Google Research unveils AI-driven global flood forecasting system

Google Research has made a significant breakthrough in global flood forecasting, utilizing artificial intelligence (AI) to expand access to reliable flood forecasts, particularly in regions where data is scarce. Floods, being the most common natural disaster, inflict approximately $50 billion in financial damages annually and affect nearly 1.5 billion people worldwide. The advent of this AI-based system marks a crucial step in improving early warning systems, potentially saving thousands of lives each year.

The research showcases the use of machine learning (ML) technologies to enhance flood forecasting on a global scale. This advancement has extended the reliability of nowcasts from zero to five days on average and has improved forecasts across Africa and Asia to levels comparable to those in Europe. The evaluation of these models was conducted in collaboration with the European Center for Medium Range Weather Forecasting (ECMWF).

Google’s Flood Hub tool, powered by the ML models, provides real-time river forecasts up to seven days in advance, covering over 80 countries. This information is vital for communities, governments, and international organizations to take proactive measures to protect vulnerable populations.

The AI-based flood forecasting initiative began in 2017 and has evolved through years of research and collaboration with academic institutions, governments, and NGOs. The system employs Long Short-Term Memory (LSTM) models, a type of state-space model, to predict river floods using publicly available weather data and physical watershed information. The models are trained on data from diverse watershed streamflow gauges and can make predictions for any river location, even those without streamflow gauges.

In comparison with the current state-of-the-art global flood forecasting system, GloFAS version 4, Google’s ML-based model has shown to provide earlier and more accurate warnings for larger and more impactful events. The system achieves reliability scores at up to 4-day or 5-day lead times that are comparable to or better than the reliability of GloFAS nowcasts.

Ultimately, Google is not only advancing the field of hydrology but also providing a lifeline to those in the path of potential floods. With the ability to predict flood events days in advance, this system offers a crucial window for preparation and evacuation, potentially saving countless lives. As Google continues to refine and expand its flood forecasting capabilities, the future looks brighter for millions who live under the constant threat of natural disasters.