Documentation of the Changes in INT2LM
Versions 1.11 to 1.18

There were several new versions of the INT2LM in the last year with various developments. After reaching a certain stability again, Version 1.18 is now distributed to the users.

Contents:

1. Modifications (grouped by subject)
• Implementation of reading data with grib_api
• Reproducibility of surface variables for isolated points
• Computation of interpolation weights
• Possibility to switch the reference atmosphere
• Interpolation of QR, QS from IFS data
2. Bug Fixes and Miscellaneous
3. Technical Adaptations
4. Changes in the Namelists
5. Changes of the Results

1. Modifications (grouped by subject)

Implementation of reading data with grib_api

In the near future, the grib_api from ECMWF will replace the DWD grib(1) library. With grib_api it is possible to read and write Grib 1 and Grib 2 data. In a first step, grib_api has been implemented in INT2LM only for reading data.

All references to grib_api routines have been encapsulated with the pragma #ifdef GRIBAPI. To activate these calls, -DGRIBAPI has to be specified for the compilation.

NOTE: This implementation must use grib_api-1.9.0 or higher.

To read grib data with grib_api can be specified by the Namelist parameters

• yinext_form_read = 'grax' (other already existing value: 'grb1', 'ncdf')
• yin_form_read = 'grax' (other already existing value: 'grb1', 'ncdf')

To read the external COSMO-Model parameters is not yet possible with grib_api. This will be implemented at a later stage.

A number of changes were necessary to implement grib_api. Here are the highlights:

• Modification of the (grib) variable table for input data in src_gribtabs.f90: Another element was introduced in the data structure ar_des_input:
  CHARACTER (LEN=30) :: ylevtyp ! name of level type
  to store the level type in a character string. While in grib1 a product is identified by tabtyp,ee,levtyp, in grib2 the name and the level type (in characters) are used.
• We already unified the different setup_vartab_xxx routines for the different input models to one routine setup_vartab_grid_in. A common setup of the variable table is defined first, and changes for the different grids are done at the end of this subroutine.
• Module io_utilities.f90:
  
  • Extensions in subroutine open_file and close_file to handle grib_api I/O
  • New subroutine read_grib_api to get grib messages using grib_api calls
  • New subroutine check_input_grid, which unifies the former subroutines check_xx_grid.
  • This subroutine now also uses grib_api calls to get grib meta data from the grib messages.
  • Modifications to subroutine check_gme_grid (which remains) to get grib meta data using grib_api calls.
• The modules src_gme_interpol.f90, src_read_coarse_grid.f90 and src_read_ext.f90 have been extended to
  • open / close grib files
  • read grib messages
  • get meta data
  • and unpack the grib records
  using grib_api calls.

To learn more about grib_api, please take a look at the manual, which you can find on the web page of ECMWF:
http://www.ecmwf.int/publications/manuals/grib_api/index.html

The implementation of grib_api in INT2LM is done using the concept of the short names. To run the INT2LM with grib_api therefore requires the usage of some grib_api tables. The tables for Version 1.9.5 are distributed with the new int2lm version as they are used by DWD. (These tables should also work with other 1.9.x versions of grib_api).

To use these tables, a grib_api environment variable has to be specified before running INT2LM:
export GRIB_DEFINITION_PATH=..../TABLES_1.9.5_COSMO/definitions.edzw:.../TABLES_1.9.5_COSMO/definitions

While testing this new INT2LM version, there were some problems recognizing certain variables, because the centre=78 (which means DWD) was specified for these variables, and when another centre has been specified in the GRIB file (what is of course done by all other COSMO centres), this did not match. Therefore we modified the tables now in a way, that it should be possible to use them in other centres. We are still in the process of learning about grib_api and maybe these tables and their usage might change in the near future.

Back to Contents

Reproducibility of surface variables for isolated points

The computation of surface values for isolated points (lakes in the land; islands in the oceans) was looking up to now for the next grid point with the same land-sea-mask. Depending on the domain decomposition, this could lead to different points.

