Documentation of the Changes in the COSMO-Model
Version 5.04f

Version 5.04f is hopefully the last step until version 5.05 can be implemented. It contains all the physical parameterizations in blocked format.

Contents:

1. Additional Changes for COSMO-ICON Physics
2. Reproducing the Behaviour of the old COSMO Physics
3. Changes in the Data Assimilation
4. New Diagnostic Output Variables to Highlight Convective Cell Tracks between Output Timesteps
5. Changes in Handling Statistically Processed Fields:
6. Changes for shortNames
7. GPU Version of TERRA
8. Changes for using RTTOV
9. Technical Changes and Bug Fixes
10. Changes to the Namelists
11. Changes of Results

1. Additional Changes for COSMO-ICON Physics

(by Matthias Raschendorfer, Jürgen Helmert, Martin Köhler, Ulrich Schättler)

A few more changes were necessary for the COSMO-ICON Physics:

• Increased allowable range for itype_heatcond to 3 (according to new TERRA)
  itype_heatcond=3: use a special (hardcoded) tuning factor to indirectly account for the impact of vegetation, which does not depend on soil moisture.
• New namelist variable idiag_snowfrac (in /PHYCTL/):
  Method for diagnosis of snow-cover fraction:
  • 1 : old parameterization depending on soil water equivalent only
  • 2, 20: more advanced parameterization depending on snow depth, accounts also for vegetation and SSO
  • 3, 30: similar to option 2, but somewhat less snow cover and limit over high vegetation
  • 4, 40: same as option 3, but even more restrictive snow cover limit over high vegetation

  Extra for 20, 30, 40:
  Artificially reduce snow-cover fraction in case of melting snow in order to reduce the ubiquitous cold bias in such situations.

  Default: 1

• New namelist variable tune_minsnowfrac (in /TUNING/)
  Minimum value to which the snow cover fraction is artificially reduced in case of melting show (in case of idiag_snowfrac = 20/30/40)

  Default: 0.125

• sfc_interface.f90: (changes results only for ALHFL_S)
  Use optional output parameters zlhfl_sfc, zshfl_sfc, zqhfl_sfc from TERRA for the fluxes lhfl_s, shfl_s, qvfl_s which are the weighted (between snow and no snow) turbulent fluxes at the boundary surface-atmosphere.
• Unified settings for ICON and COSMO in turb_data.f90 (changes results):
  • imode_rat_sea = 1: ICON value, changes in COSMO
  • loutshs = TRUE: COSMO value, changes in ICON
• turb_diffusion.f90, turb_transfer.f90: (changes results)
  Added it_start to subroutine solve_turb_budgets parameter list
• turb_utilities.f90: (changes results)
  
  • Modified initialization for fc_min
  • a_stab related correction only for first iteration
• Some more technical adaptations:
  
  • Include H_SNOW in list of restart-variables for new surface schemes
  • Include SSO_SIGMA to yvarini, if idiag_snowfrac > 1
  • Set h_snow_new_b also for landpoints to h_snow_t (output from TERRA) and include it in copyFromBlock-copylist in all cases (changes results)
• Removal of old parameterizations and obsolete namelist variables:
  The source code for the old 2-/3-layer soil model has not been ported to the blocked data format and cannot be used any more. Therefore, all associated fields, variables, namelist switches have been removed from the source code. Also two namelist variables from the former turbulence version have been eliminated:
  • eliminated fields: t_m, t_m_bd, w_g[1|2|3], w_g[1|2|3]_bd
  • eliminated switches:
    lmulti_layer, nlgw /PHYCTL/
    nlgw_ini, nlgw_bd /GRIBIN/
    wichfakt, securi /TUNING/

Back to Contents

2. Reproducing the Behaviour of the old COSMO Physics

(by Matthias Raschendorfer, Ulrich Schättler)

It is possible to reproduce the behaviour of the old non-blocked physics of COSMO rather closely with this new version. To do so, some hardcoded variables in the turbulence scheme need a different value. This is implemented in turb_data.f90 with the pragma COSMO_OLD. If you want to have results as close as possible to the non-blocked version compile the model with -DCOSMO_OLD.

