This paper introduces Pangu-Weather, an AI-based method for accurate medium-range global weather forecasting. Using three-dimensional deep networks with Earth-specific priors and a hierarchical temporal aggregation strategy, Pangu-Weather outperforms the European Centre for Medium-Range Weather Forecasts (ECMWF) operational integrated forecasting system in deterministic forecasts across all tested variables. It also excels in extreme weather and ensemble forecasting, demonstrating significantly faster computation times.
