Samudra: An AI Global Ocean Emulator for Climate

2024-12-05

arXiv.org

AI Summary

Traditional ocean circulation models suffer from high computational cost, climate drift during long-term integration, and limited fidelity in reproducing multiscale climate variability. Method: We propose the first global-scale, fully vertical-layered, zero-drift ocean AI emulator. Our approach employs an enhanced ConvNeXt-UNet architecture to jointly model sea surface height, and three-dimensional temperature, salinity, and velocity fields across all depths, augmented by physics-informed consistency losses. Results: The emulator exhibits no divergence or climate drift over century-scale free integrations; achieves markedly improved accuracy in capturing interannual variability and vertical structure; and accelerates inference by 150× relative to conventional models. Contribution: This work establishes the first high-fidelity, long-term stable, and vertically resolved global ocean AI simulator—enabling unprecedented millennial-scale climate attribution and prediction.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Diffusion Models for Vision
• Humans and AI: Emotional Intelligence

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

AI emulators for forecasting have emerged as powerful tools that can outperform conventional numerical predictions. The next frontier is to build emulators for long climate simulations with skill across a range of spatiotemporal scales, a particularly important goal for the ocean. Our work builds a skillful global emulator of the ocean component of a state-of-the-art climate model. We emulate key ocean variables, sea surface height, horizontal velocities, temperature, and salinity, across their full depth. We use a modified ConvNeXt UNet architecture trained on multidepth levels of ocean data. We show that the ocean emulator - Samudra - which exhibits no drift relative to the truth, can reproduce the depth structure of ocean variables and their interannual variability. Samudra is stable for centuries and 150 times faster than the original ocean model. Samudra struggles to capture the correct magnitude of the forcing trends and simultaneously remain stable, requiring further work.

Problem

• Emulate global ocean variables across full depth
• Ensure stability and accuracy for long climate simulations
• Improve speed and performance of ocean modeling

Innovation

• Modified ConvNeXt UNet for ocean emulation
• Multidepth ocean data training approach
• Century-stable, 150x faster than original