Hi,


Most likely I am overlooking something obvious, but I cannot figure out what. 

I am working with this dataset and have calculated averages over NUTS regions. This is how my data array now looks like:

<xarray.DataArray 'number_of_days' (nuts: 37, time: 100)>
dask.array<xarray-number_of_days, shape=(37, 100), dtype=float32, chunksize=(37, 48), chunktype=numpy.ndarray>
Coordinates:
    height   float64 ...
  * time     (time) datetime64[ns] 1986-01-01 1987-01-01 ... 2085-01-01
  * nuts     (nuts) object 'ES' 'FI' 'IS' 'FR' 'HR' ... 'ME' 'MK' 'MT' 'NL' 'NO'
Attributes:
...

So far, so good. 

Next I want to select a single year. I want to get rid of the time dimension because otherwise I end up with a time slider in my live map, which I don't want. So I thought I could use ct.cube.select(...) but no matter what I try, the time dimension persists even if I use the keyword drop=True:



darray_sub = ct.cube.select(darray,time='1987', drop=True)

<xarray.DataArray 'number_of_days' (nuts: 37, time: 1)>
dask.array<xarray-number_of_days, shape=(37, 1), dtype=float32, chunksize=(37, 1), chunktype=numpy.ndarray>
Coordinates:
    height   float64 ...
  * nuts     (nuts) object 'ES' 'FI' 'IS' 'FR' 'HR' ... 'ME' 'MK' 'MT' 'NL' 'NO'
Dimensions without coordinates: time
Attributes:
    units:                         day
    long_name:                     Ensemble members average of VBD season_len...
    type:                          real
    valid_min:                     0.0
    conventions:                   CF-1.6
    _cds_toolbox_aux_wms_layer:    nuts:nuts__level_0__resolution_medium__yea...
    _cds_toolbox_aux_wms_fids:     []
    _cds_toolbox_aux_wms_filters:  {}
    _cds_toolbox_wms_attr_map:     nuts:NUTS_ID


However, if I use ct.cube.index_select(...) the time dimension disappears even if I add the keyword drop=False:


darray_sub = ct.cube.index_select(darray,time=0,drop=False)

<xarray.DataArray 'number_of_days' (nuts: 37)>
array([ 98.37596 ,   0.442497,  20.3601  , 143.63914 ,  94.37917 ,  42.43064 ,
       102.53675 ,  25.354486, 130.1956  ,  46.885483,  45.356785,  75.993004,
        26.926882,  82.430275,  72.13661 ,   7.306597, 100.75709 , 152.26291 ,
        34.79379 , 154.937   ,  21.72802 ,  51.965862,   5.821461,  59.158653,
        12.273589,  63.25718 , 131.57185 , 117.382095,  16.603952,  13.528408,
       107.88156 ,  12.062992,  44.698086,  26.043186,  41.21859 , 123.48304 ,
         6.420299], dtype=float32)
Coordinates:
    height   float64 ...
    time     datetime64[ns] ...
  * nuts     (nuts) object 'ES' 'FI' 'IS' 'FR' 'HR' ... 'ME' 'MK' 'MT' 'NL' 'NO'
Attributes:
    units:                         day
    long_name:                     Ensemble members average of VBD season_len...
    type:                          real
    valid_min:                     0.0
    conventions:                   CF-1.6
    _cds_toolbox_aux_wms_layer:    nuts:nuts__level_0__resolution_medium__yea...
    _cds_toolbox_aux_wms_fids:     []
    _cds_toolbox_aux_wms_filters:  {}
    _cds_toolbox_wms_attr_map:     nuts:NUTS_ID


Although I could, of course, just use the index_select(...) method, I'm somewhat confused by the fact that in neither case adding drop=True or drop=False seems to make any difference at all. Why is that?

Also, I would have to work out the correct index myself which is more prone to error (smile) Is there an alternative way of getting rid of the time dimension of length 1, similar to xarray's squeeze method? 


Thanks!

Rutger

1 Comment

  1. Dear Rutger,

    The drop option in ct.cube.select and ct.cube.index_select only drops the coordinates associated to the dimension not the dimension itself.

    As far as I am aware if you really need to drop the dimension you need to use ct.cube.index_select. The index should not be an issue if you have a singleton dimension.

    Regards.

    Vivien