A new artificial intelligence weather forecasting model developed by Google DeepMind has demonstrated superior accuracy compared to conventional physics-based systems in global medium-range weather predictions. The GraphCast model, detailed in a Science journal publication, uses machine learning trained on decades of historical weather data to predict hundreds of weather variables up to 10 days in advance. In head-to-head comparisons with the European Centre for Medium-Range Weather Forecasts’ high-resolution forecasting system (HRES), GraphCast provided more accurate predictions for over 90% of 1,380 test variables. The AI system significantly outperformed HRES in tracking severe weather events, including tropical cyclones, atmospheric rivers, and extreme temperature fluctuations. Researchers note that while AI models like GraphCast offer remarkable speed and efficiency advantages—generating 10-day forecasts in under one minute on a single Google TPU—they are designed to complement rather than replace traditional numerical weather prediction systems. The model’s architecture processes current and six-hour-old global weather states to generate predictions, learning complex patterns from 39 years of reanalysis data. This development represents a major advancement in computational meteorology, potentially enabling more timely and accurate warnings for extreme weather events worldwide. Read the full article at https://sciencedaily.com/releases/2023/11/231114143522.htm