In addition you have to check the settings of the following namelist variables, which have to be set as in the old versions:

Settings for COSMO-DE from DWD before:

Back to Contents

3. Changes in the Data Assimilation

(by Christoph Schraff)

Modifications

Modifications in the redundancy check for multi-level reports:

• Spatial redundancy check limits for TEMP (and profilers) enlarged (horizontally: 0.2 → 6.0 km; vertically: 10 → 40 m) to account for differences in original BUFR and TAC (alphanumeric) reports.
• Minor change to help to select the TEMP parts B, A (in particular B, and not C, D) as active report, in order to get the obs and decoding times from Part B in the header of the composite radiosonde report.

Conversion of the (observed and simulated) wind components from the rotated to the real geographical coordinate system for output to the NetCDF feedback file ('fof'; this defines a new version of the feedback file: 1.01 → 1.02) and the VOF ASCII feedback file ('YUVERIF', this defines a new version of the VOF file: VOF Version 4 → VOF Version 5)

Observation reading and processing modules which are also used by DACE-LETKF are updated to the DACE-LETKF DWD Tag Version V1_51. This includes some technical adaptations, minor bug fixes, and modifications to make the code, notably the grid point assignment, applicable to the ICON grid.

Changes to the results:

• Changes may occur frequently, when the same radiosonde ascents are read twice, once from an original TAC report and once from an original BUFR report. Otherwise, changes are rare, but may occur (depending on processed radiosonde reports).
• The output of the wind components in the (NetCDF 'fof and ASCII 'YUVERIF') feedback files is modified.

Changes to the namelist:

None.

Note:

In a KENDA or MEC context, this COSMO version (writing feedback file Version 1.02) should be used with DACE Version 'commit 547f1ab' (as of 14 May 2017) or later (or in terms of DWD Tag Versions: V1_52 or later).

Back to Contents

4. New Diagnostic Output Variables to Highlight Convective Cell Tracks between Output Timesteps

(by Ulrich Blahak)

7 new diagnostic 2-dimensional output fields have been implemented which may be specified in the first yvarml-list. These fields contain the minima/maxima of the following quantities between the actual and the last output time step:

• UH_MAX:
  Maximum of absolute value (+ or -) of Updraft Helicity between output time steps for each grid point, in units of J/kg. UH is the vertical integral of W times vertical vorticity between 2 km and 8 km height MSL. For orography > 1500 m, the integration limits are raised to 0.5 km - 6.5 km AGL. Computation takes only updrafts into account. This means that positive values indicate cyclonically rotating updrafts, while negative values indicate anticyclonic rotation.
• VORW_CTMAX:
  Maximum of absolute value (+ or -) of low level rotation between output time steps, in units of 1/s. Low level rotation is defined here as the average vertical vorticity from the ground up to 3 km MSL or 1.5 km AGL, whichever is higher. Again, positive values indicate cyclonic rotation and negative values anticyclonic.
• W_CTMAX:
  Maximum updraft speed at each grid point from the ground to 10 km MSL and over time between output time steps, in units of m/s.
• TCOND_MAX:
  Maximum total condensate between output time steps for each grid point, in units of kg/m². Total condensate is the integral of all hydrometeors except water vapor over the total grid column. It is similar but not equal to the so-called VIL (vertically integrated liquid) from radar observations.
• TCOND10_MX:
  Similar to TCOND_MAX, but the vertical integration starts only at the -10-deg-C level upwards. It is similar but not equal to the so-called VII (vertically integrated ice) from radar observations.
• DBZ_CTMAX:
  Maximum radar reflectivity at each grid point over height and time between output time steps, in units of dBZ.
• LPI_MAX:
  Maximum Lightning Potential index at each grid point over time between output time steps, in units of J/kg.

Back to Contents

5. Changes in Handling Statistically Processed Fields

(by Ulrich Schättler)

Introduction

The new diagnostic output variables for tracking of convective cells need a modified treatment of statistically processed fields, because they should be reset after every output step.

Up to now only the following methods are possible in the COSMO-Model:
Statistical processing by

• summation or averaging:
  fields are summed up from beginning of the simulation and never reset. Only in climate mode (lbdclim=.TRUE.) the fields are reset to 0.0 after an output step.
• Min/Max values:
  • Temperatures (TMIN_2M, TMAX_2M) are reset according to namelist variable hincmxt
  • Wind speeds and gusts (e.g. VMAX_10M, VABS_10M, VGUST_DYN_10M, VGUST_CON_10M) are reset according to hincmxu
  • the new MAX-variables for tracking of convective cells should be reset after every output. The difficulty here is, that these variables could occur in different GRIBOUT namelist groups.

