Documentation of the Changes in the COSMO-Model
Version 5.05a / 5.05a_1

Contents:

0. Latest Modifications for 5.05a_1 (on 13.07.18)
1. Changes in the Dynamics
2. Changes for the Radar Forward Operator
3. Porting additional COSMO Parts to GPU
4. Technical Changes and Bug Fixes
5. Changes to the Namelists
6. Changes of Results

0. Latest Modifications for 5.05a_1 (on 31.07.18)

The Bug Fix in the Turbulence scheme (by Matthias Raschendorfer)

Similar to COSMO-Model 5.05_1, this bug fix had also to be implemented into the already existing version 5.05a.

The turbulence and surface transfer schemes of the COSMO and the ICON model have been harmonized during the last years. Unfortunately this harmonization has led to a bug in the subroutine turbtran (turb_transfer.f90) due to, roughly speaking, a confusion of the virtual potential temperature and its gradient, which can lead in very rare cases to a division by zero.

This was experienced earlier in the single precision version, but lately also in a double precision run at DWD.

This bugfix will change the model results slightly.

Back to Contents

1. Changes in the Dynamics

Corrections in the tracer advection scheme

(by Michael Baldauf (DWD) and W. Schneider (Uni Bonn))

With the namelist setting y_scalar_advect = 'BOTTDC2' one can call the 2nd order finite volume scheme of Bott (1989) together with the deformational correction method of Bott (2010). With this deformational correction the full Strang splitting (i.e. in every time step) is (with high probability) not longer necessary and a simple two-step Strang-splitting can be used. Consequently, a gain in efficiency can be achieved. With the local time stepping (by M. Baldauf) numerical reproducibility under domain decomposition is additionally fulfilled together with a slight further improvement in efficiency.

Note, that 'BOTTDC2' is a new option for the namelist variable y_scalar_advect.

Extensions for idealized test cases

(by Michael Baldauf)

Tracer advection tests with a 3D-cone-like initial condition are possible with the

• solid body rotation wind field (itype_anaprof_uv=7),
• and the Smolarkiewicz (1982) test (itype_anaprof_uv=12)

To perform these tests it is necessary to set the

One can further specify the above mentioned tests via the parameters u_infty, v_infty and w_infty in the namelist input for artificial runs. The transported quantity is qv; eventual analytic solutions are stored in qc.

Back to Contents

2. Changes for the Radar Forward Operator

(by Ulrich Blahak, DWD)

Because of changes to the interface of the subroutine calc_dbz_vec() from EMVORADO, two new namelist parameters have been implemented in the group /IOCTL/, to be able to specify different directories for input and output. Before, there was only one directory ydir_mielookup.

These namelist variables are only active, if -DRADARFWO is set during compiling.

• ydir_mielookup_read: directory for reading Mie lookup tables
• ydir_mielookup_write: directory for writing Mie lookup tables

Back to Contents

3. Porting additional COSMO Parts to GPU

(by Xavier Lapillonne, Katherine Silverthorne-Osterried, Tobias Wicky, Guy de Morsier, Philippe Marti)

Ported output, diagnostic and some remaining parts of physics to GPU. Removed all GPU-CPU synchronization in lmorg except for data assimilation. Additional model abort added for non-supported features. Fix for one of the dycore unittest.

Important changes:

• near_surface.f90: is now a module, the routine name is changed to near_surf and is ported to GPU
• src_output.f90: change loop structure in p- and z-interpolation for better mapping to GPU. All code ported
• Some utility routines (e.g. from pp_utilities, numeric_utilities) can now be called from CPU- or GPU-regions. An additional optional argument lacc has been implemented for these routines. With this argument it is possible to activate / deactivate the OpenACC statements of the routine.

Back to Contents

4. Technical Changes and Bug Fixes

Removal of iintegers from module kind_parameters.f90

The KIND-parameter iintegers has been removed from module kind_parameters.f90. Instead, the KIND-parameters i4 and i8 are used now.

Problems in the Single Precision Version

• gscp_cloudice.f90, gscp_graupel.f90:
  Set precipitation fluxes at the ground to zero, if they are less than zqmin (1.0E-15). These changes would affect the results also in double precision, therefore the modifications are encapsulated in ifdef SINGLEPRECISION at the moment.
• turb_transfer.f90:
  Is now fixed with patch 5.05_1.

Allocation of working arrays in Nudging

• src_obs_use_org.f90:
  Subroutine organize_nudging in module src_obs_use_org.f90 has been modified: Conditions for setting of lfirst=.FALSE. have been modified, so that the call to local_info_aux, which allocates some working arrays, is only done once. (PGI compiler in debug mode detected the double allocation in STEP 1).

Deactivating SPPT for GPU runs

• src_setup.f90:
  Have to de-activate SPPT (temporarily) for GPU runs because of a problem.

Back to Contents

5. Changes to the Namelists

Back to Contents

6. Changes of Results

• All modifications are only technical and do not change the results.
• If activated, the new Bott scheme with deformation correction (BOTTDC2) will change the results.

Back to Contents