Grid imprinting in write component fields using NSTD

[LarissaReames-NOAA]

Description

When comparing tmpsfc from ATM and sst from OCN, which ideally should be almost identical, @pjpegion noticed very obvious grid imprinting originating in the tmpsfc field.

Currently, tmpsfc uses the default surface bundle interpolation method which is nearest source to destination, and when Phil tried changing this manually to bilinear, the imprinting disappeared.

The problem with bilinear interpolation for surface fields though is the potential for mismatches between immutable surface properties (soil, land type, land/water) and the interpolated field. I think the solution here, which is used in chgres_cube, is using masking in the ESMF interpolation steps when a surface field uses bilinear interpolation.

To Reproduce:

Run any coupled configuration and directly compare output from ATM (tmpsfc) and OCN (sst) output.

Additional context

N/A

Output

Just one more image showing the difference between tmpsfc when using NSTD vs. Bilinear

[DeniseWorthen]

@LarissaReames-NOAA There is also the method to use bilinear_nstod, which maps bilinearly and then fills in any points not mapped w/ the nstod value. We use it in CMEPS like so

            regridmethod=ESMF_REGRIDMETHOD_BILINEAR, &
            extrapMethod=ESMF_EXTRAPMETHOD_NEAREST_STOD, &

[LarissaReames-NOAA]

@LarissaReames-NOAA There is also the method to use bilinear_nstod, which maps bilinearly and then fills in any points not mapped w/ the nstod value. We use it in CMEPS like so

            regridmethod=ESMF_REGRIDMETHOD_BILINEAR, &
            extrapMethod=ESMF_EXTRAPMETHOD_NEAREST_STOD, &

I did not know about that method, so thanks for mentioning it. I think that could be useful for islands or isolated lakes, which are always tricky when doing masked interpolation.