AxiomOcean: Forecasting the Three-Dimensional Structure of the Upper Ocean
AxiomOcean: Forecasting the Three-Dimensional Structure of the Upper Ocean
Authors: Sensen Wu, Yifan Chen, Guantao Pu, Xiaoyao Sun, Yijun Chen, Jin Qi, Ming Kong, Keyi Yang, Lichen Xu, Wenguan Wang, Xiaofeng Li, Zhenhong Du Date: 2026-05-11 Paper ID: openalex:2605.10455
Summary
AxiomOcean is a global AI forecasting model designed to address the limitations of existing AI approaches in preserving three-dimensional subsurface ocean structures. By utilizing a 3D encoder-backbone-decoder architecture, the model incorporates vertical hierarchy and cross-layer dependencies to predict temperature, salinity, and currents down to 643 meters. It demonstrates superior accuracy and physical realism compared to state-of-the-art AI benchmarks across key oceanic regions like the equatorial Pacific and the Southern Ocean.
Key Contributions
- Introduces AxiomOcean, a 3D AI-based forecasting model that explicitly models vertical hierarchies and cross-layer dependencies in the ocean column.
- Reduces day-1 RMSE by 20–35% compared to baseline AI models in 10-day global ocean forecasts at 1/12° resolution.
- Maintains physical fidelity (e.g., eddy kinetic energy, temperature/salinity variance) through improved architectural design, avoiding over-smoothing common in standard AI forecasting methods.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2605.10455
- paper_id
- 2605.10455
- paper_source
- openalex
- domain
- time-series
- tags
- time-series-forecastingphysical-consistencymultivariate-forecastingspatiotemporal-modeling
- architectures
-
- datasets
-
- concept_slugs
-
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-05-14T06:09:42Z