NeuralOGCM: Differentiable Ocean Modeling with Learnable Physics
Advancing Climate Modeling with Hybrid AI-Physics Integration for Unprecedented Accuracy and Speed
Hao Wu et al. | December 15, 2025
High-precision scientific simulation faces a long-standing trade-off between computational efficiency and physical fidelity. To address this challenge, we propose NeuralOGCM, an ocean modeling framework that fuses differentiable programming with deep learning. At the core of NeuralOGCM is a fully differentiable dynamical solver, which leverages physics knowledge as its core inductive bias. The learnable physics integration captures large-scale, deterministic physical evolution, and transforms key physical parameters (e.g., diffusion coefficients) into learnable parameters, enabling the model to autonomously optimize its physical core via end-to-end training. Concurrently, a deep neural network learns to correct for subgrid-scale processes and discretization errors not captured by the physics model. Both components work in synergy, with their outputs integrated by a unified ODE solver. Experiments demonstrate that NeuralOGCM maintains long-term stability and physical consistency, significantly outperforming traditional numerical models in speed and pure AI baselines in accuracy. Our work paves a new path for building fast, stable, and physically-plausible models for scientific computing. Our codes are available here.
Executive Impact: Bridging the Fidelity-Efficiency Gap
This research introduces a paradigm shift in ocean modeling, addressing critical limitations in climate science and offering a path to more reliable and efficient predictive capabilities for enterprise applications.
The Challenge: High-Stakes Scientific Simulation Dilemma
High-precision scientific simulation, crucial for climate science, faces a long-standing trade-off between computational efficiency and physical fidelity. Traditional General Circulation Models (GCMs) are physically faithful but prohibitively slow, while pure AI models are fast but prone to long-term instability and physical inconsistencies.
Our Solution: NeuralOGCM's Hybrid AI-Physics Approach
NeuralOGCM introduces a pioneering hybrid approach, integrating a fully differentiable physics core with a neural network corrector. This framework leverages physics knowledge as an inductive bias, allowing for end-to-end optimization of physical parameters and data-driven correction, bridging the gap between speed and fidelity.
Key Takeaways for Enterprise AI Strategy
- • Learnable Physics Paradigm: NeuralOGCM introduces a new modeling approach where key physical parameters are autonomously optimized from data via gradient descent, enhancing physical plausibility and accuracy.
- • First End-to-End Differentiable OGCM: This work presents the first fully differentiable hybrid ocean general circulation model, successfully applying this paradigm to complex real-world scientific problems with end-to-end training capabilities.
- • Superior Performance & Stability: NeuralOGCM demonstrates significant advantages in speed (orders of magnitude faster), accuracy, and long-term autoregressive stability, outperforming both traditional numerical models and pure AI baselines.
0
Speedup vs. Traditional GCMs
0
120-Day RMSE (Lower is Better)
High
Long-Term Physical Stability
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
NeuralOGCM Hybrid Architecture Flow
NeuralOGCM combines a physics-informed core with a neural network for comprehensive ocean modeling. This flowchart illustrates the synergistic workflow, from input to final prediction, ensuring both fidelity and efficiency.
Current Ocean State (yt, Ft)
→
Differentiable Physics Core (Gphys)
→
Neural Corrector (Gneural)
→
Unified ODE Solver
→
Predicted Future State (yt+Δt)
| Model | 10 Days (RMSE) | 120 Days (RMSE) | Key Advantage |
|---|---|---|---|
| NeuralOGCM | 0.919 | 1.574 |
|
| U-Net | 0.923 | 1.886 |
|
| FourCastNet | 0.959 | 3.332 |
|
| Traditional GCMs | N/A | N/A (Months of compute) |
|
Learnable
Physics Parameters (e.g., Diffusion Coefficients)
A core innovation of NeuralOGCM is its fully differentiable physics core. This enables key physical parameters, such as diffusion coefficients, to be autonomously optimized end-to-end from data, reducing uncertainty and improving physical plausibility. This mechanism partially corrects deficiencies in traditional subgrid-scale parameterization schemes.
Optimizing Physics Through End-to-End Differentiability
Problem: Traditional climate models rely on manually tuned physical parameters (e.g., diffusion) which are major sources of uncertainty. Pure AI models lack physical inductive biases, leading to inconsistent predictions.
Solution: NeuralOGCM's differentiable physics core embeds fluid dynamics equations while exposing physical parameters as learnable variables. This allows the model to optimize its physical core via end-to-end training.
Impact: This approach ensures that the model learns physically plausible parameters directly from data, leading to more accurate and reliable simulations. It addresses the 'black-box' nature of AI and the manual tuning challenges of physics-based models.
Ensuring Long-Term Physical Consistency: A 90-Day Rollout Analysis
Problem: Pure data-driven AI models (like U-Net) often suffer from severe numerical instability and error accumulation during long-term autoregressive rollouts, leading to 'hallucinations' and physically nonsensical outputs (e.g., over-saturated SST, high-frequency SSS noise) as shown in Figure 4.
Solution: NeuralOGCM, with its differentiable physics core providing strong inductive biases, successfully preserves complex spatial structures and accurate anomaly magnitudes for Sea Surface Temperature (SST) and Sea Surface Salinity (SSS) over 90-day simulations. It captures large-scale circulation patterns without significant dissipation or non-physical drift.
Impact: This robust long-term stability and physical consistency make NeuralOGCM a trustworthy tool for high-stakes scientific discovery and climate forecasting, overcoming a critical limitation of purely data-driven approaches.
Quantify Your AI-Driven Impact
Estimate the potential annual cost savings and reclaimed work hours by integrating advanced AI solutions into your enterprise operations.
Your AI Implementation Roadmap
A typical journey to integrate AI-driven climate modeling and analytics into an enterprise environment.
Discovery & Strategy
Assess current systems, identify key integration points, and define specific scientific or operational goals for AI deployment.
Model Customization & Training
Adapt NeuralOGCM or similar models to specific regional or operational data, fine-tuning learnable parameters and neural correctors.
Integration & Validation
Integrate the optimized model into existing simulation pipelines. Conduct rigorous validation against historical and real-time data.
Deployment & Monitoring
Deploy the AI system for operational use. Continuously monitor performance, accuracy, and physical consistency, with iterative improvements.
Ready to Transform Your Climate & Ocean Modeling?
Speak with our experts to discover how NeuralOGCM's innovative hybrid approach can enhance your predictive capabilities, accelerate research, and drive informed decisions.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat