When using coarsen, we often need to align the original DataArray with the coarser coordinates. For instance: ```python import xarray as xr import numpy as np

da = xr.DataArray(np.arange(4*4).reshape(4, 4), coords=[np.arange(4, 0, -1) + 0.5, np.arange(4) + 0.5], dims=['lat', 'lon'])

<xarray.DataArray (lat: 4, lon: 4)>

array([[ 0, 1, 2, 3],

[ 4, 5, 6, 7],

[ 8, 9, 10, 11],

[12, 13, 14, 15]])

Coordinates:

* lat (lat) float64 4.5 3.5 2.5 1.5

* lon (lon) float64 0.5 1.5 2.5 3.5

da.coarsen(lat=2, lon=2).mean()

<xarray.DataArray (lat: 2, lon: 2)>

array([[ 2.5, 4.5],

[10.5, 12.5]])

Coordinates:

* lat (lat) float64 4.0 2.0

* lon (lon) float64 1.0 3.0

But if the coarser coordinates are aligned like: lat: ... 5 3 1 ... lon: ... 1 3 5 ... Then directly applying `coarsen` will not work (here on the `lat` dimension). The following function extends the original DataArray so that it is aligned with the coarser coordinates:python def adjust_bbox(da, dims): """Adjust the bounding box of a DaskArray to a coarser resolution.

Args:
    da: the DaskArray to adjust.
    dims: a dictionary where keys are the name of the dimensions on which to adjust, and the values are of the form [unsigned_coarse_resolution, signed_original_resolution]
Returns:
    The DataArray bounding box adjusted to the coarser resolution.
"""
coords = {}
for k, v in dims.items():
    every, step = v
    offset = step / 2
    dim0 = da[k].values[0] - offset
    dim1 = da[k].values[-1] + offset
    if step < 0: # decreasing coordinate
        dim0 = dim0 + (every - dim0 % every) % every
        dim1 = dim1 - dim1 % every
    else: # increasing coordinate
        dim0 = dim0 - dim0 % every
        dim1 = dim1 + (every - dim1 % every) % every
    coord0 = np.arange(dim0+offset, da[k].values[0]-offset, step)
    coord1 = da[k].values
    coord2 = np.arange(da[k].values[-1]+step, dim1, step)
    coord = np.hstack((coord0, coord1, coord2))
    coords[k] = coord
return da.reindex(**coords).fillna(0)

da = adjust_bbox(da, {'lat': (2, -1), 'lon': (2, 1)})

<xarray.DataArray (lat: 6, lon: 4)>

array([[ 0., 0., 0., 0.],

[ 0., 1., 2., 3.],

[ 4., 5., 6., 7.],

[ 8., 9., 10., 11.],

[12., 13., 14., 15.],

[ 0., 0., 0., 0.]])

Coordinates:

* lat (lat) float64 5.5 4.5 3.5 2.5 1.5 0.5

* lon (lon) float64 0.5 1.5 2.5 3.5

da.coarsen(lat=2, lon=2).mean()

<xarray.DataArray (lat: 3, lon: 2)>

array([[0.25, 1.25],

[6.5 , 8.5 ],

[6.25, 7.25]])

Coordinates:

* lat (lat) float64 5.0 3.0 1.0

* lon (lon) float64 1.0 3.0

`` Nowcoarsengives the right result. Butadjust_bbox` is rather complicated and specific to this use case (evenly spaced coordinate points...). Do you know of a better/more general way of doing it?