Issue with grid.transform

[mkyoungs]

Hi, I am trying to calculate thickness weighted u velocity in MITgcm and transform it back to depth space using the time averaged density structure. See the code below.
zexpand = coords.Z.expand_dims(dim={"i": 1080,'j':310})
ztrans = grid.transform(zexpand, 'Z', coords.layer_1RHO_center.values, target_data=(ds.dens-1000))
ztrans = ztrans.rename(dens='l1_c')
ures = grid.interp(ds.LaUH1RHO.mean(dim='time')/ds.LaHw1RHO.mean(dim='time'),'X')
ures_z = grid.transform(ures,'l1_c',coords.Z,target_data=ztrans)

Everything seems to work right until the last step, when I get the error:
2607 def transform(self, da, axis, target, **kwargs):
2608 """Convert an array of data to new 1D-coordinates along 'axis'.
2609 The method takes a multidimensional array of data 'da' and
2610 transforms it onto another data_array 'target_data' in the
(...)
2688
2689 """
-> 2691 ax = self.axes[axis]
2692 return ax.transform(da, target, **kwargs)

KeyError: 'l1_c'

I am not sure what this means or how to fix it. Does anyone have any advice?

[rcaneill]

Hello Madeleine,

I think that I know what is happening: the xgcm grid contains yours axes (usually X, Y, Z, T), and these are the axes that you refer to when using the grid methods.

Here when you use grid.transform(ures,'l1_c',coords.Z,target_data=ztrans), you tell your grid to use the axis l1_c, which you have not defined in your grid.
So I guess that you will need to redefine your grid before doing the last transformation, e.g.

ds['ztrans'] = ztrans
ds['ures'] = ures
grid_with_density_axis = xgcm.Grid(ds, coords={'S':{'center':'l1_c'}}, metrics={'S':['nae_of_density_metric']})
# Use S for density axis, S as in sigma

And then use grid_with_density_axis.transform(ures,'S',coords.Z,target_data=ztrans)

I haven't tested my code snippet, so you'll probably need to adapt slightly

[mkyoungs]

Thanks for the help! I think the issue may stem from the fact that although the 1RHO coordinate (l1_c) and the adjacent outer points are categorized as grid coordinates in ds, they are not when I create the grid object. How would I just add the 1RHO coordinate to the grid object manually to deal with the lack of automatic reading in? The same issue seems to be showing up in the documentation in the MITgcm example in the section 'Creating the grid object. ' It claims to create the 1RHO axis in the grid object, but the output in the lines above don't include it.

[rcaneill]

Could you show the attributes you have for l1_c ?

Otherwise, you can specify by hand the axes, by giving the coords argument to xgcm.Grid: https://xgcm.readthedocs.io/en/latest/api.html#grid

[mkyoungs]

I can specify it manually, but then I would have to specify every axis manually. I am confused about how to do that with all of the metrics involved. Shouldn't this be updating automatically? See the dataset output. It seems to recognize the density coordinate system.

[mkyoungs]

I realized I may not have answered your question. Here are the attributes.

[rcaneill]

So the attributes don't have the mandatory axis attribute. You can add it by hand before calling the grid:

ds.l1_c.attrs['axis'] = 'S'
ds.l1_i.attrs['axis'] = 'S'
ds.l1_b.attrs['axis'] = 'S'

You will also need to add the c_grid_axis_shift attribute to -0.5 for left points, and 0.5 for right points (or whatever for inner/outer point), which I guess looking at your 1st snapshot would be:

ds.l1_i.attrs['c_grid_axis_shift'] = 0.5
ds.l1_b.attrs['c_grid_axis_shift'] = 0.5

Here is the link to the doc:
https://xgcm.readthedocs.io/en/latest/grids.html#comodo-data

[rcaneill]

Regarding metrics, I don't think that xgcm parses them automatically, you have to provide them, see Cell [4] in https://xgcm.readthedocs.io/en/latest/grid_metrics.html#Using-metrics-with-xgcm

metrics = {
    ('X',): ['dxC', 'dxG'], # X distances
    ('Y',): ['dyC', 'dyG'], # Y distances
    ('Z',): ['drW', 'drS', 'drC'], # Z distances
    ('X', 'Y'): ['rA', 'rAz', 'rAs', 'rAw'] # Areas
}
grid = Grid(ds, metrics=metrics)

And you will need to also add the metric for the density coordinate

[mkyoungs]

Yes, that worked! Thank you.