Subsetting roms grids removes needed 3d variables

[AmyMacFadyen]

What went wrong?

When I use ds.xsg.subset_bbox not all the data_vars make it through intact. The ones I needed were Cs_r and Cs_w

Operating System

Windows

Version

unsure

Python Version

3.10.18

Code to Reproduce

ds = ds.xsg.subset_bbox(bounds)

look at ds.data_vars before and after

Errors, Traceback, and Logs

[AmyMacFadyen]

@ChrisBarker-NOAA If you want to try this using libGOODS here's a code snippet. But I mainly used it just to get a xarray ds to work with in an env with xarray-subset-grid.

from libgoods.models import WCOFS
w = WCOFS(access='zarr')
w.open_ds()
ds = w.ds # here's the xarray ds to work with

Now if you look at ds.data_vars you'll see a VERY long list which includes Cs_r, Cs_w
Then do:

bounds = (-130, 42, -128, 45)
ds2 = ds.xsg.subset_bbox(bounds)

ds2.data_vars is a much smaller list:
    mask_psi  (eta_psi, xi_psi) float64 51kB dask.array<chunksize=(94, 68), meta=np.ndarray>
    Pair      (ocean_time, eta_rho, xi_rho) float32 3MB dask.array<chunksize=(1, 95, 69), meta=np.ndarray>
    Uwind     (ocean_time, eta_rho, xi_rho) float32 3MB dask.array<chunksize=(1, 95, 69), meta=np.ndarray>
    Vwind     (ocean_time, eta_rho, xi_rho) float32 3MB dask.array<chunksize=(1, 95, 69), meta=np.ndarray>
    angle     (eta_rho, xi_rho) float64 52kB dask.array<chunksize=(95, 69), meta=np.ndarray>
    f         (eta_rho, xi_rho) float64 52kB dask.array<chunksize=(95, 69), meta=np.ndarray>
    h         (eta_rho, xi_rho) float64 52kB dask.array<chunksize=(95, 69), meta=np.ndarray>
    mask_rho  (eta_rho, xi_rho) float64 52kB dask.array<chunksize=(95, 69), meta=np.ndarray>
    pm        (eta_rho, xi_rho) float64 52kB dask.array<chunksize=(95, 69), meta=np.ndarray>
    pn        (eta_rho, xi_rho) float64 52kB dask.array<chunksize=(95, 69), meta=np.ndarray>
    salt      (ocean_time, s_rho, eta_rho, xi_rho) float32 134MB dask.array<chunksize=(1, 1, 95, 69), meta=np.ndarray>
    temp      (ocean_time, s_rho, eta_rho, xi_rho) float32 134MB dask.array<chunksize=(1, 1, 95, 69), meta=np.ndarray>
    w         (ocean_time, s_w, eta_rho, xi_rho) float32 138MB dask.array<chunksize=(1, 1, 95, 69), meta=np.ndarray>
    zeta      (ocean_time, eta_rho, xi_rho) float32 3MB dask.array<chunksize=(1, 95, 69), meta=np.ndarray>
    mask_u    (eta_u, xi_u) float64 52kB dask.array<chunksize=(95, 68), meta=np.ndarray>
    u         (ocean_time, s_rho, eta_u, xi_u) float32 132MB dask.array<chunksize=(1, 1, 95, 68), meta=np.ndarray>
    ubar      (ocean_time, eta_u, xi_u) float32 3MB dask.array<chunksize=(1, 95, 68), meta=np.ndarray>
    mask_v    (eta_v, xi_v) float64 52kB dask.array<chunksize=(94, 69), meta=np.ndarray>
    v         (ocean_time, s_rho, eta_v, xi_v) float32 133MB dask.array<chunksize=(1, 1, 94, 69), meta=np.ndarray>
    vbar      (ocean_time, eta_v, xi_v) float32 3MB dask.array<chunksize=(1, 94, 69), meta=np.ndarray>
    grid      float64 8B ...

[ChrisBarker-NOAA]

Hmm -- I would think that subset_bbox shouldn't change what variables are available, unless maybe there's no way for them to be useful / correct for the subset.

I'll take a look.

[ChrisBarker-NOAA]

I have not looked yet :-(

Still needs attention

[ChrisBarker-NOAA]

Confirmed that this is still an issue.

To proceed -- it would be great to have a small, but complete example file to use for testing.

I think there are some examples in the repo somewhere, but not sure if they are complete.

@AmyMacFadyen: any chance you have (or could make) a small example?

[AmyMacFadyen]

Chris -- can you look at this soon? We can't write valid ROMS 3d subsets from xsg due to the horizontal subsetting dropping variables. I can't subset you an example file w/o causing the issue. However, you can grab a single time step CBOFS one from the libgoods-cache files.