Now, all isolated land points are identified at the beginning of the program, and for every isolated point, the next grid point in the total domain of the coarse grid is searched and saved (in subroutine external_data, Section 4.6: Test whether isolated points are present). All information on these isolated points are gathered in the variable globl_iso_points, which is of the structure struct_for_isolated, defined in data_profiles.f90.

In the subroutine interp_l in module interp_utilities.f90, the values of the special field at the isolated points are gathered from every processor and are given to all processors, if a "match" interpolation is done. In the loops over the grid points, the search for the next grid point with same land-sea-mask is then replaced by just putting the exchanged variables to the proper COSMO grid points.

This procedure changes the results for the surface variables compared to former versions and can also influence the atmosphere, because t_s is used as lower boundary condition for vertical interpolation of temperature.

Back to Contents

Computation of interpolation weights

Some problems have been reported for the options lec2lm and llm2lm which were related to the subroutine coor_coarse_grid_lm in the module src_grids.f90. Therefore, the following modifications have been implemented:

• While introducing the decomposition independent computation of the interpolation weights, it was forgotten to take care of a non-staggered grid like the one from IFS.
  (fixed by Stephan Pfahl from ETZ)
• The computations of the interpolation weights has been corrected: In some cases, when input grid point and COSMO grid point are identical, it could happen that unrealistic high interpolation weights had been computed. Now it is ensured that these weights are between 0 and 1.
• The computations of lat and lon-values for the COSMO-model grid points have been modified to use the startlat_tot, startlon_tot values instead of startlat, startlot to be domain decomposition independent.

Back to Contents

Possibility to switch the reference atmosphere

The possibility to switch the reference atmosphere from the incoming to the outgoing grid has been implemented in the module src_vert_inter_lm.f90:

The problem was:

• If the incoming model delivers pressure or pressure deviation, the pressure deviation is used and interpolated to the outgoing grid. But this pressure deviation belongs to the incoming reference atmosphere.
• If the outgoing reference atmosphere is different, the following adaptation is done (Section 5 in SR org_vert_inter_lm):
  • Compute the incoming reference atmosphere on the outgoing vertical levels (Variable zp0_lm_refmod: the corresponding formulats from SR reference_atmosphere_x are used).
  • Compute the full pressure on the outgoing grid: p_lm = zp0_lm_refmod + pp_lm
  • Compute the pressure deviation for the outgoing reference_atmosphere pp_lm = p_lm - p0_lm

Back to Contents

Interpolation of QR, QS from IFS data

Some technical adaptations have been done in src_gribtabs and src_read_coarse_grid to read and interpolate rain and snow from IFS data. Note the differences in the names (important for shortnames, when using grib_api):

Back to Contents

2. Bug Fixes and Miscellaneous

• Introduction of lake water salinity
  The possibility to process an external parameter for lake water salinity has been implemented (salt_lk_lm). If this parameter is not available, it is set to a default value of 0.0. (by Burkhardt Rockel)
• Adaptations for NetCDF I/O of 3D field HORIZON
  The NetCDF I/O has been adapted to deal with the three-dimensional external parameter HORIZON, necessary for the topographical corrections (by Anne Roches, CH).

• Possibility to use other coarse grid data
  Data from the japanese model GSM (JMA, Grib2), the US model GFS (NCEP, Grib2) and the british model UM (UKMO) can now also be used to run the COSMO Model. But only the atmospheric variables are interpolated from these models, no soil variables can be used.

Because the number of input models is increasing, the namelist parameters lgme2lm, llm2lm, lec2lm, lcm2lm, etc. have been eliminated. The variable themselves are still used as internal model variables.
A new Namelist parameter has been introduced to specify the input model with a string: yinput_model: string to identify the input model Valid options are: 'GME', 'COSMO', 'IFS', 'UM', 'GSM', 'GFS'

• Vectorization of SR moist_split:
  The subroutine moist_split has now been written for whole fields. Therefore no CALL within DO-loops is necessary and the calling routines do vectorize now. (by Oliver Fuhrer).
• Implementation of a logical switch lurban
  If lurban is set to .TRUE., an additional external parameter field fr_urban_lm is read from the external data and provided to the COSMO-Model initial fields.

