HybridOM: Hybrid Physics-Based and Data-Driven Global Ocean Modeling with Efficient Spatial Downscaling

TL;DR

HybridOM introduces a differentiable physics-based skeleton coupled with a neural flesh to achieve stable, accurate global ocean modeling while enabling efficient high-resolution regional downscaling via differentiable Flux Gating. The method unifies a Quasi-Geostrophic-augmented dynamical core with a multi-scale neural module (DSOA) and couples regional downscaling with large-scale boundaries through Neural Information Injection. Across hindcast, forecast with FuXi-2.0, and regional downscaling tasks on GLORYS12V1 and OceanBench data, HybridOM achieves state-of-the-art accuracy and maintains physical consistency, demonstrating robustness for ocean digital twins and operational forecasting. The work also provides end-to-end differentiable training and scalable inference with modest overhead, supporting kilometer-scale extensions and fully coupled ocean-atmosphere extensions in future work.

Abstract

Global ocean modeling is vital for climate science but struggles to balance computational efficiency with accuracy. Traditional numerical solvers are accurate but computationally expensive, while pure deep learning approaches, though fast, often lack physical consistency and long-term stability. To address this, we introduce HybridOM, a framework integrating a lightweight, differentiable numerical solver as a skeleton to enforce physical laws, with a neural network as the flesh to correct subgrid-scale dynamics. To enable efficient high-resolution modeling, we further introduce a physics-informed regional downscaling mechanism based on flux gating. This design achieves the inference efficiency of AI-based methods while preserving the accuracy and robustness of physical models. Extensive experiments on the GLORYS12V1 and OceanBench dataset validate HybridOM's performance in two distinct regimes: long-term subseasonal-to-seasonal simulation and short-term operational forecasting coupled with the FuXi-2.0 weather model. Results demonstrate that HybridOM achieves state-of-the-art accuracy while strictly maintaining physical consistency, offering a robust solution for next-generation ocean digital twins. Our source code is available at https://github.com/ChiyodaMomo01/HybridOM.

Figures (11)

• Figure 1: Overview of the HybridOM Architecture. (a) Global Hybrid Framework. The model couples a differentiable physical skeleton ($\mathcal{M}_{\text{phy}}$) for dynamical stability with a neural corrector ($\mathcal{N}_{\theta}$) to capture sub-grid physics. (b) Spatial Downscaling. This module employs a Flux Gating mechanism to inject coarse-scale thermodynamic constraints into high-resolution regional simulations.
• Figure 2: Global Simulation Error Analysis. Comparison of HybridOM against baselines. From left to right, the panels illustrate the errors in Zonal and Meridional Geostrophic Velocities, followed by the deviation in Upper Ocean Heat Content (OHC).
• Figure 3: Stability and Ablation Analysis of Spatial Downscaling. Time evolution of Latitude-Weighted RMSE for regional simulation in the North Pacific based on different models.
• Figure 4: HybridOM Operational Workflow. The diagram illustrates the coupling between FuXi-2.0 and HybridOM for global forecasting.
• Figure 5: Schematic Overview of the HybridOM Neural Architecture of $\mathcal{N}_{\theta}, \mathcal{N}_{\text{inv}}$.