Accurate medium-range weather prediction is of critical importance for European societies to prevent weather-induced loss of life and economic damages. In the past two years, numerical weather prediction has been revolutionised by machine learning-based models that achieve deterministic skill exceeding those of the best conventional models at a fraction of the compute and energy costs. However, probabilistic predictions are needed to reliably quantify the risks associated with the range of possible future weather states that can only be predicted.
The Artificial Intelligence/Integrated Forecasting System (AIFS) currently under development at the European Centre for Medium-Range Weather Forecasts (ECMWF) therefore includes a probabilistic ensemble forecasting system. To obtain a high-resolution and high-fidelity ensemble model that is ready for operational use, large scale training runs are needed. The substantial compute requirements for these is available on pre-exascale EuroHPC machines.
Training such a state-of-the-art ensemble model will ensure public domain and openly accessible global probabilistic numerical weather prediction and the European leadership in the domain. It will also inform probabilistic regional predictions by national meteorological services as well as the design of longer-term projection systems for subseasonal-to-seasonal and climate time scales in the near future.
Christian Lessig, European Centre for Medium-Range Weather Forecasts, United Kingdom