Coupling%20COSMO%20with%20the%20WAM%20model

1
Coupling COSMO with the WAM model
Aron Roland (TUD, Darmstadt), Mathieu Dutour
(IRB, Zagreb), Luigi Cavaleri (ISMAR, Venice),
Luciana Bertotti (ISMAR, Venice) and Lucio
Torrisi (CNMCA Rome).
2
Content

• Motivation
• Physics
• The coupling library and its methodology
• Validation of the coupling library
• Conclusion

3
Motivation

• Wind generate waves influence the atmosphere and
  determine the fluxes from the ocean to the
  atmosphere.
• Waves are driving currents and currents are
  modulating the waves.
• In order to have the full cycle we couple in this
  project the operationally used
• COSMO (Atmosphere)
• WAM (Waves)
• ROMS (Currents) models
• The leading model is here COSMO organizing the
  output, providing the forcing for the other two
  models and receiving the surface conditions from
  them.

4
Motivation
5
Some recent results of Hurricane Isabel using
coupled Ocean-Wave model (SELFE-WWMII) on
unstructured meshes driven by NARR (North
American Regional Reanalysis) winds
6
Some recent results of Isabel using coupled
Ocean-Wave model on unstructured meshes
7
Comparison of the sign. Wave height with the buoy
measurements (blue no currents, black with
currents)
8
Comparison of the average period with the buoy
measurements
9
Estimation Isabell storm surge
10
Atmospheric conditions during Xynthia courtesy
to Xavier Bertin, submitted to OM
11
Xynthia !Influence of wave induced surface drag
on the sea surface elevation courtesy to Xavier
Bertin, submitted to OM
12
Physical formulations potential for improvement
when coupling COSMO WAM

• When looking at the sea surface it becomes
  evident that the sea surface roughness depends on
  the sea state.
• When looking at a typical balance of energy
  fluxes for a growing wind sea at a constant wind
  speed of 20m/s (Janssen et al. 2002)
• At the same time Fig. 1 illustrates the role
  ocean surface waves play in the interaction of
  the atmosphere and the ocean, because on the one
  hand ocean waves receive momentum and energy from
  the atmosphere through wind input (controlling to
  some extent the drag of air flow over the
  oceans), while on the other hand, through wave
  breaking, the ocean waves transfer energy and
  momentum to the ocean thereby feeding the
  turbulent and large-scale motions of the oceans.
  (Janssen et al. 2002)

13
Impact of the atmosphere-wave coupling at ECMWF
Surface winds

• When the two-way interaction of winds and waves
  was introduced in operations on the 29th of June
  1998 there was a pronounced improvement of the
  quality of the surface wind field. (Janssen et
  al. 2002).
• The reduction of the surface wind RMS-error was
  around 10.
• It was found that with increased spatial
  resolution of the atmospheric model the influence
  of the coupling to the waves also increases.

14
The Situation at the beginning

• Source codes
• COSMO 250.000 l.o.c (simple partitioning)
• WAM 60.000 l.o.c (optimized partitioning with
  respect to the LAND/SEA mask)
• ROMS 340.000 l.o.c. (simple partitioning, more
  complicated when using nesting)
• After the study of the codes we tried to couple
  the models using the MCT (Model Coupling Toolkit)
  library according to the work of Warner et al.
  However, the library proved complicated, has no
  good manual, and does not satisfy the needs for
  realizing the interpolation in the background in
  a transparent manner.
• Therefore, we decided to develop a custom made
  MPI library that is tailored especially for these
  three models.
• We called this Library at this stage PGMCL
  (Parallel Geophysical Model Coupling Library).
  The PGMCL library has at this stage 3500 l.o.c.
  and is well documented. Easy to understand and
  nice to follow within all source code due to
  usage of CPP (C Preprocessor Pragma e.g. grep
  bwn WAT2ATM, ATM2WAV, OCN2WAV, OCN2ATM).

15
Methodology of the coupling

• The technique is that, if we have N processors we
  decompose them as
• Nocn Nwav Natm N
• Computationally, this means to split the
  MPI_COMM_WORLD into subsets by using the
  MPI_COM_SPLIT command.
• Hence, after that each model is using a
• OCN_COMM_WORLD,
• ATM_COMM_WORLD and
• WAV_COMM_WORLD.
• The coupling is done at instantaneous times and
  provide instantaneous values of the fields, i.e.
  no averaging is done. In other words the models
  are fully synchronized.

16
Interpolation Algorithm

• We want to allow different grids for each model.
  Hence some degree of interpolation is needed.
• We used linear interpolation by using the
  longitude/latitude of the grid points. Thus we
  compute a sparse matrix at the beginning of the
  run that contains the weights of the
  interpolation.
• We take care of the land/sea mask of the models
  using a direct mapping.

17
Partition methodology

• Each node of the model has access to the global
  longitude/latitude index of each model.
• Each node knows which point are computational
  nodes and which are ghost points for each node.
• As a consequence each node knows exactly what it
  gets/sends from/to the other node.
• So, we avoid a global gathering of all data on 1
  computational node and we have instead some
  MPI_INTERP_Send and corresponding
  MPI_INTERP_Recv function. This makes the
  coupler efficient on massive parallel platforms.
• Once all declarations are done, this is the only
  thing that show up in the code of the fully model
  coupled. We preferred this solution to the MCT
  library.
• This makes clear, clean, efficient and easy to
  apply exchange operations between all models.

18
Difficulties in the Development

• Beside the amount of l.o.c (lines of codes)
  750.000 we found certain weaknesses that made
  the developing/debugging of the coupled model
  more difficult
• COSMO has certain weak points in the code e.g.
• The models works even if NaN is present in the
  solution
• When switching on Netcdf output the prognostic
  arrays become NaN
• Netcdf is not working due to some errors in the
  code e.g. the same variable name is written
  frequently to the same file, which is not
  allowed.
• Unallocated arrays are initialized
• More small bugs are present in the code
• WAM has also some e.g.
• At certain place of the code same buffers have
  been used for communication, very difficult to
  trace long story
• Memory allocation/initializations weaknesses
• Certain small bugs
• However, with the help of Jean Bidlot and the
  ECWMF team most of these issues have been fixed.
• It would be very nice to have somebody from the
  COSMO team to interact with.

19
Validation of the PGMCL
Blue Cosmo Red WAM
20
Validation of the PGMCL
21
Present situation and future steps

• The COSMO model was coupled to the WAM model
  using a custom made model coupling library
  (PGMCL)
• The COSMO model receives the roughness length
  from the wave model.
• The WAM model receives air density, air
  temperature and wind velocity.
• The PGMCL library was verified up to 8 CPUs
• The next step is to investigate the impact of the
  coupling, 1st on a 0.25 grid followed by a
  operational setting at CNMCA with a spatial
  resolution of 7km for both COSMO and WAM.
• Following this we will also include the ROMS
  model in the coupling cycle and exchange more
  variables among the models e.g.
• Sea surface temperature
• Water density
• Water level elevation
• Surface currents
• As final step we will try to homogenize the
  formulation of the boundary layer, which is
  presently inconsistent among the three models.

22
Conclusions

• We have setup up the technical framework for the
  coupling of
• COSMO
• WAM and
• ROMS
• using a custom made coupling library.
• We have shown the benefits of the coupling of
  waves to the ocean and waves to the atmosphere.
• COSMO will benefit from a better representation
  of the surface roughness.
• WAM will benefit from a better representation of
  the driving wind.
• Finally, the coupling to ROMS will close the big
  circle.
• At the end we will have 1 model COSMOWAMROMS.
  This will allow to take into account the full
  interaction at the interface.