Speakers
The Domains
Weather prediction is undergoing a revolution, with data-driven techniques proving highly effective and, in many respects, surpassing the accuracy of physics-based counterparts. These systems are typically trained on reanalysis datasets such as ERA5 and initialised using operational analyses produced by data assimilation systems. Results will be presented from ECMWF’s own data-driven forecasting system, AIFS, which became operational in February 2025 (AIFS-Single) and, as an ensemble (AIFS-CRPS), in July 2025. Ongoing research explores several directions, including the development of a prototype end-to-end machine learning forecasting system trained and initialised directly on meteorological observations.