Vince's suggestion #1

See notes below:

To test something similar to the old code, in the call to
xpyp_term_ta_pdf_rhs_all from ONLY advance_xp2_xpyp_module.F90, replace
l_upwind_xpyp_ta with a hardwired .false.  This will force centered
differencing on the rhs, while the lhs will still depend on the
l_upwind_xpyp_ta flag.  Then set l_upwind_xpyp_ta flag = .true.

When Kai compared l_upwind_xpyp_ta = .true. (lhs and rhs) to
l_upwind_xpyp_ta = .false. (lhs and rhs), he didn't find much change in
global precipitation rate.   Therefore, mixing .true. and .false.
settings for l_upwind_xpyp_ta like this may not revert to the CLUBBv1
results, but it is worth a try.

modified:   components/cam/src/physics/clubb/advance_xp2_xpyp_module.F90

components/cam/src/physics/clubb/advance_xp2_xpyp_module.F90

@@ -2310,7 +2310,12 @@ subroutine xp2_xpyp_uv_rhs( solve_type, dt, l_iter, & ! In
                                    rho_ds_zt(:),                      & ! Intent(in)
                                    invrs_rho_ds_zm(:),                & ! Intent(in)
                                    gr%invrs_dzm(:),                   & ! Intent(in)
-                                   l_upwind_xpyp_ta,                  & ! Intent(in)
+!!!=== T1 === 
+!!! force centered differencing on the rhs, while the lhs will still depend on the l_upwind_xpyp_ta flag.
+!!!
+!!!                                l_upwind_xpyp_ta,                  & ! Intent(in)
+                                   .False.,                  & ! Intent(in)
+!!!=== T1 === 
                                    sgn_turbulent_vel(:),              & ! Intent(in)
                                    term_wpxp2_explicit_zm(:),         & ! Intent(in)
                                    rho_ds_zm(:),                      & ! Intent(in)