ECMWF's AIFS — Europe's Weather Agency Puts an AI Model Into Daily Operational Forecasting

The European Centre for Medium-Range Weather Forecasts (ECMWF), the intergovernmental body that runs the world's leading medium-range weather forecasting system for 35 member and cooperating states, put its Artificial Intelligence Forecasting System (AIFS) into full operational service in February 2025, with a probabilistic ensemble version following in July 2025.

AIFS aims to produce global weather forecasts of comparable or better quality than ECMWF's physics-based Integrated Forecasting System (IFS) while cutting the computing power required, run alongside (not instead of) the physics-based system.

AIFS is a deep-learning model trained on ECMWF's own decades of reanalysis data. Once trained, it produces a 10-day global forecast in a fraction of the time and computing power required by the physics-based IFS.

On independent verification scores, AIFS matches or outperforms physics-based forecasts on many measures, including gains of up to 20% on tropical cyclone track accuracy - a result confirmed in ECMWF's own published verification and covered independently by Communications of the ACM. NOAA's comparable AI ensemble model, AIGEFS, needs only about 9% of the compute of its physics-based predecessor. NOAA's own operational AI suite followed in January 2026, an early data point on cross-agency replication of the approach.

ECMWF still runs AIFS alongside, not instead of, the physics-based IFS, and forecasters caution that AI models trained on historical weather patterns are less proven for genuinely unprecedented extreme events.