Tutorial on xWRF

[lpilz]

What is your issue?

The aim of this issue is to track the progress in creating a tutorial for xWRF. Here the start of a list of features which are to be presented. Please feel free to add to this list - I'll work on implementing this over coming days.

• general parsing/coordinate transformation (what does xwrf do?)
• interface to metpy via unit CF-conventions and pint
• destaggering data using xgcm
• vertically interpolating data using xgcm
• plotting

[lpilz]

Unfortunately, I can't find a good way to interpolate the hybrid sigma coordinate from WRF to pressure using xgcm. There is apparently one method using metpy.interpolate.log_interpolate_1d, but I'm not sure if this really treats the coordinate as hybrid or as a simple log pressure coordinate. Does anybody have a good idea how to solve this? @jthielen maybe?

[jthielen]

Unfortunately, I can't find a good way to interpolate the hybrid sigma coordinate from WRF to pressure using xgcm.

Yeah, I had an idea for that (back in...2020...), but haven't been able to contribute a "log interpolate" version of xgcm's linear vertical interpolation. That would be a good thing to push forward, but I certainly wouldn't want to hold off on xWRF's tutorial for it either!

There is apparently one method using metpy.interpolate.log_interpolate_1d, but I'm not sure if this really treats the coordinate as hybrid or as a simple log pressure coordinate. Does anybody have a good idea how to solve this? @jthielen maybe?

log_interpolate_1d is generic to the true underlying "dimension coordinate," it just works locally in 1D with a generic field from which you want to extract isosurfaces. So, yes, it should work for sigma and hybrid sigma to pressure coordinate interpolation, as in this example. That being said, metpy.interpolate is a bit of a neglected module of MetPy when it comes to xarray compatibility, so you'll have to do some more manual workarounds (or use xarray.apply_ufunc in a clever way).

[kmpaul]

@dopplershift or @dcamron could probably speak to what metpy.interpolate.log_interpolate_1d is doing and whether it will work for you...

[dopplershift]

@jthielen is pretty knowledgeable on MetPy's inner workings too. For completeness, log_interpolate_1d is doing a simple linear interpolation along a single axis (e.g. z or p), but first log-transforms the passed in coordinates for that axis. @jthielen already linked to the example I was thinking of.

@lpilz MetPy isn't working with any of the hybrid/sigma values. Instead, it's taking pressure columns across those fixed hybrid surfaces, and then interpolating within that column (with a log transform) to a certain pressure value.

I echo the sentiments that it would be nice to have a version that works more with xarray. I wouldn't mind someone opening a feature request for that if you'd find it useful.

[dcherian]

You could

1. Set the appropriate formula_terms attribute
2. Teach cf_xarray to interpret the formula
3. And then call ds.cf.decode_vertical_coords() in your postprocess function to create the pressure levels.

Plugging a log transform into xgcm.transform should be fairly easy too, though it would require numba IIRC.

[jthielen]

You could

1. Set the appropriate formula_terms attribute
2. Teach cf_xarray to interpret the formula
3. And then call ds.cf.decode_vertical_coords() in your postprocess function to create the pressure levels.

I don't think this would gain us much, since the goal isn't to obtain pressure values from the coordinates (we already have those in a data variable), but instead to transform the coordinate isosurfaces from hybrid-sigma to pressure. I could be misinterpreting what decode_vertical_coords is doing, though!

Plugging a log transform into xgcm.transform should be fairly easy too, though it would require numba IIRC.

It would be a bit hacky to try a log transform externally (i.e., using a ln(p) data variable and specified levels, then taking exp of the output coordinates), but if that transform could be done internally (and a method="log" added), I think that should be fairly easy. I suppose I put that off long enough from xgcm/xgcm#222 (comment), I could try putting a PR together tomorrow, unless someone wanted to get to it first?

[dcherian]

(we already have those in a data variable),

Ah didn't realize this.

[lpilz]

Thanks for the input, everybody. Personally, I don't deem this issue to be a blocker for the tutorial, so IMO we should just publish what we have for now and add the vertical interpolation stuff once it is implemented in xgcm. This would help us get the release out of the door :) What do you think?

[kmpaul]

I agree. We can revisit later if necessary. I think this is good.

Thanks for doing this @lpilz!

[jthielen]

Personally, I don't deem this issue to be a blocker for the tutorial, so IMO we should just publish what we have for now and add the vertical interpolation stuff once it is implemented in xgcm. This would help us get the release out of the door :) What do you think?

I concur! In the mean time, to keep track of this new feature upstream, see xgcm/xgcm#483.

[jbusecke]

Just wanted to ping this here. If this is time sensitive, I could probably merge xgcm/xgcm#483 soon and we can postpone the corresponding documentation I requested there to another PR. Or do you require a release?

Seems like this would produce some great examples that could be linked/reused in the xgcm docs?