ECMWF AIFS is the Artificial Intelligence Forecasting System from the European Centre for Medium-Range Weather Forecasts. It produces global weather predictions with machine learning, running alongside the centre's physics-based forecasts. The AIFS blog documents ongoing work such as AIFS v2 with snow fields, ocean wave modelling, urban heat island signals and tropical cyclone intensity corrections. The first AIFS model weights have been released openly, and the Anemoi framework, built with national meteorological services, provides shared tools for training and running machine-learned weather models.
Operational forecasters at national meteorological services follow it to see how machine-learned models compare with physics-based systems. Climate and atmospheric scientists use the published verification studies, and machine learning researchers use open weights and the Anemoi framework to train and test their own models. Hydrologists interested in flood forecasting can follow related work such as the AIFL streamflow model. Training webinars called Discover Anemoi introduce the framework to newcomers, and an openly available ERA5 subset is prepared for model training.
Forecasting teams compare AIFS and the physics-based IFS in case studies such as severe storms and wintertime temperature verification, noting strong large-scale performance and remaining difficulty with localized wind extremes. Researchers train downscaling models, for example a flow-matching model built with Anemoi that adds spatial and temporal detail to forecasts. Others reconstruct hourly forecasts between six-hourly steps with HourGlass, correct tropical cyclone intensity, or extend forecasts to snow, waves and river flow. Meteorological services adopt Anemoi as a shared foundation for moving machine-learned models from research into operations.
