A groundbreaking development in weather forecasting has emerged as Google DeepMind's team of scientists introduced GraphCast, an artificial intelligence (AI) system that can produce highly accurate 10-day weather forecasts in just 60 seconds. This cutting-edge technology represents a significant leap forward in the field of meteorology, offering faster and more precise predictions.
Traditional weather forecasting relies on Numerical Weather Prediction (NWP), a method that involves complex physics equations translated into computer algorithms running on supercomputers. While effective, this approach is time-consuming and resource-intensive. GraphCast, on the other hand, utilizes deep learning and historical weather data to create forecasts, bypassing the need for intricate equations.
To develop GraphCast, researchers trained the AI model on four decades of historical weather data, sourced from the European Centre for Medium-Range Weather Forecasts (ECMWF). This data encompasses observations from satellites, radar, weather stations, and traditional NWP models. The AI model then learns the cause-and-effect relationships governing Earth's weather patterns.
Remarkably, despite its computational intensity during training, GraphCast produces 10-day forecasts in under a minute on a single Google TPU v4 machine. In contrast, conventional approaches can take hours on supercomputers.
The superiority of GraphCast's forecasts is highlighted by its outperformance in more than 90% of test variables and forecast lead times when compared to the industry-standard High Resolution Forecast (HRES) system. Notably, GraphCast excels in predicting severe weather events, such as cyclone tracks, atmospheric rivers, and extreme temperatures.
The AI model's ability to provide earlier warnings and more accurate forecasts offers significant benefits, potentially saving lives and minimizing the impact of extreme weather events on communities.