AIFS – ECMWF’s Data-Driven Forecasting System

Data-driven machine learning approaches have rapidly emerged as candidates for operational weather forecasting systems, delivering accurate forecasts for low computational cost, leveraging high quality reanalysis products. At the European Centre for Medium-Range Weather Forecasts, we are developing the AIFS (Artificial Intelligence Forecasting System) as a complement to our physics-based system, the IFS. Here we will provide an update on the progress towards an operational ensemble forecasting system, embracing generative ML including diffusion modelling. We will examine how ensemble AIFS systems perform for extreme events, evaluated against in-situ observations, across a range of variables. We will showcase progress towards Earth-system modelling, with representation of ocean, sea-ice, wave and land processes coupled to atmospheric dynamics.