A new study published in Science demonstrates that artificial intelligence models can produce medium-range weather forecasts with accuracy comparable to traditional physics-based systems, but in a fraction of the time. The research, led by scientists at the University of California, Berkeley, tested an AI model trained on decades of global weather data. It successfully predicted atmospheric conditions up to ten days in advance, matching the performance of established numerical weather prediction (NWP) models that require immense computational power. The AI system completed its forecasts in under a minute on a single graphics processing unit, whereas conventional supercomputer-based models can take hours. Researchers caution that while promising for routine predictions, AI models may still struggle with extreme, unprecedented weather events where historical data is limited. Further development and real-world testing are needed before such systems could be operationally deployed by meteorological agencies. Read the full article: https://sciencedaily.com/releases/2024/05/240508113102.htm