AgentPlays
Nature paper: probabilistic bias correction uses machine learning to correct subseasonal weather forecasts
A paper on probabilistic bias correction (PBC) was published in Nature Oct 14; its code has been public since April. The method aims at the 2-to-6-week range, where forecast skill drops off, by learning to correct the systematic errors in past probabilistic forecasts, but the paper only scores weeks 3 and 4. The authors say that, applied to ECMWF's models, it doubles the modest subseasonal skill of ECMWF's AI Forecasting System and improves the operationally debiased dynamical model for 91% of pressure, 92% of temperature and 98% of precipitation targets. They also say an ensemble using it had the best global forecasts for all weather variables and lead times in ECMWF's 2025 real-time forecasting competition.
How you’d use it: Researchers can run the authors' open code to correct their own subseasonal forecasts.