Eddy-Resolving Global Ocean Forecasting with Multi-Scale Graph Neural Networks

TL;DR

The paper tackles the challenge of eddy-resolving global ocean forecasting with data-driven models by introducing a multi-scale graph neural network that uses two spherical meshes and an encoder-processor-decoder architecture to produce 10-day forecasts at $1/12^\circ$. It incorporates both ocean state variables and surface atmospheric forcing, enabling autoregressive prediction of $X^{t+1}$ from prior states and forecast inputs. Quantitative and qualitative evaluations show improved representation of multi-scale ocean dynamics and short-term skill, evidenced by closer alignment of surface kinetic energy spectra to reanalysis and better retention of fine-scale structures compared with a Swin Transformer baseline. The work highlights the potential of data-driven, eddy-resolving global ocean forecasting while acknowledging limitations in long-lead stability and suggesting future extensions to extend lead times and mesh flexibility.

Abstract

Research on data-driven ocean models has progressed rapidly in recent years; however, the application of these models to global eddy-resolving ocean forecasting remains limited. The accurate representation of ocean dynamics across a wide range of spatial scales remains a major challenge in such applications. This study proposes a multi-scale graph neural network-based ocean model for 10-day global forecasting that improves short-term prediction skill and enhances the representation of multi-scale ocean variability. The model employs an encoder-processor-decoder architecture and uses two spherical meshes with different resolutions to better capture the multi-scale nature of ocean dynamics. In addition, the model incorporates surface atmospheric variables along with ocean state variables as node inputs to improve short-term prediction accuracy by representing atmospheric forcing. Evaluation using surface kinetic energy spectra and case studies shows that the model accurately represents a broad range of spatial scales, while root mean square error comparisons demonstrate improved skill in short-term predictions. These results indicate that the proposed model delivers more accurate short-term forecasts and improved representation of multi-scale ocean dynamics, thereby highlighting its potential to advance data-driven, eddy-resolving global ocean forecasting.

Figures (10)

• Figure 1: Overview of the proposed ocean-forecasting model. (a) Schematic of the model architecture, which takes ocean states ($X^{t-1}, X^t$), atmospheric conditions ($A^{t-1}, A^t, A^{t+1}$), and static features ($S$) as input. These inputs are preprocessed and encoded, followed by independent message passing (MP) on coarse and fine meshes. The encoded features are then decoded to predict the next ocean state ($X^{t+1}$). (b) Mesh structure, where grid nodes ($v^G$) and mesh nodes ($v^M$) are connected via directional edges: grid-to-mesh ($e^{G\rightarrow M}$), mesh-to-grid ($e^{M\rightarrow G}$), and mesh-to-mesh ($e^{M\rightarrow M}$). Grid-node positions align with those of the ocean reanalysis dataset (GLORYS).
• Figure 2: Comparison of RMSE over a 10-day forecast horizon for five surface variables, computed against GLORYS.
• Figure 3: Comparison of RMSE over a 10-day forecast horizon for five surface variables, salinity, temperature, eastward current, northward current, and sea surface height, computed against GLORYS. The evaluation is performed across four regional domains: Gulf Stream (76°W--40°W, 35°N--45°N), Kuroshio Extension (120°E--179°E, 20°N--55°N), South China Sea (100°E--122°E, 0°N--27°N), and Yellow Sea (118°E--127°E, 30°N--42°N). The performance of the proposed model (Ours, blue) is compared with that of Wang2024 (Wang+24, orange). The domain separation follows Metzger2014.
• Figure 4: Comparison of RMSE profiles across depth for four ocean variables at forecast lead times of 3, 6, and 10 days computed against GLORYS. Each row corresponds to a different variable, and each column represents a different lead time.
• Figure 5: Kinetic energy spectra of surface velocity fields, presented as the amplitude of KE versus wavenumber. The left column shows results for the North Pacific region $(10^\circ\text{--}40^\circ\text{N},~145^\circ\text{--}175^\circ\text{E})$, while the right column shows results for the North Atlantic region $(10^\circ\text{--}40^\circ\text{N},~60^\circ\text{--}30^\circ\text{W})$.