Here we archive the process of building cfgrib engine for xarray in python3.7 (Anaconda). Although there still seems to be some problems, the eccodes and cfgrib have been installed sucessfully.
First, we found using conda install cannot install eccodes due to dependency problems (might comes from python version issue, we did not check then).
Then we install the eccodes from source. The process was smooth, just install cmake prior to the eccodes installation.
When install python interface, try both --install-option and export ECCODES_DIR=/path/to/where/you/install/eccodes.
After that:
pip install --force-reinstall cfgrib
Note conda does not work in my case.
Set all possible environmental varibales:
export ECCODES_DIR=/home/yhuangci/lee/soft/eccodes-2.17.0
export PATH=$ECCODES_DIR:$PATH
export LIBECCODES=$ECCODES_DIR/lib
export LD_LIBRARY_PATH=$ECCODES_DIR/lib:$LD_LIBRARY_PATH
xr.open_dataset('example.grib', engine='cfgrib') works with some data-caused errors.
Error things like:
cfgrib.dataset.DatasetBuildError: key present and new value is different
See this post. Add a backend_kwargs to overwrite the “strict” option will fix the problem:
ds_grib = xr.open_dataset('../../data/era5/sfc202005.grib', engine='cfgrib', backend_kwargs={'errors': 'ignore'})
Updated 2020-05-19