Lorenz '96 with GAN parameterization

Reqiured Python Libraries:

• numpy
• scipy
• matplotlib
• xarray
• pandas
• jupyter
• keras
• tensorflow<=1.15.0
• scikit-learn
• netcdf4
• pyyaml
• numba

Installation

The Anaconda or Miniconda Python distribution is recommended for managing the dependencies for the lorenz_gan library. Install the library by searching for miniconda and downloading the appropriate install script for your OS. Python 3.5 or later is recommended.

To install the dependencies required for this package, use the following command:

>>> conda install numpy scipy matplotlib xarray pandas ipython jupyter keras tensorflow scikit-learn netcdf4 pyyaml

Once the dependencies are installed, install the lorenz_gan package in your Python environment with the following commands

>>> cd ~/lorenz_gan
>>> pip install .

Running the Lorenz '96 Model

The Lorenz '96 model can be run in both "truth" mode with a numerically resolved subgrid layer, and in "forecast" mode with a parameterized subgrid forcing term.

train_lorenz_gan.py

The program train_lorenz_gan.py runs the Lorenz truth model and fits all of the parameterizations, including the GAN. To run the program

>>> cd ~/lorenz_gan/
>>> python train_lorenz_gan.py config/lorenz_gan_travis.yaml -g

This option will run the Lorenz truth model and train a random number updater, histogram parameterization, polynomial regression parameterization, and a GAN parameterization. If you wish to re-train the parameterizations without re-running the truth model, then you can add a -r argument after the config file. The -g option runs the GAN training. If you do not wish to train the GAN, which can take a few minutes to run, then leave it off.

run_lorenz_forecast.py

The program run_lorenz_forecast.py runs an ensemble of Lorenz forecast models with the parameterization specified in the config file.

>>> python run_lorenz_forecast.py config/forecast_gan_travis.yaml -p 2

The -p argument specifies the number of processors to be used for generating forecasts. Each forecast member in the ensemble is run in a separate process.