AI framework improves subseasonal weather forecasts up to six weeks

Researchers have developed a machine learning framework called Probabilistic Bias Correction (PBC) that significantly improves the accuracy of weather forecasts for the two-to-six-week window, a period known as subseasonal forecasting. The framework, detailed in a paper published in Nature on 14 October 2026, corrects systematic errors in existing dynamical and artificial intelligence models, offering more reliable predictions for extreme weather events.

How Probabilistic Bias Correction Works

Subseasonal forecasting is difficult because small errors in initial conditions compound over time, leading to significant inaccuracies. PBC addresses this by learning from a model’s past mistakes. "Our new pipeline (PBC) takes your favorite model as input and learns in what ways it tends to make errors. It then corrects those errors, yielding a better output forecast for the sub-seasonal range," said Lester Mackey, senior principal researcher at Microsoft Research New England.

The framework operates by analyzing historical forecasts and observations to identify systematic biases. It uses two complementary machine learning components: Debias++, which applies location- and date-specific corrections, and Persistence++, which blends forecasts with recent weather trends. These corrections are then combined to produce a probabilistic forecast that accounts for uncertainty.

Performance in Global Competitions

The effectiveness of PBC was demonstrated in the 2025 AI Weather Quest, a real-time competition organized by the European Centre for Medium-Range Weather Forecasts (ECMWF). A team named MicroDuet, which applied PBC to ECMWF’s dynamical model and a hybrid model called POET, achieved the best results for all weather variables and lead times in the inaugural season.

According to the Nature paper, MicroDuet outperformed the debiased dynamical forecasts of six operational forecasting centers (ECMWF, NOAA, ECCC, HMCR, KMA, and JMA), an international multimodel ensemble, and the forecasting systems of 34 teams worldwide. In the September–October–November season, MicroDuet was the only model to significantly outperform the debiased ECMWF model for all variables and lead times.

The framework also showed strong results when applied to ECMWF’s AI Forecasting System (AIFS-SUBS). PBC doubled the mean ranked probability skill score (RPSS) of AIFS-SUBS across all variables and lead times in 2025. When applied to the dynamical ECMWF model, PBC improved skill for 91% of pressure, 92% of temperature, and 98% of precipitation targets compared to operational debiasing protocols.

Real-World Applications in Kenya

Beyond competitive benchmarks, PBC is being deployed in real-world scenarios to aid agricultural planning. Genevieve Flaspohler, who runs Rhiza Research, is working with the Kenya Meteorological Service Authority (KMSA) to test PBC for monsoon forecasting. "A good weather forecast, delivered at the right time, on the right communication channel, can be a big differentiator for agricultural outcomes," Flaspohler said.

Dr. Phillip Okello, a principal meteorologist at KMSA, noted the framework's effectiveness in the region. "Most bias correction just adjusts how much rain a forecast predicts. PBC corrects the probabilities themselves, and across the Greater Horn of Africa it improved our forecasts more than any other method we tested," Okello said. The tool is helping local forecasters identify the likelihood of very heavy or very light rain, which is critical for farmers who rely on rainfall for their crops.

Why it matters

Accurate subseasonal forecasts are essential for managing risks associated with extreme weather, such as wildfires, floods, and heatwaves. By improving the reliability of predictions 2-6 weeks in advance, PBC can help utilities, farmers, and emergency responders make better decisions. Judah Cohen, a climatologist at MIT and co-author of the study, emphasized the societal benefits: "Extreme weather has the largest impact on our society and improved forecasts of these events can yield large societal benefits." The framework is open-source, making it accessible for broader adoption in regions with limited forecasting infrastructure.

---

Sources

• nature.com: Advancing subseasonal forecasting with machine learning | Nature
• microsoft.com: New AI framework bridges a critical gap in mid-range weather forecasts - Microsoft Research
• gizmodo.com: Forget the 10-Day Forecast. AI Just Improved Extreme Weather Warnings Up to 6 Weeks Out

This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.