oocgcm does not trivially apply to mitgcm outputs

[serazing]

I have worked on adding the grid of the mitgcm, it requires a few hack to make it work with the current API:

• I do not have a mask with the mitgcm files so I used the 'cell_vertical_fraction_at_t_location' instead. It seems to work for the moment even if it is not the best solution
• I skipped the chunk checking
• I have relaxed the condition on the required array and have added a function to generate the the cell_size if needed

I have successfully computed some gradients but the laplacian does not work. I think there is some work to do to test if oocgcm apply well to different models.

I have pushed a branch to keep working on it, @rabernat you might be interested in.

[rabernat]

I would be thrilled if you could get this working! I would be happy to help as much as possible.

My preferred approach is to have oocgcm rely only on the dataset metadata, without special logic for different models. So we may need to change certain aspects of xmitgcm metadata construction.

As for the mask, we could easily add mask variables to xmitgcm based on the hfac variables. If you let me know what names / attributes they should have, I will make a PR asap.

[bderembl]

Is your implementation for mitgcm available somewhere? I'd be glad to try it.
I've been using Ryan's package so far which works great. But having one single tool for nemo and mitgcm would be easier for me.

[serazing]

@bderembl, I have created a dedicated branch on the github repo: mitgcm. You are welcome to try and contribute to it.

@rabernat, NEMO uses the following binary mask for the C grid:

_nemo_keymap_byte_mask = {
    'tmask': 'sea_binary_mask_at_t_location',
    'umask': 'sea_binary_mask_at_u_location',
    'vmask': 'sea_binary_mask_at_v_location',
    'fmask': 'sea_binary_mask_at_f_location',

What I have done for the MITGCM part, is to use hFacC, hFacV and hFacS as binary masks (which are not really binaries):

        self.variables["sea_binary_mask_at_t_location"] = \
                        self._get(self.coordinate_file, "hFacC",
                                  chunks=self.chunks, grid_location='t')
        self.variables["sea_binary_mask_at_u_location"] = \
                        self._get(self.coordinate_file, "hFacW",
                                  chunks=self.chunks, grid_location='u')
        self.variables["sea_binary_mask_at_v_location"] = \
                        self._get(self.coordinate_file, "hFacS",
                                  chunks=self.chunks, grid_location='v')

If you can generate binary masks in xmitgcm, it would be great. The names do not matter since they are easily changeable inside the code. What is important, I think, is to have the correct comodo-like attributes (ex: "sea_binary_mask_at_t_location")

[bderembl]

thanks.
for some reason, mitgcm doesn't use the same naming convention for netcdf and binary outputs (e.g. HFacW in netcdf and hFacW in binary).

In your first post, you mention that your successfully computed gradients, etc; what kind of outputs were you using? Can you post your script?

[serazing]

It seems feasible to add a if test to read either HFacW or hFacW, probably not the best solution though. Note that using xmitgcm to read binaries returns a Dataset with hFacW.

[rabernat]

@bderembl: are you aware of xmitgcm?
http://xmitgcm.readthedocs.io/en/latest/
xmitgcm is a python package for reading MITgcm binary MDS files into xarray data structures.

It is complementary to oocgcm and provides a good way to load your data if you use MDS format. There are several of us working on and maintaining xmitgcm, and I think you will really like it.

Regarding the variable names, we have explicitly tried to use the same naming conventions in our generated xarray datasets as MITgcm's netCDF output. But if you find further inconsistencies, please don't hesitate to open a new issue: https://github.com/xgcm/xmitgcm/issues

[rabernat]

I should also mention that we are working on xgcm:
http://xgcm.readthedocs.io/en/latest/
whose goals are much more overlapping with oocgcm. We hope to some day merge these packages. For now xgcm is very primitive, but it provides some useful low-level routines for working with mitgcm output. Here is an example notebook:
https://gist.github.com/rabernat/ff43c9726803568ccabd2ed5ac6048e4

[bderembl]

thanks.
I wasn't aware of xgcm and it seems to complement xmitgcm really well.

about the naming issue, I think your notation is consistent with the name of the binary files. the issue is probably in mitgcm itself (cf. mnc_cw_write_grid_info.F and write_grid.F)