  NOTE:
  For the INT2LM, as it is provided, this is only a dummy-switch, because no Urban-Module is provided. But there are several groups who add this module on their own and can work with this switch now.

• src_read_ext.f90:
  broadcast irefatm_in for NetCDF input to all tasks
• src_vert_interpol.f90, src_coarse_interpol.f90
  introduced additional debug output
• src_lm_output.f90
  corrected the attribute "flag_values" in NetCDF for soiltyp added characteristic 'i' for inland water lakes
• src_lm_field.f90
  corrected the vertical dimension of qg in call to SR vert_int_lm
• src_gribtabs.f90
  added characteristic 'i' for inland water lakes
• external_data.f90
  initialize the full soiltyp field for option lec2lm
• src_read_coarse_grid
  For reading / computing the vertical coordinate parameters of the input grid, the parameter nlevskip (used for IFS) had been forgotten.
• src_namelists.f90
  The default value for nbitmap_d has been doubled because of larger GME grid.
• src_read_coarse_grid.f90
  Checking EPS characteristics has now also be implemented for grib_api
• src_lm_output.f90
  For the new reference atmosphere, there were undefined values in the gds between the vertical coordinate and the 2 additional parameters delta_t, h_scal (if SLEVE coordinate is not used). These values have now be set to 0 For running in ensemble mode, the number of total ensemble members has erroneously be set to 0. This has been removed.
• src_read_coarse_grid.f90
  The fields zt0, zp0hl in SR org_read_coarse_grid, Section 6, must be allocated with verical dimensions ke_in, ke1in. (reported by Christian Grams, KIT)
• src_lm_fields.f90
  The bottom boundary condition for the vertical interpolation of QR, QS was not set correctly. This led to non-reproducible results.

Back to Contents

3. Technical Adaptations

• A new information module info_int2lm has been introduced into the INT2LM This module has been introduced in all DWD libraries and models and gives information on the compiler, compiler options, etc. For more information on this module and how to use it, see the misc.global in the DOCS direcotory of INT2LM.
• New structure of the Makefile
  Some parts of the Makefile have been "outsourced" to external files:
  • Fopts: this file includes the compiler calls, options and libraries
  • ObjFiles: this file defines the list of object files that have to be compiled
  • ObjDependencies: this file contains the dependencies that have to be considered when compiling
  The user now only has to provide a file Fopts for a special compiler. Some example files are contained in the new subdirectory LOCAL

Back to Contents

---

4. Changes of the Namelists

There were the following changes for the Namelist variables:

Group	Name	Meaning	Default
/CONTRL/	lgme2lm llm2lm lec2lm etc.	These variables have been eliminated. To specify the incoming coarse grid model, a new variable has to be specified now.
	yinput_model	To specify the incoming coarse grid model with a string. Valid settings are: 'GME', 'COSMO', 'IFS', 'UM', 'GSM', 'GFS'	' '
/DATA/	yinext_form_read	A new option has been added to specify reading external parameters from the coarse grid model with grib_api. Valid settings are: 'grb1', 'grax', 'ncdf'	'grb1'
	yin_form_read	new option has been added to specify reading coarse grid model data with grib_api. Valid settings are: 'grb1', 'grax', 'ncdf'	'grb1'

Back to Contents

---

5. Changes of the Results

Because of the changes in the subroutine coor_coarse_grid_lm, the results can change numerically.

The values for surface values can change because of the new treatment of isolated points.

The values for QR and QS do change because of a bug fix in the bottom boundary condition for the vertical interpolation.

Back to Contents