A more generalized way has been implemented now in the COSMO-Model, how the resetting of statistically processed fields is organized:

All statistically processed fields are grouped into four categories.

For every category the type of resetting can be chosen. For that there are four new Namelist variables in /GRIBOUT/ to control the type of resetting:

   ireset_temps, ireset_winds, ireset_sums, ireset_cells

Allowed values for these variables are

0. no resetting at all (default for sums and averages)
1. reset after given interval (only for temps and winds; this is the default)
2. reset after every output (default for cells).

The default values of the four new namelist variables in /GRIBOUT/ are in a way that the current behaviour of the COSMO-Model regarding statistically processed fields is retained.

New Subroutines in organize_data.f90

Two new subroutines are included in module organize_data.f90 to control the resetting of the statistically processed fields:

• setup_stat_proc:
  Initializes a table list_stat_proc, which contains organizational information on statistical processing of field. The table contains
  • yname: short name of the field
  • iloc[1|2|3]: location in the variable table
  • itype_reset: when to reset the fields:
    0. no resetting at all
    1. resetting according to some other namelist values (hincmxu, hincmxt)
    2. resetting after every output
  • [h|n]lastout, [h|n]nextout: to organize the resetting
  • nuout: (unit_of_time) if a field is contained in several /GRIBOUT/ namelist groups, the nunit_of_time-value with smallest time increment has to be taken.
• reset_stat_proc:
  Scans the table list_stat_proc and resets the fields depending on itype_reset, if required.

Changes to current COSMO-Model implementation

The resetting of temperatures in near_surface.f90 and of winds in initialize_loops (lmorg.f90) has been removed. Resetting of statistically processed fields is now done in organize_data.f90 after output of fields by calling the new subroutine reset_satt_proc.

To control resetting "after every output", an additional entry in the structure ar_des for the variable table has been added: lout_done
After output, this logical flag is set to .TRUE., the routine reset_stat_proc checks this flag and resets the variable, if itype_reset==2. lout_done is then set to .FALSE. again.

Back to Contents

6. Changes for shortNames

(by Ulrich Schättler)

There were changes in the shortnames for output fields from radiation, the multi-layer snow model and the wind gusts due to a unification with ICON names:

Having the 10M specification in the short names for the dynamical and convective wind gusts has the advantage that the corresponding GRIB2 specifications can be put to the shortName.def.

Another necessary change is to increase the length of the short names in COSMO to 13 characters. This has implications to the formatting of YUCHKDAT.

There are new fields, which are only active, if the radar forward operator is activated (which will be added in one of the next versions):

• DBZCMP_SIM: simulated radar reflectivity composite from radar forward operator
• DBZCMP_OBS: observed radar reflectivity composite

For development purposes an extra GRIB (1) table 254 has been added with dummy variables. These can be used during development to write special model variables which do not have an official Grib coding. This table 254 is coded with COSMO internal number 13 in lst_gribtabs.

Back to Contents

7. GPU Version of TERRA

(by Clement Valentin, MCH)

OpenACC statements have been implemented in TERRA (module sfc_terra.f90) in order to run also this scheme on GPUs. For a better optimization on GPUs the index lists for rock and ice have been eliminated.

The changes are bit reproducible, at least when using GNU compiler and Cray compiler with the flag -hflex_mp=intolerant. With -hflex_mp=conservative the results are modified due to different compiler optimizations.

Back to Contents

8. Changes for using RTTOV

(by Axel Hutt, Robin Faulwetter, Ulrich Schättler)

Usage of RTTOV12 has been implemented in the model: To use RTTOV12 the COSMO-Model has to be compiled with pragma -DRTTOV12.

Usage of RTTOV9 has been eliminated.

Vertical interpolation of COSMO model levels to RTTOV pressure levels:

Up to now this vertical interpolation has been done with RTTOV internal routines. To do that, the argument addinterp for calling rttov_fill_input in module src_sat_rttov.f90 has been set to .TRUE. But these routines are very expensive in terms of CPU time, therefore, COSMO routines are now used to do the vertical interpolation. A new subroutine "prepare_rttovlevels" has been added to compute all input values on the RTTOV pressure levels. The routine rttov_fill_input is now called with addinterp=.FALSE.

