interpolation method

[wwieder]

Hey, sorry to ping you out of the blue with this comment, but Rosie shared your repo. The work here looks good! Very encouraging.

I wanted to ask about the interpolation method being used? I'll note that I'm not an expert on such matters, but the CLM diagnostics package uses area-conservative regridding to go from the ne30 to f09 grids, whereas it looks like currently this is set up to do bilinear interpolation. Not sure what's right, but I wanted to start a discussion.

xesmf_clm_fates_diagnostic/src/xesmf_clm_fates_diagnostic/regrid_and_plot_w_timeseries.py

Line 65 in f7c3e69

regridder_obsdata = xe.Regridder(ds, ds_out_gpp, 'bilinear', periodic=True)

[maritsandstad]

Thanks for commenting. Yes, it uses bilinear for the moment. There was no deep thought that went into that, but sometimes the requirements for area conserving are a bit stricter, so first order this was going to just work quickly and easily so plots can be reviewed. Happy to test out area conserving as well, I agree that it's probably more correct, but I find (being no expert in anything) that often for things like this, where absolute precision might not be the primary goal, bilinear is good enough.

[maritsandstad]

@wwieder I've now updated to use conservative_normed which seemed to me the most appropriate alternative, to retain the area conservation, but avoid coastal artefacts which can be a problem with conservative regridding, particularly for land only data. Testing with this seems alright to me, but I don't really have great intuition on what the data should look like, so I can't guarantee, anyways it's really easy to change, and it can even be made a dynamical option, but I doubt that would be worth it for most use cases...

[maritsandstad]

New update, so I had a bug, and actually the conservative_normed requires weight values on boundaries that I don't have, so it's going to be bilinear for now...

[wwieder]

Hi Marit,

I ended up taking your code and doing a deep dive into the 'best' way to conserve fluxes and make nice maps.
The results are being integrated into our CUPiD workflow (through a branch of the ADF, atmosphere diagnostic package being developed at NCAR. You can see the PR here https://github.com/wwieder/ADF/tree/clm-test.

Notably, I'm relying heavily on your script, which you'll recognize here https://github.com/wwieder/ADF/blob/clm-test/scripts/regridding/regrid_se_to_fv.py.

I also spend a long time looking at coastlines and regional sums using this notebook https://github.com/wwieder/ADF/blob/clm-test/scripts/regridding/regrid_conservative.ipynb.

The short end of the story is that I'm:

• weighting source fluxes by landfrac,
• using conservative regridding (and a weight file created accordingly)
• dividing the regridded flux by the regridded landfrac (note this results in chubby coastlines on the destination grid)
• calculating regional or global fluxes (using regridded landfrac and destination area)
• mapping after applying the destination grid landmask.