grixraz: mch GRIB archives as virtual Zarr stores

This is a proof-of-concept library that uses kerchunk to represent MeteoSwiss GRIB archives of NWP data as virtual Zarr stores: no extracting, moving or copying.

You can find a demo in notebooks/demo.ipynb and an example of a reference file in notebooks/example_reference_file.json.

Installation

On a CSCS machine:

• Run source install_micromamba.sh in order to install micromamba.
• Inside your source code folder, run micromamba create -f environment.yaml in order to setup the micromamba environment.
• Activate your environment by micromamba activate grixraz. Now you have installed eccodes and poetry.
• Install the dependencies managed by poetry: poetry install
• Setup the eccodes COSMO definitions env var: source setup_env.sh

As a developer

• also install the pre-commit hooks: pre-commit install

Opening the reference file from the demo

If you're in a rush and don't want to go through the demo, you can also just install fsspec (+ xarray and zarr) in your python environment and read the reference file from the example (it's on tsa).

import json
import fsspec
import xarray as xr

with open("/scratch/fzanetta/grixraz/examples/kenda_anasurf_202306.json", "r") as f:
    ref = json.load(f)

fs = fsspec.filesystem("reference", fo=ref)
ds = xr.open_dataset(fs.get_mapper(""), engine="zarr", backend_kwargs=dict(consolidated=False),
                      chunks={'valid_time':1})