This paper introduces FuXi, a cascaded machine learning (ML) weather forecasting system capable of producing 15-day global forecasts at a 6-hour temporal and 0.25° spatial resolution. Trained on 39 years of ECMWF ERA5 reanalysis data, FuXi's performance is comparable to the ECMWF ensemble mean (EM) for 15-day forecasts, surpassing the ECMWF HRES in skillful forecast lead time. FuXi's ensemble, generated by perturbing initial conditions and model parameters, provides forecast uncertainty estimates.
Publisher
npj Climate and Atmospheric Science
Published On
Jul 26, 2023
Authors
Lei Chen, Xiaohui Zhong, Feng Zhang, Yuan Cheng, Yinghui Xu, Yuan Qi, Hao Li
Tags
weather forecasting
machine learning
ensemble forecasting
ECMWF
forecast uncertainty
global predictions
temporal resolution
Related Publications
Explore these studies to deepen your understanding of the subject.
