PeterHjortLauritzen
commented
2 years ago Solution: @JulioTBacmeister
Set LANDFRAC=0 only when elevation is 0. That should remove inland LANDFRAC=0 points (unless inland lakes are at zero elevation!)
This is done with the following code
landfrac_local = 1.0_r8 do ip=1,6 do j=1,ncube do i=1,ncube if (ABS(terr(i,j,ip))<1.0E-12_r8.and.ABS(landfrac(i,j,ip))<1E-12_r8) then landfrac_local(i,j,ip) = 0.0_r8 end if end do end do end do
This leads to the following landfraction (mapped to the ne30pg3 grid):
Looks like the inland LANDFRAC=0 problem has been solved.
Please note that the scope of this work is to have zero elevation in atmosphere model in coastal regions (for better coupling between atmosphere and ocean model). This approach does not make sure that elevation is at the height of inland lakes (e.g. Caspian Sea, Great Lakes).
The difference between not smoothing over ocean (using the above algorithm) and smoothing everywhere is here in terms of PHIS:
Difference between not smoothing over ocean (using the above algorithm) and namelist default (/glade/p/cesmdata/inputdata/atm/cam/topo/se/ne30pg3_nc3000_Co060_Fi001_PF_nullRR_Nsw042_20171014.nc"):
Differences PHIS at coast lines is obviously the zeroing out of Laplacian where there is ocean. The inland differences look like the differences in smoothing operator (over steep orography very minor changes in smoothing operator will show larger differences).
Latest and greatest topo file (with zeroing out smoothing over ocean) is here:
/glade/p/cgd/amp/pel/topo/files/ne30pg3_gmted2010_bedmachine_nc3000_Laplace0100_20220520.nc
Plot below shows the difference between smoothing everywhere and not smoothing when LANDFRAC=0 with ncube=3000 intermediate cubed-sphere grid (~3km):
and ncube=540 (~12km)
There are significant differences inland. Why?
The problem: Binning from ~1km LANDFRAC to 3km cubed-sphere picks up on isolated "inland" ocean points that are not picked up when using ncube=540 intermediate cubed-sphere grid. See this regional plots over Himalayas of LANDFRAC when using ncube=3000:
and ncube=540:
There are sporadic LANDFRAC=0 points (dark blue points) where the Laplacian smoother will not be applied in the ncube=3000 LANDFRAC but close to none in ncube=540 LANDFRAC.