Ok, so for now I roll with this:

```python def shrink_dataarray(da, dims=None): """remove nodata borders from spatial dims of dataarray""" dims = set(dims) if dims else set(da.dims)

if len(dims) != 2:
    raise IndexError

# non-spatial dims (carry over, only shrink spatial dims)
nsd = set(da.dims) - dims
nsd_indexers = {d: range(len(da[d])) for d in nsd}

indexers = {d: (da.count(dim=dims - set([d])|nsd).cumsum() != 0) * 
               (da.count(dim=dims - set([d])|nsd)[::-1].cumsum()[::-1] != 0)
            for d in dims}

indexers.update(nsd_indexers)

return da.isel(**indexers)

```

Is it possible to identify non-spatial dims with plain xarray dataarrays (non cf-xarray)? And is there maybe a way to detect unlimited dims (usually the time dim)?

@fujiisoup Thanks, that’s great and much cleaner than my previous numpy code. I’ll run with that and maybe try to pack that in a general function. Not sure is this a common enough problem to have in xarray itself?

@cwerner ```python In [40]: idx = (da.count('y').cumsum() != 0) * (da.count('y')[::-1].cumsum()[::- ...: 1] != 0)

In [42]: da.isel(x=idx)
Out[42]: <xarray.DataArray (x: 3, y: 4)> array([[nan, 0., 2., nan], [nan, nan, nan, nan], [nan, 2., 0., nan]]) Dimensions without coordinates: x, y ``` Maybe this works, but I have no cleaner solution.

@keewis @fujiisoup @shoyer thanks. this does indeed not work for my used case if there's a all-nan stretch between parts of the array (think UK and the channel and the northern coast of France) - I simply want to get rid of extra space around a geographic domain (i.e. the nan edges)

``` data = np.array([ [np.nan, np.nan, np.nan, np.nan], [np.nan, 0, 2, np.nan], [np.nan, np.nan, np.nan, np.nan], [np.nan, 2, 0, np.nan], [np.nan, np.nan, np.nan, np.nan], ]) da = xr.DataArray(data, dims=("x", "y"))

this also results in a 2x2 array, but should be 3x2

```

Another way to write this currently would be with where: array.where(array.notnull(), drop=True)

@keewis I think it is close to da.dropna(how='all') python In [12]: da.dropna('x', how='all').dropna('y', how='all') Out[12]: <xarray.DataArray (x: 2, y: 2)> array([[0., 2.], [2., 0.]]) Dimensions without coordinates: x, y I think supporting multiple dimensions for dropna is totally in our scope. Currently, dropna only works with a single dimension and da.dropna(how='all') does not work.

@cwerner Is it close to your example? If you don't want to drop all nans but only those located at the edges, the above example does not work.

you should be able to emulate that using: ```python In [2]: data = np.array([ ...: [np.nan, np.nan, np.nan, np.nan], ...: [np.nan, 0, 2, np.nan], ...: [np.nan, 2, 0, np.nan], ...: [np.nan, np.nan, np.nan, np.nan], ...: ]) ...: da = xr.DataArray(data, dims=("x", "y"))

In [3]: def shrink(arr): ...: notnull = da.notnull() ...: indexers = { ...: dim: notnull.any(dim=set(da.dims) - set([dim])) ...: for dim in da.dims ...: } ...: return arr.sel(**indexers) ...:

In [4]: da Out[4]: <xarray.DataArray (x: 4, y: 4)> array([[nan, nan, nan, nan], [nan, 0., 2., nan], [nan, 2., 0., nan], [nan, nan, nan, nan]]) Dimensions without coordinates: x, y

In [5]: shrink(da) Out[5]: <xarray.DataArray (x: 2, y: 2)> array([[0., 2.], [2., 0.]]) Dimensions without coordinates: x, y ``` @pydata/xarray: is this common enough to add it to the API?