inconsistent 1D interp on 1D and ND DataArrays with NaNs

delgadom commented 3 weeks ago

What happened?

May be a duplicate of #5852, and as stated there, interp with nans isn't really supported. But I noticed that in some cases xr.DataArray.interp drops valid data if there are NaNs in the array, and that this behavior is different depending on the number of dimensions, even if only a single dimension is specified (so interp1d should be used in every case).

Not sure if this warrants a different issue, but the NaN handling appears to be different depending on the number of dimensions, even if we're interpolating along a single dimension (so should be essentially broadcasting interp1d:

Here's a MRE:

import xarray as xr
import numpy as np

a = xr.DataArray(
    [np.nan, 1, 2],
    dims=["x"],
    coords=[range(0, 5, 2)],
)

print("a:\n")
print(a)

b = xr.DataArray(
    [[np.nan, 1, 2]],
    dims=["y", "x"],
    coords=[["first"], range(0, 5, 2)],
)

print("\n\nb:\n")
print(b)

print("\n\na.interp(x=range(5)):\n")
print(a.interp(x=range(5)))

print("\n\nb.interp(x=range(5)):\n")
print(b.interp(x=range(5)))

print("\n\nxarray version:")
print(xr.__version__)

gives

a:

<xarray.DataArray (x: 3)> Size: 24B
array([nan,  1.,  2.])
Coordinates:
  * x        (x) int64 24B 0 2 4

b:

<xarray.DataArray (y: 1, x: 3)> Size: 24B
array([[nan,  1.,  2.]])
Coordinates:
  * y        (y) <U5 20B 'first'
  * x        (x) int64 24B 0 2 4

a.interp(x=range(5)):

<xarray.DataArray (x: 5)> Size: 40B
array([nan, nan, 1. , 1.5, 2. ])
Coordinates:
  * x        (x) int64 40B 0 1 2 3 4

b.interp(x=range(5)):

<xarray.DataArray (y: 1, x: 5)> Size: 40B
array([[nan, nan, nan, 1.5, 2. ]])
Coordinates:
  * y        (y) <U5 20B 'first'
  * x        (x) int64 40B 0 1 2 3 4

xarray version:
2024.10.0

Note that the second interpolation case, in which the array is 2D, drops a valid value (1, in position x=2), even though it is present in the original array. The first case correctly interpolates between 1 and 2 and does not drop the first value.

Apologies - I don't have time now to diagnose whether this is an xarray issue with the new 2D interp handling or simply a scipy interp1d broadcasting issue, but it was certainly unexpected!

What did you expect to happen?

No response

Minimal Complete Verifiable Example

No response

MVCE confirmation

[X] Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
[X] Complete example — the example is self-contained, including all data and the text of any traceback.
[X] Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
[X] New issue — a search of GitHub Issues suggests this is not a duplicate.
[X] Recent environment — the issue occurs with the latest version of xarray and its dependencies.

Relevant log output

No response

Anything else we need to know?

No response

Environment

INSTALLED VERSIONS ------------------ commit: None python: 3.11.4 | packaged by conda-forge | (main, Jun 10 2023, 18:08:41) [Clang 15.0.7 ] python-bits: 64 OS: Darwin OS-release: 24.0.0 machine: arm64 processor: arm byteorder: little LC_ALL: None LANG: en_US.UTF-8 LOCALE: ('en_US', 'UTF-8') libhdf5: 1.14.0 libnetcdf: 4.9.2 xarray: 2024.10.0 pandas: 2.2.3 numpy: 1.24.3 scipy: 1.14.1 netCDF4: 1.6.4 pydap: None h5netcdf: None h5py: 3.8.0 zarr: 2.18.3 cftime: 1.6.2 nc_time_axis: None iris: None bottleneck: 1.3.7 dask: 2024.8.0 distributed: 2024.8.0 matplotlib: 3.7.1 cartopy: 0.24.0 seaborn: 0.12.2 numbagg: None fsspec: 2023.6.0 cupy: None pint: None sparse: None flox: 9.11 numpy_groupies: 0.11.2 setuptools: 67.7.2 pip: 23.1.2 conda: None pytest: None mypy: None IPython: 8.14.0 sphinx: None

dcherian commented 2 weeks ago

@hollymandel do you have time to take a quick look here? Simply diagnosing the core problem would be quite helpful

hollymandel commented 2 weeks ago

This behavior seems to come from scipy:

import scipy

print(scipy.interpolate.interp1d(x = [0,2,4], y =[np.nan, 1,2])([0,1,2,3,4])) # [nan nan 1.  1.5 2. ]
print(scipy.interpolate.interp1d(x = [0,2,4], y =[[np.nan, 1,2]])([0,1,2,3,4])) # [[nan nan nan 1.5 2. ]]

hollymandel commented 2 weeks ago

I think adding a warning (as suggested in #5852) is reasonable, can do the PR if others agree.

dcherian commented 2 weeks ago

Nice, I think a warning that suggests .dropna would be good.

hollymandel commented 2 days ago

Here is a draft PR: #9813

I guess the maintainers should decide whether this is too noisy? I'm not really sure.

pydata / xarray