NEWS  AI  November 14, 2023  Reading time: ~1 minute)

Max (RS editor)

Google DeepMind's latest AI model, GraphCast, has emerged as a notable technology in weather forecasting, surpassing traditional methods in both speed and accuracy, especially for the analysis of extreme weather conditions. The AI accurately predicted Hurricane Lee's landfall in Nova Scotia a remarkable three days earlier than conventional models.

Amid a surge in unpredictable extreme weather events exacerbated by climate change, the ability to predict these occurrences swiftly and precisely is crucial for effective disaster preparedness and saving lives. GraphCast, detailed in today's Science publication, outshines the current gold standard, the European Centre for Medium-Range Weather Forecasts, in over 90% of more than 1,300 test areas.

Notably, GraphCast's prowess extends to forecasting conditions up to 10 days in advance, outperforming the ECMWF's model on over 99% of weather variables for Earth's troposphere. This includes crucial factors like rain and air temperature. Crucially, GraphCast provides meteorologists with early warnings of extreme conditions, such as high temperatures and cyclone paths, surpassing traditional models. 

GraphCast's methodology involves leveraging four decades of historical weather data and utilizing graph neural networks, mapping Earth's surface into over a million grid points. This approach enables the model to predict various conditions, including temperature, wind speed, direction, and humidity, quickly and accurately.

The impact of AI on weather forecasting has been profound, with models like GraphCast prompting a reevaluation of AI's role in meteorology. 

In a generous move, Google DeepMind has made GraphCast open source, a step applauded by the scientific community. It's important to emphasizes the importance of large organizations contributing to the scientific community, especially in the face of rising climate change challenges.

(Cover image by Google Deepmind | Article Image by Freepik)