Add basic handling for X,Y,band,time dimensions for transpose and rep…

[jzvolensky]

Hi all,

Corresponding issue which this will cover as well: #243.

I am currently working on adding the align parameter to the resample_spatial function as per the OpenEO processes documentation.

Part of this also involves handling cubes with different dimensions setups. For now, I edited the original fixed transpose to check the shape in the 4 basic dimensions.

Currently I added the align as a dummy parameter which does not do anything but allows us to use the function.

Couple of questions and ideas:

1. Regarding the transpose/handling of the cube dimensions, how could we go about handling other dimensions since there are a lot of possibilities?
2. How should the align parameter be implemented? what should the result be by using different align options?
3. Since we already use ODC reprojection there is parameter anchor available which appears to do a similar operation to align link to the docs

Any feedback, ideas and suggestions are welcome!

Thanks!

[clausmichele]

It would be good to have feedbacks from the VITO team (@jdries @soxofaan) about how they implemented resample_spatial on their side. Do you handle N-dimensional datacubes? Like x,y,bands,t + other dimensions that can be added using add_dimension?

[jdries]

we don't do adding other dimensions, resample spatial just acts upon the x and y dimensions. The t and bands dimension are of course supported and basically left untouched.

[clausmichele]

I didn't mean that you add dimensions within this process, but I was wondering if you support as input datacubes with arbitrary number of dimensions (additional to x and y), which seems so from your reply! Thanks

[jdries]

to be clear: our backend does not at all support adding arbitrary number of dimensions.
We mainly support these combinations for raster cubes:
x,y
x,y,bands
x,y,bands,t
x,y,t

[ValentinaHutter]

Thanks for your inputs on the process and adding the align parameter!
I think, we can keep using the data.openeo here, as this checks the dimensions in the dataset - and for example uses t instead of time if the temporal dimension is called like that.

[jzvolensky]

Hi, sure but using data.openeo limits you to using band , time, x and y only now no? Because the moment you add a dimension and run resample it fails. We are looking into ways to extend the support of other dimensions. We can certainly reuse data.openeo for something but ultimately extend the resample_spatial dim support.

[jzvolensky]

we could do something like this maybe:

    dims = list(data.dims)

    known_dims = [
        data.openeo.band_dims[0],
        data.openeo.temporal_dims[0],
        data.openeo.y_dim,
        data.openeo.x_dim,
    ]

    other_dims = [dim for dim in dims if dim not in known_dims]

    data_cp = data.transpose(*other_dims, *known_dims)

This way we can keep the data.openeo features plus add other dimensions. Of course, some limits and checks would need to be performed on the other dimensions as we cannot realistically allow and support everything.

[jzvolensky]

Hi Valentina!

I updated the latest commit based on your suggestion to keep the data.openeo functionality. The dummy align is set as well to prevent errors from occurring. Let me know what you think!

[ValentinaHutter]

LGTM!

The tests are currently failing because you need to pre-commit run --all-files on the branch.

[ValentinaHutter]

Thanks for the contribution! Will add this to the latest release!