Why time indices are ignored in FieldSet.from_netcdf?

[bbarcelollull]

I have a set of netcdf files with a date overlap: the last date of one file is the first date of the next file. I have one file per day and hourly outputs, and instead of having 24 indices per file, I have 25 indices. Because of this, I get an error when creating the fieldset with FieldSet.from_netcdf. To solve this, I need to read only the first 24 indices of each file. However, when doing an easy test like this for only one file:

indices = {'depth': np.arange(2).tolist(),
              'lat': np.arange(2).tolist(), 
              'lon': np.arange(2).tolist(), 
              'time': np.arange(2).tolist()} 

fieldset = FieldSet.from_netcdf(filenames, variables, dimensions, indices, deferred_load=False)

I get the following warning: "WARNING: time dimension in indices is not necessary anymore. It is then ignored."

And the shape of the U field (fieldset.U.data.shape) is (25, 2, 2, 2).

Hence, indices works for all dimensions except for time. Why this happens? Can I force the code to read the time indices?

[erikvansebille]

Hi @bbarcelollull, thanks for posting this question. You are right that parcels doesn't support indexing for the time-dimensions. That would have take a lot of development effort, and in most use-cases so far there is one snapshot per file so users can easily subselect by changing the list of files they provide when creating a fieldset.

An alternative is to use FieldSet.from_xarray_dataset(), which is much more flexible because it allows you to pass xarray datasets directly