Resolving sources of uncertainty in AI weather forecasting
Published in arXiv, 2025
We introduce Pangu-Bayes, a probabilistic AI weather forecasting framework that explicitly resolves atmospheric-state uncertainty and learned-model uncertainty.
Pangu-Bayes treats these two uncertainty sources as distinct stochastic variables and investigates their source-dependent effects on global probabilistic weather forecasting and tropical cyclone prediction.
* Equal contribution.
Recommended citation: Wenbo Hu*, Xinlei Xiong*, Shuxun Zhou, Kaifeng Bi, Lingxi Xie, Jun Zhu, Richang Hong, and Qi Tian. (2025). "Resolving sources of uncertainty in AI weather forecasting." arXiv preprint arXiv:2511.14218.
Download Paper