A new artificial intelligence weather forecasting model developed by Google DeepMind, called GraphCast, has demonstrated superior performance compared to the world’s leading traditional numerical weather prediction system. The AI model, described in a recent Science paper, can generate accurate 10-day global weather predictions in under one minute on a single Google TPU v4 machine. GraphCast significantly outperformed the European Centre for Medium-Range Weather Forecasts’ high-resolution forecast (HRES) system on over 90% of 1,380 test variables, including temperature, pressure, wind speed, and humidity at multiple atmospheric levels. The model was trained on nearly four decades of historical weather data from ECMWF’s ERA5 reanalysis dataset, learning to predict future weather conditions based on current and past states. While traditional systems rely on complex physics-based equations solved on supercomputers, GraphCast uses a machine learning approach that is computationally cheaper once trained. Researchers note that AI models like GraphCast should complement rather than replace traditional methods, as they currently lack the ability to simulate novel climate extremes not present in training data. The model’s code has been made publicly available to advance weather prediction science. For the full article, visit https://sciencedaily.com/releases/2023/11/231114143522.htm