New Wine in an Old Bottle -- NeuralPOM: A Hybrid Ocean Forecast Model with Differentiable Dynamics and Staged Learning
Abstract
Coarse-grid ocean circulation models rely on empirical subgrid parameterizations that struggle to capture mesoscale eddies, whereas purely data-driven neural forecasters accumulate trajectory drift, violate tracer conservation, and damp small-scale variability over extended integrations. Here we present NeuralPOM, a hybrid ocean forecast model that couples differentiable explicit mode-split and semi-implicit free-surface implementations of the Princeton Ocean Model (POM) to a conservation-constrained neural corrector. The three-dimensional primitive-equation solver advances the ocean state, while the neural network periodically updates the physical fields at high-frequency (1 h) or low-frequency (1 day) intervals under volume, heat, and salt conservation constraints. Using a fine-grid reference dataset exceeding 100,000 model days in a wind-driven double-gyre basin, we introduce a three-stage training procedure comprising offline single-step pretraining, multi-step differentiable training through the ocean solver, and multi-day closed-loop rollout training. Over 100-day forecasts, NeuralPOM reduces errors across all prognostic fields relative to coarse-grid physical closures and direct neural forecasters, preserving mesoscale kinetic energy while improving eddy phase alignment. Controlled ablations show that closed-loop training suppresses multi-day drift, whereas backpropagating gradients through the differentiable solver stabilizes high-frequency coupling. Finally, because high-frequency tracer updates can introduce density anomalies via the equation of state that excite spurious baroclinic pressure gradients, we introduce Density-Normal Damping (DND) to attenuate the density-modifying (diapycnal) tracer component while retaining the density-compensated (isopycnal) correction.
Related
- Uncertainty analysis of a machine learning-aided ensemble forecast for coastal flood maps
- Sensitivity Fidelity Reveals Physical Dependencies Beyond Forecast Accuracy in Deep-Learning Earth-System Models
- A unified deep learning framework for subseasonal wind forecast post-processing
- Seasonal hydropower forecast framework using a combination of deep learning and physics-based methods
- Structured Neural Modeling of Daily Arctic Sea-Ice Concentration Evolution: Physical-Trajectory-Driven Learning and Forecast-Domain Adaptation
BibTeX
@misc{shu2026wine,
title = {{New Wine in an Old Bottle -- NeuralPOM: A Hybrid Ocean Forecast Model with Differentiable Dynamics and Staged Learning}},
author = {Ruiqi Shu and Guangliang Liu and Ruijian Gou and Tianrun Gao and Haomin Wen and Hao Li and Xiaomeng Huang},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.02460},
eprint = {2611.02460},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.02460},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - New Wine in an Old Bottle -- NeuralPOM: A Hybrid Ocean Forecast Model with Differentiable Dynamics and Staged Learning
AU - Ruiqi Shu
AU - Guangliang Liu
AU - Ruijian Gou
AU - Tianrun Gao
AU - Haomin Wen
AU - Hao Li
AU - Xiaomeng Huang
PY - 2026
DA - 2026/11/02
PB - arXiv
DO - 10.48550/arxiv.2611.02460
UR - https://arxiv.org/abs/2611.02460
AB - Coarse-grid ocean circulation models rely on empirical subgrid parameterizations that struggle to capture mesoscale eddies, whereas purely data-driven neural forecasters accumulate trajectory drift, violate tracer conservation, and damp small-scale variability over extended integrations. Here we present NeuralPOM, a hybrid ocean forecast model that couples differentiable explicit mode-split and semi-implicit free-surface implementations of the Princeton Ocean Model (POM) to a conservation-constrained neural corrector. The three-dimensional primitive-equation solver advances the ocean state, while the neural network periodically updates the physical fields at high-frequency (1 h) or low-frequency (1 day) intervals under volume, heat, and salt conservation constraints. Using a fine-grid reference dataset exceeding 100,000 model days in a wind-driven double-gyre basin, we introduce a three-stage training procedure comprising offline single-step pretraining, multi-step differentiable training through the ocean solver, and multi-day closed-loop rollout training. Over 100-day forecasts, NeuralPOM reduces errors across all prognostic fields relative to coarse-grid physical closures and direct neural forecasters, preserving mesoscale kinetic energy while improving eddy phase alignment. Controlled ablations show that closed-loop training suppresses multi-day drift, whereas backpropagating gradients through the differentiable solver stabilizes high-frequency coupling. Finally, because high-frequency tracer updates can introduce density anomalies via the equation of state that excite spurious baroclinic pressure gradients, we introduce Density-Normal Damping (DND) to attenuate the density-modifying (diapycnal) tracer component while retaining the density-compensated (isopycnal) correction.
ER -