issue_comments

@mathause Thanks! Shall we close this issue?

@mathause For .values, if I delete vectorize=True, I got this error: File "/home/xin/miniconda3/envs/satpy/lib/python3.7/site-packages/scipy/interpolate/interpolate.py", line 455, in __init__ raise ValueError("the x array must have exactly one dimension.") ValueError: the x array must have exactly one dimension. Then, I keep vectorize=True deleted and use the np.interp, I got this error: File "/mnt/d/Github/s5p-wrfchem/s5p_utils.py", line 264, in interp1d_np return np.interp(xi, x, data) File "<__array_function__ internals>", line 6, in interp File "/home/xin/miniconda3/envs/satpy/lib/python3.7/site-packages/numpy/lib/function_base.py", line 1412, in interp return interp_func(x, xp, fp, left, right) ValueError: object too deep for desired array If I let vectorize=True shows again and use the np.interp, I got the error mentioned before: File "/home/xin/miniconda3/envs/satpy/lib/python3.7/site-packages/numpy/lib/function_base.py", line 1830, in _update_dim_sizes % (dim, size, dim_sizes[dim])) ValueError: inconsistent size for core dimension 'dim0': 2 vs 39

For the one without .values, this is the result of repr(s5p['p']): <xarray.DataArray (bottom_top: 25, y: 389, x: 450)> dask.array<where, shape=(25, 389, 450), dtype=float32, chunksize=(25, 389, 450), chunktype=numpy.ndarray> Coordinates: * bottom_top (bottom_top) int32 0 1 2 3 4 5 6 7 8 ... 17 18 19 20 21 22 23 24 vertices int32 0 crs object +proj=latlong +datum=WGS84 +ellps=WGS84 +type=crs Dimensions without coordinates: y, x Attributes: name: p resolution: None calibration: None polarization: None level: None modifiers: () units: hPa After the bottom_up in renamed to new_dim, it works without error for both scipy and numpy interpolation function.

I tested again with a subset of my data: subset_no2 = regrid_vars['no2'].isel(x=277, y=[212, 213]) subset_p = regrid_vars['p'].isel(x=277, y=[212, 213]) subset_interp = s5p['p'].isel(x=277, y=[212, 213]) interped = xr.apply_ufunc( interp1d_np, subset_no2, subset_p, subset_interp, input_core_dims=[["bottom_top"], ["bottom_top"], ["new_dim"]], output_core_dims=[["new_dim"]], exclude_dims=set(("bottom_top",)), vectorize=True, )

Error without .values: File "/home/xin/miniconda3/envs/satpy/lib/python3.7/site-packages/xarray/core/computation.py", line 508, in broadcast_compat_data list(core_dims), missing_core_dims ValueError: operand to apply_ufunc has required core dimensions ['new_dim'], but some of these dimensions are absent on an input variable: ['new_dim']

Error with .values: File "/home/xin/miniconda3/envs/satpy/lib/python3.7/site-packages/numpy/lib/function_base.py", line 1830, in _update_dim_sizes % (dim, size, dim_sizes[dim])) ValueError: inconsistent size for core dimension 'dim0': 2 vs 39

@dcherian If .values is removed, I got this error: File "/home/xin/miniconda3/envs/satpy/lib/python3.7/site-packages/xarray/core/computation.py", line 508, in broadcast_compat_data list(core_dims), missing_core_dims ValueError: operand to apply_ufunc has required core dimensions ['new_dim'], but some of these dimensions are absent on an input variable: ['new_dim']

Here's the information of regrid_vars['no2'], regrid_vars['p'] and s5p['p']: <xarray.DataArray 'no2' (bottom_top: 39, y: 389, x: 450)> <xarray.DataArray (bottom_top: 39, y: 389, x: 450)> <xarray.DataArray (bottom_top: 25, y: 389, x: 450)>

BTW, I have nan values in regrid_vars['no2'] and regrid_vars['p']. I think that wouldn't cause that error.

@mathause Thanks! It works well. Here's the solution:

Code

``` def interp1d_np(data, x, xi): from scipy import interpolate # return np.interp(xi, x, data) f = interpolate.interp1d(x, data, fill_value='extrapolate') return f(xi)

interped = xr.apply_ufunc( interp1d_np, # first the function bottom_up, # now arguments in the order expected by 'interp1_np' pressure.values, # as above interp_p.values, # as above input_core_dims=[["z"], ["z"], ["new_z"]], # list with one entry per arg output_core_dims=[["new_z"]], # returned data has one dimension exclude_dims=set(("z",)), # dimensions allowed to change size. Must be a set! vectorize=True, # loop over non-core dims ) interped = interped.rename({"new_z": "z"})

print(np.testing.assert_allclose(output.values, interped.values)) ```

Result:

None

However, when I apply it to my real data, I got some errors:

Code

``` def interp1d_np(data, x, xi): from scipy import interpolate f = interpolate.interp1d(x, data, fill_value='extrapolate') return f(xi)

interped = xr.apply_ufunc(
    interp1d_np,
    regrid_vars['no2'],
    regrid_vars['p'].values,
    s5p['p'].values,
    input_core_dims=[["bottom_top"], ["bottom_top"], ["new_dim"]],
    output_core_dims=[["new_dim"]],
    exclude_dims=set(("bottom_top",)),
    vectorize=True,
)

```

Error:

File "/home/xin/miniconda3/envs/satpy/lib/python3.7/site-packages/numpy/lib/function_base.py", line 1830, in _update_dim_sizes % (dim, size, dim_sizes[dim])) ValueError: inconsistent size for core dimension 'dim0': 450 vs 39 Here's the output of print(regrid_vars['no2'].shape, regrid_vars['p'].values.shape, s5p['p'].values.shape): (39, 389, 450) (39, 389, 450) (25, 389, 450) The shape looks fine.