Documentation of the Changes in COSMO-Model 4.4

This new version contains changes in the Tiedtke convection scheme (up to now only in affect with the Leapfrog scheme), some optimizations in the advection of the Runge-Kutta scheme, some optimizations for vector processors and (again) technical adaptations and bug fixes.

Contents:

1. Documentation of the Changes
• Tiedtke Convection
• Advection in Runge-Kutta Dynamics
• Sub-grid Scale Orography Scheme
• Vectorization
• Technical Adaptations
2. Summary of new and obsolete Namelist Variables
3. Changes of Results

1. Documentation of the Changes

Tiedtke Convection (by Dmitrii Mironov)

A number of changes in the Tiedtke cumulus convection scheme, that are related to the treatment of convective cloud condensate as a mixed water-ice phase and of detrained convective cloud condensate are made. These are:

• the relative humidity threshold for the evaporation of convective precipitation in the sub-cloud layer is set to 80% (95% in the previous operational formulation),
• cloud water-cloud ice mixed phase in convective clouds is allowed over a certain temperature range (in the previous operational formulation, convective cloud condensate is either water or ice, depending upon the air temperature being higher or lower than the water freezing point), and
• detrained convective cloud condensate is saved as tendencies of the cloud water and of the cloud ice, these tendencies are then passed to the COSMO-model physics and dynamics schemes for further processing (the previous operational formulation assumes instantaneous evaporation of detrained convective cloud condensate).

The fraction of cloud ice is computed as a function of temperature. The lower and the upper bounds of the temperature range, where convective cloud water and convective cloud ice are allowed to co-exist and the form of the interpolation function are the same as in the ECMWF IFS (IFS Documentation 2006).

These changes affect the results of the Leapfrog applications in combination with the Tiedtke convection schemes.

Back to Contents

Advection in Runge-Kutta Dynamics (by Michael Baldauf)

The structure of the advection routines in the Runge-Kutta dynamics has been changed for higher efficiency.

These changes affect the results of the Runge-Kutta applications.

Back to Contents

Sub-grid Scale Orography Scheme (by Jan-Peter Schulz)

The sub-grid scale orography (SSO) scheme by Lott and Miller (1997) has been implemented in the COSMO model. It is also included in the DWD global model and works here well. The scheme deals explicitly with a low-level flow which is blocked when the sub-grid scale orography is sufficiently high. For this blocked flow separation occurs at the mountain flanks, resulting in a form drag. The upper part of the low-level flow goes over the orography and generates gravity waves.

Verification at DWD had shown that the forecasted surface pressure in the COSMO-EU model shows a systematic bias. In particular, in wintertime high pressure systems tend to develop a positive pressure bias (by 1-2 hPa after 48h), low pressure systems a negative bias ("highs too high, lows too low"). Assumed causes are an underestimation of the cross-isobar flow in PBL, caused by too little surface drag or too weak blocking at the orography.

Tests with an envelope orography which exhibits considerably higher values in the mountains have shown a high sensitivity of the model to this change in orography with respect to blocking (of cyclones). Consequently, low pressure systems were filled more efficiently, the negative pressure bias was reduced. But an obvious disadvantage of the envelope orography is that the precipitation fields are altered in an unfavourable way. An alternative way to increase the surface drag and the blocking while not altering the precipitation fields is the use of a SSO scheme.

Back to Contents

Vectorization (by Uli Schättler and Jens-Olaf Beismann)

Some more routines have been optimized for vectorization:

• Subroutine cal_conv_ind of module pp_utilities (for the output of convective indices) has been vectorized.
• src_turbdiff: Elimination of POINTERS in SR stab_funct.
• src_lheat_nudge: Vectorization of subroutine horizontal_filtering by reordering some loops.
• src_lheating: Introduction of compiler directives.
• src_leapfrog: Moved a SQRT function outside a loop.
• numeric_utilities: Vectorization of mean_over_box and mean_cov_over_box by loop reordering.
• fast_waves: Introduction of compiler directives and loop re-organization.

All these changes do not affect the results at all!

Back to Contents

Technical Adaptations

• Reorganization of time_utilities:

  The option to calculate the sum of the timings over the whole forecast has been introduced. The Namelist switches ltime_mean, ltime_proc have been eliminated and itype_timing has been introduced instead (see Section 2 for a detailed description).

  The search algorithm in subroutine get_timings has been eliminated. A direct mapping with fixed integers is used now instead.

  The computation of the forecast time in seconds is done with double precision integers.

• Adaptations for some compiler complaints: Some compilers are more restrict (or less flexible) regarding syntax. The following 2 issues were changed in several subroutines:

  Declaration of variables in an argument list of a subroutine: The dimensions of fields have to be declared before the field, i.e:

  INTEGER	(KIND=iintegers)	::	idim, jdim
  REAL	(KIND=ireals)	::	field(idim,jdim)

  and not

  REAL	(KIND=ireals)	::	field(idim,jdim)
  INTEGER	(KIND=iintegers)	::	idim, jdim

  Some compilers produce a program interuption, if a variable in an argument list is not initialized, even if this variable is not used in a certain application. So these variables are initialized in any case.

• Bug Fixes:

  Initialization of hd_mask in src_setup: The initialization of the variables i_west, i_east, j_south, j_north was not reproducible. For certain configurations, these values could be different, depending on the special MPI-task. They are now computed by a global operation.

  Fix in the subroutine multiplicative_filling, to avoid negative integrals which could appear in special cases.

  Fix in the initialization of the blacklist field for the latent heat nudging

  Fix in the initialization of nincrad in SR input_phyctl (organize_physics) If hincrad is given in the Namelist input, the variable nincrad was initialized with 0 and not adapted to hincrad. This led to a wrong computation of the solar zenith angle, which should be computed for the middle of the time interval between 2 radiation steps. This Bug Fix changes the results of all applications!!

• Miscellaneous:

  Values smaller than E-300 could appear in the subroutine tautsp2D, if the value of Z_VEC is smaller than E-100. To avoid such small values (which lead to underflow on certain machines), values of Z_VEC < 1E-50 are treated as zero.

  Some adaptations to avoid "invalid operations" on the NEC in the module src_mult_local.f90

  Decoding of vertical coordinate parameter nfltvc (for SLEVE) with Grib function REFSTF

  Adaptations for running the Runge-Kutta scheme with DFI (initialization of rho_snow in dfi_initializations, consideration of sign of dt in several RK routines)

  2 new GRIB fields have been introduced for use in Switzerland: Swiss00, Swiss12

Back to Contents