An end-to-end machine-learning system for global eddy-resolving ocean forecast
Machine learning (ML) models, especially those based on deep neural networks, have achieved competitive forecast accuracy and greatly reduced computational burden compared to traditional numerical global ocean forecast systems (GOFSs). However, it remains difficult for these ML models to operate on their own, as they rely on the initial conditions generated by numerical GOFSs. In this study, we present WenHai-E2E, an end-to-end ML GOFS that replaces the entire pipeline of numerical GOFSs at deployment time and makes eddy-resolving global ocean forecast directly from multisource heterogeneous ocean observations, without requiring operational initial conditions from numerical GOFSs. WenHai-E2E outperforms a state-of-the-art numerical GOFS in the forecast accuracy. Our results demonstrate the promise of ML GOFSs to surpass numerical GOFSs and provide insight into the pathway for developing next generation of GOFSs.
Authors
- Zhao Jing (ORCID: https://orcid.org/0000-0002-8430-9149)
- Guangnan Wu (ORCID: https://orcid.org/0000-0002-0841-5791)
- Lixin Wu (ORCID: https://orcid.org/0000-0002-4694-5531)
- Yingzhe Cui (ORCID: https://orcid.org/0000-0003-1595-0298)
- Xiang Zhang (ORCID: https://orcid.org/0009-0007-4541-0475)
- Furong Li
- Shenghui Zhou (ORCID: https://orcid.org/0009-0001-1962-2585)
Institutions
- Shandong University (CN)
- Qingdao Academy of Intelligent Industries (CN)
- Haier Group (China) (CN)
- Laoshan Laboratory (CN)
- Ocean University of China (CN)
Publication Details
- Journal
- Science Advances
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1126/sciadv.aeh9058
- Primary Topic
- Oceanographic and Atmospheric Processes
- Type
- article
- Field-Weighted Citation Impact
- 0.00