[论文解读] XiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting

Xiang Wang, Renzhi Wang+17 more|arXiv (Cornell University)|Feb 5, 2024

Oceanographic and Atmospheric Processes被引用 20

一句话总结

XiHe 是首个数据驱动的1/12°全球海洋涡旋分辨预测模型，性能超过领先的 GOFS，运行速度约比传统系统快1000倍。

ABSTRACT

The leading operational Global Ocean Forecasting Systems (GOFSs) use physics-driven numerical forecasting models that solve the partial differential equations with expensive computation. Recently, specifically in atmosphere weather forecasting, data-driven models have demonstrated significant potential for speeding up environmental forecasting by orders of magnitude, but there is still no data-driven GOFS that matches the forecasting accuracy of the numerical GOFSs. In this paper, we propose the first data-driven 1/12° resolution global ocean eddy-resolving forecasting model named XiHe, which is established from the 25-year France Mercator Ocean International's daily GLORYS12 reanalysis data. XiHe is a hierarchical transformer-based framework coupled with two special designs. One is the land-ocean mask mechanism for focusing exclusively on the global ocean circulation. The other is the ocean-specific block for effectively capturing both local ocean information and global teleconnection. Extensive experiments are conducted under satellite observations, in situ observations, and the IV-TT Class 4 evaluation framework of the world's leading operational GOFSs from January 2019 to December 2020. The results demonstrate that XiHe achieves stronger forecast performance in all testing variables than existing leading operational numerical GOFSs including Mercator Ocean Physical SYstem (PSY4), Global Ice Ocean Prediction System (GIOPS), BLUElinK OceanMAPS (BLK), and Forecast Ocean Assimilation Model (FOAM). Particularly, the accuracy of ocean current forecasting of XiHe out to 60 days is even better than that of PSY4 in just 10 days. Additionally, XiHe is able to forecast the large-scale circulation and the mesoscale eddies. Furthermore, it can make a 10-day forecast in only 0.35 seconds, which accelerates the forecast speed by thousands of times compared to the traditional numerical GOFSs.

研究动机与目标

• Develop a data-driven global ocean forecast model at 1/12° resolution capable of resolving mesoscale eddies and large-scale circulation.
• Enable rapid forecasts that dramatically reduce computation time compared to physics-driven GOFSs.
• Learn ocean dynamics by integrating local and global spatial information through specialized transformer blocks.
• Evaluate performance against leading operational GOFSs using IV-TT Class 4 framework and external observations.

提出的方法

• Propose XiHe, a hierarchical transformer-based framework that processes 1/12° global ocean data into token embeddings and uses an ocean-specific transformer stack to capture local and teleconnected-global dynamics.
• Introduce an ocean-land masking mechanism to exclude land regions and reduce computation.
• Design an ocean-specific block with local spatial information extraction (SIE) via window-attention and a global SIE module using group vectors for cross-region teleconnections.
• Use a patch partition/restore pipeline inspired by vision transformers to manage high-resolution input and produce non-auto-regressive multi-step forecasts.
• Train with MSE loss on 5 target variables across 23 depth levels plus SSH, using GLORYS12 reanalysis as input and ERA5 OSTIA SST for SST component, with data from 1993–2020.

实验结果

研究问题

• RQ1 Can a data-driven model achieve superior accuracy to leading numerical GOFS at 1/12° resolution for multi-variable global ocean forecasts?
• RQ2 How do land/ocean separation and dedicated local/global spatial modules affect learning of ocean dynamics and forecast skill?
• RQ3 Is the model capable of forecasting up to 60 days with accuracy surpassing conventional GOFSs, and how fast can forecasts be produced?
• RQ4 Do the global structures learned by the model capture both large-scale circulation and mesoscale eddy activity?

主要发现

• XiHe achieves stronger forecast performance across testing variables than leading operational numerical GOFSs (PSY4, GIOPS, BLK, FOAM).
• Ocean current forecasts up to 60 days are more accurate than PSY4 at 10 days.
• XiHe can produce a 10-day forecast in about 0.36 seconds on a single GPU, enabling thousands offold faster forecasting than traditional GOFSs.
• The model demonstrates capability to forecast large-scale circulation and mesoscale eddies.
• The architecture effectively combines local detail with global teleconnections via the ocean-specific blocks and the ocean-land masking mechanism.