Advancing subseasonal forecasting with machine learning.
10 affiliations
- 1Microsoft Research, Cambridge, MA, USA.
- 2Harvard College, Cambridge, MA, USA.
- 3University of Toronto, Toronto, Ontario, Canada.
- 4Instituto de Matemática Pura e Aplicada, Rio de Janeiro, Brazil.
- 5Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, Cambridge, MA, USA. [email protected].
- 6Janus Research Group, Lexington, MA, USA. [email protected].
- 7Microsoft, Redmond, WA, USA.
- 8Rhiza Research, Oakland, CA, USA.
- 9European Centre for Medium-Range Weather Forecasts, Reading, UK.
- 10Microsoft Research, Cambridge, MA, USA. [email protected].
Abstract
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 degradation, we introduce probabilistic bias correction, a machine learning framework that substantially reduces systematic error by learning to correct historical probabilistic forecasts. When applied to the leading dynamical and artificial intelligence models from the European Centre for Medium-Range Weather Forecasts (ECMWF), probabilistic bias correction doubles the modest subseasonal skill of the AI Forecasting System¹⁶ and improves the skill of the operationally debiased dynamical model¹⁷ for 91% of pressure, 92% of temperature and 98% of precipitation targets. We designed probabilistic bias correction for operational deployment, and, in the 2025 real-time forecasting competition of ECMWF¹⁸, its global forecasts were the best for all weather variables and lead times, outperforming the dynamical models from six operational forecasting centres, an international dynamical multimodel ensemble, the AI Forecasting System of ECMWF and the forecasting systems of 34 teams worldwide. These probabilistic skill gains translate into more accurate prediction of extreme events and have the potential to improve agricultural planning, energy management and disaster preparedness in vulnerable communities¹⁹.
Related
- Advancing Near-Term Water Quality Forecasting with Explainable Machine Learning and Probabilistic Uncertainty Quantification
- KNOU-Flood: Advancing global streamflow forecasting with lead-aware probabilistic machine learning
- Advancing Methods to Monitor Greenland Ice-Marginal Lake Area Fluctuations and Outburst Floods Using Machine Learning and Multispectral Optical Imagery
- Skillful Data-Driven Subseasonal Soil Moisture Forecasting: Prospects and Limits for Flash Drought Prediction
- Dynamical-Model-Informed Stochastic Flow Matching for Global Probabilistic Subseasonal Forecasting of Means and Extremes
BibTeX
@article{guan2026advancing,
title = {{Advancing subseasonal forecasting with machine learning.}},
author = {Hannah Guan and Soukayna Mouatadid and Paulo Orenstein and Judah Cohen and Haiyu Dong and Zekun Ni and Jeremy Berman and Genevieve Flaspohler and Alex Lu and Jakob Schloer and Joshua Talib and Jonathan A Weyn and Lester Mackey},
journal = {Nature},
year = {2026},
month = oct,
doi = {10.1038/s41586-026-11126-x},
url = {https://europepmc.org/article/MED/42875498},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Advancing subseasonal forecasting with machine learning.
AU - Hannah Guan
AU - Soukayna Mouatadid
AU - Paulo Orenstein
AU - Judah Cohen
AU - Haiyu Dong
AU - Zekun Ni
AU - Jeremy Berman
AU - Genevieve Flaspohler
AU - Alex Lu
AU - Jakob Schloer
AU - Joshua Talib
AU - Jonathan A Weyn
AU - Lester Mackey
PY - 2026
DA - 2026/10/14
JO - Nature
DO - 10.1038/s41586-026-11126-x
UR - https://europepmc.org/article/MED/42875498
AB - 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 degradation, we introduce probabilistic bias correction, a machine learning framework that substantially reduces systematic error by learning to correct historical probabilistic forecasts. When applied to the leading dynamical and artificial intelligence models from the European Centre for Medium-Range Weather Forecasts (ECMWF), probabilistic bias correction doubles the modest subseasonal skill of the AI Forecasting System¹⁶ and improves the skill of the operationally debiased dynamical model¹⁷ for 91% of pressure, 92% of temperature and 98% of precipitation targets. We designed probabilistic bias correction for operational deployment, and, in the 2025 real-time forecasting competition of ECMWF¹⁸, its global forecasts were the best for all weather variables and lead times, outperforming the dynamical models from six operational forecasting centres, an international dynamical multimodel ensemble, the AI Forecasting System of ECMWF and the forecasting systems of 34 teams worldwide. These probabilistic skill gains translate into more accurate prediction of extreme events and have the potential to improve agricultural planning, energy management and disaster preparedness in vulnerable communities¹⁹.
SN - 0028-0836
SN - 1476-4687
ER -