Masking issue with nearest_s2d interpolation

[sidekock]

I am trying to reproject satellite data to a custom grid using nearest_s2d interpolation.

Since the satellite does not fully cover the domain, the interpolation will take the closest value. I wanted to mask that part in my target, so I calculated my target grid's mask (no data=0, data = 1) and also added a mask to my satellite dataset (all one since the satellite has data in its projection). The resulting regridder however does not mask the area given by the mask in the target. Am I doing something wrong or is this not the way the masking is meant to be used?

[aulemahal]

Hi @sidekock,

Would you be able to share a code example and maybe some figures showing the (erroneous) output of the regridder ? I think the masking is supposed to work as you expected it, and a quick test on my side seems to confirm it.

import xarray as xr
import xesmf as xe

ds = xr.tutorial.open_dataset("air_temperature")

# Input grid is a 2x2 subset in the middle of the grid
dsin = ds.isel(time=0, drop=True).isel(lat=slice(10, 12),  lon=slice(20, 22))

# Output grid
dso = ds.isel(time=0, drop=True)

# No masking
reg1 = xe.Regridder(dsin, dso, method='nearest_s2d')
o1 = reg1(dsin)
o1.air.plot()

# With masking, simply mask a "halo" on the east and southern borders
dso2 = dso.copy()
dso2['mask'] = ((dso2.lon < 300) & (dso2.lat > 30)).T
reg2 = xe.Regridder(dsin, dso2, method='nearest_s2d')
o2 = reg2(dsin)
o2.air.plot()

Figure 1 : No masking

Figure 2 : with masking

Figure 3 : The mask dso2.mask

[sidekock]

Hi @aulemahal

I just moved the files over to a local system to give you some examples; indeed, I get the correct output now. I need to check if this was an issue with some files not being overwritten or if the problem was something else. The versions of xESMF on the HPC are different from the one I am using locally, but since the HPC version is 0.8.8 and mine is 0.8.2, I don't think that is the issue. I'll run some checks tomorrow to see if I can produce the same result on the cluster. Thanks for the quick response!

Kind regards

Simon