question about stencils

[alanjhewitt]

there is something called "tile_stencil" in jule_exp_kernel_mod.F90 that is used in the calculation of frac_surft. it is not present in the ukca equivalent kernel when calculating frac_surft.

i could do with some help getting tile_stencil into the ukca kernel. for some reason, the jules kernel does not have an associated algorithm layer

From Jules jules_exp_kernel_mod.F90

do l = 1, land_field
coast%fland(l) = flandg(ainfo%land_index(l),1)
do n = 1, n_land_tile
! Jules requires fractions with respect to the land area
ainfo%frac_surft(l, n) = real(tile_fraction(tile_stencil(1,1,ainfo%land_index(l))+n-1), r_um) &
/ coast%fland(l)
end do
end do

From UKCA aerosol_ukca_kernel_mod.F90

_do i = 1, n_land_pts
  do m = 1, n_land_tile

    frac_surft( i, m ) =                                                   &
     real( tile_fraction( map_tile( 1, land_index(i) ) + m - 1 ), r_um ) / &
     frac_land( land_index(i),1 )

  end do
end do_

There is also some stencil code in the following modules (the first of which calls the jules kernel):
science/um_physics_interface/source/psy/psykal_lite_phys_mod.F90
science/gungho/source/psy/psykal_lite_mod.F90

[stevemullerworth]

It's not clear whether you need a stencil here. The Jules calculation seems to use only element 1 of the stencil (that's the second index of the tile_stencil array). Element 1 of the stencil is the central point of the stencil, so it is the same as map_tile. You can think of map_tile being a stencil of extent 1.

While stencils are not yet supported for domain kernels (hence the need for psykal-lite code), normal maps are supported.

[alanjhewitt]

I have now solved the problem. This was a red herring, the stencils were required for Jules because of halos and not for UKCA.