NeuralPOM
Differentiable ocean dynamics and staged learning.
NeuralPOM combines the Princeton Ocean Model with constrained neural corrections. This guide connects the physical formulation, learning method, and code to the experiments in the paper.
Understand the model
- Physical framework
Primitive equations, grid, and dynamical cores.
- Constrained corrections
How neural increments update the ocean state.
- Staged learning
Pretrain, Midtrain, and Posttrain.
- Density-normal damping
The tracer projection, explained interactively.
Work with the code
- Installation and quick start
Set up a configuration and inspect the training plan.
- Data and calibration
Reference ensembles, preparation, and model inputs.
- Reproduce the experiments
Training, frozen forecasts, and source tables.
- Implementation guide
Find the relevant modules and entry points.