The preprocessor extract_trajectory doesn't create coordinate bounds, and confuses lat/lon coordinates

[ledm]

The preprocessor extract_trajectory is built to interpolate the model data between two points at a prescribed number of points.

This means that it creates a new set of coordinates between those two points. However, the new coordinates do not include the coordinate attribute, bounds.

ie:

The preprocessor

  prep_transect:
    extract_trajectory:
      latitudes: [0., 0. ]
      longitudes: [-5.,  5.]
      number_points: 25

Produces the follow latitude

print(cube.cooord('latitude')
AuxCoord(masked_array(data=[-5.1491704 , -4.4021463 , -4.044439  , -3.6941166 ,
                   -3.3493495 , -3.0086572 , -2.3350284 , -2.0004609 ,
                   -1.6666366 , -1.3331935 , -0.9999002 , -0.33332792,
                    0.        ,  0.33332792,  0.9999002 ,  1.3331935 ,
                    1.6666366 ,  2.0004609 ,  2.3350284 ,  3.0086572 ,
                    3.3493495 ,  3.6941166 ,  4.044439  ,  4.4021463 ,
                    5.1491704 ],
             mask=False,
       fill_value=1e+20,
            dtype=float32)

However, you'll also notice that it actually switches latitudes and longitudes:

https://docs.esmvaltool.org/projects/esmvalcore/en/latest/_modules/esmvalcore/preprocessor/_volume.html#extract_trajectory

        longitudes = np.linspace(minlat, maxlat, num=number_points)
        latitudes = np.linspace(minlon, maxlon, num=number_points)

should be:

        longitudes = np.linspace(minlon, maxlon, num=number_points)
        latitudes = np.linspace(minlat, maxlat, num=number_points)

Whoops!

(This was first an issue on ESMValTool, instead of ESMValCore: ESMValGroup/ESMValTool#2174)

[ledm]

Note that PR #1174 doesn't address the issue with the missing boundaries.

[valeriupredoi]

would you be willing to fix this (well, just the bounds bit) @ledm pls 🍺

[ledm]

I came up with a workaround at the diagnostic level, but sure it would work in the preprocessor...

[zklaus]

I am not sure there should be bounds here. In the example, the points happen to lie on the same latitude, so on the face of it, bounds seem to have a clear interpretation, but what if the trajectory is not aligned like this. Does it really make sense to build rectangular cells around the points in that case? Should those touch at corners, or would you want to add parameters to the preprocessor to define something like a halo size in all involved directions?