To do the vertical interpolation, it is necessary to know the RTTOV pressure levels. nlev_rttov is a new variable in data_satellites to specify the number of RTTOV pressure levels and pres_rttov is a variable, which is preset with the correct pressure levels, depending on the RTTOV version chosen.

Back to Contents

9. Technical Changes and Bug Fixes

• conv_interface.f90: (by Ulrich Schättler)
  
  • Update host before and update device after calling Tiedtke convection, because this is not yet GPU able.
  • Include ztke_b in acc create and acc delete lists.
• lmorg.f90:
  
  • Added itype_vdif to serialization.
  • Some corrections for time measurement.
  • More detailed timings for additional computations.
• organize_data.f90:
  
  • include H_SNOW in list of restart-variables for new surface schemes
  • write hhl_prof also for non-binary input data to YUSPECIF (for better comparability)
  • Include SSO_SIGMA to yvarini, if idiag_snowfrac > 1
  • New grib 1 table 254 (internal number 13 in lst_gribtabs) for local use and dummy variables in GRIB1 (also in src_setup_vartab)
• organize_physics.f90:
  
  • Removed i_add_computations (replaced once with i_phy_computations)
  • Added update host for sobs_b, thbs_b, pabs_b after radiation
• radiation_utilities.f90:
  
  • Removed t_s from acc present list.
  • Compute zcosphi by COS(zrlat): there are problems in single precision when using 1.0-SIN(zrlat)**2.
• sfc_interface.f90:
  
  • set h_snow_new_b also for landpoints to h_snow_t (output from TERRA) and include it in copyFromBlock-copylist in all cases (changes results!)
  • Bug fix in hail variable for TWOMOM_SB
• src_allocation.f90:
  Allocate prh_gsp in any case because it is used in copytoBlock structure.
• src_artifdata.f90:
  Initialize the new isoiltype from soiltype.
• src_cpp_dycore.f90:
  Added itype_vdif to serialization.
• src_gridpoints.f90:
  Changed format for writing hours in lgpshort with 2 decimals.
• src_input.f90:
  
  • Bug fix in computation of vcoord%kflat
  • Fix in ncdf-input for checking the date: adapted to a timestep dt that does not fit to a full hour.
• src_lheat_nudge.f90:
  Increased length of character variables for directory names.
• src_output.f90:
  Adaptation of global communication for helper fields of LPI.
• src_runge_kutta.f90:
  Added dqvdt, qvfl_s, lhfl_s to serialization.
• src_slow_tendencies_rk.f90:
  
  • Added shfl_s, umfl_s, vmfl_s to serialization.
  • Bug fix for lprog_tke.
• src_tracer.f90:
  Added entry lout_done to the variable table.
• src_traj.f90:
  Bug fix in call to subroutine curl.
• sso_interface.f90:
  Added hsurf_b to the acc present list.
• time_utilities.f90:
  More detailed timings for additional computations.
• utilities.f90:
  Implemented additional subroutine trapz_integral.

Back to Contents

10. Changes to the Namelists

Back to Contents

11. Changes of Results

The following modifications in the COSMO-ICON physics do change the results:

• sfc_interface.f90:
  Set h_snow_new_b also for landpoints to h_snow_t (output from TERRA) and include it in copyFromBlock-copylist in all cases. (This is a bug fix).
• turb_data.f90:
  Using imode_rat_sea = 1.
• turb_utilities.f90:
  
  • Modified initialization for fc_min.
  • a_stab-related correction only for first iteration (with adaptations in turb_diffusion, turb_transfer).
• radiation_utilities.f90:
  Compute zcosphi by COS(zrlat): there are problems in single precision when using 1.0-SIN(zrlat)**2.

Eventually results are changed by the modifications in the data assimilation (see above).

All other changes are bit reproducible with GNU compiler and with Cray compiler using -hflex_mp=intolerant

Changes to ASCII output

• Formatting of YUCHKDAT, because shortnames are now 13 characters long.
• Short meteograph output: time variables are now written with 2 digits after the decimal point.

Back to Contents