AIFS

Numerical weather prediction (NWP) plays a critical role for European societies to avoid weather-induced harm and improve benefits, e.g. in the agricultural sector or for renewable energies. Very recently, AI-based NWP prototype models by companies such as NVIDIA, Huawei and Google have demonstrated deterministic NWP skill exceeding those of the best conventional global models, such as the Integrated Forecasting System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF).

These developments challenge the leading role of Europe in NWP.  This motivated ECMWF to develop its own AI-based NWP model, which is still under development but already attains skill comparable to the best models in the literature.

With the requested compute time, the project wants to ensure that Europe has a leading AI-based NWP model. In particular, the team want to extend the AIFS to 9 km resolution and develop an ensemble system, two aspects that have so far not been demonstrated in the literature but that are important for operational use. ECMWF is well positioned for this research through its expertise in conventional modelling, its extensive data archive, and its know-how on the evaluation of NWP forecasts.

The requested compute time will efficiently use EuroHPC resources to  strengthen Europe's expertise on large scale machine learning and ensure public domain models and expertise for AI-based numerical weather prediction.