Advancing subseasonal forecasting with machine learning.

Probabilistic bias correction doubles the subseasonal skill of AI weather forecasts and improves dynamical model accuracy for 91% of pressure, 92% of temperature, and 98% of precipitation targets.

• Why it matters: Accurate subseasonal weather forecasts are crucial for decision-making in agriculture, wildfire management, water and energy allocation, and disaster preparedness, but current models struggle at these timescales due to errors and chaos in the atmosphere.
• What they did: The authors developed a machine learning framework called probabilistic bias correction that learns to reduce systematic errors by correcting historical probabilistic forecasts, and applied it to ECMWF models.
• The result: In the 2025 ECMWF real-time forecasting competition, the corrected models outperformed six operational centers, an international ensemble, ECMWF’s AI system, and 34 other teams, enabling more precise predictions of extreme events.