Google DeepMind and Google Research have introduced Weather Lab, an innovative interactive platform designed to enhance tropical cyclone prediction using advanced artificial intelligence. The centerpiece is an experimental AI model, built on stochastic neural networks, capable of forecasting a cyclone’s formation, track, intensity, size, and shape by generating 50 potential scenarios up to 15 days ahead. Preliminary internal evaluations suggest that this AI model matches or even surpasses the accuracy of current physics-based methods for both cyclone track and intensity. For instance, its 5-day track predictions average 140 km closer to actual cyclone locations than the ECMWF’s ENS model, a significant 1.5-day improvement. Furthermore, it demonstrated superior intensity prediction compared to NOAA’s HAFS system. Google DeepMind is actively collaborating with the U.S. National Hurricane Center (NHC), providing their expert forecasters with live AI predictions to augment current forecasting capabilities and support more timely disaster preparedness and evacuations. While Weather Lab serves as a research tool and not for official warnings, it aims to provide invaluable data to weather agencies and emergency services, with ongoing partnerships extending to organizations like CIRA and the UK Met Office.
https://deepmind.google/discover/blog/weather-lab-cyclone-predictions-with-ai/