This package provides a framework for calibrating the parameters of vertical physics schemes of ocean circulation models using variational optimization. The parameters are calibrated through the minimization of an 'objective function' which compares model predictions with 'Large Eddy Simulations' (LES). Tunax is written in JAX in order to use automatic differentiation for computing the gradient of the objective function with respect to model parameters.
The source code of tunax is in the folder tunax/. The two main modules are tunax/model.py which contains the implementation of the forward Single Column Model and tunax/fitter.py which contains the implementation of the calibration part. The various physical closures equations that Tunax can calibrate are implemented in the folder tunax/closures/. The other modules implement the object used by the model and the callibrator and they are described in the documentation.
The folder notebooks/ contains some notebooks of demonstration of tunax.
Tunax is installed by default with JAX on CPU.
Tunax is pip installable the following command will install the lastest stable release :
pip install tunaxThe most recent additions to Tunax since the last release are implemented in the main branch of this repository. To benefit from them one can clone or fork this repository and install Tunax with the package manager poetry :
poetry install --no-devTo use the notebooks with the specific packages that they use, one can clone or fork the branch main or another of this repository and install Tunax with the package manager poetry :
poetry install