具有亚日级和涡旋分辨分辨率的数据驱动全球海洋预报模型
A data-driven global ocean forecasting model with sub-daily and eddy-resolving resolution
AI总结：
本文提出全球数据驱动海洋预报模型 TianHai，以6小时、1/12°涡旋分辨分辨率和1,500米垂直深度实现全数据驱动海气耦合预报，并在1至10天多变量基准中达到最先进性能。
AI中文摘要：
在高空间和时间分辨率下进行高保真海洋预报，对于捕捉精细尺度动力特征至关重要，并对灾害预测、海上航行和可持续海洋管理具有深远意义。尽管传统数值模型能够生成亚日级、涡旋分辨的预报，但它们需要大量计算资源，而且往往难以在如此精细的尺度上保持预测技巧。数据驱动模型以显著更高的计算效率提供了一种有前景的替代方案；然而，大多数模型受限于每日输出，并且在扩展到亚日级时间尺度时表现出准确性的快速衰减。在此，我们介绍 TianHai，这是首个同类全球数据驱动6小时预报模型，可提供1/12°涡旋分辨分辨率的预测，垂直范围深至1,500米。TianHai的一个关键特征是通过数据驱动大气预报系统 FuXi-Atmosphere 整合大气强迫，从而能够显式表征海气耦合效应。与传统方法不同，TianHai不依赖数值大气模型或外部气象预报，使其成为用于耦合预测的完全数据驱动框架。基准实验表明，在1至10天的预报提前期范围内，TianHai在温度和盐度剖面、纬向和经向海流、海表温度以及海平面异常预报方面均达到最先进性能。
英文摘要：
High-fidelity ocean forecasting at high spatial and temporal resolution is essential for capturing fine-scale dynamical features, with profound implications for hazard prediction, maritime navigation, and sustainable ocean management. While conventional numerical models can generate sub-daily, eddy-resolving forecasts, they demand substantial computational resources and often struggle to maintain predictive skill at such fine scales. Data-driven models offer a promising alternative with significantly higher computational efficiency; however, most are constrained to daily outputs and show a rapid decay in accuracy when extended to sub-daily timescales. Here, we introduce TianHai, the first-of-its-kind global data-driven 6-hour forecasting model, which delivers predictions at 1/12° eddy-resolving resolution with a vertical extent down to 1,500 m. A key feature of TianHai is the integration of atmospheric forcings through FuXi-Atmosphere, a data-driven atmospheric forecasting system, which enables the explicit representation of air-sea coupling effects. Unlike conventional approaches, TianHai does not rely on numerical atmospheric models or external meteorological forecasts, making it a fully data-driven framework for coupled prediction. Benchmark experiments demonstrate that TianHai delivers state-of-the-art performance in forecasting temperature and salinity profiles, zonal and meridional currents, sea surface temperature, and sea level anomalies for lead times ranging from 1 to 10 days.
