Generative assimilation and prediction for weather and climate
14 affiliations
- 1Complexity Science, Potsdam Institute for Climate Impact Research, Potsdam, Germany
- 2Earth System Modelling, School of Engineering and Design, Technical University of Munich, Munich, Germany
- 3Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
- 4School of Atmospheric Sciences and Key Laboratory of Mesoscale Severe Weather/Ministry of Education, Nanjing University, Nanjing, China
- 5Chinese Academy of Meteorological Sciences, Beijing, China
- 6College of Meteorology and Oceanography, National University of Defense Technology, Changsha, China
- 7Academy of Advanced Interdisciplinary Studies, Wuhan University, Wuhan, China
- 8Hubei Center for Applied Mathematics, Wuhan University, Wuhan, China
- 9School of Atmospheric Sciences, Sun Yat-Sen University, Zhuhai, China
- 10Department of Atmospheric and Oceanic Sciences, School of Physics, Peking University, Beijing, China
- 11Institute of Ocean Research, School of Physics, Peking University, Beijing, China
- 12Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
- 13China Meteorological Administration Key Laboratory for Climate Prediction Studies, National Climate Center, Beijing, China
- 14Xiong’an Institute of Meteorological Artificial Intelligence, Xiong’an, China
Abstract
Machine learning models have shown great success in predicting weather up to two weeks ahead, outperforming process-based benchmarks 1 , 2 , 3 , 4 , 5 . However, most existing approaches do not fully integrate data assimilation into the prediction workflow 6 , 7 , and often suffer from error accumulation in autoregressive computation, limiting their applicability to predictions beyond the medium range 8 and climate projections 9 . We introduce Generative Assimilation and Prediction (GAP), a unified deep generative framework that connects data assimilation, weather forecasting, and climate simulation through a shared probabilistic representation of atmospheric states. GAP learns the climatological distribution from reanalysis data, and adapts it to different tasks by incorporating observational, predictive, and forcing constraints. Evaluated across a range of weather-climate tasks, GAP produces probabilistic weather forecasts competitive with leading forecasting systems, using its own assimilated initial states derived from 5% synthetic perfect observations, with particular skill in capturing extreme events such as atmospheric rivers, heatwaves and the Madden-Julian Oscillation. At seasonal timescales, GAP shows promising predictive skill when constrained by oceanic boundary forcing. Moreover, GAP supports stable millennial-scale climate simulations that reproduce atmospheric variability from daily weather fluctuations to decadal oscillations. These results suggest that generative modeling of the climatological distribution offers a viable foundation for seamless weather and climate prediction, thereby opening a path toward integrated, observation-to-climate workflows.
Related
- Skillful Seasonal Tropical Cyclone Prediction with AI Weather--Climate Models Forced by Dynamically Predicted SSTs
- Accounting for model error in ensemble variational data assimilation through learned generative dynamics
- Xaurora: Generative Weather Forecasting with Denoising Stochastic Interpolants from a Foundation Model Prior
- Climate-informed cryospheric reanalysis via hierarchical Bayesian data assimilation
- Weather forecasting in a changing climate: the rise of AI and Machine learning?
BibTeX
@article{yang2026generative,
title = {{Generative assimilation and prediction for weather and climate}},
author = {Shangshang Yang and Congyi Nai and Xinyan Liu and Weidong Li and Jie Chao and Jingnan Wang and Leyi Wang and Xichen Li and Xichen Li and Xi Chen and Bo Lu and Xin Li and Ziniu Xiao and Niklas Boers and Huiling Yuan and Baoxiang Pan},
journal = {npj Climate and Atmospheric Science},
year = {2026},
month = nov,
doi = {10.1038/s41612-026-01585-4},
url = {https://doi.org/10.1038/s41612-026-01585-4},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Generative assimilation and prediction for weather and climate
AU - Shangshang Yang
AU - Congyi Nai
AU - Xinyan Liu
AU - Weidong Li
AU - Jie Chao
AU - Jingnan Wang
AU - Leyi Wang
AU - Xichen Li
AU - Xichen Li
AU - Xi Chen
AU - Bo Lu
AU - Xin Li
AU - Ziniu Xiao
AU - Niklas Boers
AU - Huiling Yuan
AU - Baoxiang Pan
PY - 2026
DA - 2026/11/02
JO - npj Climate and Atmospheric Science
DO - 10.1038/s41612-026-01585-4
UR - https://doi.org/10.1038/s41612-026-01585-4
AB - Machine learning models have shown great success in predicting weather up to two weeks ahead, outperforming process-based benchmarks 1 , 2 , 3 , 4 , 5 . However, most existing approaches do not fully integrate data assimilation into the prediction workflow 6 , 7 , and often suffer from error accumulation in autoregressive computation, limiting their applicability to predictions beyond the medium range 8 and climate projections 9 . We introduce Generative Assimilation and Prediction (GAP), a unified deep generative framework that connects data assimilation, weather forecasting, and climate simulation through a shared probabilistic representation of atmospheric states. GAP learns the climatological distribution from reanalysis data, and adapts it to different tasks by incorporating observational, predictive, and forcing constraints. Evaluated across a range of weather-climate tasks, GAP produces probabilistic weather forecasts competitive with leading forecasting systems, using its own assimilated initial states derived from 5% synthetic perfect observations, with particular skill in capturing extreme events such as atmospheric rivers, heatwaves and the Madden-Julian Oscillation. At seasonal timescales, GAP shows promising predictive skill when constrained by oceanic boundary forcing. Moreover, GAP supports stable millennial-scale climate simulations that reproduce atmospheric variability from daily weather fluctuations to decadal oscillations. These results suggest that generative modeling of the climatological distribution offers a viable foundation for seamless weather and climate prediction, thereby opening a path toward integrated, observation-to-climate workflows.
SN - 2397-3722
ER -