Add functionality to write an OutputVar to a netcdf file

[oalcabes]

Currently we can read an OutputVar from a netcdf file, but we can't write to a netcdf file. This is super important and helpful for folks doing post-processing on CliMA data.

[AmanKashyap0807]

Hi @oalcabes and @ph-kev, I am Aman, a 3rd year student from India. I have been exploring the CliMA repositories recently as I am interested in contributing and potentially proposing a project for Julia GSoC 2026.

I am very interested in solving this issue. I noticed it is already assigned to @ph-kev, but I would love to work on this if it's still available!
My high level idea is to implement a write_var function (using NCDatasets.jl) that takes the attributes, dims, and data from the OutputVar and writes them into a standard NetCDF file. I am still getting familiar with the specifics of the NetCDF format, but I think I can figure out the mapping by looking at how the current reading logic works in reverse.

Could you let me know if I can start on this, or if there are any specific design requirements I should follow? If this is already being handled, please let me know if there are other issues I can help with.

[ph-kev]

Hi @AmanKashyap0807, this is definitely an issue that you can tackle! One thing that I want to note is that the NetCDF files produced from ClimaDiagnostics does not follow the typical conventions of a NetCDF file and there is some liberty that ClimaAnalysis does with the preprocessing of the OutputVar which might make this issue a little difficult to tackle. In general, we will not expect conversion from OutputVar to NetCDF files to preserve all the metadata and data exactly, but this is okay. Lastly, there might be better function name than write_var, since we may want to write to another file format in the future. Let me know if there is anything else that you need from me to get started on this issue.

[AmanKashyap0807]

Thank you for assigning it to me and your support.
I am starting working on this, I will ask if i get stuck anywhere.

[AmanKashyap0807]

While testing, I also looked into scalar (0D) round-trips. I left the related test commented out in the PR, since the failure originates from the existing read constructor rather than the new write logic, and felt out of scope for this issue. I added a short note in the PR description for visibility.

Happy to follow up on this separately if that would be useful.

[AmanKashyap0807]

Hi @ph-kev @oalcabes,

Just a quick update, I have opened PR #370 to fix this issue, but the CI workflows are currently waiting for maintainer approval. When you get a chance, could you please review and approve the workflows so the necessary tests can be executed?

Let me know if you have any feedback or any change required in pr.