Installation

Recommended: Install from conda-forge

Given that some ufs2arco dependencies are only available on conda-forge, it’s recommended that users install using this method:

conda install -c conda-forge ufs2arco

This will install all underlying dependencies, including mpi4py which by default installs mpich for MPI support. If a different MPI distribution is desired, for example Intel MPI, then this can be installed via:

conda install -c conda-forge ufs2arco impi_rt

in the exact same way as directed in this section of the mpi4py documentation.

Install from conda-forge without MPI

If you do not want to install MPI through conda-forge, for instance if you want to use an MPI distribution that is already built on a system you’re using, then it is recommended to install the nompi build on conda forge as follows:

conda install -c conda-forge ufs2arco=*=nompi*

Then, one can install mpi4py from pip using these instructions, or following instructions specific to your machine.

Install with pip

It is possible to install ufs2arco from pypi via:

pip install ufs2arco

However, this will not come with the MPI or xesmf dependencies, since these need to be installed from conda-forge.

Running Example Notebooks or Building the Documentation Locally

Due to the way pandoc is installed via pip as detailed here it is recommended to create an environment with conda in order to build the documentation locally. This is also recommended for running any of the example notebooks locally, since there are a couple of additional dependencies required. To do this, first download this environment.yaml file, then create the conda environment:

conda env create -f environment.yaml
conda activate ufs2arco

Note that you will then want to add ufs2arco to the jupyter kernel:

python -m ipykernel install --user --name=ufs2arco

so that it can be accessed from jupyter lab / notebook.