Documentation of the Changes in the COSMO-Model
Version 5.4b

Version 5.4b is the next development version, which will not be officially released. Main changes are the convection schemes in blocked data format and the new boundary condition module.

Contents:

1. Implementation of new Boundary Condition Module
2. Implementation of Convection Schemes in Blocked Data Format
3. Changes in C++ DyCore
4. Changes to the Namelists
5. Changes of Results

1. Implementation of new Boundary Condition Module

(by O.Fuhrer, P. Spoerri et al.)

A new module has been implemented in the COSMO-Model, which contains subroutines to set special boundary conditions in the model. With implementing this module and applying the methods in the model, some inconsistent settings of boundary conditions have been modified. This will change the results, but verification at MCH has shown only neutral effects.

Subroutines (methods) included in src_lbc.f90:

• lbc_value:
  Implementation of the copy value boundary condition: A single value is copied to all boundary points.
• lbc_copy:
  Implementation of the copy field boundary condition: A source field is copied to all boundary points.
• lbc_zerograd:
  Implementation of the zero-gradient boundary condition: Points in the boundary zone are copied from corresponding points in the inner domain in order to impose a zero-gradient boundary condition.
• lbc_interpolate:
  Implementation of the interpolate boundary condition: Points in the boundary zone are computed as a linear combination of two boundary fields bd1 and bd2.
• lbc_tendency:
  Implementation of the tendency boundary condition: Points in the boundary zone are computed by adding corresponding values of a boundary tendency field with a factor dts.
• lbc_compute_tendency:
  Computes a boundary tendency field from two boundary fields with a time increment dt.

There are also two internal subroutines (methods):

• check_setup:
  General backend for applying boundary conditions both at u-, v-, and mass-point for a 3d field.
• get_bc_ranges:
  Computes boundary range for a 2d block.

The methods of src_lbc.f90 have been implemented in several modules. Here we document the parts which change the results of the model:

• fast_waves_rk, fast_waves_sc and src_runge_kutta:
  Up to now:
  For the fast-waves solver only the inner-most boundary line has been set in the way it is used by fast_waves_rk or fast_waves_sc. The advection then sees different kinds of boundary conditions, because it uses not only the inner-most line.
  Now tendency conditions are set for the full boundary zone.
  Additionally, the boundary conditions for u, v in fast_waves_sc and are now set after the computation of u, and v. Previously these boundary conditions were set before the computation of u and v creating an inconsistency of the computed values near the boundary.
• src_advection_rk:
  Up to now:
  The subroutines [western|eastern|southern|northern]_boundary set a zero-gradient condition for rho and tke. And lateral_boundaries_zerograd does the same for tracers in use. These routines were called once before starting the advection operators.
  This is now done by the subroutine advection_pd_lbc. But this routine is called before every advection operator, so that every operator works with updated boundary conditions.
• lmorg:
  A new subroutine rlx_bc has been implemented which is called after the relaxation. It sets (again) the values on the boundaries, but the values for w are changed, because they are now set with a zerograd-condition, which was not the case before.

Another change related to boundaries has been implemented in src_relaxation:
The computation of the saturation adjustment has been restricted to the inner domain: The saturation adjustment on the boundaries has already be computed in the dynamics, so it does not have to be done again here.

Back to Contents

2. Implementation of Convection Schemes in Blocked Data Format

(by Xavier Lapillonne, Jochen Förstner, Ulrich Schättler)

The two convection schemes used in COSMO (Tiedtke and shallow convection) and their interfaces have been adapted to run in the blocked data format. In addition the IFS scheme by Tiedtke-Bechtold, which is used in ICON, has also been implemented. The following new source modules are now available:

• conv_interface.f90: Interface module with subroutines
  

  conv_init	basic initializations of the schemes
  conv_init_copy	set up the copy-list for blocked data
  conv_prepare	perform some calculations necessary on ijk structure
  conv_organize	call the chosen schemes
  conv_finalize	clean up
  conv_in_wkarr_alloc	allocation of local memory for the interface
  conv_in_wkarr_dealloc	deallocation of local memory for the interface (used with _OPENACC)

  All routines are written that they work for every scheme. Which scheme really is used can be determined by itype_conv and is taken care of in the routines.
• conv_data.f90:
  Module that defines data used in the schemes. This modules also defines local memory for the schemes, which can be used when _OPENACC is specified. Two subroutines conv_wkarr_alloc and conv_wkarr_dealloc are provided to manage the local memory if _OPENACC is used.
• conv_shallow.f90:
  Computational part of former module src_conv_shallow, adapted to the blocked data structure.
• conv_tiedtke.f90:
  Computational part of former module src_conv_tiedtke, adapted to the blocked data structure.
• conv_cumaster.f90:
  (and additional modules: conv_cuascn.f90, conv_cudescn.f90, conv_cuflxtends.f90, conv_cufunctions.f90, conv_cuinit.f90, conv_cuparameters.f90) Modules for Tiedtke-Bechtold convection scheme.

The convection scheme is chosen with the namelist variable itype_conv, which has the following possibilities:

0. old Tiedtke scheme
1. not available
2. Tiedtke-Bechtold scheme (from IFS) !! NEW Option
3. shallow convection scheme

Back to Contents

3. Changes in C++ DyCore

(by Pascal Spoerri)

Take imode_turb from turb_data now (was in data_runcontrol before).

Back to Contents

4. Changes to the Namelists

There is a new option for itype_conv: with "2" the Tiedtke-Bechtold scheme is chosen. And there are new switches in /PHYCTL/ and in /TUNING/. The new variables in /TUNING/ are only in effect, when Tiedtke-Bechtold scheme is chosen.

Back to Contents

5. Changes of Results

The implementation of the convection schemes in blocked data format are (in principle) bit reproducible compared to the former implementation in src_conv_shallow and src_conv_tiedtke. Only the Cray-compiler with option hflex_mp=conservative changes the results. With hflex_mp=intolerant the results are really bit-reproducible (also with GNU Compiler).

With the implementation of the new boundary condition module also some boundary conditions have been slightly modified compared to the former version, to provide more consistent boundary conditions. This changes the results, but from a meteorological point of view the differences are neutral.

Back to Contents