Read levels into separated datasets

[julienvienne]

Is your feature request related to a problem? Please describe.

Hi,
I am currently struggling for reading a mixed parameter grib file which contains (among others) temperature parameter ('t') at 100m and 200m. I would like these to be read in separated datasets, but it seems that the cfgrib.open_datasets() method automatically merge the record into the heightAboveLevel dimension (see example below).
Is there a workaround to read all grib record independantly and create one dataset for each, letting enduser merge the datasets by himself ?

Describe the solution you'd like

• use level filtering key to create separated datasets
• make the squeeze=False option working for this
• add new method or option to read all records without merging

Describe alternatives you've considered

I have tried to include the level keys for making separated datasets, but the levels are merged anyway into the same variable :

                datasets = cfgrib.open_datasets(
                    grib_file,
                    chunks={'latitude': 2000, 'longitude': 2000},
                    backend_kwargs={'indexpath': idx_path,
                                    'read_keys': ['shortName', 'typeOfLevel', 'level'],
                                    'squeeze': False,
                                    'filter_by_keys': {
                                        'typeOfLevel': 'heightAboveGround',
                                        "level": [0,2,100,200]
                                    }
                                    }
                )

Additional context

No response

Organisation

Weathernews

[DWesl]

Does this do what you want?

[cfgrib.open_dataset(
     grib_file, backend_kwargs={"filter_by_keys": {"typeOfLevel": "heightAboveGround", "level": level}})
    for level in [0, 2, 100, 200]]

I am curious that you got it to load multiple values of a single key: when I tried doing that ("shortName", to skip warnings about other variables not being available at all levels), it didn't work for me: is that new behavior, or only present when using the filter_by_keys keyword-argument through the backend_kwargs keyword-argument, rather than passing it directly to cfgrib.open_dataset*?