Truace
Subscribe

Advancing subseasonal forecasting with machine learning

Decision-makers rely on weather forecasts to plant crops, manage wildfires, allocate water and energy and prepare for weather extremes1-3. Today, such forecasts enjoy unprecedented accuracy as far as 2 weeks thanks to steady advances in physics-based dynamical models and data-driven artificial intelligence models4-10. However, model skill drops precipitously at subseasonal timescales (2-6 weeks ahead) due to compounding errors, systemic model biases and the chaotic nature of the atmosphere11-15. To counter this…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1465· cite & verify
Advancing subseasonal forecasting with machine learning
Polar-Orbiting Satellites are Key to Accurate Hurricane Forecasts (NESDIS 2012-12-10) by NOAA. Public domain

In brief

Researchers present probabilistic bias correction, a machine learning framework trained on historical probabilistic forecasts to remove systematic errors. Applied to ECMWF's AI Forecasting System, it doubled that model's modest skill at 2-6 weeks ahead. It also improved the operationally debiased dynamical model for most pressure, temperature and precipitation targets.

Forecast skill falls sharply beyond about two weeks, which is the range that matters for crop planting, water and energy allocation and extreme-event preparation. A method built for operational deployment that wins a real-time competition suggests practical value. The agricultural, energy and disaster-preparedness benefits are described as potential rather than measured outcomes.

Main points

  1. The method is called probabilistic bias correction and learns from historical probabilistic forecasts to reduce systematic error.
  2. It improved the operationally debiased dynamical model for 91% of pressure, 92% of temperature and 98% of precipitation targets.
  3. In ECMWF's 2025 real-time competition its global forecasts were best for all weather variables and lead times.
  4. Competitors it outperformed included dynamical models from six operational centres, a multimodel ensemble and 34 teams worldwide.
  5. Skill gains translate into more accurate prediction of extreme events.

The gain

A machine learning bias-correction layer doubled the subseasonal skill of ECMWF's AI model and beat leading operational and team-built systems in a 2025 real-time competition.

The rundown

Weather forecasts are already highly accurate up to about two weeks, thanks to physics-based dynamical models and data-driven AI models. At subseasonal timescales of 2-6 weeks, skill drops because of compounding errors, systemic model biases and the chaotic atmosphere.

The framework was applied to ECMWF's leading dynamical model and its AI Forecasting System. For the AI model, skill at these lead times was described as modest before correction and roughly doubled after it.

Beyond retrospective tests, the approach was entered in ECMWF's 2025 real-time forecasting competition, where it outperformed every competing system on all variables and lead times. The authors say the skill gains improve extreme-event prediction and could aid agricultural planning, energy management and disaster preparedness in vulnerable communities.

Sources

  1. Peer-reviewedNature2026-10-14

The debate