Google's Research division has dramatically expanded its flood forecasting capabilities, reaching 700 million people across 100 countries, up from 460 million in 80 countries. The enhanced system features improved accuracy, providing seven-day warnings with the same precision as previous five-day forecasts.
"Our team at Google Research has pioneered scalable global flood forecasting models that bring near real-time data to aid organisations and communities around the world," said Yossi Matias, VP Engineering & Research and Crisis Response Lead at Google.
The expansion includes several key developments:
- Launch of a new API pilot programme for researchers and partners
- Introduction of "virtual gauges" for regions lacking physical sensors
- Release of historical flood forecasting data dating back to 1981
- Addition of 250,000 forecast points across 150 countries in Google's Flood Hub
The system has already proved vital in recent disasters. During floods in Rio Grande do Sul, Brazil, Google collaborated with the Geological Service of Brazil to monitor over 200 new locations through Flood Hub. This data helped public authorities and local residents develop crisis response plans. World Vision Brazil and its regional partners used the flood forecasts to predict affected regions and distribute essential supplies within two days.
A combination of the following achieved the breakthrough. Tripling training locations, rebuilding the model architecture from the ground up, and incorporating advanced weather forecasting data alongside more extensively labelled datasets.