Samudra 2: Scaling Ocean Emulators across Resolutions

2026-05-24

AI Summary

This work addresses the high computational cost of traditional ocean circulation models, which hinders large-scale ensemble simulations, and overcomes limitations of existing neural ocean simulators that struggle to simultaneously achieve high resolution and multi-year autoregressive forecasting while suffering from variance collapse and imprinting artifacts. The authors propose a resolution-scalable autoregressive neural ocean simulator based on an enhanced wide U-Net architecture incorporating modified ConvNeXt blocks, a dynamic channel reweighting loss, and a multi-resolution training strategy. The method enables efficient, multi-year, high-resolution global ocean prediction on a single GPU, improving the R² of upper-ocean temperature from 0.56 to 0.87 at 1° resolution, reducing deep-ocean errors by approximately sevenfold, and successfully scaling to 1/4° resolution—accurately reproducing mesoscale eddies and western boundary currents, thereby supporting large ensemble studies of sea-level rise and ocean heat uptake.

Abstract

Ocean general circulation models (OGCMs) are essential to climate science but computationally expensive, limiting ensemble size and forcing scenarios. Neural emulators promise orders-of-magnitude speedups, yet existing ocean emulators have not combined fine spatial resolution with multi-year autoregressive rollouts. Samudra, the first autoregressive neural ocean emulator to produce multi-decade global rollouts, is limited to $1^\circ$ resolution and exhibits two long-horizon failure modes: \emph{variance collapse}, the loss of temporal variability, and \emph{imprinting artifacts}, in which velocity patterns leak into deep-ocean fields. We present Samudra 2, which introduces a wider U-Net backbone with modified ConvNeXt-style blocks and a reduced block-internal expansion factor, together with a dynamic loss that reweights output channels according to their prediction errors, strengthening gradients for slow-evolving deep-ocean fields. At $1^\circ$, Samudra 2 increases upper-ocean global-mean temperature $R^2$ from 0.56 to 0.87 and reduces deep-ocean temperature error by roughly sevenfold. The same architecture scales to $1/2^\circ$ and $1/4^\circ$ over approximately 8-year autoregressive rollouts, recovering mesoscale eddies and sharp western boundary currents. Running on a single GPU, Samudra 2 enables larger ensembles for sea-level projections, ocean heat uptake, and climate variability studies. We provide code, documentation, and benchmark resources at https://openathena.ai/Ocean_Emulator/.

Problem

• ocean emulator
• autoregressive rollout
• spatial resolution
• variance collapse
• imprinting artifacts

Innovation

• neural emulator
• autoregressive rollout
• dynamic loss
• high-resolution ocean modeling
• ConvNeXt-style blocks