Add denormalisation routine for the gw_convect_dp net.

src/physics/cam/gw_ml.F90

@@ -122,11 +122,8 @@ subroutine gw_drag_convect_dp_ml(ncol, dt, &
   call torch_delete(net_input_tensors)
   call torch_delete(net_output_tensors)
 
-  ! Extract data and return
-  do i = 1, ncol
-      utgw(i, :) = net_outputs(1:pver, i)
-      vtgw(i, :) = net_outputs(pver+1:2*pver, i)
-  end do
+  ! Denormalise outputs and extract the data
+  call denormalise_data(ncol, utgw, vtgw, net_outputs)
 
 end subroutine gw_drag_convect_dp_ml
 
@@ -393,5 +390,20 @@ subroutine normalise_data(ncol, u, v, t, dse, nm, netdt, zm, rhoi, ps, lat, lon,
 
 end subroutine normalise_data
 
+subroutine denormalise_data(ncol, utgw, vtgw, nn_output)
+
+  integer, intent(in) :: ncol
+  real(r8), intent(out) :: utgw(ncol,pver), vtgw(ncol,pver)
+  real(r8), intent(in) :: nn_output(2*pver, ncol)
+
+  integer :: i
+
+  ! Extract data, denormalise, and deconcatenate from NN output tensor
+  do i = 1, ncol
+      utgw(i, :) = (nn_output(1:pver, i) * utgw_std(:)) + utgw_mean(:)
+      vtgw(i, :) = (nn_output(pver+1:2*pver, i) * vtgw_std(:)) + vtgw_mean(:)
+  end do
+
+end subroutine denormalise_data
 
 end module gw_ml