learn different hydrological behaviors and predict floods in various types of watersheds.
Google Research conducted a pilot early warning system in the Ganges-Brahmaputra river basin in India, which later expanded to include more regions and refine the flood forecasting model. By training a single model with all available river data, the AI can make predictions for any location, regardless of available data. This makes it possible to significantly improve flood predictions in the short term.
Nature has advocated for sharing forecasts in real time and publicly without costs or access barriers to help mitigate the impact of floods and other natural disasters across the globe. By doing so, it becomes easier for people to prepare for potential disasters and take necessary precautions to minimize damage caused by these events.
